Merge branch 'master' into master

comfy/context_windows.py

@@ -8,6 +8,8 @@ from abc import ABC, abstractmethod
 import logging
 import comfy.model_management
 import comfy.patcher_extension
+import comfy.utils
+import comfy.conds
 if TYPE_CHECKING:
     from comfy.model_base import BaseModel
     from comfy.model_patcher import ModelPatcher
@@ -51,12 +53,18 @@ class ContextHandlerABC(ABC):
 
 
 class IndexListContextWindow(ContextWindowABC):
-    def __init__(self, index_list: list[int], dim: int=0, total_frames: int=0):
+    def __init__(self, index_list: list[int], dim: int=0, total_frames: int=0, modality_windows: dict=None, context_overlap: int=0):
         self.index_list = index_list
         self.context_length = len(index_list)
+        self.context_overlap = context_overlap
         self.dim = dim
         self.total_frames = total_frames
         self.center_ratio = (min(index_list) + max(index_list)) / (2 * total_frames)
+        self.modality_windows = modality_windows  # dict of {mod_idx: IndexListContextWindow}
+        self.guide_frames_indices: list[int] = []
+        self.guide_overlap_info: list[tuple[int, int]] = []
+        self.guide_kf_local_positions: list[int] = []
+        self.guide_downscale_factors: list[int] = []
 
     def get_tensor(self, full: torch.Tensor, device=None, dim=None, retain_index_list=[]) -> torch.Tensor:
         if dim is None:
@@ -85,6 +93,11 @@ class IndexListContextWindow(ContextWindowABC):
         region_idx = int(self.center_ratio * num_regions)
         return min(max(region_idx, 0), num_regions - 1)
 
+    def get_window_for_modality(self, modality_idx: int) -> 'IndexListContextWindow':
+        if modality_idx == 0:
+            return self
+        return self.modality_windows[modality_idx]
+
 
 class IndexListCallbacks:
     EVALUATE_CONTEXT_WINDOWS = "evaluate_context_windows"
@@ -148,6 +161,172 @@ def slice_cond(cond_value, window: IndexListContextWindow, x_in: torch.Tensor, d
     return cond_value._copy_with(sliced)
 
 
+def compute_guide_overlap(guide_entries: list[dict], keyframe_idxs: torch.Tensor, temporal_downscale_ratio: int, window_index_list: list[int]):
+    """Compute which concatenated guide frames overlap with a context window.
+
+    Each guide's latent-space start is derived from its first token's pixel-t-start
+    in keyframe_idxs (shape (B, [t,h,w], num_tokens, [start, end])), divided by the
+    model's temporal_downscale_ratio.
+
+    Args:
+        guide_entries: list of guide_attention_entry dicts
+        keyframe_idxs: per-token pixel coords cond tensor for the modality
+        temporal_downscale_ratio: model's pixel-to-latent temporal compression ratio
+        window_index_list: the window's frame indices into the video portion
+
+    Returns:
+        suffix_indices: indices into the guide_frames tensor for frame selection
+        overlap_info: list of (entry_idx, overlap_count) for guide_attention_entries adjustment
+        kf_local_positions: window-local frame positions for keyframe_idxs regeneration
+        total_overlap: total number of overlapping guide frames
+    """
+    window_set = set(window_index_list)
+    window_list = list(window_index_list)
+    suffix_indices = []
+    overlap_info = []
+    kf_local_positions = []
+    suffix_base = 0
+    token_offset = 0
+
+    for entry_idx, entry in enumerate(guide_entries):
+        first_t_pixel = int(keyframe_idxs[0, 0, token_offset, 0].item())
+        latent_start = (first_t_pixel + temporal_downscale_ratio - 1) // temporal_downscale_ratio
+        guide_len = entry["latent_shape"][0]
+        entry_overlap = 0
+
+        for local_offset in range(guide_len):
+            video_pos = latent_start + local_offset
+            if video_pos in window_set:
+                suffix_indices.append(suffix_base + local_offset)
+                kf_local_positions.append(window_list.index(video_pos))
+                entry_overlap += 1
+
+        if entry_overlap > 0:
+            overlap_info.append((entry_idx, entry_overlap))
+        suffix_base += guide_len
+        token_offset += entry["pre_filter_count"]
+
+    return suffix_indices, overlap_info, kf_local_positions, len(suffix_indices)
+
+
+@dataclass
+class WindowingState:
+    """Per-modality context windowing state for each step,
+    built using IndexListContextHandler._build_window_state().
+    For non-multimodal models the lists are length 1
+    """
+    latents: list[torch.Tensor]                  # per-modality working latents (guide frames stripped)
+    guide_latents: list[torch.Tensor | None]     # per-modality guide frames stripped from latents
+    guide_entries: list[list[dict] | None]       # per-modality guide_attention_entry metadata
+    keyframe_idxs: list[torch.Tensor | None]     # per-modality keyframe_idxs tensor for guide latent_start derivation
+    latent_shapes: list | None                   # original packed shapes for unpack/pack (None if not multimodal)
+    dim: int = 0                                 # primary modality temporal dim for context windowing
+    is_multimodal: bool = False
+    temporal_downscale_ratio: int = 1            # model's pixel-to-latent temporal compression ratio
+
+    def prepare_window(self, window: IndexListContextWindow, model) -> IndexListContextWindow:
+        """Reformat window for multimodal contexts by deriving per-modality index lists.
+        Non-multimodal contexts return the input window unchanged.
+        """
+        if not self.is_multimodal:
+            return window
+
+        x = self.latents[0]
+        primary_total = self.latent_shapes[0][self.dim]
+        primary_overlap = window.context_overlap
+        map_shapes = self.latent_shapes
+        if x.size(self.dim) != primary_total:
+            map_shapes = list(self.latent_shapes)
+            video_shape = list(self.latent_shapes[0])
+            video_shape[self.dim] = x.size(self.dim)
+            map_shapes[0] = torch.Size(video_shape)
+        try:
+            per_modality_indices = model.map_context_window_to_modalities(
+                window.index_list, map_shapes, self.dim)
+        except AttributeError:
+            raise NotImplementedError(
+                f"{type(model).__name__} must implement map_context_window_to_modalities for multimodal context windows.")
+        modality_windows = {}
+        for mod_idx in range(1, len(self.latents)):
+            modality_total_frames = self.latents[mod_idx].shape[self.dim]
+            ratio = modality_total_frames / primary_total if primary_total > 0 else 1
+            modality_overlap = max(round(primary_overlap * ratio), 0)
+            modality_windows[mod_idx] = IndexListContextWindow(
+                per_modality_indices[mod_idx], dim=self.dim,
+                total_frames=modality_total_frames,
+                context_overlap=modality_overlap)
+        return IndexListContextWindow(
+            window.index_list, dim=self.dim, total_frames=x.shape[self.dim],
+            modality_windows=modality_windows, context_overlap=primary_overlap)
+
+    def slice_for_window(self, window: IndexListContextWindow, retain_index_list: list[int], device=None) -> tuple[list[torch.Tensor], list[int]]:
+        """Slice latents for a context window, injecting guide frames where applicable.
+        For multimodal contexts, uses the modality-specific windows derived in prepare_window().
+        """
+        sliced = []
+        guide_frame_counts = []
+        for idx in range(len(self.latents)):
+            modality_window = window.get_window_for_modality(idx)
+            retain = retain_index_list if idx == 0 else []
+            s = modality_window.get_tensor(self.latents[idx], device, retain_index_list=retain)
+            if self.guide_entries[idx] is not None:
+                s, ng = self._inject_guide_frames(s, modality_window, modality_idx=idx)
+            else:
+                ng = 0
+            sliced.append(s)
+            guide_frame_counts.append(ng)
+        return sliced, guide_frame_counts
+
+    def strip_guide_frames(self, out_per_modality: list[list[torch.Tensor]], guide_frame_counts: list[int], window: IndexListContextWindow):
+        """Strip injected guide frames from per-cond, per-modality outputs in place."""
+        for idx in range(len(self.latents)):
+            if guide_frame_counts[idx] > 0:
+                window_len = len(window.get_window_for_modality(idx).index_list)
+                for ci in range(len(out_per_modality)):
+                    out_per_modality[ci][idx] = out_per_modality[ci][idx].narrow(self.dim, 0, window_len)
+
+    def _inject_guide_frames(self, latent_slice: torch.Tensor, window: IndexListContextWindow, modality_idx: int = 0) -> tuple[torch.Tensor, int]:
+        guide_entries = self.guide_entries[modality_idx]
+        guide_frames = self.guide_latents[modality_idx]
+        keyframe_idxs = self.keyframe_idxs[modality_idx]
+        suffix_idx, overlap_info, kf_local_pos, guide_frame_count = compute_guide_overlap(
+            guide_entries, keyframe_idxs, self.temporal_downscale_ratio, window.index_list)
+        # Shift keyframe positions to account for causal_window_fix anchor occupying sub-pos 0.
+        anchor_idx = getattr(window, 'causal_anchor_index', None)
+        if anchor_idx is not None and anchor_idx >= 0:
+            kf_local_pos = [p + 1 for p in kf_local_pos]
+        window.guide_frames_indices = suffix_idx
+        window.guide_overlap_info = overlap_info
+        window.guide_kf_local_positions = kf_local_pos
+
+        # Derive per-overlap-entry latent_downscale_factor from guide entry latent_shape vs guide frame spatial dims.
+        # guide_frames has full (post-dilation) spatial dims; entry["latent_shape"] has pre-dilation dims.
+        guide_downscale_factors = []
+        if guide_frame_count > 0:
+            full_H = guide_frames.shape[3]
+            for entry_idx, _ in overlap_info:
+                entry_H = guide_entries[entry_idx]["latent_shape"][1]
+                guide_downscale_factors.append(full_H // entry_H)
+        window.guide_downscale_factors = guide_downscale_factors
+
+        if guide_frame_count > 0:
+            idx = tuple([slice(None)] * self.dim + [suffix_idx])
+            return torch.cat([latent_slice, guide_frames[idx]], dim=self.dim), guide_frame_count
+        return latent_slice, 0
+
+    def patch_latent_shapes(self, sub_conds, new_shapes):
+        if not self.is_multimodal:
+            return
+
+        for cond_list in sub_conds:
+            if cond_list is None:
+                continue
+            for cond_dict in cond_list:
+                model_conds = cond_dict.get('model_conds', {})
+                if 'latent_shapes' in model_conds:
+                    model_conds['latent_shapes'] = comfy.conds.CONDConstant(new_shapes)
+
+
 @dataclass
 class ContextSchedule:
     name: str
@@ -162,7 +341,7 @@ ContextResults = collections.namedtuple("ContextResults", ['window_idx', 'sub_co
 class IndexListContextHandler(ContextHandlerABC):
     def __init__(self, context_schedule: ContextSchedule, fuse_method: ContextFuseMethod, context_length: int=1, context_overlap: int=0, context_stride: int=1,
                  closed_loop: bool=False, dim:int=0, freenoise: bool=False, cond_retain_index_list: list[int]=[], split_conds_to_windows: bool=False,
-                 causal_window_fix: bool=True):
+                 latent_retain_index_list: list[int]=[], causal_window_fix: bool=True):
         self.context_schedule = context_schedule
         self.fuse_method = fuse_method
         self.context_length = context_length
@@ -174,17 +353,118 @@ class IndexListContextHandler(ContextHandlerABC):
         self.freenoise = freenoise
         self.cond_retain_index_list = [int(x.strip()) for x in cond_retain_index_list.split(",")] if cond_retain_index_list else []
         self.split_conds_to_windows = split_conds_to_windows
+        self.latent_retain_index_list = [int(x.strip()) for x in latent_retain_index_list.split(",")] if latent_retain_index_list else []
         self.causal_window_fix = causal_window_fix
 
         self.callbacks = {}
 
+    @staticmethod
+    def _get_latent_shapes(conds):
+        for cond_list in conds:
+            if cond_list is None:
+                continue
+            for cond_dict in cond_list:
+                model_conds = cond_dict.get('model_conds', {})
+                if 'latent_shapes' in model_conds:
+                    return model_conds['latent_shapes'].cond
+        return None
+
+    @staticmethod
+    def _get_guide_entries(conds):
+        for cond_list in conds:
+            if cond_list is None:
+                continue
+            for cond_dict in cond_list:
+                model_conds = cond_dict.get('model_conds', {})
+                entries = model_conds.get('guide_attention_entries')
+                if entries is not None and hasattr(entries, 'cond') and entries.cond:
+                    return entries.cond
+        return None
+
+    @staticmethod
+    def _get_keyframe_idxs(conds):
+        for cond_list in conds:
+            if cond_list is None:
+                continue
+            for cond_dict in cond_list:
+                model_conds = cond_dict.get('model_conds', {})
+                kf = model_conds.get('keyframe_idxs')
+                if kf is not None and hasattr(kf, 'cond') and kf.cond is not None:
+                    return kf.cond
+        return None
+
+    def _apply_freenoise(self, noise: torch.Tensor, conds: list[list[dict]], seed: int) -> torch.Tensor:
+        """Apply FreeNoise shuffling, scaling context length/overlap per-modality by frame ratio.
+        If guide frames are present on the primary modality, only the video portion is shuffled.
+        """
+        guide_entries = self._get_guide_entries(conds)
+        guide_count = sum(e["latent_shape"][0] for e in guide_entries) if guide_entries else 0
+
+        latent_shapes = self._get_latent_shapes(conds)
+        if latent_shapes is not None and len(latent_shapes) > 1:
+            modalities = comfy.utils.unpack_latents(noise, latent_shapes)
+            primary_total = latent_shapes[0][self.dim]
+            primary_video_count = modalities[0].size(self.dim) - guide_count
+            apply_freenoise(modalities[0].narrow(self.dim, 0, primary_video_count), self.dim, self.context_length, self.context_overlap, seed)
+            for i in range(1, len(modalities)):
+                mod_total = latent_shapes[i][self.dim]
+                ratio = mod_total / primary_total if primary_total > 0 else 1
+                mod_ctx_len = max(round(self.context_length * ratio), 1)
+                mod_ctx_overlap = max(round(self.context_overlap * ratio), 0)
+                modalities[i] = apply_freenoise(modalities[i], self.dim, mod_ctx_len, mod_ctx_overlap, seed)
+            noise, _ = comfy.utils.pack_latents(modalities)
+            return noise
+        video_count = noise.size(self.dim) - guide_count
+        apply_freenoise(noise.narrow(self.dim, 0, video_count), self.dim, self.context_length, self.context_overlap, seed)
+        return noise
+
+    def _build_window_state(self, x_in: torch.Tensor, conds: list[list[dict]], model: BaseModel) -> WindowingState:
+        """Build windowing state for the current step, including unpacking latents and extracting guide frame info from conds."""
+        latent_shapes = self._get_latent_shapes(conds)
+        is_multimodal = latent_shapes is not None and len(latent_shapes) > 1
+        unpacked_latents = comfy.utils.unpack_latents(x_in, latent_shapes) if is_multimodal else [x_in]
+
+        unpacked_latents_list = list(unpacked_latents)
+        guide_latents_list = [None] * len(unpacked_latents)
+        guide_entries_list = [None] * len(unpacked_latents)
+        keyframe_idxs_list = [None] * len(unpacked_latents)
+
+        extracted_guide_entries = self._get_guide_entries(conds)
+        extracted_keyframe_idxs = self._get_keyframe_idxs(conds)
+
+        # Strip guide frames (only from first modality for now)
+        if extracted_guide_entries is not None:
+            guide_count = sum(e["latent_shape"][0] for e in extracted_guide_entries)
+            if guide_count > 0:
+                x = unpacked_latents[0]
+                latent_count = x.size(self.dim) - guide_count
+                unpacked_latents_list[0] = x.narrow(self.dim, 0, latent_count)
+                guide_latents_list[0] = x.narrow(self.dim, latent_count, guide_count)
+                guide_entries_list[0] = extracted_guide_entries
+                keyframe_idxs_list[0] = extracted_keyframe_idxs
+
+
+        return WindowingState(
+            latents=unpacked_latents_list,
+            guide_latents=guide_latents_list,
+            guide_entries=guide_entries_list,
+            keyframe_idxs=keyframe_idxs_list,
+            latent_shapes=latent_shapes,
+            dim=self.dim,
+            is_multimodal=is_multimodal,
+            temporal_downscale_ratio=model.latent_format.temporal_downscale_ratio)
+
     def should_use_context(self, model: BaseModel, conds: list[list[dict]], x_in: torch.Tensor, timestep: torch.Tensor, model_options: dict[str]) -> bool:
-        # for now, assume first dim is batch - should have stored on BaseModel in actual implementation
-        if x_in.size(self.dim) > self.context_length:
-            logging.info(f"Using context windows {self.context_length} with overlap {self.context_overlap} for {x_in.size(self.dim)} frames.")
+        window_state = self._build_window_state(x_in, conds, model) # build window_state to check frame counts, will be built again in execute
+        total_frame_count = window_state.latents[0].size(self.dim)
+        if total_frame_count > self.context_length:
+            logging.info(f"\nUsing context windows: Context length {self.context_length} with overlap {self.context_overlap} for {total_frame_count} frames.")
             if self.cond_retain_index_list:
                 logging.info(f"Retaining original cond for indexes: {self.cond_retain_index_list}")
+            if self.latent_retain_index_list:
+                logging.info(f"Retaining original latent for indexes: {self.latent_retain_index_list}")
             return True
+        logging.info(f"\nNot using context windows since context length ({self.context_length}) exceeds input frames ({total_frame_count}).")
         return False
 
     def prepare_control_objects(self, control: ControlBase, device=None) -> ControlBase:
@@ -275,7 +555,9 @@ class IndexListContextHandler(ContextHandlerABC):
         return resized_cond
 
     def set_step(self, timestep: torch.Tensor, model_options: dict[str]):
-        mask = torch.isclose(model_options["transformer_options"]["sample_sigmas"], timestep[0], rtol=0.0001)
+        sample_sigmas = model_options["transformer_options"]["sample_sigmas"]
+        current_timestep = timestep[0].to(sample_sigmas.dtype)
+        mask = torch.isclose(sample_sigmas, current_timestep, rtol=0.0001)
         matches = torch.nonzero(mask)
         if torch.numel(matches) == 0:
             return  # substep from multi-step sampler: keep self._step from the last full step
@@ -284,54 +566,98 @@ class IndexListContextHandler(ContextHandlerABC):
     def get_context_windows(self, model: BaseModel, x_in: torch.Tensor, model_options: dict[str]) -> list[IndexListContextWindow]:
         full_length = x_in.size(self.dim) # TODO: choose dim based on model
         context_windows = self.context_schedule.func(full_length, self, model_options)
-        context_windows = [IndexListContextWindow(window, dim=self.dim, total_frames=full_length) for window in context_windows]
+        context_windows = [IndexListContextWindow(window, dim=self.dim, total_frames=full_length, context_overlap=self.context_overlap) for window in context_windows]
         return context_windows
 
     def execute(self, calc_cond_batch: Callable, model: BaseModel, conds: list[list[dict]], x_in: torch.Tensor, timestep: torch.Tensor, model_options: dict[str]):
         self._model = model
         self.set_step(timestep, model_options)
-        context_windows = self.get_context_windows(model, x_in, model_options)
-        enumerated_context_windows = list(enumerate(context_windows))
 
-        conds_final = [torch.zeros_like(x_in) for _ in conds]
+        window_state = self._build_window_state(x_in, conds, model)
+        num_modalities = len(window_state.latents)
+
+        context_windows = self.get_context_windows(model, window_state.latents[0], model_options)
+        enumerated_context_windows = list(enumerate(context_windows))
+        total_windows = len(enumerated_context_windows)
+
+        # Initialize per-modality accumulators (length 1 for single-modality)
+        accum = [[torch.zeros_like(m) for _ in conds] for m in window_state.latents]
         if self.fuse_method.name == ContextFuseMethods.RELATIVE:
-            counts_final = [torch.ones(get_shape_for_dim(x_in, self.dim), device=x_in.device) for _ in conds]
+            counts = [[torch.ones(get_shape_for_dim(m, self.dim), device=m.device) for _ in conds] for m in window_state.latents]
         else:
-            counts_final = [torch.zeros(get_shape_for_dim(x_in, self.dim), device=x_in.device) for _ in conds]
-        biases_final = [([0.0] * x_in.shape[self.dim]) for _ in conds]
+            counts = [[torch.zeros(get_shape_for_dim(m, self.dim), device=m.device) for _ in conds] for m in window_state.latents]
+        biases = [[([0.0] * m.shape[self.dim]) for _ in conds] for m in window_state.latents]
 
         for callback in comfy.patcher_extension.get_all_callbacks(IndexListCallbacks.EXECUTE_START, self.callbacks):
             callback(self, model, x_in, conds, timestep, model_options)
 
