diff --git a/comfy/ldm/modules/attention.py b/comfy/ldm/modules/attention.py
index 2411aff5c..e6500cff4 100644
--- a/comfy/ldm/modules/attention.py
+++ b/comfy/ldm/modules/attention.py
@@ -709,7 +709,7 @@ def attention3_sage(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
     return out
 
 try:
-    @torch.library.custom_op("flash_attention::flash_attn", mutates_args=())
+    @torch.library.custom_op("comfy::flash_attn", mutates_args=())
     def flash_attn_wrapper(q: torch.Tensor, k: torch.Tensor, v: torch.Tensor,
                     dropout_p: float = 0.0, causal: bool = False, softmax_scale: float = -1.0) -> torch.Tensor:
         softmax_scale_arg = None if softmax_scale == -1.0 else softmax_scale
diff --git a/comfy/ldm/seedvr/vae.py b/comfy/ldm/seedvr/vae.py
index c9f430184..7a8070b65 100644
--- a/comfy/ldm/seedvr/vae.py
+++ b/comfy/ldm/seedvr/vae.py
@@ -30,7 +30,7 @@ from enum import Enum
 import logging
 import comfy.model_management
 import comfy.ops
-ops = comfy.ops.disable_weight_init
+ops = comfy.ops.manual_cast
 
 
 def _seedvr2_temporal_slicing_min_size(temporal_size, temporal_overlap, temporal_scale=1):
@@ -103,11 +103,10 @@ def tiled_vae(
     storage_device = vae_model.device
     result = None
     count = None
-    def run_temporal_chunks(spatial_tile, model=vae_model, device=storage_device):
-        device = torch.device(device)
-        t_chunk = spatial_tile.to(device=device, dtype=next(model.parameters()).dtype, non_blocking=True).contiguous()
+    def run_temporal_chunks(spatial_tile, model=vae_model):
+        t_chunk = spatial_tile.contiguous()
         old_device = getattr(model, "device", None)
-        model.device = device
+        model.device = t_chunk.device
         old_slicing_min_size = getattr(model, slicing_attr, None)
         if old_slicing_min_size is not None and slicing_min_size is not None:
             if slicing_min_size <= 0:
@@ -397,7 +396,7 @@ class Attention(nn.Module):
 
 def causal_norm_wrapper(norm_layer: nn.Module, x: torch.Tensor) -> torch.Tensor:
     input_dtype = x.dtype
-    if isinstance(norm_layer, (ops.LayerNorm, ops.RMSNorm)):
+    if isinstance(norm_layer, (nn.LayerNorm, nn.RMSNorm)):
         if x.ndim == 4:
             x = x.permute(0, 2, 3, 1)
             x = norm_layer(x)
@@ -408,14 +407,14 @@ def causal_norm_wrapper(norm_layer: nn.Module, x: torch.Tensor) -> torch.Tensor:
             x = norm_layer(x)
             x = x.permute(0, 4, 1, 2, 3)
             return x.to(input_dtype)
-    if isinstance(norm_layer, (ops.GroupNorm, nn.BatchNorm2d, nn.SyncBatchNorm)):
+    if isinstance(norm_layer, (nn.GroupNorm, nn.BatchNorm2d, nn.SyncBatchNorm)):
         if x.ndim <= 4:
             return norm_layer(x).to(input_dtype)
         if x.ndim == 5:
             b, c, t, h, w = x.shape
             x = x.transpose(1, 2).reshape(b * t, c, h, w)
             memory_occupy = x.numel() * x.element_size() / 1024**3
-            if isinstance(norm_layer, ops.GroupNorm) and memory_occupy > get_norm_limit():
+            if isinstance(norm_layer, nn.GroupNorm) and memory_occupy > get_norm_limit():
                 num_chunks = min(BYTEDANCE_GN_CHUNKS_FP16 if x.element_size() == 2 else BYTEDANCE_GN_CHUNKS_FP32, norm_layer.num_groups)
                 if norm_layer.num_groups % num_chunks != 0:
                     raise ValueError(
@@ -423,9 +422,9 @@ def causal_norm_wrapper(norm_layer: nn.Module, x: torch.Tensor) -> torch.Tensor:
                     )
                 num_groups_per_chunk = norm_layer.num_groups // num_chunks
 