+        # accumulate results from each context window
         for enum_window in enumerated_context_windows:
-            results = self.evaluate_context_windows(calc_cond_batch, model, x_in, conds, timestep, [enum_window], model_options)
+            results = self.evaluate_context_windows(
+                calc_cond_batch, model, x_in, conds, timestep, [enum_window],
+                model_options, window_state=window_state, total_windows=total_windows)
             for result in results:
-                self.combine_context_window_results(x_in, result.sub_conds_out, result.sub_conds, result.window, result.window_idx, len(enumerated_context_windows), timestep,
-                                            conds_final, counts_final, biases_final)
+                # result.sub_conds_out is per-cond, per-modality: list[list[Tensor]]
+                for mod_idx in range(num_modalities):
+                    mod_out = [result.sub_conds_out[ci][mod_idx] for ci in range(len(conds))]
+                    modality_window = result.window.get_window_for_modality(mod_idx)
+                    self.combine_context_window_results(
+                        window_state.latents[mod_idx], mod_out, result.sub_conds, modality_window,
+                        result.window_idx, total_windows, timestep,
+                        accum[mod_idx], counts[mod_idx], biases[mod_idx])
+
+        # fuse accumulated results into final conds
         try:
-            # finalize conds
-            if self.fuse_method.name == ContextFuseMethods.RELATIVE:
-                # relative is already normalized, so return as is
-                del counts_final
-                return conds_final
-            else:
-                # normalize conds via division by context usage counts
-                for i in range(len(conds_final)):
-                    conds_final[i] /= counts_final[i]
-                del counts_final
-                return conds_final
+            result_out = []
+            for ci in range(len(conds)):
+                finalized = []
+                for mod_idx in range(num_modalities):
+                    if self.fuse_method.name != ContextFuseMethods.RELATIVE:
+                        accum[mod_idx][ci] /= counts[mod_idx][ci]
+                    f = accum[mod_idx][ci]
+
+                    # if guide frames were injected, append them to the end of the fused latents for the next step
+                    if window_state.guide_latents[mod_idx] is not None:
+                        f = torch.cat([f, window_state.guide_latents[mod_idx]], dim=self.dim)
+                    finalized.append(f)
+
+                # pack modalities together if needed
+                if window_state.is_multimodal and len(finalized) > 1:
+                    packed, _ = comfy.utils.pack_latents(finalized)
+                else:
+                    packed = finalized[0]
+
+                result_out.append(packed)
+            return result_out
         finally:
             for callback in comfy.patcher_extension.get_all_callbacks(IndexListCallbacks.EXECUTE_CLEANUP, self.callbacks):
                 callback(self, model, x_in, conds, timestep, model_options)
 
-    def evaluate_context_windows(self, calc_cond_batch: Callable, model: BaseModel, x_in: torch.Tensor, conds, timestep: torch.Tensor, enumerated_context_windows: list[tuple[int, IndexListContextWindow]],
-                                model_options, device=None, first_device=None):
+    def evaluate_context_windows(self, calc_cond_batch: Callable, model: BaseModel, x_in: torch.Tensor, conds,
+                                timestep: torch.Tensor, enumerated_context_windows: list[tuple[int, IndexListContextWindow]],
+                                model_options, window_state: WindowingState, total_windows: int = None,
+                                device=None, first_device=None):
+        """Evaluate context windows and return per-cond, per-modality outputs in ContextResults.sub_conds_out
+
+        For each window:
+        1. Builds windows (for each modality if multimodal)
+        2. Slices window for each modality
+        3. Injects concatenated latent guide frames where present
+        4. Packs together if needed and calls model
+        5. Unpacks and strips any guides from outputs
+        """
+        x = window_state.latents[0]
+
         results: list[ContextResults] = []
         for window_idx, window in enumerated_context_windows:
             # allow processing to end between context window executions for faster Cancel
             comfy.model_management.throw_exception_if_processing_interrupted()
 
-            # causal_window_fix: prepend a pre-window frame that will be stripped post-forward
+            # prepare the window accounting for multimodal windows
+            window = window_state.prepare_window(window, model)
+
+            # causal_window_fix: prepend a pre-window frame that will be stripped post-forward.
+            # Set anchor before slice_for_window so the latent slice and downstream cond slices both pick it up.
             anchor_applied = False
             if self.causal_window_fix:
                 anchor_idx = window.index_list[0] - 1
@@ -339,27 +665,46 @@ class IndexListContextHandler(ContextHandlerABC):
                     window.causal_anchor_index = anchor_idx
                     anchor_applied = True
 
+            # slice the window for each modality, injecting guide frames where applicable
+            sliced, guide_frame_counts_per_modality = window_state.slice_for_window(window, self.latent_retain_index_list, device)
+
             for callback in comfy.patcher_extension.get_all_callbacks(IndexListCallbacks.EVALUATE_CONTEXT_WINDOWS, self.callbacks):
                 callback(self, model, x_in, conds, timestep, model_options, window_idx, window, model_options, device, first_device)
 
-            # update exposed params
+            logging.info(f"Context window {window_idx + 1}/{total_windows or len(enumerated_context_windows)}: frames {window.index_list[0]}-{window.index_list[-1]} of {x.shape[self.dim]}"
+                         + (f" (+{guide_frame_counts_per_modality[0]} guide frames)" if guide_frame_counts_per_modality[0] > 0 else "")
+                         )
+
+            # if multimodal, pack modalities together
+            if window_state.is_multimodal and len(sliced) > 1:
+                sub_x, sub_shapes = comfy.utils.pack_latents(sliced)
+            else:
+                sub_x, sub_shapes = sliced[0], [sliced[0].shape]
+
+            # get resized conds for window
             model_options["transformer_options"]["context_window"] = window
-            # get subsections of x, timestep, conds
-            sub_x = window.get_tensor(x_in, device)
-            sub_timestep = window.get_tensor(timestep, device, dim=0)
-            sub_conds = [self.get_resized_cond(cond, x_in, window, device) for cond in conds]
+            sub_timestep = window.get_tensor(timestep, dim=0)
+            sub_conds = [self.get_resized_cond(cond, x, window) for cond in conds]
 
+            # if multimodal, patch latent_shapes in conds for correct unpacking in model
+            window_state.patch_latent_shapes(sub_conds, sub_shapes)
+
+            # call model on window
             sub_conds_out = calc_cond_batch(model, sub_conds, sub_x, sub_timestep, model_options)
-            if device is not None:
-                for i in range(len(sub_conds_out)):
-                    sub_conds_out[i] = sub_conds_out[i].to(x_in.device)
 
-            # strip causal_window_fix anchor if applied
+            # unpack outputs
+            out_per_modality = [comfy.utils.unpack_latents(sub_conds_out[i], sub_shapes) for i in range(len(sub_conds_out))]
+
+            # strip causal_window_fix anchor from primary modality before guide strip so window_len math stays correct
             if anchor_applied:
-                for i in range(len(sub_conds_out)):
-                    sub_conds_out[i] = sub_conds_out[i].narrow(self.dim, 1, sub_conds_out[i].shape[self.dim] - 1)
+                for ci in range(len(out_per_modality)):
+                    t = out_per_modality[ci][0]
+                    out_per_modality[ci][0] = t.narrow(self.dim, 1, t.shape[self.dim] - 1)
 
-            results.append(ContextResults(window_idx, sub_conds_out, sub_conds, window))
+            # strip injected guide frames
+            window_state.strip_guide_frames(out_per_modality, guide_frame_counts_per_modality, window)
+
+            results.append(ContextResults(window_idx, out_per_modality, sub_conds, window))
         return results
 
 
@@ -383,7 +728,7 @@ class IndexListContextHandler(ContextHandlerABC):
                     biases_final[i][idx] = bias_total + bias
         else:
             # add conds and counts based on weights of fuse method
-            weights = get_context_weights(window.context_length, x_in.shape[self.dim], window.index_list, self, sigma=timestep)
+            weights = get_context_weights(window.context_length, x_in.shape[self.dim], window.index_list, self, sigma=timestep, context_overlap=window.context_overlap)
             weights_tensor = match_weights_to_dim(weights, x_in, self.dim, device=x_in.device)
             for i in range(len(sub_conds_out)):
                 window.add_window(conds_final[i], sub_conds_out[i] * weights_tensor)
@@ -393,16 +738,22 @@ class IndexListContextHandler(ContextHandlerABC):
             callback(self, x_in, sub_conds_out, sub_conds, window, window_idx, total_windows, timestep, conds_final, counts_final, biases_final)
 
 
-def _prepare_sampling_wrapper(executor, model, noise_shape: torch.Tensor, *args, **kwargs):
-    # limit noise_shape length to context_length for more accurate vram use estimation
+def _prepare_sampling_wrapper(executor, model, noise_shape: torch.Tensor, conds, *args, **kwargs):
+    # Scale noise_shape to a single context window so VRAM estimation budgets per-window.
     model_options = kwargs.get("model_options", None)
     if model_options is None:
         raise Exception("model_options not found in prepare_sampling_wrapper; this should never happen, something went wrong.")
     handler: IndexListContextHandler = model_options.get("context_handler", None)
     if handler is not None:
         noise_shape = list(noise_shape)
-        noise_shape[handler.dim] = min(noise_shape[handler.dim], handler.context_length)
-    return executor(model, noise_shape, *args, **kwargs)
+        is_packed = len(noise_shape) == 3 and noise_shape[1] == 1
+        if is_packed:
+            # TODO: latent_shapes cond isn't attached yet at this point, so we can't compute a
+            # per-window flat latent here. Skipping the clamp over-estimates but prevents immediate OOM.
+            pass
+        elif handler.dim < len(noise_shape) and noise_shape[handler.dim] > handler.context_length:
+            noise_shape[handler.dim] = min(noise_shape[handler.dim], handler.context_length)
+    return executor(model, noise_shape, conds, *args, **kwargs)
 
 
 def create_prepare_sampling_wrapper(model: ModelPatcher):
@@ -422,11 +773,12 @@ def _sampler_sample_wrapper(executor, guider, sigmas, extra_args, callback, nois
         raise Exception("context_handler not found in sampler_sample_wrapper; this should never happen, something went wrong.")
     if not handler.freenoise:
         return executor(guider, sigmas, extra_args, callback, noise, *args, **kwargs)
-    noise = apply_freenoise(noise, handler.dim, handler.context_length, handler.context_overlap, extra_args["seed"])
+
+    conds = [guider.conds.get('positive', guider.conds.get('negative', []))]
+    noise = handler._apply_freenoise(noise, conds, extra_args["seed"])
 
     return executor(guider, sigmas, extra_args, callback, noise, *args, **kwargs)
 
-
 def create_sampler_sample_wrapper(model: ModelPatcher):
     model.add_wrapper_with_key(
         comfy.patcher_extension.WrappersMP.SAMPLER_SAMPLE,
@@ -434,7 +786,6 @@ def create_sampler_sample_wrapper(model: ModelPatcher):
         _sampler_sample_wrapper
     )
 
-
 def match_weights_to_dim(weights: list[float], x_in: torch.Tensor, dim: int, device=None) -> torch.Tensor:
     total_dims = len(x_in.shape)
     weights_tensor = torch.Tensor(weights).to(device=device)
@@ -580,8 +931,9 @@ def get_matching_context_schedule(context_schedule: str) -> ContextSchedule:
     return ContextSchedule(context_schedule, func)
 
 
-def get_context_weights(length: int, full_length: int, idxs: list[int], handler: IndexListContextHandler, sigma: torch.Tensor=None):
-    return handler.fuse_method.func(length, sigma=sigma, handler=handler, full_length=full_length, idxs=idxs)
+def get_context_weights(length: int, full_length: int, idxs: list[int], handler: IndexListContextHandler, sigma: torch.Tensor=None, context_overlap: int=None):
+    context_overlap = handler.context_overlap if context_overlap is None else context_overlap
+    return handler.fuse_method.func(length, sigma=sigma, handler=handler, full_length=full_length, idxs=idxs, context_overlap=context_overlap)
 
 
 def create_weights_flat(length: int, **kwargs) -> list[float]:
@@ -599,18 +951,18 @@ def create_weights_pyramid(length: int, **kwargs) -> list[float]:
         weight_sequence = list(range(1, max_weight, 1)) + [max_weight] + list(range(max_weight - 1, 0, -1))
     return weight_sequence
 
-def create_weights_overlap_linear(length: int, full_length: int, idxs: list[int], handler: IndexListContextHandler, **kwargs):
+def create_weights_overlap_linear(length: int, full_length: int, idxs: list[int], context_overlap: int, **kwargs):
     # based on code in Kijai's WanVideoWrapper: https://github.com/kijai/ComfyUI-WanVideoWrapper/blob/dbb2523b37e4ccdf45127e5ae33e31362f755c8e/nodes.py#L1302
     # only expected overlap is given different weights
     weights_torch = torch.ones((length))
     # blend left-side on all except first window
     if min(idxs) > 0:
-        ramp_up = torch.linspace(1e-37, 1, handler.context_overlap)
-        weights_torch[:handler.context_overlap] = ramp_up
+        ramp_up = torch.linspace(1e-37, 1, context_overlap)
+        weights_torch[:context_overlap] = ramp_up
     # blend right-side on all except last window
     if max(idxs) < full_length-1:
-        ramp_down = torch.linspace(1, 1e-37, handler.context_overlap)
-        weights_torch[-handler.context_overlap:] = ramp_down
+        ramp_down = torch.linspace(1, 1e-37, context_overlap)
+        weights_torch[-context_overlap:] = ramp_down
     return weights_torch
 
 class ContextFuseMethods:

comfy/ldm/cosmos/predict2.py

@@ -515,7 +515,7 @@ class Block(nn.Module):
             h=H,
             w=W,
         )
-        x_B_T_H_W_D = x_B_T_H_W_D + gate_self_attn_B_T_1_1_D.to(residual_dtype) * result_B_T_H_W_D.to(residual_dtype)
+        x_B_T_H_W_D = torch.addcmul(x_B_T_H_W_D, gate_self_attn_B_T_1_1_D.to(residual_dtype), result_B_T_H_W_D.to(residual_dtype))
 
         def _x_fn(
             _x_B_T_H_W_D: torch.Tensor,
@@ -548,7 +548,7 @@ class Block(nn.Module):
             shift_cross_attn_B_T_1_1_D,
             transformer_options=transformer_options,
         )
-        x_B_T_H_W_D = result_B_T_H_W_D.to(residual_dtype) * gate_cross_attn_B_T_1_1_D.to(residual_dtype) + x_B_T_H_W_D
+        x_B_T_H_W_D = torch.addcmul(x_B_T_H_W_D, gate_cross_attn_B_T_1_1_D.to(residual_dtype), result_B_T_H_W_D.to(residual_dtype))
 
         normalized_x_B_T_H_W_D = _fn(
             x_B_T_H_W_D,
@@ -557,7 +557,7 @@ class Block(nn.Module):
             shift_mlp_B_T_1_1_D,
         )
         result_B_T_H_W_D = self.mlp(normalized_x_B_T_H_W_D.to(compute_dtype))
-        x_B_T_H_W_D = x_B_T_H_W_D + gate_mlp_B_T_1_1_D.to(residual_dtype) * result_B_T_H_W_D.to(residual_dtype)
+        x_B_T_H_W_D = torch.addcmul(x_B_T_H_W_D, gate_mlp_B_T_1_1_D.to(residual_dtype), result_B_T_H_W_D.to(residual_dtype))
         return x_B_T_H_W_D
 
 

comfy/ldm/krea2/model.py

@@ -0,0 +1,290 @@
+"""Krea 2 (K2) — single-stream MMDiT.
+
+Text tokens produced by a Qwen3-VL-4B 12-layer ``txtfusion`` adapter and patchified image tokens are
+concatenated into one sequence and run through ``layers`` shared transformer blocks with
+AdaLN-single modulation, GQA + per-head QK-norm + sigmoid-gated attention, SwiGLU MLP, and 3-axis RoPE.
+"""
+
+from typing import Optional
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from einops import rearrange
+
+import comfy.model_management
+import comfy.patcher_extension
+import comfy.ldm.common_dit
+from comfy.ldm.flux.layers import EmbedND, timestep_embedding
+from comfy.ldm.flux.math import apply_rope
+from comfy.ldm.modules.attention import optimized_attention_masked
+
+
+class RMSNorm(nn.Module):
+    """RMSNorm with the reference ``(1 + scale)`` weight convention (scale stored zero-centered)."""
+
+    def __init__(self, features: int, eps: float = 1e-5, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.eps = eps
+        self.scale = nn.Parameter(torch.empty(features, device=device, dtype=dtype))
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        dtype = x.dtype
+        weight = comfy.model_management.cast_to(self.scale, dtype=torch.float32, device=x.device) + 1.0
+        return F.rms_norm(x.float(), (x.shape[-1],), weight=weight, eps=self.eps).to(dtype)
+
+
+class QKNorm(nn.Module):
+    def __init__(self, dim: int, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.qnorm = RMSNorm(dim, device=device, dtype=dtype, operations=operations)
+        self.knorm = RMSNorm(dim, device=device, dtype=dtype, operations=operations)
+
+    def forward(self, q, k):
+        return self.qnorm(q), self.knorm(k)
+
+
+class SwiGLU(nn.Module):
+    def __init__(self, features: int, multiplier: int, bias: bool = False, multiple: int = 128,
+                 device=None, dtype=None, operations=None):
+        super().__init__()
+        mlpdim = int(2 * features / 3) * multiplier
+        mlpdim = multiple * ((mlpdim + multiple - 1) // multiple)
+        self.gate = operations.Linear(features, mlpdim, bias=bias, device=device, dtype=dtype)
+        self.up = operations.Linear(features, mlpdim, bias=bias, device=device, dtype=dtype)
+        self.down = operations.Linear(mlpdim, features, bias=bias, device=device, dtype=dtype)
+
+    def forward(self, x):
+        return self.down(F.silu(self.gate(x)).mul_(self.up(x)))
+
+
+class Attention(nn.Module):
+    def __init__(self, dim: int, heads: int, kvheads: Optional[int] = None, bias: bool = False,
+                 device=None, dtype=None, operations=None):
+        super().__init__()
+        self.heads = heads
+        self.kvheads = kvheads if kvheads is not None else heads
+        self.headdim = dim // self.heads
+        self.wq = operations.Linear(dim, self.headdim * self.heads, bias=bias, device=device, dtype=dtype)
+        self.wk = operations.Linear(dim, self.headdim * self.kvheads, bias=bias, device=device, dtype=dtype)
+        self.wv = operations.Linear(dim, self.headdim * self.kvheads, bias=bias, device=device, dtype=dtype)
+        self.gate = operations.Linear(dim, dim, bias=bias, device=device, dtype=dtype)
+        self.qknorm = QKNorm(self.headdim, device=device, dtype=dtype, operations=operations)
+        self.wo = operations.Linear(dim, dim, bias=bias, device=device, dtype=dtype)
+
+    def forward(self, x, freqs=None, mask=None, transformer_options={}):
+        q, k, v, gate = self.wq(x), self.wk(x), self.wv(x), self.gate(x)
+        q = rearrange(q, "B L (H D) -> B H L D", H=self.heads)
+        k = rearrange(k, "B L (H D) -> B H L D", H=self.kvheads)
+        v = rearrange(v, "B L (H D) -> B H L D", H=self.kvheads)
+        q, k = self.qknorm(q, k)
+        if freqs is not None:
+            q, k = apply_rope(q, k, freqs)
+        if self.kvheads != self.heads:
+            rep = self.heads // self.kvheads
+            k = k.repeat_interleave(rep, dim=1)
+            v = v.repeat_interleave(rep, dim=1)
+        out = optimized_attention_masked(q, k, v, self.heads, mask=mask, skip_reshape=True,
+                                         transformer_options=transformer_options)
+        return self.wo(out * F.sigmoid(gate))
+
+
+class SimpleModulation(nn.Module):
+    def __init__(self, dim: int, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.lin = nn.Parameter(torch.empty(2, dim, device=device, dtype=dtype))
+
+    def forward(self, vec):
+        out = vec + comfy.model_management.cast_to(self.lin, dtype=vec.dtype, device=vec.device).unsqueeze(0)
+        scale, shift = out.chunk(2, dim=1)
+        return scale, shift
+
+
+class DoubleSharedModulation(nn.Module):
+    def __init__(self, dim: int, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.lin = nn.Parameter(torch.empty(6 * dim, device=device, dtype=dtype))
+
+    def forward(self, vec):
+        out = vec + comfy.model_management.cast_to(self.lin, dtype=vec.dtype, device=vec.device)
+        return out.chunk(6, dim=-1)
+
+
+class TextFusionBlock(nn.Module):
+    def __init__(self, features, heads, multiplier, bias=False, kvheads=None, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.prenorm = RMSNorm(features, device=device, dtype=dtype, operations=operations)
+        self.postnorm = RMSNorm(features, device=device, dtype=dtype, operations=operations)
+        self.attn = Attention(features, heads, kvheads=kvheads, bias=bias, device=device, dtype=dtype, operations=operations)
+        self.mlp = SwiGLU(features, multiplier, bias, device=device, dtype=dtype, operations=operations)
+
+    def forward(self, x, mask=None, transformer_options={}):
+        x = x + self.attn(self.prenorm(x), mask=mask, transformer_options=transformer_options)
+        x = x + self.mlp(self.postnorm(x))
+        return x
+
+
+class TextFusionTransformer(nn.Module):
+    def __init__(self, num_txt_layers, txt_dim, heads, multiplier, bias=False, kvheads=None, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.layerwise_blocks = nn.ModuleList([
+            TextFusionBlock(txt_dim, heads, multiplier, bias, kvheads, device=device, dtype=dtype, operations=operations)
+            for _ in range(2)
+        ])
+        self.projector = operations.Linear(num_txt_layers, 1, bias=False, device=device, dtype=dtype)
+        self.refiner_blocks = nn.ModuleList([
+            TextFusionBlock(txt_dim, heads, multiplier, bias, kvheads, device=device, dtype=dtype, operations=operations)
+            for _ in range(2)
+        ])
+
+    def forward(self, x, mask=None, transformer_options={}):
+        b, l, n, d = x.shape
+        x = x.reshape(b * l, n, d)
+        for block in self.layerwise_blocks:
+            x = block(x.contiguous(), mask=None, transformer_options=transformer_options)
+        x = rearrange(x, "(b l) n d -> b l d n", b=b, l=l)
+        x = self.projector(x).squeeze(-1)
+        for block in self.refiner_blocks:
+            x = block(x, mask=mask, transformer_options=transformer_options)
+        return x
+
+
+class SingleStreamBlock(nn.Module):
+    def __init__(self, features, heads, multiplier, bias=False, kvheads=None, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.mod = DoubleSharedModulation(features, device=device, dtype=dtype, operations=operations)
+        self.prenorm = RMSNorm(features, device=device, dtype=dtype, operations=operations)
+        self.postnorm = RMSNorm(features, device=device, dtype=dtype, operations=operations)
+        self.attn = Attention(features, heads, kvheads=kvheads, bias=bias, device=device, dtype=dtype, operations=operations)
+        self.mlp = SwiGLU(features, multiplier, bias, device=device, dtype=dtype, operations=operations)
+
+    def forward(self, x, vec, freqs, mask=None, transformer_options={}):
+        prescale, preshift, pregate, postscale, postshift, postgate = self.mod(vec)
+        x = x + pregate * self.attn((1 + prescale) * self.prenorm(x) + preshift, freqs, mask, transformer_options=transformer_options)
+        x = x + postgate * self.mlp((1 + postscale) * self.postnorm(x) + postshift)
+        return x
+
+
+class LastLayer(nn.Module):
+    def __init__(self, features, patch, channels, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.norm = RMSNorm(features, device=device, dtype=dtype, operations=operations)
+        self.linear = operations.Linear(features, patch * patch * channels, bias=True, device=device, dtype=dtype)
+        self.modulation = SimpleModulation(features, device=device, dtype=dtype, operations=operations)
+
+    def forward(self, x, tvec):
+        scale, shift = self.modulation(tvec)
+        x = (1 + scale) * self.norm(x) + shift
+        return self.linear(x)
+
+
+class SingleStreamDiT(nn.Module):
+    def __init__(self, features=6144, tdim=256, txtdim=2560, heads=48, kvheads=12, multiplier=4,
+                 layers=28, patch=2, channels=16, bias=False, theta=1e3, txtlayers=12,
+                 txtheads=20, txtkvheads=20, image_model=None,
+                 device=None, dtype=None, operations=None, **kwargs):
+        super().__init__()
+        self.dtype = dtype
+        self.patch = patch
+        self.channels = channels
+        self.tdim = tdim
+        self.heads = heads
+        self.txtdim = txtdim
+        self.txtlayers = txtlayers
+
+        headdim = features // heads
+        axes = [headdim - 12 * (headdim // 16), 6 * (headdim // 16), 6 * (headdim // 16)]
+        assert sum(axes) == headdim, f"axes {axes} sum != headdim {headdim}"
+        self.pe_embedder = EmbedND(dim=headdim, theta=int(theta), axes_dim=axes)
+
+        self.first = operations.Linear(channels * patch ** 2, features, bias=True, device=device, dtype=dtype)
+        self.blocks = nn.ModuleList([
+            SingleStreamBlock(features, heads, multiplier, bias, kvheads, device=device, dtype=dtype, operations=operations)
+            for _ in range(layers)
+        ])
+        self.tmlp = nn.Sequential(
+            operations.Linear(tdim, features, device=device, dtype=dtype),
+            nn.GELU(approximate="tanh"),
+            operations.Linear(features, features, device=device, dtype=dtype),
+        )
+        self.txtfusion = TextFusionTransformer(txtlayers, txtdim, txtheads, multiplier, bias, txtkvheads,
+                                               device=device, dtype=dtype, operations=operations)
+        self.txtmlp = nn.Sequential(
+            RMSNorm(txtdim, device=device, dtype=dtype, operations=operations),
+            operations.Linear(txtdim, features, device=device, dtype=dtype),
+            nn.GELU(approximate="tanh"),
+            operations.Linear(features, features, device=device, dtype=dtype),
+        )
+        self.last = LastLayer(features, patch, channels, device=device, dtype=dtype, operations=operations)
+        self.tproj = nn.Sequential(
+            nn.GELU(approximate="tanh"),
+            operations.Linear(features, features * 6, device=device, dtype=dtype),
+        )
+
+    def forward(self, x, timesteps, context, attention_mask=None, transformer_options={}, **kwargs):
+        return comfy.patcher_extension.WrapperExecutor.new_class_executor(
+            self._forward,
+            self,
+            comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, transformer_options),
+        ).execute(x, timesteps, context, attention_mask, transformer_options, **kwargs)
+
+    def _forward(self, x, timesteps, context, attention_mask=None, transformer_options={}, **kwargs):
+        temporal = x.ndim == 5
+        if temporal:
+            b5, c5, t5, h5, w5 = x.shape
+            x = x.reshape(b5 * t5, c5, h5, w5)
+        bs, c, H_orig, W_orig = x.shape
+        patch = self.patch
+        # Pad the latent up to a multiple of patch (as Flux/Lumina/QwenImage do); crop back at the end.
+        x = comfy.ldm.common_dit.pad_to_patch_size(x, (patch, patch))
+        H, W = x.shape[-2], x.shape[-1]
+        h_, w_ = H // patch, W // patch
+
+        # context arrives as (B, seq, txtlayers*txtdim); reshape to (B, txtlayers, seq, txtdim).
+        context = self._unpack_context(context)
+
+        img = rearrange(x, "b c (h ph) (w pw) -> b (h w) (c ph pw)", ph=patch, pw=patch)
+        img = self.first(img)
+
+        t = self.tmlp(timestep_embedding(timesteps, self.tdim).unsqueeze(1).to(img.dtype))
+        tvec = self.tproj(t)
+
+        context = self.txtfusion(context, mask=None, transformer_options=transformer_options)
+        context = self.txtmlp(context)
+
+        txtlen, imglen = context.shape[1], img.shape[1]
+        combined = torch.cat((context, img), dim=1)
+
+        # Position ids: text at 0, image at (0, h_idx, w_idx).
+        device = combined.device
+        txtpos = torch.zeros(bs, txtlen, 3, device=device, dtype=torch.float32)
+        imgids = torch.zeros(h_, w_, 3, device=device, dtype=torch.float32)
+        imgids[..., 1] = torch.arange(h_, device=device, dtype=torch.float32)[:, None]
+        imgids[..., 2] = torch.arange(w_, device=device, dtype=torch.float32)[None, :]
+        imgpos = imgids.reshape(1, h_ * w_, 3).repeat(bs, 1, 1)
+        pos = torch.cat((txtpos, imgpos), dim=1)
+
+        freqs = self.pe_embedder(pos)
+
+        for block in self.blocks:
+            combined = block(combined, tvec, freqs, None, transformer_options=transformer_options)
+
+        final = self.last(combined, t)
+        out = final[:, txtlen:txtlen + imglen, :]
+        out = rearrange(out, "b (h w) (c ph pw) -> b c (h ph) (w pw)",
+                        h=h_, w=w_, ph=patch, pw=patch, c=self.channels)
+        out = out[:, :, :H_orig, :W_orig]  # crop padding back off
+        if temporal:
+            out = out.reshape(b5, t5, self.channels, H_orig, W_orig).movedim(1, 2)
+        return out
+
+    def _unpack_context(self, context):
+        # context: (B, seq, txtlayers*txtdim) -> (B, seq, txtlayers, txtdim).
+        b, seq, fused = context.shape
+        if fused != self.txtlayers * self.txtdim:
+            raise ValueError(
+                f"Krea2 expects conditioning with {self.txtlayers}x{self.txtdim}={self.txtlayers * self.txtdim} "
+                f"features (a {self.txtlayers}-layer Qwen3-VL stack) but got {fused}. "
+                f"Load the text encoder with CLIPLoader type 'krea2'."
+            )
+        return context.reshape(b, seq, self.txtlayers, self.txtdim)

comfy/ldm/lightricks/model.py

@@ -1085,7 +1085,7 @@ class LTXVModel(LTXBaseModel):
         )
 
         grid_mask = None
-        if keyframe_idxs is not None:
+        if keyframe_idxs is not None and keyframe_idxs.shape[2] > 0:
             additional_args.update({ "orig_patchified_shape": list(x.shape)})
             denoise_mask = self.patchifier.patchify(denoise_mask)[0]
             grid_mask = ~torch.any(denoise_mask < 0, dim=-1)[0]
@@ -1330,7 +1330,7 @@ class LTXVModel(LTXBaseModel):
         x = x * (1 + scale) + shift
         x = self.proj_out(x)
 
-        if keyframe_idxs is not None:
+        if keyframe_idxs is not None and keyframe_idxs.shape[2] > 0:
             grid_mask = kwargs["grid_mask"]
             orig_patchified_shape = kwargs["orig_patchified_shape"]
             full_x = torch.zeros(orig_patchified_shape, dtype=x.dtype, device=x.device)

comfy/lora.py

@@ -326,6 +326,17 @@ def model_lora_keys_unet(model, key_map={}):
                 key_map["transformer.{}".format(key_lora)] = k
                 key_map["lycoris_{}".format(key_lora.replace(".", "_"))] = k #SimpleTuner lycoris format
 
+    if isinstance(model, comfy.model_base.Krea2):
+        diffusers_keys = comfy.utils.krea2_to_diffusers(model.model_config.unet_config, output_prefix="diffusion_model.")
+        for k in diffusers_keys:
+            if k.endswith(".weight"):
+                to = diffusers_keys[k]
+                key_lora = k[:-len(".weight")]
+                key_map["diffusion_model.{}".format(key_lora)] = to
+                key_map["transformer.{}".format(key_lora)] = to
+                key_map["lycoris_{}".format(key_lora.replace(".", "_"))] = to
+                key_map[key_lora] = to
+
     if isinstance(model, comfy.model_base.Lumina2):
         diffusers_keys = comfy.utils.z_image_to_diffusers(model.model_config.unet_config, output_prefix="diffusion_model.")
         for k in diffusers_keys:

comfy/model_base.py

@@ -21,6 +21,7 @@ import comfy.ldm.hunyuan3dv2_1.hunyuandit
 import torch
 import logging
 import comfy.ldm.lightricks.av_model
+import comfy.ldm.lightricks.symmetric_patchifier
 import comfy.context_windows
 from comfy.ldm.modules.diffusionmodules.openaimodel import UNetModel, Timestep
 from comfy.ldm.cascade.stage_c import StageC
@@ -58,6 +59,7 @@ import comfy.ldm.omnigen.omnigen2
 import comfy.ldm.boogu.model
 import comfy.ldm.qwen_image.model
 import comfy.ldm.ideogram4.model
+import comfy.ldm.krea2.model
 import comfy.ldm.kandinsky5.model
 import comfy.ldm.anima.model
 import comfy.ldm.ace.ace_step15
@@ -1205,6 +1207,127 @@ class LTXAV(BaseModel):
     def scale_latent_inpaint(self, sigma, noise, latent_image, **kwargs):
         return latent_image
 
+    def map_context_window_to_modalities(self, primary_indices, latent_shapes, dim):
+        result = [primary_indices]
+        if len(latent_shapes) < 2:
+            return result
+
+        video_total = latent_shapes[0][dim]
+
+        for i in range(1, len(latent_shapes)):
+            mod_total = latent_shapes[i][dim]
+            # Map each primary index to its proportional range of modality indices and
+            # concatenate in order. Preserves wrapped/strided geometry so the modality
+            # attends to the same temporal regions as the primary window.
+            mod_indices = []
+            seen = set()
+            for v_idx in primary_indices:
+                a_start = min(int(round(v_idx * mod_total / video_total)), mod_total - 1)
+                a_end = min(int(round((v_idx + 1) * mod_total / video_total)), mod_total)
+                if a_end <= a_start:
+                    a_end = a_start + 1
+                for a in range(a_start, a_end):
+                    if a not in seen:
+                        seen.add(a)
+                        mod_indices.append(a)
+            result.append(mod_indices)
+
+        return result
+
+    @staticmethod
+    def _get_guide_entries(conds):
+        for cond_list in conds:
+            if cond_list is None:
+                continue
+            for cond_dict in cond_list:
+                model_conds = cond_dict.get('model_conds', {})
+                entries = model_conds.get('guide_attention_entries')
+                if entries is not None and hasattr(entries, 'cond') and entries.cond:
+                    return entries.cond
+        return None
+
+    def resize_cond_for_context_window(self, cond_key, cond_value, window, x_in, device, retain_index_list=[]):
+        # Audio denoise mask — slice using audio modality window
+        if cond_key == "audio_denoise_mask" and hasattr(window, 'modality_windows') and window.modality_windows:
+            audio_window = window.modality_windows.get(1)
+            if audio_window is not None and hasattr(cond_value, "cond") and isinstance(cond_value.cond, torch.Tensor):
+                sliced = audio_window.get_tensor(cond_value.cond, device, dim=2)
+                return cond_value._copy_with(sliced)
+
+        # Video denoise mask — split into video + guide portions, slice each
+        if cond_key == "denoise_mask" and hasattr(cond_value, "cond") and isinstance(cond_value.cond, torch.Tensor):
+            cond_tensor = cond_value.cond
+            guide_count = cond_tensor.size(window.dim) - x_in.size(window.dim)
+            if guide_count > 0:
+                T_video = x_in.size(window.dim)
+                video_mask = cond_tensor.narrow(window.dim, 0, T_video)
+                guide_mask = cond_tensor.narrow(window.dim, T_video, guide_count)
+                sliced_video = window.get_tensor(video_mask, device, retain_index_list=retain_index_list)
+                suffix_indices = window.guide_frames_indices
+                if suffix_indices:
+                    idx = tuple([slice(None)] * window.dim + [suffix_indices])
+                    sliced_guide = guide_mask[idx].to(device)
+                    return cond_value._copy_with(torch.cat([sliced_video, sliced_guide], dim=window.dim))
+                else:
+                    return cond_value._copy_with(sliced_video)
+
+        # Keyframe indices — regenerate pixel coords for window, select guide positions
+        if cond_key == "keyframe_idxs":
+            kf_local_pos = window.guide_kf_local_positions
+            if not kf_local_pos:
+                return cond_value._copy_with(cond_value.cond[:, :, :0, :])  # empty
+            H, W = x_in.shape[3], x_in.shape[4]
+            window_len = len(window.index_list)
+            # account for causal_window_fix anchor in coord space size
+            anchor_idx = getattr(window, 'causal_anchor_index', None)
+            if anchor_idx is not None and anchor_idx >= 0:
+                window_len += 1
+            patchifier = self.diffusion_model.patchifier
+            latent_coords = patchifier.get_latent_coords(window_len, H, W, 1, cond_value.cond.device)
+            scale_factors = self.diffusion_model.vae_scale_factors
+            pixel_coords = comfy.ldm.lightricks.symmetric_patchifier.latent_to_pixel_coords(
+                latent_coords,
+                scale_factors,
+                causal_fix=self.diffusion_model.causal_temporal_positioning)
+            tokens = []
+            for pos in kf_local_pos:
+                tokens.extend(range(pos * H * W, (pos + 1) * H * W))
+            pixel_coords = pixel_coords[:, :, tokens, :]
+
+            # Adjust spatial end positions for dilated (downscaled) guides.
+            # Each guide entry may have a different downscale factor; expand the
+            # per-entry factor to cover all tokens belonging to that entry.
+            downscale_factors = window.guide_downscale_factors
+            overlap_info = window.guide_overlap_info
+            if downscale_factors:
+                per_token_factor = []
+                for (entry_idx, overlap_count), dsf in zip(overlap_info, downscale_factors):
+                    per_token_factor.extend([dsf] * (overlap_count * H * W))
+                factor_tensor = torch.tensor(per_token_factor, device=pixel_coords.device, dtype=pixel_coords.dtype)
+                spatial_end_offset = (factor_tensor.unsqueeze(0).unsqueeze(0).unsqueeze(-1) - 1) * torch.tensor(
+                    scale_factors[1:], device=pixel_coords.device, dtype=pixel_coords.dtype,
+                ).view(1, -1, 1, 1)
+                pixel_coords[:, 1:, :, 1:] += spatial_end_offset
+
+            B = cond_value.cond.shape[0]
+            if B > 1:
+                pixel_coords = pixel_coords.expand(B, -1, -1, -1)
+            return cond_value._copy_with(pixel_coords)
+
+        # Guide attention entries — adjust per-guide counts based on window overlap
+        if cond_key == "guide_attention_entries":
+            overlap_info = window.guide_overlap_info
+            H, W = x_in.shape[3], x_in.shape[4]
+            new_entries = []
+            for entry_idx, overlap_count in overlap_info:
+                e = cond_value.cond[entry_idx]
+                new_entries.append({**e,
+                    "pre_filter_count": overlap_count * H * W,
+                    "latent_shape": [overlap_count, H, W]})
+            return cond_value._copy_with(new_entries)
+
+        return None
+
 class HunyuanVideo(BaseModel):
     def __init__(self, model_config, model_type=ModelType.FLOW, device=None):
         super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.hunyuan_video.model.HunyuanVideo)
@@ -2162,6 +2285,17 @@ class Ideogram4(BaseModel):
             out['c_crossattn'] = comfy.conds.CONDRegular(cross_attn)
         return out
 
+class Krea2(BaseModel):
+    def __init__(self, model_config, model_type=ModelType.FLUX, device=None):
+        super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.krea2.model.SingleStreamDiT)
+
+    def extra_conds(self, **kwargs):
+        out = super().extra_conds(**kwargs)
+        cross_attn = kwargs.get("cross_attn", None)
+        if cross_attn is not None:
+            out['c_crossattn'] = comfy.conds.CONDRegular(cross_attn)
+        return out
+
 class HunyuanImage21(BaseModel):
     def __init__(self, model_config, model_type=ModelType.FLOW, device=None):
         super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.hunyuan_video.model.HunyuanVideo)

comfy/model_detection.py

@@ -876,6 +876,21 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
         dit_config["num_layers"] = count_blocks(state_dict_keys, '{}layers.'.format(key_prefix) + '{}.')
         return dit_config
 
+    if '{}txtfusion.projector.weight'.format(key_prefix) in state_dict_keys:  # Krea 2 (K2)
+        dit_config = {}
+        dit_config["image_model"] = "krea2"
+        head_dim = 128
+        first_w = state_dict['{}first.weight'.format(key_prefix)]  # (features, channels*patch^2)
+        dit_config["features"] = first_w.shape[0]
+        dit_config["channels"] = first_w.shape[1] // (2 * 2)  # patch=2
+        dit_config["patch"] = 2
+        dit_config["layers"] = count_blocks(state_dict_keys, '{}blocks.'.format(key_prefix) + '{}.')
+        dit_config["heads"] = state_dict['{}blocks.0.attn.wq.weight'.format(key_prefix)].shape[0] // head_dim
+        dit_config["kvheads"] = state_dict['{}blocks.0.attn.wk.weight'.format(key_prefix)].shape[0] // head_dim
+        dit_config["txtlayers"] = state_dict['{}txtfusion.projector.weight'.format(key_prefix)].shape[1]
+        dit_config["txtdim"] = state_dict['{}txtfusion.layerwise_blocks.0.prenorm.scale'.format(key_prefix)].shape[0]
+        return dit_config
+
     if '{}visual_transformer_blocks.0.cross_attention.key_norm.weight'.format(key_prefix) in state_dict_keys: # Kandinsky 5
         dit_config = {}
         model_dim = state_dict['{}visual_embeddings.in_layer.bias'.format(key_prefix)].shape[0]

comfy/sd.py

@@ -58,6 +58,7 @@ import comfy.text_encoders.omnigen2
 import comfy.text_encoders.qwen_image
 import comfy.text_encoders.hunyuan_image
 import comfy.text_encoders.z_image
+import comfy.text_encoders.krea2
 import comfy.text_encoders.ideogram4
 import comfy.text_encoders.ovis
 import comfy.text_encoders.kandinsky5
@@ -1317,6 +1318,7 @@ class CLIPType(Enum):
     PIXELDIT = 29
     IDEOGRAM4 = 30
     BOOGU = 31
+    KREA2 = 32
 
 
 