+                weights = comfy.ops.cast_to_input(norm_layer.weight, x).chunk(num_chunks, dim=0)
+                biases = comfy.ops.cast_to_input(norm_layer.bias, x).chunk(num_chunks, dim=0)
                 x = list(x.chunk(num_chunks, dim=1))
-                weights = norm_layer.weight.chunk(num_chunks, dim=0)
-                biases = norm_layer.bias.chunk(num_chunks, dim=0)
                 for i, (w, bias) in enumerate(zip(weights, biases)):
                     x[i] = F.group_norm(x[i], num_groups_per_chunk, w, bias, norm_layer.eps)
                     x[i] = x[i].to(input_dtype)
@@ -1459,7 +1458,6 @@ class VideoAutoencoderKLWrapper(VideoAutoencoderKL):
     def _encode_with_raw_latent(self, x):
         if x.ndim == 4:
             x = x.unsqueeze(2)
-        x = x.to(dtype=next(self.parameters()).dtype)
         self.device = x.device
         p = super().encode(x)
         z = p.squeeze(2)
diff --git a/comfy/quant_ops.py b/comfy/quant_ops.py
index b1aabdc93..15f9b1fdb 100644
--- a/comfy/quant_ops.py
+++ b/comfy/quant_ops.py
@@ -48,7 +48,7 @@ try:
     # older Triton lacks libdevice.rint on the HIP backend and hard-crashes the INT8 path.
     if args.disable_triton_backend:
         ck.registry.disable("triton")
-    elif args.enable_triton_backend or (torch.version.hip is not None and _rocm_kitchen_arch_supported()):
+    elif args.enable_triton_backend: # or (torch.version.hip is not None and _rocm_kitchen_arch_supported()):
         try:
             import triton
             triton_version = tuple(int(v) for v in triton.__version__.split(".")[:2])
diff --git a/comfy_extras/nodes_audio.py b/comfy_extras/nodes_audio.py
index 6adcc95fa..4ac5ced53 100644
--- a/comfy_extras/nodes_audio.py
+++ b/comfy_extras/nodes_audio.py
@@ -298,6 +298,7 @@ class PreviewAudio(IO.ComfyNode):
             search_aliases=["play audio"],
             display_name="Preview Audio",
             category="audio",
+            description="Preview the audio without saving it to the ComfyUI output directory.",
             inputs=[
                 IO.Audio.Input("audio"),
             ],
diff --git a/comfy_extras/nodes_load_3d.py b/comfy_extras/nodes_load_3d.py
index 6ef9a1ca3..a9df557c2 100644
--- a/comfy_extras/nodes_load_3d.py
+++ b/comfy_extras/nodes_load_3d.py
@@ -92,6 +92,7 @@ class Preview3D(IO.ComfyNode):
             search_aliases=["view mesh", "3d viewer"],
             display_name="Preview 3D & Animation",
             category="3d",
+            description="Preview a 3D model file without saving it to the ComfyUI output directory.",
             is_experimental=True,
             is_output_node=True,
             inputs=[
@@ -136,6 +137,7 @@ class Preview3DAdvanced(IO.ComfyNode):
             display_name="Preview 3D (Advanced)",
             search_aliases=["preview 3d", "3d viewer", "view mesh", "frame 3d", "3d camera output"],
             category="3d",
+            description="Preview a 3D model file without saving it to the ComfyUI output directory.",
             is_experimental=True,
             is_output_node=True,
             inputs=[
@@ -193,6 +195,7 @@ class PreviewGaussianSplat(IO.ComfyNode):
             node_id="PreviewGaussianSplat",
             display_name="Preview Splat",
             category="3d",
+            description="Preview a gaussian splat 3D file without saving it to the ComfyUI output directory.",
             is_experimental=True,
             is_output_node=True,
             search_aliases=[
@@ -261,6 +264,7 @@ class PreviewPointCloud(IO.ComfyNode):
             node_id="PreviewPointCloud",
             display_name="Preview Point Cloud",
             category="3d",
+            description="Preview a point cloud 3D file without saving it to the ComfyUI output directory.",
             is_experimental=True,
             is_output_node=True,
             search_aliases=[
diff --git a/comfy_extras/nodes_mask.py b/comfy_extras/nodes_mask.py
index 76af338de..3fae7221f 100644
--- a/comfy_extras/nodes_mask.py
+++ b/comfy_extras/nodes_mask.py
@@ -419,17 +419,18 @@ class MaskPreview(IO.ComfyNode):
             search_aliases=["show mask", "view mask", "inspect mask", "debug mask"],
             display_name="Preview Mask",
             category="image/mask",
-            description="Saves the input images to your ComfyUI output directory.",
+            description="Preview the masks without saving them to the ComfyUI output directory.",
             inputs=[
                 IO.Mask.Input("mask"),
             ],
             hidden=[IO.Hidden.prompt, IO.Hidden.extra_pnginfo],
             is_output_node=True,
+            outputs=[IO.Mask.Output(display_name="mask")]
         )
 
     @classmethod
     def execute(cls, mask, filename_prefix="ComfyUI") -> IO.NodeOutput:
-        return IO.NodeOutput(ui=UI.PreviewMask(mask))
+        return IO.NodeOutput(mask, ui=UI.PreviewMask(mask))
 
 
 class MaskExtension(ComfyExtension):
diff --git a/comfy_extras/nodes_preview_any.py b/comfy_extras/nodes_preview_any.py
index 1070a69d0..d985f3287 100644
--- a/comfy_extras/nodes_preview_any.py
+++ b/comfy_extras/nodes_preview_any.py
@@ -18,6 +18,7 @@ class PreviewAny():
 