@@ -1642,6 +1644,10 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
                 clip_data[0] = comfy.utils.state_dict_prefix_replace(clip_data[0], {"model.language_model.": "model.", "model.visual.": "visual.", "lm_head.": "model.lm_head."})
                 clip_target.clip = comfy.text_encoders.boogu.te(**llama_detect(clip_data))
                 clip_target.tokenizer = comfy.text_encoders.boogu.BooguTokenizer
+            elif clip_type == CLIPType.KREA2 and te_model == TEModel.QWEN3VL_4B:  # Krea2: full Qwen3-VL-4B (12-layer tap for conditioning + multimodal generate).
+                clip_data[0] = comfy.utils.state_dict_prefix_replace(clip_data[0], {"model.language_model.": "model.", "model.visual.": "visual.", "lm_head.": "model.lm_head."})
+                clip_target.clip = comfy.text_encoders.krea2.te(**llama_detect(clip_data))
+                clip_target.tokenizer = comfy.text_encoders.krea2.Krea2Tokenizer
             elif clip_type in (CLIPType.FLUX, CLIPType.FLUX2):  # Flux2 Klein reuses the Qwen3-VL LM (3-layer tap -> 12288); visual unused.
                 klein_model_type = "qwen3_8b" if te_model == TEModel.QWEN3VL_8B else "qwen3_4b"
                 clip_target.clip = comfy.text_encoders.flux.klein_te(**llama_detect(clip_data), model_type=klein_model_type)

comfy/supported_models.py

@@ -26,6 +26,7 @@ import comfy.text_encoders.kandinsky5
 import comfy.text_encoders.z_image
 import comfy.text_encoders.ideogram4
 import comfy.text_encoders.boogu
+import comfy.text_encoders.krea2
 import comfy.text_encoders.anima
 import comfy.text_encoders.ace15
 import comfy.text_encoders.longcat_image
@@ -1827,6 +1828,35 @@ class Ideogram4(supported_models_base.BASE):
         hunyuan_detect = comfy.text_encoders.hunyuan_video.llama_detect(state_dict, "{}qwen3vl_8b.transformer.".format(pref))
         return supported_models_base.ClipTarget(comfy.text_encoders.ideogram4.Ideogram4Tokenizer, comfy.text_encoders.ideogram4.te(**hunyuan_detect))
 
+
+class Krea2(supported_models_base.BASE):
+    unet_config = {
+        "image_model": "krea2",
+    }
+
+    sampling_settings = {
+        "multiplier": 1.0,
+        "shift": 1.15,
+    }
+
+    memory_usage_factor = 2.2
+
+    latent_format = latent_formats.Wan21
+
+    supported_inference_dtypes = [torch.bfloat16, torch.float16, torch.float32]
+
+    vae_key_prefix = ["vae."]
+    text_encoder_key_prefix = ["text_encoders."]
+
+    def get_model(self, state_dict, prefix="", device=None):
+        out = model_base.Krea2(self, device=device)
+        return out
+
+    def clip_target(self, state_dict={}):
+        pref = self.text_encoder_key_prefix[0]
+        hunyuan_detect = comfy.text_encoders.hunyuan_video.llama_detect(state_dict, "{}qwen3vl_4b.transformer.".format(pref))
+        return supported_models_base.ClipTarget(comfy.text_encoders.krea2.Krea2Tokenizer, comfy.text_encoders.krea2.te(**hunyuan_detect))
+
 class QwenImage(supported_models_base.BASE):
     unet_config = {
         "image_model": "qwen_image",
@@ -2335,6 +2365,7 @@ models = [
     Boogu,
     QwenImage,
     Ideogram4,
+    Krea2,
     Flux2,
     Lens,
     Kandinsky5Image,

comfy/text_encoders/krea2.py

@@ -0,0 +1,84 @@
+"""Krea 2 (K2) text encoder: Qwen3-VL-4B, 12-layer tap.
+
+K2 conditions on a stack of hidden states from 12 layers of Qwen3-VL-4B
+(reference taps ``hidden_states[2,5,8,...,35]``), kept as a ``(B, 12, seq, 2560)`` tensor and
+consumed by the DiT's internal ``txtfusion`` adapter. Comfy carries conditioning as a 3D tensor,
+so the 12-layer stack is flattened to ``(B, seq, 12*2560)`` here and unpacked inside the model.
+"""
+
+import numbers
+
+import torch
+
+import comfy.text_encoders.qwen3vl
+from comfy import sd1_clip
+
+# tap k == hidden_states[k] (no offset).
+KREA2_TAP_LAYERS = [2, 5, 8, 11, 14, 17, 20, 23, 26, 29, 32, 35]
+
+# Identical system template to Qwen-Image; Krea2 strips the system+user-opening prefix.
+KREA2_TEMPLATE = "<|im_start|>system\nDescribe the image by detailing the color, shape, size, texture, quantity, text, spatial relationships of the objects and background:<|im_end|>\n<|im_start|>user\n{}<|im_end|>\n<|im_start|>assistant\n"
+
+
+class Krea2Tokenizer(comfy.text_encoders.qwen3vl.Qwen3VLTokenizer):
+    def __init__(self, embedding_directory=None, tokenizer_data={}):
+        super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data, model_type="qwen3vl_4b")
+        self.llama_template = KREA2_TEMPLATE  # conditioning template; image text-gen uses qwen3vl's default image template.
+
+    def tokenize_with_weights(self, text, return_word_ids=False, llama_template=None, images=[], prevent_empty_text=False, thinking=True, **kwargs):
+        # Krea2 conditions on the no-think template; thinking=True drops the empty <think> block qwen3vl adds.
+        return super().tokenize_with_weights(text, return_word_ids=return_word_ids, llama_template=llama_template, images=images, prevent_empty_text=prevent_empty_text, thinking=thinking, **kwargs)
+
+
+class Krea2Qwen3VLClipModel(comfy.text_encoders.qwen3vl.Qwen3VLClipModel):
+    def __init__(self, device="cpu", dtype=None, attention_mask=True, model_options={}):
+        super().__init__(device=device, layer=KREA2_TAP_LAYERS, layer_idx=None, dtype=dtype,
+                         attention_mask=attention_mask, model_options=model_options, model_type="qwen3vl_4b")
+
+
+class Krea2TEModel(sd1_clip.SD1ClipModel):
+    def __init__(self, device="cpu", dtype=None, model_options={}):
+        super().__init__(device=device, dtype=dtype, name="qwen3vl_4b", clip_model=Krea2Qwen3VLClipModel, model_options=model_options)
+
+    def encode_token_weights(self, token_weight_pairs, template_end=-1):
+        out, pooled, extra = super().encode_token_weights(token_weight_pairs)  # out: (B, 12, seq, 2560)
+        tok_pairs = token_weight_pairs["qwen3vl_4b"][0]
+
+        # Strip the system + user-opening prefix
+        count_im_start = 0
+        if template_end == -1:
+            for i, v in enumerate(tok_pairs):
+                elem = v[0]
+                if not torch.is_tensor(elem) and isinstance(elem, numbers.Integral):
+                    if elem == 151644 and count_im_start < 2:
+                        template_end = i
+                        count_im_start += 1
+            if out.shape[2] > (template_end + 3):
+                if tok_pairs[template_end + 1][0] == 872:      # "user"
+                    if tok_pairs[template_end + 2][0] == 198:   # "\n"
+                        template_end += 3
+
+        out = out[:, :, template_end:]
+
+        b, n, seq, h = out.shape
+        # Flatten the 12-layer axis into the feature dim: (B, seq, 12*2560). Unpacked in the model.
+        out = out.permute(0, 2, 1, 3).reshape(b, seq, n * h)
+
+        if "attention_mask" in extra:
+            extra["attention_mask"] = extra["attention_mask"][:, template_end:]
+            if extra["attention_mask"].sum() == torch.numel(extra["attention_mask"]):
+                extra.pop("attention_mask")
+
+        return out, pooled, extra
+
+
+def te(dtype_llama=None, llama_quantization_metadata=None):
+    class Krea2TEModel_(Krea2TEModel):
+        def __init__(self, device="cpu", dtype=None, model_options={}):
+            if llama_quantization_metadata is not None:
+                model_options = model_options.copy()
+                model_options["quantization_metadata"] = llama_quantization_metadata
+            if dtype_llama is not None:
+                dtype = dtype_llama
+            super().__init__(device=device, dtype=dtype, model_options=model_options)
+    return Krea2TEModel_

comfy/utils.py

@@ -828,6 +828,43 @@ def twinflow_z_image_key_mapping(state_dict, key):
         if new_key not in state_dict:
             state_dict[new_key] = state_dict.pop(key)
     return state_dic
+def krea2_to_diffusers(mmdit_config, output_prefix=""):
+    n_layers = mmdit_config.get("layers", 0)
+    n_txt_layerwise = 2  # TextFusionTransformer hardcodes 2 layerwise + 2 refiner blocks
+    n_txt_refiner = 2
+    key_map = {}
+
+    def add_block(prefix_to, prefix_from):
+        block_map = {
+            "attn.to_q": "attn.wq", "attn.to_k": "attn.wk", "attn.to_v": "attn.wv",
+            "attn.to_gate": "attn.gate", "attn.to_out.0": "attn.wo",
+            "attn.to_out": "attn.wo",  # some tools drop the ".0" on to_out
+            "ff.gate": "mlp.gate", "ff.up": "mlp.up", "ff.down": "mlp.down",
+        }
+        for d, c in block_map.items():
+            key_map["{}.{}.weight".format(prefix_to, d)] = "{}{}.{}.weight".format(output_prefix, prefix_from, c)
+
+    for i in range(n_layers):
+        add_block("transformer_blocks.{}".format(i), "blocks.{}".format(i))
+    for i in range(n_txt_layerwise):
+        add_block("text_fusion.layerwise_blocks.{}".format(i), "txtfusion.layerwise_blocks.{}".format(i))
+    for i in range(n_txt_refiner):
+        add_block("text_fusion.refiner_blocks.{}".format(i), "txtfusion.refiner_blocks.{}".format(i))
+
+    MAP_BASIC = [
+        ("img_in", "first"),
+        ("time_embed.linear_1", "tmlp.0"),
+        ("time_embed.linear_2", "tmlp.2"),
+        ("time_mod_proj", "tproj.1"),
+        ("txt_in.linear_1", "txtmlp.1"),
+        ("txt_in.linear_2", "txtmlp.3"),
+        ("text_fusion.projector", "txtfusion.projector"),
+        ("final_layer.linear", "last.linear"),
+    ]
+    for d, c in MAP_BASIC:
+        key_map["{}.weight".format(d)] = "{}{}.weight".format(output_prefix, c)
+
+    return key_map
 
 def repeat_to_batch_size(tensor, batch_size, dim=0):
     if tensor.shape[dim] > batch_size:

comfy_api_nodes/apis/bytedance.py

@@ -163,15 +163,27 @@ class SeedanceVirtualLibraryCreateAssetRequest(BaseModel):
     asset_type: str | None = Field(None, description="BytePlus asset type. Defaults to Image server-side when omitted.")
 
 
-# Dollars per 1K tokens, keyed by (model_id, has_video_input).
+# Dollars per 1K tokens, keyed by (model_id, has_video_input, resolution).
 SEEDANCE2_PRICE_PER_1K_TOKENS = {
-    ("dreamina-seedance-2-0-260128", False): 0.007,
-    ("dreamina-seedance-2-0-260128", True): 0.0043,
-    ("dreamina-seedance-2-0-fast-260128", False): 0.0056,
-    ("dreamina-seedance-2-0-fast-260128", True): 0.0033,
+    ("dreamina-seedance-2-0-260128", False, "480p"): 0.007,
+    ("dreamina-seedance-2-0-260128", True, "480p"): 0.0043,
+    ("dreamina-seedance-2-0-260128", False, "720p"): 0.007,
+    ("dreamina-seedance-2-0-260128", True, "720p"): 0.0043,
+    ("dreamina-seedance-2-0-260128", False, "1080p"): 0.0077,
+    ("dreamina-seedance-2-0-260128", True, "1080p"): 0.0047,
+    ("dreamina-seedance-2-0-260128", False, "4k"): 0.004,
+    ("dreamina-seedance-2-0-260128", True, "4k"): 0.0024,
+    ("dreamina-seedance-2-0-fast-260128", False, "480p"): 0.0056,
+    ("dreamina-seedance-2-0-fast-260128", True, "480p"): 0.0033,
+    ("dreamina-seedance-2-0-fast-260128", False, "720p"): 0.0056,
+    ("dreamina-seedance-2-0-fast-260128", True, "720p"): 0.0033,
 }
 
 
+def seedance2_price_per_1k_tokens(model_id: str, has_video_input: bool, resolution: str) -> float | None:
+    return SEEDANCE2_PRICE_PER_1K_TOKENS.get((model_id, has_video_input, resolution))
+
+
 RECOMMENDED_PRESETS = [
     ("1024x1024 (1:1)", 1024, 1024),
     ("864x1152 (3:4)", 864, 1152),

comfy_api_nodes/nodes_bytedance.py

@@ -15,7 +15,6 @@ from comfy_api_nodes.apis.bytedance import (
     RECOMMENDED_PRESETS_SEEDREAM_4_0,
     RECOMMENDED_PRESETS_SEEDREAM_4_5,
     RECOMMENDED_PRESETS_SEEDREAM_5_LITE,
-    SEEDANCE2_PRICE_PER_1K_TOKENS,
     SEEDANCE2_REF_VIDEO_PIXEL_LIMITS,
     VIDEO_TASKS_EXECUTION_TIME,
     GetAssetResponse,
@@ -40,6 +39,7 @@ from comfy_api_nodes.apis.bytedance import (
     TaskVideoContentUrl,
     Text2ImageTaskCreationRequest,
     Text2VideoTaskCreationRequest,
+    seedance2_price_per_1k_tokens,
 )
 from comfy_api_nodes.util import (
     ApiEndpoint,
@@ -141,7 +141,7 @@ SEEDANCE2_RATIO_WH = {
     "9:16": (9, 16),
     "21:9": (21, 9),
 }
-SEEDANCE2_RES_SHORT_SIDE = {"480p": 480, "720p": 720, "1080p": 1080}
+SEEDANCE2_RES_SHORT_SIDE = {"480p": 480, "720p": 720, "1080p": 1080, "4k": 2160}
 
 
 def _seedance2_target_dims(resolution: str, ratio: str, image: torch.Tensor) -> tuple[int, int]:
@@ -377,9 +377,9 @@ async def _seedance_virtual_library_upload_video_asset(
     return f"asset://{create_resp.asset_id}"
 
 
-def _seedance2_price_extractor(model_id: str, has_video_input: bool):
+def _seedance2_price_extractor(model_id: str, has_video_input: bool, resolution: str):
     """Returns a price_extractor closure for Seedance 2.0 poll_op."""
-    rate = SEEDANCE2_PRICE_PER_1K_TOKENS.get((model_id, has_video_input))
+    rate = seedance2_price_per_1k_tokens(model_id, has_video_input, resolution)
     if rate is None:
         return None
 
@@ -1621,7 +1621,7 @@ class ByteDance2TextToVideoNode(IO.ComfyNode):
                 IO.DynamicCombo.Input(
                     "model",
                     options=[
-                        IO.DynamicCombo.Option("Seedance 2.0", _seedance2_text_inputs(["480p", "720p", "1080p"])),
+                        IO.DynamicCombo.Option("Seedance 2.0", _seedance2_text_inputs(["480p", "720p", "1080p", "4k"])),
                         IO.DynamicCombo.Option("Seedance 2.0 Fast", _seedance2_text_inputs(["480p", "720p"])),
                     ],
                     tooltip="Seedance 2.0 for maximum quality; Seedance 2.0 Fast for speed optimization.",
@@ -1660,11 +1660,15 @@ class ByteDance2TextToVideoNode(IO.ComfyNode):
                   $rate480 := 10044;
                   $rate720 := 21600;
                   $rate1080 := 48800;
+                  $rate4k := 195200;
                   $m := widgets.model;
-                  $pricePer1K := $contains($m, "fast") ? 0.008008 : 0.01001;
                   $res := $lookup(widgets, "model.resolution");
                   $dur := $lookup(widgets, "model.duration");
-                  $rate := $res = "1080p" ? $rate1080 :
+                  $pricePer1K := $res = "4k"    ? 0.00572 :
+                                 $res = "1080p" ? 0.011011 :
+                                 $contains($m, "fast") ? 0.008008 : 0.01001;
+                  $rate := $res = "4k"    ? $rate4k :
+                           $res = "1080p" ? $rate1080 :
                            $res = "720p"  ? $rate720 :
                                             $rate480;
                   $cost := $dur * $rate * $pricePer1K / 1000;
@@ -1703,7 +1707,7 @@ class ByteDance2TextToVideoNode(IO.ComfyNode):
             ApiEndpoint(path=f"{BYTEPLUS_SEEDANCE2_TASK_STATUS_ENDPOINT}/{initial_response.id}"),
             response_model=TaskStatusResponse,
             status_extractor=lambda r: r.status,
-            price_extractor=_seedance2_price_extractor(model_id, has_video_input=False),
+            price_extractor=_seedance2_price_extractor(model_id, has_video_input=False, resolution=model["resolution"]),
             poll_interval=9,
         )
         return IO.NodeOutput(await download_url_to_video_output(response.content.video_url))
@@ -1724,7 +1728,7 @@ class ByteDance2FirstLastFrameNode(IO.ComfyNode):
                     options=[
                         IO.DynamicCombo.Option(
                             "Seedance 2.0",
-                            _seedance2_text_inputs(["480p", "720p", "1080p"], default_ratio="adaptive"),
+                            _seedance2_text_inputs(["480p", "720p", "1080p", "4k"], default_ratio="adaptive"),
                         ),
                         IO.DynamicCombo.Option(
                             "Seedance 2.0 Fast",
@@ -1791,11 +1795,15 @@ class ByteDance2FirstLastFrameNode(IO.ComfyNode):
                   $rate480 := 10044;
                   $rate720 := 21600;
                   $rate1080 := 48800;
+                  $rate4k := 195200;
                   $m := widgets.model;
-                  $pricePer1K := $contains($m, "fast") ? 0.008008 : 0.01001;
                   $res := $lookup(widgets, "model.resolution");
                   $dur := $lookup(widgets, "model.duration");
-                  $rate := $res = "1080p" ? $rate1080 :
+                  $pricePer1K := $res = "4k"    ? 0.00572 :
+                                 $res = "1080p" ? 0.011011 :
+                                 $contains($m, "fast") ? 0.008008 : 0.01001;
+                  $rate := $res = "4k"    ? $rate4k :
+                           $res = "1080p" ? $rate1080 :
                            $res = "720p"  ? $rate720 :
                                             $rate480;
                   $cost := $dur * $rate * $pricePer1K / 1000;
@@ -1913,7 +1921,7 @@ class ByteDance2FirstLastFrameNode(IO.ComfyNode):
             ApiEndpoint(path=f"{BYTEPLUS_SEEDANCE2_TASK_STATUS_ENDPOINT}/{initial_response.id}"),
             response_model=TaskStatusResponse,
             status_extractor=lambda r: r.status,
-            price_extractor=_seedance2_price_extractor(model_id, has_video_input=False),
+            price_extractor=_seedance2_price_extractor(model_id, has_video_input=False, resolution=model["resolution"]),
             poll_interval=9,
         )
         return IO.NodeOutput(await download_url_to_video_output(response.content.video_url))
@@ -2010,7 +2018,7 @@ class ByteDance2ReferenceNode(IO.ComfyNode):
                     options=[
                         IO.DynamicCombo.Option(
                             "Seedance 2.0",
-                            _seedance2_reference_inputs(["480p", "720p", "1080p"], default_ratio="adaptive"),
+                            _seedance2_reference_inputs(["480p", "720p", "1080p", "4k"], default_ratio="adaptive"),
                         ),
                         IO.DynamicCombo.Option(
                             "Seedance 2.0 Fast",
@@ -2056,13 +2064,19 @@ class ByteDance2ReferenceNode(IO.ComfyNode):
                   $rate480 := 10044;
                   $rate720 := 21600;
                   $rate1080 := 48800;
+                  $rate4k := 195200;
                   $m := widgets.model;
                   $hasVideo := $lookup(inputGroups, "model.reference_videos") > 0;
-                  $noVideoPricePer1K := $contains($m, "fast") ? 0.008008 : 0.01001;
-                  $videoPricePer1K := $contains($m, "fast") ? 0.004719 : 0.006149;
                   $res := $lookup(widgets, "model.resolution");
                   $dur := $lookup(widgets, "model.duration");
-                  $rate := $res = "1080p" ? $rate1080 :
+                  $noVideoPricePer1K := $res = "4k"    ? 0.00572 :
+                                        $res = "1080p" ? 0.011011 :
+                                        $contains($m, "fast") ? 0.008008 : 0.01001;
+                  $videoPricePer1K := $res = "4k"    ? 0.003432 :
+                                      $res = "1080p" ? 0.006721 :
+                                      $contains($m, "fast") ? 0.004719 : 0.006149;
+                  $rate := $res = "4k"    ? $rate4k :
+                           $res = "1080p" ? $rate1080 :
                            $res = "720p"  ? $rate720 :
                                             $rate480;
                   $noVideoCost := $dur * $rate * $noVideoPricePer1K / 1000;
@@ -2258,7 +2272,9 @@ class ByteDance2ReferenceNode(IO.ComfyNode):
             ApiEndpoint(path=f"{BYTEPLUS_SEEDANCE2_TASK_STATUS_ENDPOINT}/{initial_response.id}"),
             response_model=TaskStatusResponse,
             status_extractor=lambda r: r.status,
-            price_extractor=_seedance2_price_extractor(model_id, has_video_input=has_video_input),
+            price_extractor=_seedance2_price_extractor(
+                model_id, has_video_input=has_video_input, resolution=model["resolution"]
+            ),
             poll_interval=9,
         )
         return IO.NodeOutput(await download_url_to_video_output(response.content.video_url))

comfy_api_nodes/nodes_gemini.py

@@ -5,7 +5,6 @@ See: https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/infer
 
 import base64
 import os
-from enum import Enum
 from fnmatch import fnmatch
 from io import BytesIO
 from typing import Any, Literal
@@ -78,15 +77,6 @@ GEMINI_IMAGE_2_PRICE_BADGE = IO.PriceBadge(
 )
 
 
-class GeminiImageModel(str, Enum):
-    """
-    Gemini Image Model Names allowed by comfy-api
-    """
-
-    gemini_2_5_flash_image_preview = "gemini-2.5-flash-image-preview"
-    gemini_2_5_flash_image = "gemini-2.5-flash-image"
-
-
 async def create_image_parts(
     cls: type[IO.ComfyNode],
     images: Input.Image | list[Input.Image],
@@ -243,21 +233,15 @@ def calculate_tokens_price(response: GeminiGenerateContentResponse) -> float | N
     if not response.modelVersion:
         return None
     # Define prices (Cost per 1,000,000 tokens), see https://cloud.google.com/vertex-ai/generative-ai/pricing
-    if response.modelVersion in ("gemini-2.5-pro-preview-05-06", "gemini-2.5-pro"):
+    if response.modelVersion == "gemini-2.5-pro":
         input_tokens_price = 1.25
         output_text_tokens_price = 10.0
         output_image_tokens_price = 0.0
-    elif response.modelVersion in (
-        "gemini-2.5-flash-preview-04-17",
-        "gemini-2.5-flash",
-    ):
+    elif response.modelVersion == "gemini-2.5-flash":
         input_tokens_price = 0.30
         output_text_tokens_price = 2.50
         output_image_tokens_price = 0.0
-    elif response.modelVersion in (
-        "gemini-2.5-flash-image-preview",
-        "gemini-2.5-flash-image",
-    ):
+    elif response.modelVersion == "gemini-2.5-flash-image":
         input_tokens_price = 0.30
         output_text_tokens_price = 2.50
         output_image_tokens_price = 30.0
@@ -455,8 +439,6 @@ class GeminiNode(IO.ComfyNode):
                 IO.Combo.Input(
                     "model",
                     options=[
-                        "gemini-2.5-pro-preview-05-06",
-                        "gemini-2.5-flash-preview-04-17",
                         "gemini-2.5-pro",
                         "gemini-2.5-flash",
                         "gemini-3-pro-preview",
@@ -904,8 +886,7 @@ class GeminiImage(IO.ComfyNode):
                 ),
                 IO.Combo.Input(
                     "model",
-                    options=GeminiImageModel,
-                    default=GeminiImageModel.gemini_2_5_flash_image,
+                    options=["gemini-2.5-flash-image"],
                     tooltip="The Gemini model to use for generating responses.",
                 ),
                 IO.Int.Input(

comfy_api_nodes/nodes_grok.py

@@ -30,7 +30,7 @@ from comfy_api_nodes.util import (
 
 
 _GROK_VIDEO_MODEL_API_IDS = {
-    "grok-imagine-video-1.5": "grok-imagine-video-1.5-preview",
+    "grok-imagine-video-1.5": "grok-imagine-video-1.5",
 }
 
 
@@ -521,8 +521,8 @@ class GrokVideoNode(IO.ComfyNode):
                 ),
                 IO.Combo.Input(
                     "resolution",
-                    options=["480p", "720p"],
-                    tooltip="The resolution of the output video.",
+                    options=["480p", "720p", "1080p"],
+                    tooltip="The resolution of the output video. 1080p is only available for grok-imagine-video-1.5.",
                 ),
                 IO.Combo.Input(
                     "aspect_ratio",
@@ -570,11 +570,12 @@ class GrokVideoNode(IO.ComfyNode):
                 (
                   $is15 := $contains(widgets.model, "1.5");
                   $rate := $is15
-                    ? (widgets.resolution = "720p" ? 0.2002 : 0.1144)
+                    ? (widgets.resolution = "1080p" ? 0.25 : (widgets.resolution = "720p" ? 0.14 : 0.08))
                     : (widgets.resolution = "720p" ? 0.07 : 0.05);
-                  $imgCost := $is15 ? 0.0143 : 0.002;
+                  $imgCost := $is15 ? 0.01 : 0.002;
                   $base := $rate * widgets.duration;
-                  {"type":"usd","usd": inputs.image.connected ? $base + $imgCost : $base}
+                  $total := inputs.image.connected ? $base + $imgCost : $base;
+                  {"type":"usd","usd": $is15 ? $total * 1.43 : $total}
                 )
                 """,
             ),
@@ -593,6 +594,8 @@ class GrokVideoNode(IO.ComfyNode):
     ) -> IO.NodeOutput:
         if image is None and model == "grok-imagine-video-1.5":
             raise ValueError(f"The '{model}' model requires an input image; connect one to the 'image' input.")
+        if resolution == "1080p" and model != "grok-imagine-video-1.5":
+            raise ValueError(f"1080p resolution is only available for grok-imagine-video-1.5, not '{model}'.")
         image_url = None
         if image is not None:
             if get_number_of_images(image) != 1:

comfy_api_nodes/nodes_wan.py

@@ -48,10 +48,13 @@ from comfy_api_nodes.util import (
     upload_image_to_comfyapi,
     upload_video_to_comfyapi,
     validate_audio_duration,
+    validate_image_aspect_ratio,
+    validate_image_dimensions,
     validate_string,
     validate_video_duration,
 )
 
+
 RES_IN_PARENS = re.compile(r"\((\d+)\s*[x×]\s*(\d+)\)")
 
 
@@ -1657,6 +1660,44 @@ class HappyHorseTextToVideoApi(IO.ComfyNode):
                 IO.DynamicCombo.Input(
                     "model",
                     options=[
+                        IO.DynamicCombo.Option(
+                            "happyhorse-1.1-t2v",
+                            [
+                                IO.String.Input(
+                                    "prompt",
+                                    multiline=True,
+                                    default="",
+                                    tooltip="Prompt describing the elements and visual features. "
+                                    "Supports English and Chinese.",
+                                ),
+                                IO.Combo.Input(
+                                    "resolution",
+                                    options=["720P", "1080P"],
+                                ),
+                                IO.Combo.Input(
+                                    "ratio",
+                                    options=[
+                                        "16:9",
+                                        "9:16",
+                                        "1:1",
+                                        "4:3",
+                                        "3:4",
+                                        "21:9",
+                                        "9:21",
+                                        "5:4",
+                                        "4:5",
+                                    ],
+                                ),
+                                IO.Int.Input(
+                                    "duration",
+                                    default=5,
+                                    min=3,
+                                    max=15,
+                                    step=1,
+                                    display_mode=IO.NumberDisplay.number,
+                                ),
+                            ],
+                        ),
                         IO.DynamicCombo.Option(
                             "happyhorse-1.0-t2v",
                             [
@@ -1719,7 +1760,9 @@ class HappyHorseTextToVideoApi(IO.ComfyNode):
                 (
                   $res := $lookup(widgets, "model.resolution");
                   $dur := $lookup(widgets, "model.duration");
-                  $ppsTable := { "720p": 0.14, "1080p": 0.24 };
+                  $ppsTable := $contains(widgets.model, "1.1")
+                    ? { "720p": 0.2002, "1080p": 0.2574 }
+                    : { "720p": 0.14, "1080p": 0.24 };
                   $pps := $lookup($ppsTable, $res);
                   { "type": "usd", "usd": $pps * $dur }
                 )
@@ -1781,6 +1824,30 @@ class HappyHorseImageToVideoApi(IO.ComfyNode):
                 IO.DynamicCombo.Input(
                     "model",
                     options=[
+                        IO.DynamicCombo.Option(
+                            "happyhorse-1.1-i2v",
+                            [
+                                IO.String.Input(
+                                    "prompt",
+                                    multiline=True,
+                                    default="",
+                                    tooltip="Prompt describing the elements and visual features. "
+                                    "Supports English and Chinese.",
+                                ),
+                                IO.Combo.Input(
+                                    "resolution",
+                                    options=["720P", "1080P"],
+                                ),
+                                IO.Int.Input(
+                                    "duration",
+                                    default=5,
+                                    min=3,
+                                    max=15,
+                                    step=1,
+                                    display_mode=IO.NumberDisplay.number,
+                                ),
+                            ],
+                        ),
                         IO.DynamicCombo.Option(
                             "happyhorse-1.0-i2v",
                             [
@@ -1843,7 +1910,9 @@ class HappyHorseImageToVideoApi(IO.ComfyNode):
                 (
                   $res := $lookup(widgets, "model.resolution");
                   $dur := $lookup(widgets, "model.duration");
-                  $ppsTable := { "720p": 0.14, "1080p": 0.24 };
+                  $ppsTable := $contains(widgets.model, "1.1")
+                    ? { "720p": 0.2002, "1080p": 0.2574 }
+                    : { "720p": 0.14, "1080p": 0.24 };
                   $pps := $lookup($ppsTable, $res);
                   { "type": "usd", "usd": $pps * $dur }
                 )
@@ -1859,6 +1928,8 @@ class HappyHorseImageToVideoApi(IO.ComfyNode):
         seed: int,
         watermark: bool,
     ):
+        validate_image_dimensions(first_frame, min_width=300, min_height=300)
+        validate_image_aspect_ratio(first_frame, (1, 2.5), (2.5, 1), strict=False)
         media = [
             Wan27MediaItem(
                 type="first_frame",
@@ -2053,6 +2124,62 @@ class HappyHorseReferenceVideoApi(IO.ComfyNode):
                 IO.DynamicCombo.Input(
                     "model",
                     options=[
+                        IO.DynamicCombo.Option(
+                            "happyhorse-1.1-r2v",
+                            [
+                                IO.String.Input(
+                                    "prompt",
+                                    multiline=True,
+                                    default="",
+                                    tooltip="Prompt describing the video. Use identifiers such as 'character1' and "
+                                    "'character2' to refer to the reference characters.",
+                                ),
+                                IO.Combo.Input(
+                                    "resolution",
+                                    options=["720P", "1080P"],
+                                ),
+                                IO.Combo.Input(
+                                    "ratio",
+                                    options=[
+                                        "16:9",
+                                        "9:16",
+                                        "1:1",
+                                        "4:3",
+                                        "3:4",
+                                        "21:9",
+                                        "9:21",
+                                        "5:4",
+                                        "4:5",
+                                    ],
+                                ),
+                                IO.Int.Input(
+                                    "duration",
+                                    default=5,
+                                    min=3,
+                                    max=15,
+                                    step=1,
+                                    display_mode=IO.NumberDisplay.number,
+                                ),
+                                IO.Autogrow.Input(
+                                    "reference_images",
+                                    template=IO.Autogrow.TemplateNames(
+                                        IO.Image.Input("reference_image"),
+                                        names=[
+                                            "image1",
+                                            "image2",
+                                            "image3",
+                                            "image4",
+                                            "image5",
+                                            "image6",
+                                            "image7",
+                                            "image8",
+                                            "image9",
+                                        ],
+                                        min=1,
+                                    ),
+                                ),
+                            ],
+                        ),
                         IO.DynamicCombo.Option(
                             "happyhorse-1.0-r2v",
                             [
@@ -2133,7 +2260,9 @@ class HappyHorseReferenceVideoApi(IO.ComfyNode):
                 (
                   $res := $lookup(widgets, "model.resolution");
                   $dur := $lookup(widgets, "model.duration");
-                  $ppsTable := { "720p": 0.14, "1080p": 0.24 };
+                  $ppsTable := $contains(widgets.model, "1.1")
+                    ? { "720p": 0.2002, "1080p": 0.2574 }
+                    : { "720p": 0.14, "1080p": 0.24 };
                   $pps := $lookup($ppsTable, $res);
                   { "type": "usd", "usd": $pps * $dur }
                 )
@@ -2149,8 +2278,11 @@ class HappyHorseReferenceVideoApi(IO.ComfyNode):
         watermark: bool,
     ):
         validate_string(model["prompt"], strip_whitespace=False, min_length=1)
-        media = []
         reference_images = model.get("reference_images", {})
+        for key in reference_images:
+            validate_image_dimensions(reference_images[key], min_width=400, min_height=400)
+            validate_image_aspect_ratio(reference_images[key], (1, 2.5), (2.5, 1), strict=False)
+        media = []
         for key in reference_images:
             media.append(
                 Wan27MediaItem(
@@ -2159,7 +2291,7 @@ class HappyHorseReferenceVideoApi(IO.ComfyNode):
                 )
             )
         if not media:
-            raise ValueError("At least one reference reference image must be provided.")
+            raise ValueError("At least one reference image must be provided.")
 
         initial_response = await sync_op(
             cls,

comfy_execution/jobs.py

@@ -4,11 +4,22 @@ Provides normalization and helper functions for job status tracking.
 """
 
 import uuid
-from typing import Optional
+from typing import Callable, Optional
 
 from comfy_api.internal import prune_dict
 
 
+# Result of classifying a job for cancellation.
+# 'running'  -> job is currently executing (interrupt it)
+# 'pending'  -> job is queued but not started (dequeue it)
+# 'terminal' -> job already finished (present in history); cancel is a no-op
+# 'unknown'  -> job id is not present anywhere
+CANCEL_RUNNING = 'running'
+CANCEL_PENDING = 'pending'
+CANCEL_TERMINAL = 'terminal'
+CANCEL_UNKNOWN = 'unknown'
+
+
 class JobStatus:
     """Job status constants."""
     PENDING = 'pending'
@@ -407,3 +418,71 @@ def get_all_jobs(
         jobs = jobs[:limit]
 
     return (jobs, total_count)
+
+
+def classify_job_for_cancel(prompt_id: str, running: list, queued: list, history: dict) -> str:
+    """Classify a job id for cancellation.
+
+    Returns one of CANCEL_RUNNING, CANCEL_PENDING, CANCEL_TERMINAL, CANCEL_UNKNOWN.
+
+    Queue items are tuples whose second element (index 1) is the prompt_id.
+    History is a dict keyed by prompt_id, so a job present there has already
+    finished and cancelling it is a no-op.
+    """
+    for item in running:
+        if item[1] == prompt_id:
+            return CANCEL_RUNNING
+    for item in queued:
+        if item[1] == prompt_id:
+            return CANCEL_PENDING
+    if prompt_id in history:
+        return CANCEL_TERMINAL
+    return CANCEL_UNKNOWN
+
+
+def cancel_job(
+    prompt_id: str,
+    running: list,
+    queued: list,
+    history: dict,
+    interrupt: Callable[[str], bool],
+    dequeue: Callable[[str], bool],
+) -> str:
+    """Cancel a single job by id, regardless of state.
+
+    Maps the cancel onto the runtime's existing mechanics:
+      - a running job is interrupted via ``interrupt``
+      - a pending job is removed from the queue via ``dequeue``
+      - a job that already finished (terminal) is a no-op
+      - an unknown id is a no-op (callers that need fail-fast behaviour should
+        validate ids up front with ``classify_job_for_cancel``)
+
+    Both ``interrupt`` and ``dequeue`` take the prompt id and return whether
+    they acted on a job that was *actually* in that state, so the value returned
+    here reflects what truly happened rather than the (possibly stale)
+    classification. This matters around the narrow TOCTOU windows where a job
+    changes state between the caller's snapshot and the action:
+
+      - a job classified RUNNING may have finished before ``interrupt`` fires:
+        ``interrupt`` returns False and this returns CANCEL_UNKNOWN (no-op).
+      - a job classified PENDING may have started executing before ``dequeue``
+        fires: ``dequeue`` returns False, ``interrupt`` then catches the now-
+        running job and this returns CANCEL_RUNNING. If it had simply finished
+        instead, both return False and this returns CANCEL_UNKNOWN.
+
+    ``interrupt`` must be atomic — interrupt the job only if it is still the one
+    running — so a cancel can never land on an unrelated prompt that started in
+    the meantime (see ``execution.PromptQueue.interrupt_if_running``).
+    """
+    classification = classify_job_for_cancel(prompt_id, running, queued, history)
+    if classification == CANCEL_RUNNING:
+        return CANCEL_RUNNING if interrupt(prompt_id) else CANCEL_UNKNOWN
+    if classification == CANCEL_PENDING:
+        if dequeue(prompt_id):
+            return CANCEL_PENDING
+        # Left the pending queue between classification and dequeue: if it
+        # started executing, interrupt the now-running job; otherwise it has
+        # already finished and the cancel is a genuine no-op.
+        return CANCEL_RUNNING if interrupt(prompt_id) else CANCEL_UNKNOWN
+    # CANCEL_TERMINAL and CANCEL_UNKNOWN are intentional no-ops.
+    return classification

comfy_extras/nodes_audio.py

@@ -158,7 +158,7 @@ class SaveAudio(IO.ComfyNode):
         return IO.Schema(
             node_id="SaveAudio",
             search_aliases=["export flac"],
-            display_name="Save Audio (FLAC) (Deprecated)",
+            display_name="Save Audio (FLAC) (DEPRECATED)",
             category="audio",
             essentials_category="Audio",
             inputs=[
@@ -166,8 +166,9 @@ class SaveAudio(IO.ComfyNode):
                 IO.String.Input("filename_prefix", default="audio/ComfyUI"),
             ],
             hidden=[IO.Hidden.prompt, IO.Hidden.extra_pnginfo],
-            is_output_node=True,
             is_deprecated=True,
+            is_output_node=True,
+            outputs=[IO.Audio.Output("audio")]
         )
 
     @classmethod
@@ -175,11 +176,10 @@ class SaveAudio(IO.ComfyNode):
         if audio is None:
             raise ValueError("SaveAudio: input audio is None (source video may have no audio track).")
         return IO.NodeOutput(
+            audio,
             ui=UI.AudioSaveHelper.get_save_audio_ui(audio, filename_prefix=filename_prefix, cls=cls, format=format)
         )
 
-    save_flac = execute  # TODO: remove
-
 
 class SaveAudioMP3(IO.ComfyNode):
     @classmethod
@@ -187,7 +187,7 @@ class SaveAudioMP3(IO.ComfyNode):
         return IO.Schema(
             node_id="SaveAudioMP3",
             search_aliases=["export mp3"],
-            display_name="Save Audio (MP3) (Deprecated)",
+            display_name="Save Audio (MP3) (DEPRECATED)",
             category="audio",
             essentials_category="Audio",
             inputs=[
@@ -196,8 +196,9 @@ class SaveAudioMP3(IO.ComfyNode):
                 IO.Combo.Input("quality", options=["V0", "128k", "320k"], default="V0"),
             ],
             hidden=[IO.Hidden.prompt, IO.Hidden.extra_pnginfo],
-            is_output_node=True,
             is_deprecated=True,
+            is_output_node=True,
+            outputs=[IO.Audio.Output("audio")]
         )
 
     @classmethod
@@ -205,13 +206,12 @@ class SaveAudioMP3(IO.ComfyNode):
         if audio is None:
             raise ValueError("SaveAudioMP3: input audio is None (source video may have no audio track).")
         return IO.NodeOutput(
+            audio,
             ui=UI.AudioSaveHelper.get_save_audio_ui(
                 audio, filename_prefix=filename_prefix, cls=cls, format=format, quality=quality
             )
         )
 
-    save_mp3 = execute  # TODO: remove
-
 
 class SaveAudioOpus(IO.ComfyNode):
     @classmethod
@@ -219,7 +219,7 @@ class SaveAudioOpus(IO.ComfyNode):
         return IO.Schema(
             node_id="SaveAudioOpus",
             search_aliases=["export opus"],
-            display_name="Save Audio (Opus) (Deprecated)",
+            display_name="Save Audio (Opus) (DEPRECATED)",
             category="audio",
             inputs=[
                 IO.Audio.Input("audio"),
@@ -227,8 +227,9 @@ class SaveAudioOpus(IO.ComfyNode):
                 IO.Combo.Input("quality", options=["64k", "96k", "128k", "192k", "320k"], default="128k"),
             ],
             hidden=[IO.Hidden.prompt, IO.Hidden.extra_pnginfo],
-            is_output_node=True,
             is_deprecated=True,
+            is_output_node=True,
+            outputs=[IO.Audio.Output("audio")]
         )
 
     @classmethod
@@ -236,13 +237,12 @@ class SaveAudioOpus(IO.ComfyNode):
         if audio is None:
             raise ValueError("SaveAudioOpus: input audio is None (source video may have no audio track).")
         return IO.NodeOutput(
+            audio,
             ui=UI.AudioSaveHelper.get_save_audio_ui(
                 audio, filename_prefix=filename_prefix, cls=cls, format=format, quality=quality
             )
         )
 
-    save_opus = execute  # TODO: remove
-
 
 class SaveAudioAdvanced(IO.ComfyNode):
     @classmethod
@@ -258,10 +258,7 @@ class SaveAudioAdvanced(IO.ComfyNode):
                 IO.String.Input(
                     "filename_prefix",
                     default="audio/ComfyUI",
-                    tooltip=(
-                        "The prefix for the file to save. May include formatting tokens "
-                        "such as %date:yyyy-MM-dd%."
-                    ),
+                    tooltip=("The prefix for the file to save. May include formatting tokens such as %date:yyyy-MM-dd%."),
                 ),
                 IO.DynamicCombo.Input(
                     "format",
@@ -279,6 +276,7 @@ class SaveAudioAdvanced(IO.ComfyNode):
             ],
             hidden=[IO.Hidden.prompt, IO.Hidden.extra_pnginfo],
             is_output_node=True,
+            outputs=[IO.Audio.Output("audio")],
         )
 
     @classmethod
@@ -289,7 +287,7 @@ class SaveAudioAdvanced(IO.ComfyNode):
             ui=UI.AudioSaveHelper.get_save_audio_ui(audio, filename_prefix=filename_prefix, cls=cls, format=file_format, quality=quality)
         else:
             ui=UI.AudioSaveHelper.get_save_audio_ui(audio, filename_prefix=filename_prefix, cls=cls, format=file_format)
-        return IO.NodeOutput(ui=ui)
+        return IO.NodeOutput(audio, ui=ui)
 
 
 class PreviewAudio(IO.ComfyNode):
@@ -305,13 +303,14 @@ class PreviewAudio(IO.ComfyNode):
             ],
             hidden=[IO.Hidden.prompt, IO.Hidden.extra_pnginfo],
             is_output_node=True,
+            outputs=[IO.Audio.Output("audio")]
         )
 
     @classmethod
     def execute(cls, audio) -> IO.NodeOutput:
         if audio is None:
             raise ValueError("PreviewAudio: input audio is None (source video may have no audio track).")
-        return IO.NodeOutput(ui=UI.PreviewAudio(audio, cls=cls))
+        return IO.NodeOutput(audio, ui=UI.PreviewAudio(audio, cls=cls))
 
     save_flac = execute  # TODO: remove
 

comfy_extras/nodes_context_windows.py

@@ -13,21 +13,22 @@ class ContextWindowsManualNode(io.ComfyNode):
             description="Manually set context windows.",
             inputs=[
                 io.Model.Input("model", tooltip="The model to apply context windows to during sampling."),
-                io.Int.Input("context_length", min=1, default=16, tooltip="The length of the context window.", advanced=True),
-                io.Int.Input("context_overlap", min=0, default=4, tooltip="The overlap of the context window.", advanced=True),
+                io.Int.Input("context_length", min=1, default=16, tooltip="The length of the context window."),
+                io.Int.Input("context_overlap", min=0, default=4, tooltip="The overlap of the context window."),
                 io.Combo.Input("context_schedule", options=[
                     comfy.context_windows.ContextSchedules.STATIC_STANDARD,
                     comfy.context_windows.ContextSchedules.UNIFORM_STANDARD,
                     comfy.context_windows.ContextSchedules.UNIFORM_LOOPED,
                     comfy.context_windows.ContextSchedules.BATCHED,
-                    ], tooltip="The stride of the context window."),
-                io.Int.Input("context_stride", min=1, default=1, tooltip="The stride of the context window; only applicable to uniform schedules.", advanced=True),
+                    ], default=comfy.context_windows.ContextSchedules.STATIC_STANDARD, tooltip="Step-dependent scheduling algorithm for context windows."),
+                io.Int.Input("context_stride", min=1, default=1, tooltip="The stride of the context window; only applicable to uniform schedules."),
                 io.Boolean.Input("closed_loop", default=False, tooltip="Whether to close the context window loop; only applicable to looped schedules."),
                 io.Combo.Input("fuse_method", options=comfy.context_windows.ContextFuseMethods.LIST_STATIC, default=comfy.context_windows.ContextFuseMethods.PYRAMID, tooltip="The method to use to fuse the context windows."),
                 io.Int.Input("dim", min=0, max=5, default=0, tooltip="The dimension to apply the context windows to."),
                 io.Boolean.Input("freenoise", default=False, tooltip="Whether to apply FreeNoise noise shuffling, improves window blending."),
-                io.String.Input("cond_retain_index_list", default="", tooltip="List of latent indices to retain in the conditioning tensors for each window, for example setting this to '0' will use the initial start image for each window."),
+                io.String.Input("cond_retain_index_list", default="", tooltip="List of latent indices to retain in the conditioning tensors for each window. For concat-style I2V models (e.g. Wan I2V, HunyuanVideo I2V, Cosmos I2V, SVD) the encoded start image lives in the c_concat conditioning channels; setting this to '0' will retain that start image content at sub-pos 0 of every window."),
                 io.Boolean.Input("split_conds_to_windows", default=False, tooltip="Whether to split multiple conditionings (created by ConditionCombine) to each window based on region index."),
+                io.String.Input("latent_retain_index_list", default="", tooltip="List of latent indices to retain in the noise latent itself for each window. Use for workflows where reference content (e.g. a start image) lives directly in the noise latent rather than in separate conditioning channels (e.g. inplace-style I2V like LTXV, AnimateDiff). Independent of cond_retain_index_list."),
                 io.Boolean.Input("causal_window_fix", default=True, tooltip="Whether to add a causal fix frame to non-0-indexed context windows."),
             ],
             outputs=[
@@ -38,7 +39,7 @@ class ContextWindowsManualNode(io.ComfyNode):
 
     @classmethod
     def execute(cls, model: io.Model.Type, context_length: int, context_overlap: int, context_schedule: str, context_stride: int, closed_loop: bool, fuse_method: str, dim: int, freenoise: bool,
-                cond_retain_index_list: list[int]=[], split_conds_to_windows: bool=False, causal_window_fix: bool=True) -> io.Model:
+                cond_retain_index_list: list[int]=[], split_conds_to_windows: bool=False, latent_retain_index_list: list[int]=[], causal_window_fix: bool=True) -> io.Model:
         model = model.clone()
         model.model_options["context_handler"] = comfy.context_windows.IndexListContextHandler(
             context_schedule=comfy.context_windows.get_matching_context_schedule(context_schedule),
@@ -51,6 +52,7 @@ class ContextWindowsManualNode(io.ComfyNode):
             freenoise=freenoise,
             cond_retain_index_list=cond_retain_index_list,
             split_conds_to_windows=split_conds_to_windows,
+            latent_retain_index_list=latent_retain_index_list,
             causal_window_fix=causal_window_fix,
         )
         # make memory usage calculation only take into account the context window latents
@@ -65,33 +67,71 @@ class WanContextWindowsManualNode(ContextWindowsManualNode):
         schema = super().define_schema()
         schema.node_id = "WanContextWindowsManual"
         schema.display_name = "WAN Context Windows (Manual)"
-        schema.description = "Manually set context windows for WAN-like models (dim=2)."
+        schema.display_name = "Wan Context Windows"
+        schema.description = "Set context windows for Wan-like models."
         schema.category="model/patch/wan"
         schema.inputs = [
             io.Model.Input("model", tooltip="The model to apply context windows to during sampling."),
-                io.Int.Input("context_length", min=1, max=nodes.MAX_RESOLUTION, step=4, default=81, tooltip="The length of the context window.", advanced=True),
-                io.Int.Input("context_overlap", min=0, default=30, tooltip="The overlap of the context window.", advanced=True),
+                io.Int.Input("context_length", min=1, max=nodes.MAX_RESOLUTION, step=4, default=81, tooltip="The length of the context window in real frames. Must be 4*n + 1."),
+                io.Int.Input("context_overlap", min=0, default=30, tooltip="The overlap of the context window in real frames."),
                 io.Combo.Input("context_schedule", options=[
                     comfy.context_windows.ContextSchedules.STATIC_STANDARD,
                     comfy.context_windows.ContextSchedules.UNIFORM_STANDARD,
                     comfy.context_windows.ContextSchedules.UNIFORM_LOOPED,
                     comfy.context_windows.ContextSchedules.BATCHED,
-                    ], tooltip="The stride of the context window."),
+                    ], default=comfy.context_windows.ContextSchedules.UNIFORM_STANDARD, tooltip="Step-dependent scheduling algorithm for context windows."),
                 io.Int.Input("context_stride", min=1, default=1, tooltip="The stride of the context window; only applicable to uniform schedules.", advanced=True),
-                io.Boolean.Input("closed_loop", default=False, tooltip="Whether to close the context window loop; only applicable to looped schedules."),
+                io.Boolean.Input("closed_loop", default=False, tooltip="Whether to close the context window loop; only applicable to looped schedules.", advanced=True),
                 io.Combo.Input("fuse_method", options=comfy.context_windows.ContextFuseMethods.LIST_STATIC, default=comfy.context_windows.ContextFuseMethods.PYRAMID, tooltip="The method to use to fuse the context windows."),
-                io.Boolean.Input("freenoise", default=False, tooltip="Whether to apply FreeNoise noise shuffling, improves window blending."),
-                #io.String.Input("cond_retain_index_list", default="", tooltip="List of latent indices to retain in the conditioning tensors for each window, for example setting this to '0' will use the initial start image for each window."),
-                #io.Boolean.Input("split_conds_to_windows", default=False, tooltip="Whether to split multiple conditionings (created by ConditionCombine) to each window based on region index."),
+                io.Boolean.Input("freenoise", default=True, tooltip="Whether to apply FreeNoise noise shuffling, improves window blending.", advanced=True),
+                io.Boolean.Input("retain_first_frame", default=False, tooltip="Retain the first I2V frame in every context window (may help retain initial reference)."),
+                io.Boolean.Input("split_conds_to_windows", default=False, tooltip="Whether to split multiple conditionings (created by ConditionCombine) to each window based on region index.", advanced=True),
         ]
         return schema
 
     @classmethod
     def execute(cls, model: io.Model.Type, context_length: int, context_overlap: int, context_schedule: str, context_stride: int, closed_loop: bool, fuse_method: str, freenoise: bool,
-                cond_retain_index_list: list[int]=[], split_conds_to_windows: bool=False) -> io.Model:
-        context_length = max(((context_length - 1) // 4) + 1, 1)  # at least length 1
-        context_overlap = max(((context_overlap - 1) // 4) + 1, 0)  # at least overlap 0
-        return super().execute(model, context_length, context_overlap, context_schedule, context_stride, closed_loop, fuse_method, dim=2, freenoise=freenoise, cond_retain_index_list=cond_retain_index_list, split_conds_to_windows=split_conds_to_windows)
+                retain_first_frame: bool=False, split_conds_to_windows: bool=False) -> io.Model:
+        context_length = max(((context_length - 1) // 4) + 1, 1) # at least length 1
+        context_overlap = max(context_overlap // 4, 0)  # at least overlap 0
+        retain_index_list = "0" if retain_first_frame else ""
+        return super().execute(model, context_length, context_overlap, context_schedule, context_stride, closed_loop, fuse_method, dim=2, freenoise=freenoise, cond_retain_index_list=retain_index_list, split_conds_to_windows=split_conds_to_windows)
+
+
+class LTXVContextWindowsNode(ContextWindowsManualNode):
+    @classmethod
+    def define_schema(cls) -> io.Schema:
+        schema = super().define_schema()
+        schema.node_id = "LTXVContextWindows"
+        schema.display_name = "LTXV Context Windows"
+        schema.description = "Set context windows for LTXV-like models."
+        schema.inputs = [
+            io.Model.Input("model", tooltip="The model to apply context windows to during sampling."),
+            io.Int.Input("context_length", min=1, max=nodes.MAX_RESOLUTION, step=8, default=145, tooltip="The length of the context window in real frames. Must be 8*n + 1."),
+            io.Int.Input("context_overlap", min=0, step=8, default=40, tooltip="The overlap of the context window in real frames."),
+            io.Combo.Input("context_schedule", options=[
+                comfy.context_windows.ContextSchedules.STATIC_STANDARD,
+                comfy.context_windows.ContextSchedules.UNIFORM_STANDARD,
+                comfy.context_windows.ContextSchedules.UNIFORM_LOOPED,
+                comfy.context_windows.ContextSchedules.BATCHED,
+                ], default=comfy.context_windows.ContextSchedules.UNIFORM_STANDARD, tooltip="Step-dependent scheduling algorithm for context windows."),
+            io.Int.Input("context_stride", min=1, default=1, tooltip="The stride of the context window; only applicable to uniform schedules.", advanced=True),
+            io.Boolean.Input("closed_loop", default=False, tooltip="Whether to close the context window loop; only applicable to looped schedules.", advanced=True),
+            io.Combo.Input("fuse_method", options=comfy.context_windows.ContextFuseMethods.LIST_STATIC, default=comfy.context_windows.ContextFuseMethods.PYRAMID, tooltip="The method to use to fuse the context windows."),
+            io.Boolean.Input("freenoise", default=True, tooltip="Whether to apply FreeNoise noise shuffling, improves window blending.", advanced=True),
+            io.Boolean.Input("retain_first_frame", default=False, tooltip="Retain the first latent frame in every context window (may help retain initial reference)."),
+            io.Boolean.Input("split_conds_to_windows", default=False, tooltip="Whether to split multiple conditionings (created by ConditionCombine) to each window based on region index.", advanced=True),
+        ]
+        return schema
+
+    @classmethod
+    def execute(cls, model: io.Model.Type, context_length: int, context_overlap: int, context_schedule: str, fuse_method: str, freenoise: bool,
+                retain_first_frame: bool=False, split_conds_to_windows: bool=False, context_stride: int=1, closed_loop: bool=False) -> io.Model:
+        context_length = max(((context_length - 1) // 8) + 1, 1)  # at least length 1
+        context_overlap = max(context_overlap // 8, 0)  # at least overlap 0
+        retain_index_list = "0" if retain_first_frame else ""
+        return super().execute(model, context_length, context_overlap, context_schedule, context_stride, closed_loop, fuse_method, dim=2, freenoise=freenoise,
+                               cond_retain_index_list=retain_index_list, latent_retain_index_list=retain_index_list, split_conds_to_windows=split_conds_to_windows)
 
 
 class ContextWindowsExtension(ComfyExtension):
@@ -99,6 +139,7 @@ class ContextWindowsExtension(ComfyExtension):
         return [
             ContextWindowsManualNode,
             WanContextWindowsManualNode,
+            LTXVContextWindowsNode,
         ]
 
 def comfy_entrypoint():

comfy_extras/nodes_frame_interpolation.py

@@ -77,7 +77,7 @@ class FrameInterpolate(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="FrameInterpolate",
-            display_name="Frame Interpolate",
+            display_name="Run Frame Interpolation Model",
             category="video",
             search_aliases=["rife", "film", "frame interpolation", "slow motion", "interpolate frames", "vfi"],
             inputs=[

comfy_extras/nodes_images.py

@@ -214,11 +214,13 @@ class SaveAnimatedWEBP(IO.ComfyNode):
             ],
             hidden=[IO.Hidden.prompt, IO.Hidden.extra_pnginfo],
             is_output_node=True,
+            outputs=[IO.Image.Output(display_name="images")]
         )
 
     @classmethod
     def execute(cls, images, fps, filename_prefix, lossless, quality, method, num_frames=0) -> IO.NodeOutput:
         return IO.NodeOutput(
+            images,
             ui=UI.ImageSaveHelper.get_save_animated_webp_ui(
                 images=images,
                 filename_prefix=filename_prefix,
@@ -230,8 +232,6 @@ class SaveAnimatedWEBP(IO.ComfyNode):
             )
         )
 
-    save_images = execute  # TODO: remove
-
 
 class SaveAnimatedPNG(IO.ComfyNode):
 
@@ -249,11 +249,13 @@ class SaveAnimatedPNG(IO.ComfyNode):
             ],
             hidden=[IO.Hidden.prompt, IO.Hidden.extra_pnginfo],
             is_output_node=True,
+            outputs=[IO.Image.Output(display_name="images")]
         )
 
     @classmethod
     def execute(cls, images, fps, compress_level, filename_prefix="ComfyUI") -> IO.NodeOutput:
         return IO.NodeOutput(
+            images,
             ui=UI.ImageSaveHelper.get_save_animated_png_ui(
                 images=images,
                 filename_prefix=filename_prefix,
@@ -263,8 +265,6 @@ class SaveAnimatedPNG(IO.ComfyNode):
             )
         )
 
-    save_images = execute  # TODO: remove
-
 
 class ImageStitch(IO.ComfyNode):
     """Upstreamed from https://github.com/kijai/ComfyUI-KJNodes"""
@@ -513,6 +513,7 @@ class SaveSVGNode(IO.ComfyNode):
             ],
             hidden=[IO.Hidden.prompt, IO.Hidden.extra_pnginfo],
             is_output_node=True,
+            outputs=[IO.SVG.Output("svg")],
         )
 
     @classmethod
@@ -562,9 +563,7 @@ class SaveSVGNode(IO.ComfyNode):
 
             results.append(UI.SavedResult(filename=file, subfolder=subfolder, type=IO.FolderType.output))
             counter += 1
-        return IO.NodeOutput(ui={"images": results})
-
-    save_svg = execute  # TODO: remove
+        return IO.NodeOutput(svg, ui={"images": results})
 
 
 class GetImageSize(IO.ComfyNode):
@@ -1157,40 +1156,27 @@ class SaveImageAdvanced(IO.ComfyNode):
                 IO.String.Input(
                     "filename_prefix",
                     default="ComfyUI",
-                    tooltip=(
-                        "The prefix for the file to save. May include formatting tokens "
-                        "such as %date:yyyy-MM-dd% or %Empty Latent Image.width%."
-                    ),
+                    tooltip=("The prefix for the file to save. May include formatting tokens such as %date:yyyy-MM-dd% or %Empty Latent Image.width%."),
                 ),
                 IO.DynamicCombo.Input(
                     "format",
                     options=[
                         IO.DynamicCombo.Option("png", [
-                            IO.Combo.Input("bit_depth", options=["8-bit", "16-bit"],
-                                           default="8-bit", advanced=True),
-                            IO.Combo.Input("input_color_space", options=["sRGB"],
-                                           default="sRGB", advanced=True),
+                            IO.Combo.Input("bit_depth", options=["8-bit", "16-bit"], default="8-bit", advanced=True),
+                            IO.Combo.Input("input_color_space", options=["sRGB"], default="sRGB", advanced=True),
                         ]),
                         IO.DynamicCombo.Option("exr", [
-                            IO.Combo.Input("bit_depth", options=["32-bit float"],
-                                           default="32-bit float", advanced=True),
+                            IO.Combo.Input("bit_depth", options=["32-bit float"], default="32-bit float", advanced=True),
                             IO.Combo.Input(
                                 "input_color_space",
                                 options=["sRGB", "HDR", "linear"],
                                 default="sRGB",
                                 advanced=True,
                                 tooltip=(
-                                    "Colorspace of the input tensor. The EXR is "
-                                    "always written as scene-linear in the matching "
-                                    "gamut.\n"
-                                    "  'sRGB'   — input is sRGB-encoded Rec.709; "
-                                    "the inverse sRGB EOTF is applied.\n"
-                                    "  'HDR'    — input is HLG-encoded Rec.2020 "
-                                    "(BT.2100); the inverse HLG OETF is applied "
-                                    "to get scene-linear light.\n"
-                                    "  'linear' — input is already scene-linear "
-                                    "(Rec.709 primaries); written through unchanged. "
-                                    "Use this for renderer/compositor output."
+                                    "Colorspace of the input tensor. The EXR is always written as scene-linear in the matching gamut.\n"
+                                    "sRGB — input is sRGB-encoded Rec.709; the inverse sRGB EOTF is applied.\n"
+                                    "HDR — input is HLG-encoded Rec.2020 (BT.2100); the inverse HLG OETF is applied to get scene-linear light.\n"
+                                    "linear — input is already scene-linear (Rec.709 primaries); written through unchanged. Use this for renderer/compositor output."
                                 ),
                             ),
                         ]),
@@ -1200,6 +1186,7 @@ class SaveImageAdvanced(IO.ComfyNode):
             ],
             hidden=[IO.Hidden.prompt, IO.Hidden.extra_pnginfo],
             is_output_node=True,
+            outputs=[IO.Image.Output(display_name="images")]
         )
 
     @classmethod
@@ -1237,7 +1224,7 @@ class SaveImageAdvanced(IO.ComfyNode):
             results.append({"filename": file, "subfolder": subfolder, "type": "output"})
             counter += 1
 
-        return IO.NodeOutput(ui={"images": results})
+        return IO.NodeOutput(images, ui={"images": results})
 
 
 class ImagesExtension(ComfyExtension):

comfy_extras/nodes_load_3d.py

@@ -317,11 +317,74 @@ class PreviewPointCloud(IO.ComfyNode):
         )
 
 
+MESH_EXTENSIONS = {'.gltf', '.glb', '.obj', '.fbx', '.stl'}
+
+
+class Load3DAdvanced(IO.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        input_dir = os.path.join(folder_paths.get_input_directory(), "3d")
+        os.makedirs(input_dir, exist_ok=True)
+
+        input_path = Path(input_dir)
+        base_path = Path(folder_paths.get_input_directory())
+
+        files = [
+            normalize_path(str(file_path.relative_to(base_path)))
+            for file_path in input_path.rglob("*")
+            if file_path.suffix.lower() in MESH_EXTENSIONS
+        ]
+        return IO.Schema(
+            node_id="Load3DAdvanced",
+            display_name="Load 3D (Advanced)",
+            category="3d",
+            search_aliases=[
+                "load mesh",
+                "load gltf",
+                "load glb",
+                "load obj",
+                "load fbx",
+                "load stl",
+            ],
+            is_experimental=True,
+            inputs=[
+                IO.Combo.Input("model_file", options=["none"] + sorted(files), upload=IO.UploadType.model),
+                IO.Load3D.Input("viewport_state"),
+                IO.Int.Input("width", default=1024, min=1, max=4096, step=1),
+                IO.Int.Input("height", default=1024, min=1, max=4096, step=1),
+            ],
+            outputs=[
+                IO.File3DAny.Output(display_name="model_3d"),
+                IO.Load3DModelInfo.Output(display_name="model_3d_info"),
+                IO.Load3DCamera.Output(display_name="camera_info"),
+                IO.Int.Output(display_name="width"),
+                IO.Int.Output(display_name="height"),
+            ],
+        )
+
+    @classmethod
+    def validate_inputs(cls, model_file, **kwargs) -> bool | str:
+        if not model_file or model_file == "none":
+            return True
+        if not folder_paths.exists_annotated_filepath(model_file):
+            return f"Invalid 3D model file: {model_file}"
+        return True
+
+    @classmethod
+    def execute(cls, model_file, viewport_state, width: int, height: int, **kwargs) -> IO.NodeOutput:
+        file_3d = None
+        if model_file and model_file != "none":
+            file_3d = Types.File3D(folder_paths.get_annotated_filepath(model_file))
+        model_3d_info = viewport_state.get('model_3d_info', [])
+        return IO.NodeOutput(file_3d, model_3d_info, viewport_state['camera_info'], width, height)
+
+
 class Load3DExtension(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[IO.ComfyNode]]:
         return [
             Load3D,
+            Load3DAdvanced,
             Preview3D,
             Preview3DAdvanced,
             PreviewGaussianSplat,

comfy_extras/nodes_logic.py

@@ -89,7 +89,8 @@ class SwitchNode(io.ComfyNode):
         template = io.MatchType.Template("switch")
         return io.Schema(
             node_id="ComfySwitchNode",
-            display_name="Switch",
+            search_aliases=["if", "then", "switch", "conditional", "branch"],
+            display_name="If/Else Switch",
             category="utilities/logic",
             is_experimental=True,
             inputs=[

comfy_extras/nodes_model_merging_model_specific.py

@@ -337,6 +337,36 @@ class ModelMergeQwenImage(comfy_extras.nodes_model_merging.ModelMergeBlocks):
 
         return {"required": arg_dict}
 
+class ModelMergeKrea2(comfy_extras.nodes_model_merging.ModelMergeBlocks):
+    CATEGORY = "model/merging/model specific"
+
+    @classmethod
+    def INPUT_TYPES(s):
+        arg_dict = { "model1": ("MODEL",),
+                              "model2": ("MODEL",)}
+
+        argument = ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01})
+
+        arg_dict["first."] = argument
+        arg_dict["tmlp."] = argument
+        arg_dict["txtmlp."] = argument
+        arg_dict["tproj."] = argument
+
+        for i in range(2):
+            arg_dict["txtfusion.layerwise_blocks.{}.".format(i)] = argument
+
+        arg_dict["txtfusion.projector."] = argument
+
+        for i in range(2):
+            arg_dict["txtfusion.refiner_blocks.{}.".format(i)] = argument
+
+        for i in range(28):
+            arg_dict["blocks.{}.".format(i)] = argument
+
+        arg_dict["last."] = argument
+
+        return {"required": arg_dict}
+
 NODE_CLASS_MAPPINGS = {
     "ModelMergeSD1": ModelMergeSD1,
     "ModelMergeSD2": ModelMergeSD1, #SD1 and SD2 have the same blocks
@@ -353,4 +383,5 @@ NODE_CLASS_MAPPINGS = {
     "ModelMergeCosmosPredict2_2B": ModelMergeCosmosPredict2_2B,
     "ModelMergeCosmosPredict2_14B": ModelMergeCosmosPredict2_14B,
     "ModelMergeQwenImage": ModelMergeQwenImage,
+    "ModelMergeKrea2": ModelMergeKrea2,
 }

comfy_extras/nodes_primitive.py

@@ -10,12 +10,11 @@ class String(io.ComfyNode):
         return io.Schema(
             node_id="PrimitiveString",
             search_aliases=["text", "string", "text box", "prompt"],
-            display_name="Text String",
+            display_name="Text String (DEPRECATED)",
             category="utilities/primitive",
-            inputs=[
-                io.String.Input("value"),
-            ],
+            inputs=[io.String.Input("value")],
             outputs=[io.String.Output()],
+            is_deprecated=True
         )
 
     @classmethod
@@ -29,12 +28,10 @@ class StringMultiline(io.ComfyNode):
         return io.Schema(
             node_id="PrimitiveStringMultiline",
             search_aliases=["text", "string", "text multiline", "string multiline", "text box", "prompt"],
-            display_name="Text String (Multiline)",
+            display_name="Input Text",
             category="utilities/primitive",
             essentials_category="Basics",
-            inputs=[
-                io.String.Input("value", multiline=True),
-            ],
+            inputs=[io.String.Input("value", multiline=True)],
             outputs=[io.String.Output()],
         )
 

comfy_extras/nodes_seed.py

@@ -0,0 +1,33 @@
+import sys
+from typing_extensions import override
+
+from comfy_api.latest import ComfyExtension, io
+
+
+class SeedNode(io.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        return io.Schema(
+            node_id="SeedNode",
+            display_name="Seed",
+            search_aliases=["seed", "random"],
+            category="utilities",
+            inputs=[
+                io.Int.Input("seed", min=0, max=sys.maxsize, control_after_generate=io.ControlAfterGenerate.fixed),
+            ],
+            outputs=[io.Int.Output(display_name="seed")],
+        )
+
+    @classmethod
+    def execute(cls, seed: int) -> io.NodeOutput:
+        return io.NodeOutput(seed)
+
+
+class SeedExtension(ComfyExtension):
+    @override
+    async def get_node_list(self) -> list[type[io.ComfyNode]]:
+        return [SeedNode]
+
+
+async def comfy_entrypoint() -> SeedExtension:
+    return SeedExtension()

comfy_extras/nodes_video.py

@@ -27,6 +27,7 @@ class SaveWEBM(io.ComfyNode):
             ],
             hidden=[io.Hidden.prompt, io.Hidden.extra_pnginfo],
             is_output_node=True,
+            outputs=[io.Image.Output(display_name="images")]
         )
 
     @classmethod
@@ -69,7 +70,7 @@ class SaveWEBM(io.ComfyNode):
         container.mux(stream.encode())
         container.close()
 
-        return io.NodeOutput(ui=ui.PreviewVideo([ui.SavedResult(file, subfolder, io.FolderType.output)]))
+        return io.NodeOutput(images, ui=ui.PreviewVideo([ui.SavedResult(file, subfolder, io.FolderType.output)]))
 
 class SaveVideo(io.ComfyNode):
     @classmethod
@@ -89,6 +90,7 @@ class SaveVideo(io.ComfyNode):
             ],
             hidden=[io.Hidden.prompt, io.Hidden.extra_pnginfo],
             is_output_node=True,
+            outputs=[io.Video.Output("video")],
         )
 
     @classmethod
@@ -117,7 +119,7 @@ class SaveVideo(io.ComfyNode):
             metadata=saved_metadata
         )
 
-        return io.NodeOutput(ui=ui.PreviewVideo([ui.SavedResult(file, subfolder, io.FolderType.output)]))
+        return io.NodeOutput(video, ui=ui.PreviewVideo([ui.SavedResult(file, subfolder, io.FolderType.output)]))
 
 
 class CreateVideo(io.ComfyNode):
@@ -233,13 +235,8 @@ class VideoSlice(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="Video Slice",
-            display_name="Video Slice",
-            search_aliases=[
-                "trim video duration",
-                "skip first frames",
-                "frame load cap",
-                "start time",
-            ],
+            display_name="Trim Video",
+            search_aliases=["trim video duration", "skip first frames", "frame load cap", "start time"],
             category="video",
             essentials_category="Video Tools",
             inputs=[

comfyui_version.py

@@ -1,3 +1,3 @@
 # This file is automatically generated by the build process when version is
 # updated in pyproject.toml.
-__version__ = "0.25.0"
+__version__ = "0.26.0"

execution.py

@@ -1308,6 +1308,25 @@ class PromptQueue:
             queued = copy.copy(self.queue)
             return (running, queued)
 
+    def interrupt_if_running(self, prompt_id):
+        """Interrupt the running prompt with this id, atomically.
+
+        Checks the live running set and signals the interrupt under the queue
+        mutex, so the worker cannot move the job to done (and start the next
+        prompt) in between. Returns True if a matching job was running and an
+        interrupt was signalled, False otherwise. The atomicity is what keeps a
+        cancel from landing on an unrelated prompt that started after a separate
+        is-running check: the global interrupt flag is reset at the start of
+        every prompt (execute_async), so a job that finishes before consuming
+        the flag cannot leak the interrupt onto its successor.
+        """
+        with self.mutex:
+            for item in self.currently_running.values():
+                if item[1] == prompt_id:
+                    nodes.interrupt_processing()
+                    return True
+        return False
+
     def get_tasks_remaining(self):
         with self.mutex:
             return len(self.queue) + len(self.currently_running)

extra_model_paths.yaml.example

@@ -8,21 +8,37 @@
 #     # You can use is_default to mark that these folders should be listed first, and used as the default dirs for eg downloads
 #     #is_default: true
 #     checkpoints: models/checkpoints/
+#     configs: models/configs/
+#     loras: models/loras/
+#     vae: models/vae/
 #     text_encoders: |
 #          models/text_encoders/
-#          models/clip/  # legacy location still supported
-#     clip_vision: models/clip_vision/
-#     configs: models/configs/
-#     controlnet: models/controlnet/
+#          models/clip/
 #     diffusion_models: |
-#                  models/diffusion_models
-#                  models/unet
+#          models/unet/
+#          models/diffusion_models/
+#     clip_vision: models/clip_vision/
+#     style_models: models/style_models/
 #     embeddings: models/embeddings/
-#     loras: models/loras/
+#     diffusers: models/diffusers/
+#     vae_approx: models/vae_approx/
+#     controlnet: |
+#          models/controlnet/
+#          models/t2i_adapter/
+#     gligen: models/gligen/
 #     upscale_models: models/upscale_models/
-#     vae: models/vae/
-#     audio_encoders: models/audio_encoders/
+#     latent_upscale_models: models/latent_upscale_models/
+#     custom_nodes: custom_nodes/
+#     hypernetworks: models/hypernetworks/
+#     photomaker: models/photomaker/
+#     classifiers: models/classifiers/
 #     model_patches: models/model_patches/
+#     audio_encoders: models/audio_encoders/
+#     background_removal: models/background_removal/
+#     frame_interpolation: models/frame_interpolation/
+#     geometry_estimation: models/geometry_estimation/
+#     optical_flow: models/optical_flow/
+#     detection: models/detection/
 
 
 #config for a1111 ui
@@ -45,8 +61,7 @@
 #     controlnet: models/ControlNet
 
 
-# For a full list of supported keys (style_models, vae_approx, hypernetworks, photomaker,
-# model_patches, audio_encoders, classifiers, etc.) see folder_paths.py.
+# For the canonical list of supported keys and extensions, see folder_paths.py.
 
 #other_ui:
 #    base_path: path/to/ui

main.py

@@ -557,8 +557,13 @@ if __name__ == "__main__":
         logging.warning("WARNING: You are using a python version older than 3.10, please upgrade to a newer one. 3.12 and above is recommended.")
 
     if args.disable_dynamic_vram:
-        logging.warning("Dynamic vram disabled with argument. If you have any issues with dynamic vram enabled please give us a detailed reports as this argument will be removed soon.")
-
+        logging.warning(
+            "Dynamic vram disabled with argument. If you have any issues with "
+            "dynamic vram enabled please give us a detailed reports as this "
+            "argument will be removed soon. If you use gguf we recommend keeping "
+            "dynamic vram enabled and using native ComfyUI model formats instead. "
+            "ComfyUI native formats like fp8 will be faster even if they are larger than your memory."
+        )
     event_loop, _, start_all_func = start_comfyui()
     try:
         x = start_all_func()

nodes.py

@@ -480,11 +480,13 @@ class SaveLatent:
 
     @classmethod
     def INPUT_TYPES(s):
-        return {"required": { "samples": ("LATENT", ),
-                              "filename_prefix": ("STRING", {"default": "latents/ComfyUI"})},
-                "hidden": {"prompt": "PROMPT", "extra_pnginfo": "EXTRA_PNGINFO"},
-                }
-    RETURN_TYPES = ()
+        return { "required": {
+            "samples": ("LATENT",),
+            "filename_prefix": ("STRING", {"default": "latents/ComfyUI"})},
+            "hidden": {"prompt": "PROMPT", "extra_pnginfo": "EXTRA_PNGINFO"},
+        }
+    RETURN_TYPES = ("LATENT",)
+    RETURN_NAMES = ("samples",)
     FUNCTION = "save"
 
     OUTPUT_NODE = True
@@ -522,7 +524,7 @@ class SaveLatent:
         output["latent_format_version_0"] = torch.tensor([])
 
         comfy.utils.save_torch_file(output, file, metadata=metadata)
-        return { "ui": { "latents": results } }
+        return { "ui": { "latents": results }, "result": (samples,) }
 
 
 class LoadLatent:
@@ -967,7 +969,7 @@ class CLIPLoader:
     @classmethod
     def INPUT_TYPES(s):
         return {"required": { "clip_name": (folder_paths.get_filename_list("text_encoders"), ),
-                              "type": (["stable_diffusion", "stable_cascade", "sd3", "stable_audio", "mochi", "ltxv", "pixart", "cosmos", "lumina2", "wan", "hidream", "chroma", "ace", "omnigen2", "qwen_image", "hunyuan_image", "flux2", "ovis", "longcat_image", "cogvideox", "lens", "pixeldit", "ideogram4", "boogu"], ),
+                              "type": (["stable_diffusion", "stable_cascade", "sd3", "stable_audio", "mochi", "ltxv", "pixart", "cosmos", "lumina2", "wan", "hidream", "chroma", "ace", "omnigen2", "qwen_image", "hunyuan_image", "flux2", "ovis", "longcat_image", "cogvideox", "lens", "pixeldit", "ideogram4", "boogu", "krea2"], ),
                               },
                 "optional": {
                               "device": (["default", "cpu"], {"advanced": True}),
@@ -1627,14 +1629,18 @@ class SaveImage:
         return {
             "required": {
                 "images": ("IMAGE", {"tooltip": "The images to save."}),
-                "filename_prefix": ("STRING", {"default": "ComfyUI", "tooltip": "The prefix for the file to save. This may include formatting information such as %date:yyyy-MM-dd% or %Empty Latent Image.width% to include values from nodes."})
+                "filename_prefix": ("STRING", {
+                    "default": "ComfyUI",
+                    "tooltip": "The prefix for the file to save. This may include formatting information such as %date:yyyy-MM-dd% or %Empty Latent Image.width% to include values from nodes."
+                })
             },
             "hidden": {
                 "prompt": "PROMPT", "extra_pnginfo": "EXTRA_PNGINFO"
             },
         }
 
-    RETURN_TYPES = ()
+    RETURN_TYPES = ("IMAGE",)
+    RETURN_NAMES = ("images",)
     FUNCTION = "save_images"
 
     OUTPUT_NODE = True
@@ -1670,7 +1676,7 @@ class SaveImage:
             })
             counter += 1
 
-        return { "ui": { "images": results } }
+        return { "ui": { "images": results }, "result" : (images,) }
 
 class PreviewImage(SaveImage):
     def __init__(self):
@@ -2467,6 +2473,7 @@ async def init_builtin_extra_nodes():
         "nodes_gaussian_splat.py",
         "nodes_triposplat.py",
         "nodes_depth_anything_3.py",
+        "nodes_seed.py",
     ]
 
     import_failed = []

openapi.yaml

@@ -55,6 +55,12 @@ components:
                     description: URL for asset preview/thumbnail
                     format: uri
                     type: string
+                short_url:
+                    description: Durable, owner-gated short link to this asset's content (relative `/api/s/{id}` path). Stable across the underlying signed URL's expiry — resolving it re-mints a fresh signed URL on every request — so it is safe to persist or share into chat, unlike `preview_url`. Only the minting user can resolve it. Omitted when the short-link surface is disabled or the asset has no resolvable content hash.
+                    nullable: true
+                    type: string
+                    x-runtime:
+                        - cloud
                 size:
                     description: Size of the asset in bytes
                     format: int64
@@ -1686,6 +1692,12 @@ paths:
                             schema:
                                 $ref: '#/components/schemas/ErrorResponse'
                     description: Unsupported media type
+                "422":
+                    content:
+                        application/json:
+                            schema:
+                                $ref: '#/components/schemas/ErrorResponse'
+                    description: Validation error (e.g., disallowed model_type tag)
                 "500":
                     content:
                         application/json:
@@ -2131,6 +2143,12 @@ paths:
                             schema:
                                 $ref: '#/components/schemas/ErrorResponse'
                     description: Source asset with given hash not found
+                "422":
+                    content:
+                        application/json:
+                            schema:
+                                $ref: '#/components/schemas/ErrorResponse'
+                    description: Validation error (e.g., disallowed model_type tag)
                 "500":
                     content:
                         application/json:
@@ -2351,6 +2369,10 @@ paths:
             description: |
                 Returns a list of model folders available in the system.
                 This is an experimental endpoint that replaces the legacy /models endpoint.
+                Each folder's name is the identifier to pass to /api/experiment/models/{folder}.
+                Once the model_type migration is active the names are model_type folder_names
+                (e.g. `ultralytics_bbox`); a folder with no folder_name mapping is returned by
+                its directory path.
             operationId: getModelFolders
             responses:
                 "200":
@@ -2981,6 +3003,17 @@ paths:
                   schema:
                     format: uuid
                     type: string
+                - description: |
+                    When present, each output item in the response receives a `short_url` field containing a short link for that asset. Omit this parameter (the default) to receive a response identical to the no-param baseline. The value selects the link's lifetime and auth model: use `ephemeral_tool_chain` for short-lived (≤5 minute) machine-to-machine handoffs — these are public bearer links where the link ID itself is the credential, so anyone holding the link can resolve it (intended for pasting into an agent/MCP tool chain); use `default` for durable (30 day) human-revisitable links, which are owner-gated and resolvable only by the authenticated owner. Links are always minted under the authenticated request owner's identity; the auth model is selected by the server and is never settable by the caller.
+                  in: query
+                  name: short_link
+                  schema:
+                    enum:
+                        - ephemeral_tool_chain
+                        - default
+                    type: string
+                  x-runtime:
+                    - cloud
             responses:
                 "200":
                     content:

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ComfyUI"
-version = "0.25.0"
+version = "0.26.0"
 readme = "README.md"
 license = { file = "LICENSE" }
 requires-python = ">=3.10"

requirements.txt

@@ -1,6 +1,6 @@
-comfyui-frontend-package==1.45.15
-comfyui-workflow-templates==0.10.0
-comfyui-embedded-docs==0.5.4
+comfyui-frontend-package==1.45.19
+comfyui-workflow-templates==0.10.2
+comfyui-embedded-docs==0.5.5
 torch
 torchsde
 torchvision

server.py

@@ -8,7 +8,15 @@ import time
 import nodes
 import folder_paths
 import execution
-from comfy_execution.jobs import JobStatus, get_job, get_all_jobs, validate_job_id
+from comfy_execution.jobs import (
+    JobStatus,
+    get_job,
+    get_all_jobs,
+    validate_job_id,
+    cancel_job,
+    CANCEL_PENDING,
+    CANCEL_RUNNING,
+)
 import uuid
 import urllib
 import json
@@ -899,6 +907,107 @@ class PromptServer():
 
             return web.json_response(job)
 
+        def _cancel_job_by_id(job_id):
+            """Cancel a single job by id using the queue's existing mechanics.
+
+            Running jobs are interrupted (same mechanism as /interrupt); pending
+            jobs are dequeued (same mechanism as /queue {"delete": [...]}).
+            Already-finished or unknown ids are no-ops. State-agnostic.
+
+            Returns True when a cancel was actually dispatched (running or
+            pending job), False when the call was a no-op (terminal/unknown id).
+            """
+            running, queued = self.prompt_queue.get_current_queue()
+            history = self.prompt_queue.get_history()
+
+            def interrupt(prompt_id):
+                logging.info(f"Cancelling running prompt {prompt_id}")
+                # Atomic: only interrupts if the job is still the one running,
+                # so a cancel can't land on a prompt that started in the gap
+                # since the snapshot above. Returns whether it actually fired.
+                return self.prompt_queue.interrupt_if_running(prompt_id)
+
+            def dequeue(prompt_id):
+                logging.info(f"Cancelling pending prompt {prompt_id}")
+                return self.prompt_queue.delete_queue_item(lambda a: a[1] == prompt_id)
+
+            classification = cancel_job(job_id, running, queued, history, interrupt, dequeue)
+            return classification in (CANCEL_RUNNING, CANCEL_PENDING)
+
+        @routes.post("/api/jobs/{job_id}/cancel")
+        async def cancel_job_by_id(request):
+            """Cancel a single job by id, regardless of state.
+
+            Idempotent: cancelling a job that has already finished, or an id
+            that is not known, returns 200 with {"cancelled": false} rather
+            than an error.
+            """
+            job_id = request.match_info.get("job_id", None)
+            if not job_id:
+                return web.json_response(
+                    {"error": "job_id is required"},
+                    status=400
+                )
+
+            cancelled = _cancel_job_by_id(job_id)
+            return web.json_response({"cancelled": cancelled})
+
+        @routes.post("/api/jobs/cancel")
+        async def cancel_jobs_batch(request):
+            """Cancel a batch of jobs by id.
+
+            Body: {"job_ids": ["<uuid>", ...]}
+
+            Best-effort and idempotent: every well-formed id is cancelled if it
+            is running or pending; ids that are already finished or unknown are
+            no-ops, not errors. A batch of all no-ops still returns 200 with
+            {"cancelled": false}. This matches the single-cancel endpoint and
+            means "cancel all" still cancels the in-progress jobs even if some
+            finished between the client's snapshot and the request. Malformed
+            ids are still rejected up front with 400 (see below).
+            """
+            try:
+                json_data = await request.json()
+            except json.JSONDecodeError:
+                return web.json_response(
+                    {"error": "Request body must be valid JSON"},
+                    status=400
+                )
+
+            job_ids = json_data.get("job_ids") if isinstance(json_data, dict) else None
+            if not isinstance(job_ids, list):
+                return web.json_response(
+                    {"error": "job_ids must be a list"},
+                    status=400
+                )
+
+            # Validate that every element is a well-formed job id before doing
+            # anything else.  An unhashable element (e.g. a nested dict or list)
+            # would cause a TypeError when used as a history dict key; a
+            # non-string or non-UUID value is never a valid id.  Reject early
+            # with 400 rather than letting the classify loop raise 500.
+            invalid_ids = []
+            for jid in job_ids:
+                try:
+                    validate_job_id(jid)
+                except (ValueError, AttributeError):
+                    invalid_ids.append(jid if isinstance(jid, str) else repr(jid))
+            if invalid_ids:
+                return web.json_response(
+                    {"error": "job_ids contains invalid id(s)", "invalid_ids": invalid_ids},
+                    status=400,
+                )
+
+            # Best-effort: cancel each id that is still running/pending; an id
+            # that has finished or never existed is a no-op rather than a reason
+            # to fail the whole batch.
+            cancelled = False
+            for jid in job_ids:
+                if _cancel_job_by_id(jid):
+                    cancelled = True
+
+            return web.json_response({"cancelled": cancelled})
+
         @routes.get("/history")
         async def get_history(request):
             max_items = request.rel_url.query.get("max_items", None)

tests-unit/jobs_cancel_test/jobs_cancel_test.py

@@ -0,0 +1,453 @@
+"""Tests for the jobs-namespace cancel endpoints.
+
+Covers both layers:
+
+* the pure cancel helpers in ``comfy_execution.jobs``
+  (``classify_job_for_cancel`` / ``cancel_job``), which hold the business
+  logic of mapping a cancel onto interrupt-vs-dequeue, and
+
+* the HTTP contract of ``POST /api/jobs/{job_id}/cancel`` and
+  ``POST /api/jobs/cancel`` (status codes, single-cancel idempotency, and
+  best-effort batch cancellation that treats unknown/finished ids as no-ops
+  while still rejecting malformed ids with 400).
+
+The HTTP layer is exercised against a small aiohttp app whose handlers are a
+faithful copy of the wiring in ``server.py`` driven by a fake queue that
+mirrors ``execution.PromptQueue`` (``get_current_queue`` / ``get_history`` /
+``delete_queue_item``). This keeps the test free of the heavy ComfyUI runtime
+(torch, nodes, ...) while still testing the real cancel logic.
+"""
+
+import json
+
+import pytest
+from aiohttp import web
+
+from comfy_execution.jobs import (
+    CANCEL_PENDING,
+    CANCEL_RUNNING,
+    CANCEL_TERMINAL,
+    CANCEL_UNKNOWN,
+    cancel_job,
+    classify_job_for_cancel,
+    validate_job_id,
+)
+
+# Classifications for which a cancel was actually dispatched (vs a no-op).
+_CANCELLED = (CANCEL_RUNNING, CANCEL_PENDING)
+
+# Canonical UUID ids for HTTP-layer tests (the batch endpoint validates UUID format).
+_UUID_A = "aaaaaaaa-aaaa-4aaa-aaaa-aaaaaaaaaaaa"
+_UUID_B = "bbbbbbbb-bbbb-4bbb-bbbb-bbbbbbbbbbbb"
+_UUID_C = "cccccccc-cccc-4ccc-cccc-cccccccccccc"
+_UUID_D = "dddddddd-dddd-4ddd-dddd-dddddddddddd"
+_UUID_MISSING = "ffffffff-ffff-4fff-ffff-ffffffffffff"
+
+
+def make_queue_item(prompt_id, number=0):
+    """Build a queue tuple shaped like the real ones: index 1 is the id."""
+    return (number, prompt_id, {}, {}, [])
+
+
+class FakePromptQueue:
+    """Minimal stand-in for execution.PromptQueue for the cancel paths.
+
+    Tracks interrupts and dequeues so tests can assert side effects.
+    """
+
+    def __init__(self, running=None, pending=None, history=None):
+        self._running = list(running or [])
+        self._pending = list(pending or [])
+        self._history = dict(history or {})
+        self.interrupt_count = 0
+
+    def get_current_queue(self):
+        return (list(self._running), list(self._pending))
+
+    def get_history(self, prompt_id=None):
+        if prompt_id is None:
+            return dict(self._history)
+        if prompt_id in self._history:
+            return {prompt_id: self._history[prompt_id]}
+        return {}
+
+    def delete_queue_item(self, function):
+        for i, item in enumerate(self._pending):
+            if function(item):
+                self._pending.pop(i)
+                return True
+        return False
+
+    def interrupt_if_running(self, prompt_id):
+        # Mirrors execution.PromptQueue.interrupt_if_running: only signals an
+        # interrupt when the id is actually in the running set.
+        if any(item[1] == prompt_id for item in self._running):
+            self.interrupt_count += 1
+            return True
+        return False
+
+
+def build_app(queue):
+    """Build an aiohttp app exposing the cancel routes against ``queue``.
+
+    Handler bodies mirror server.py exactly.
+    """
+
+    def _cancel_job_by_id(job_id):
+        running, pending = queue.get_current_queue()
+        history = queue.get_history()
+
+        def interrupt(prompt_id):
+            return queue.interrupt_if_running(prompt_id)
+
+        def dequeue(prompt_id):
+            return queue.delete_queue_item(lambda a: a[1] == prompt_id)
+
+        classification = cancel_job(
+            job_id, running, pending, history, interrupt, dequeue
+        )
+        return classification in _CANCELLED
+
+    async def cancel_job_by_id(request):
+        job_id = request.match_info.get("job_id", None)
+        if not job_id:
+            return web.json_response({"error": "job_id is required"}, status=400)
+        cancelled = _cancel_job_by_id(job_id)
+        return web.json_response({"cancelled": cancelled})
+
+    async def cancel_jobs_batch(request):
+        try:
+            json_data = await request.json()
+        except json.JSONDecodeError:
+            return web.json_response(
+                {"error": "Request body must be valid JSON"}, status=400
+            )
+
+        job_ids = json_data.get("job_ids") if isinstance(json_data, dict) else None
+        if not isinstance(job_ids, list):
+            return web.json_response({"error": "job_ids must be a list"}, status=400)
+
+        invalid_ids = []
+        for jid in job_ids:
+            try:
+                validate_job_id(jid)
+            except (ValueError, AttributeError):
+                invalid_ids.append(jid if isinstance(jid, str) else repr(jid))
+        if invalid_ids:
+            return web.json_response(
+                {"error": "job_ids contains invalid id(s)", "invalid_ids": invalid_ids},
+                status=400,
+            )
+
+        cancelled = False
+        for jid in job_ids:
+            if _cancel_job_by_id(jid):
+                cancelled = True
+        return web.json_response({"cancelled": cancelled})
+
+    app = web.Application()
+    app.router.add_post("/api/jobs/{job_id}/cancel", cancel_job_by_id)
+    app.router.add_post("/api/jobs/cancel", cancel_jobs_batch)
+    return app
+
+
+# ---------------------------------------------------------------------------
+# Pure helper tests: classification + cancel side effects
+# ---------------------------------------------------------------------------
+
+
+class TestClassifyJobForCancel:
+    def test_running(self):
+        running = [make_queue_item("a")]
+        assert classify_job_for_cancel("a", running, [], {}) == CANCEL_RUNNING
+
+    def test_pending(self):
+        pending = [make_queue_item("b")]
+        assert classify_job_for_cancel("b", [], pending, {}) == CANCEL_PENDING
+
+    def test_terminal(self):
+        history = {"c": {"prompt": make_queue_item("c"), "outputs": {}, "status": {}}}
+        assert classify_job_for_cancel("c", [], [], history) == CANCEL_TERMINAL
+
+    def test_unknown(self):
+        assert classify_job_for_cancel("z", [], [], {}) == CANCEL_UNKNOWN
+
+
+class TestCancelJobHelper:
+    """``interrupt`` and ``dequeue`` both take the id and return whether they
+    actually acted, so cancel_job's return reflects the real outcome."""
+
+    def test_running_is_interrupted_not_dequeued(self):
+        interrupts = []
+        dequeues = []
+        result = cancel_job(
+            "a", [make_queue_item("a")], [], {},
+            interrupt=lambda pid: interrupts.append(pid) or True,
+            dequeue=lambda pid: dequeues.append(pid) or True,
+        )
+        assert result == CANCEL_RUNNING
+        assert interrupts == ["a"]
+        assert dequeues == []
+
+    def test_pending_is_dequeued_not_interrupted(self):
+        interrupts = []
+        dequeues = []
+        result = cancel_job(
+            "b", [], [make_queue_item("b")], {},
+            interrupt=lambda pid: interrupts.append(pid) or True,
+            dequeue=lambda pid: dequeues.append(pid) or True,
+        )
+        assert result == CANCEL_PENDING
+        assert dequeues == ["b"]
+        assert interrupts == []
+
+    def test_terminal_is_noop(self):
+        history = {"c": {"prompt": make_queue_item("c"), "outputs": {}, "status": {}}}
+        interrupts = []
+        dequeues = []
+        result = cancel_job(
+            "c", [], [], history,
+            interrupt=lambda pid: interrupts.append(pid) or True,
+            dequeue=lambda pid: dequeues.append(pid) or True,
+        )
+        assert result == CANCEL_TERMINAL
+        assert interrupts == []
+        assert dequeues == []
+
+    def test_unknown_is_noop(self):
+        interrupts = []
+        dequeues = []
+        result = cancel_job(
+            "z", [], [], {},
+            interrupt=lambda pid: interrupts.append(pid) or True,
+            dequeue=lambda pid: dequeues.append(pid) or True,
+        )
+        assert result == CANCEL_UNKNOWN
+        assert interrupts == []
+        assert dequeues == []
+
+    def test_running_but_finished_before_interrupt_returns_unknown(self):
+        """Classified RUNNING from a stale snapshot, but the job finished before
+        the atomic interrupt fired (interrupt returns False). cancel_job reports
+        UNKNOWN rather than claiming a cancel that did not happen — and the
+        atomic interrupt guarantees no unrelated job was hit."""
+        interrupts = []
+        result = cancel_job(
+            "a", [make_queue_item("a")], [], {},
+            interrupt=lambda pid: interrupts.append(pid) or False,
+            dequeue=lambda pid: True,
+        )
+        assert result == CANCEL_UNKNOWN
+        assert interrupts == ["a"]  # interrupt was attempted atomically
+
+    def test_pending_started_running_is_interrupted(self):
+        """Pending->running race: the job leaves the queue (dequeue False)
+        because it started executing. The atomic interrupt catches the now-
+        running job, so cancel_job interrupts it and reports CANCEL_RUNNING."""
+        interrupts = []
+        dequeues = []
+        result = cancel_job(
+            "b", [], [make_queue_item("b")], {},
+            interrupt=lambda pid: interrupts.append(pid) or True,
+            dequeue=lambda pid: (dequeues.append(pid), False)[1],
+        )
+        assert result == CANCEL_RUNNING
+        assert dequeues == ["b"]    # dequeue attempted first
+        assert interrupts == ["b"]  # then the now-running job was interrupted
+
+    def test_pending_dequeue_miss_not_running_returns_unknown(self):
+        """Dequeue miss where the job is not running anymore (it finished): the
+        atomic interrupt finds nothing to interrupt and returns False, so
+        cancel_job is a no-op reporting UNKNOWN — never reporting a cancel that
+        did not happen, and never interrupting a bystander."""
+        interrupts = []
+        dequeues = []
+        result = cancel_job(
+            "b", [], [make_queue_item("b")], {},
+            interrupt=lambda pid: interrupts.append(pid) or False,
+            dequeue=lambda pid: (dequeues.append(pid), False)[1],
+        )
+        assert result == CANCEL_UNKNOWN
+        assert dequeues == ["b"]
+        assert interrupts == ["b"]  # interrupt attempted, found nothing running
+
+
+# ---------------------------------------------------------------------------
+# HTTP contract tests: POST /api/jobs/{job_id}/cancel
+# ---------------------------------------------------------------------------
+
+
+class TestSingleCancelEndpoint:
+    @pytest.mark.asyncio
+    async def test_cancel_running_job_interrupts(self, aiohttp_client):
+        queue = FakePromptQueue(running=[make_queue_item("a")])
+        client = await aiohttp_client(build_app(queue))
+
+        resp = await client.post("/api/jobs/a/cancel")
+
+        assert resp.status == 200
+        assert (await resp.json()) == {"cancelled": True}
+        assert queue.interrupt_count == 1
+
+    @pytest.mark.asyncio
+    async def test_cancel_pending_job_dequeues(self, aiohttp_client):
+        queue = FakePromptQueue(pending=[make_queue_item("b")])
+        client = await aiohttp_client(build_app(queue))
+
+        resp = await client.post("/api/jobs/b/cancel")
+
+        assert resp.status == 200
+        assert (await resp.json()) == {"cancelled": True}
+        # Pending job removed from the queue; nothing interrupted.
+        assert queue.get_current_queue()[1] == []
+        assert queue.interrupt_count == 0
+
+    @pytest.mark.asyncio
+    async def test_cancel_terminal_job_is_idempotent_noop(self, aiohttp_client):
+        history = {"c": {"prompt": make_queue_item("c"), "outputs": {}, "status": {}}}
+        queue = FakePromptQueue(history=history)
+        client = await aiohttp_client(build_app(queue))
+
+        resp = await client.post("/api/jobs/c/cancel")
+
+        # Already-finished job: 200 no-op (cancelled=false), not an error.
+        assert resp.status == 200
+        assert (await resp.json()) == {"cancelled": False}
+        assert queue.interrupt_count == 0
+
+    @pytest.mark.asyncio
+    async def test_cancel_unknown_id_is_200_noop(self, aiohttp_client):
+        queue = FakePromptQueue()
+        client = await aiohttp_client(build_app(queue))
+
+        resp = await client.post("/api/jobs/does-not-exist/cancel")
+
+        # Single-cancel of an unknown id is treated as an idempotent no-op.
+        assert resp.status == 200
+        assert (await resp.json()) == {"cancelled": False}
+        assert queue.interrupt_count == 0
+
+    @pytest.mark.asyncio
+    async def test_cancel_pending_that_started_running_interrupts(self, aiohttp_client):
+        """Pending->running race end to end: the job is pending at snapshot time
+        but starts executing by the time we dequeue (delete misses). The live
+        re-check sees it running and interrupts it, so the cancel is not dropped
+        and the caller still gets cancelled=True."""
+
+        class RacingQueue(FakePromptQueue):
+            def delete_queue_item(self, function):
+                # The worker picked the job up just before we removed it: it
+                # leaves the pending queue (delete misses) and is now running.
+                self._running = list(self._pending)
+                self._pending = []
+                return False
+
+        queue = RacingQueue(pending=[make_queue_item("b")])
+        client = await aiohttp_client(build_app(queue))
+
+        resp = await client.post("/api/jobs/b/cancel")
+
+        assert resp.status == 200
+        assert (await resp.json()) == {"cancelled": True}
+        assert queue.interrupt_count == 1
+
+
+# ---------------------------------------------------------------------------
+# HTTP contract tests: POST /api/jobs/cancel (batch)
+# ---------------------------------------------------------------------------
+
+
+class TestBatchCancelEndpoint:
+    @pytest.mark.asyncio
+    async def test_batch_happy_path(self, aiohttp_client):
+        queue = FakePromptQueue(
+            running=[make_queue_item(_UUID_A)],
+            pending=[make_queue_item(_UUID_B, number=1)],
+        )
+        client = await aiohttp_client(build_app(queue))
+
+        resp = await client.post("/api/jobs/cancel", json={"job_ids": [_UUID_A, _UUID_B]})
+
+        assert resp.status == 200
+        assert (await resp.json()) == {"cancelled": True}
+        assert queue.interrupt_count == 1            # running job interrupted
+        assert queue.get_current_queue()[1] == []    # pending job dequeued
+
+    @pytest.mark.asyncio
+    async def test_batch_best_effort_skips_unknown_id(self, aiohttp_client):
+        """An unknown id in the batch is a no-op, not a reason to abort: the
+        running and pending jobs are still cancelled (200, cancelled=true). This
+        is the "cancel all as a job finishes" case from review."""
+        queue = FakePromptQueue(
+            running=[make_queue_item(_UUID_A)],
+            pending=[make_queue_item(_UUID_B, number=1)],
+        )
+        client = await aiohttp_client(build_app(queue))
+
+        resp = await client.post(
+            "/api/jobs/cancel", json={"job_ids": [_UUID_A, _UUID_MISSING, _UUID_B]}
+        )
+
+        assert resp.status == 200
+        assert (await resp.json()) == {"cancelled": True}
+        assert queue.interrupt_count == 1            # running job interrupted
+        assert queue.get_current_queue()[1] == []    # pending job dequeued
+
+    @pytest.mark.asyncio
+    async def test_batch_all_terminal_is_idempotent_noop(self, aiohttp_client):
+        history = {
+            _UUID_C: {"prompt": make_queue_item(_UUID_C), "outputs": {}, "status": {}},
+            _UUID_D: {"prompt": make_queue_item(_UUID_D), "outputs": {}, "status": {}},
+        }
+        queue = FakePromptQueue(history=history)
+        client = await aiohttp_client(build_app(queue))
+
+        resp = await client.post("/api/jobs/cancel", json={"job_ids": [_UUID_C, _UUID_D]})
+
+        # All known but terminal: 200 with cancelled=false, nothing dispatched.
+        assert resp.status == 200
+        assert (await resp.json()) == {"cancelled": False}
+        assert queue.interrupt_count == 0
+
+    @pytest.mark.asyncio
+    async def test_batch_missing_job_ids_is_400(self, aiohttp_client):
+        queue = FakePromptQueue()
+        client = await aiohttp_client(build_app(queue))
+
+        resp = await client.post("/api/jobs/cancel", json={})
+
+        assert resp.status == 400
+
+    @pytest.mark.asyncio
+    async def test_batch_unhashable_element_is_400_not_500(self, aiohttp_client):
+        """An unhashable element such as a dict or list must yield 400, not 500.
+
+        Previously, passing e.g. {"job_ids": [{}]} would reach the classify
+        loop where ``prompt_id in history`` raises TypeError on an unhashable
+        type, resulting in an unhandled 500.  The input-validation guard must
+        catch this before any queue or history access.
+        """
+        queue = FakePromptQueue()
+        client = await aiohttp_client(build_app(queue))
+
+        resp = await client.post("/api/jobs/cancel", json={"job_ids": [{}]})
+
+        assert resp.status == 400
+        body = await resp.json()
+        assert "invalid_ids" in body
+        # No queue side effects.
+        assert queue.interrupt_count == 0
+
+    @pytest.mark.asyncio
+    async def test_batch_non_uuid_string_element_is_400(self, aiohttp_client):
+        """A string that is not a valid UUID must be rejected with 400."""
+        queue = FakePromptQueue()
+        client = await aiohttp_client(build_app(queue))
+
+        resp = await client.post(
+            "/api/jobs/cancel", json={"job_ids": ["not-a-uuid"]}
+        )
+
+        assert resp.status == 400
+        body = await resp.json()
+        assert "invalid_ids" in body