     CATEGORY = "utilities"
     SEARCH_ALIASES = ["show output", "inspect", "debug", "print value", "show text"]
+    DESCRIPTION = "Preview any input value as text."
 
     def main(self, source=None):
         torch.set_printoptions(edgeitems=6)
diff --git a/comfy_extras/nodes_primitive.py b/comfy_extras/nodes_primitive.py
index 7f90daf14..35761863f 100644
--- a/comfy_extras/nodes_primitive.py
+++ b/comfy_extras/nodes_primitive.py
@@ -10,11 +10,10 @@ class String(io.ComfyNode):
         return io.Schema(
             node_id="PrimitiveString",
             search_aliases=["text", "string", "text box", "prompt"],
-            display_name="Text String (DEPRECATED)",
+            display_name="Text",
             category="utilities/primitive",
             inputs=[io.String.Input("value")],
-            outputs=[io.String.Output()],
-            is_deprecated=True
+            outputs=[io.String.Output()]
         )
 
     @classmethod
@@ -28,7 +27,7 @@ class StringMultiline(io.ComfyNode):
         return io.Schema(
             node_id="PrimitiveStringMultiline",
             search_aliases=["text", "string", "text multiline", "string multiline", "text box", "prompt"],
-            display_name="Input Text",
+            display_name="Text (Multiline)",
             category="utilities/primitive",
             essentials_category="Basics",
             inputs=[io.String.Input("value", multiline=True)],
diff --git a/comfy_extras/nodes_video.py b/comfy_extras/nodes_video.py
index d3acc9ad0..3bfd00be4 100644
--- a/comfy_extras/nodes_video.py
+++ b/comfy_extras/nodes_video.py
@@ -81,7 +81,7 @@ class SaveVideo(io.ComfyNode):
             display_name="Save Video",
             category="video",
             essentials_category="Basics",
-            description="Saves the input images to your ComfyUI output directory.",
+            description="Saves the input videos to your ComfyUI output directory.",
             inputs=[
                 io.Video.Input("video", tooltip="The video to save."),
                 io.String.Input("filename_prefix", default="video/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."),
diff --git a/nodes.py b/nodes.py
index 31602e582..883258bd1 100644
--- a/nodes.py
+++ b/nodes.py
@@ -1709,6 +1709,7 @@ class PreviewImage(SaveImage):
         self.compress_level = 1
 
     SEARCH_ALIASES = ["preview", "preview image", "show image", "view image", "display image", "image viewer"]
+    DESCRIPTION = "Preview the images without saving them to the ComfyUI output directory."
 
     @classmethod
     def INPUT_TYPES(s):
diff --git a/tests-unit/comfy_test/test_seedvr2_dtype.py b/tests-unit/comfy_test/test_seedvr2_dtype.py
index 8e08b6dde..d743cc848 100644
--- a/tests-unit/comfy_test/test_seedvr2_dtype.py
+++ b/tests-unit/comfy_test/test_seedvr2_dtype.py
@@ -1,4 +1,5 @@
 import torch
+import torch.nn as nn
 
 from comfy.cli_args import args as cli_args
 
@@ -48,3 +49,31 @@ def test_seedvr2_vae_decode_memory_covers_full_frame_lab_transfer():
     assert estimate == 101 * 960 * 1280 * 160
     assert estimate > 15 * 1024 ** 3
     assert estimate > old_estimate * 100
+
+
+def test_seedvr2_vae_encode_preserves_compute_dtype(monkeypatch):
+    wrapper = seedvr_vae.VideoAutoencoderKLWrapper.__new__(seedvr_vae.VideoAutoencoderKLWrapper)
+    nn.Module.__init__(wrapper)
+    wrapper._dummy = nn.Parameter(torch.empty(1, dtype=torch.float16))
+    input_dtype = None
+
+    def encode(self, x):
+        nonlocal input_dtype
+        input_dtype = x.dtype
+        return x
+
+    monkeypatch.setattr(seedvr_vae.VideoAutoencoderKL, "encode", encode)
+
+    x = torch.zeros((1, 3, 1, 8, 8), dtype=torch.float32)
+    wrapper._encode_with_raw_latent(x)
+
+    assert input_dtype == torch.float32
+
+
+def test_seedvr2_vae_ops_cast_weights_to_compute_dtype():
+    attention = seedvr_vae.Attention(query_dim=4, heads=1, dim_head=4).to(torch.float16)
+    hidden_states = torch.zeros((1, 2, 4), dtype=torch.float32)
+
+    output = attention(hidden_states)
+
+    assert output.dtype == torch.float32
diff --git a/tests-unit/comfy_test/test_seedvr2_vae_tiled.py b/tests-unit/comfy_test/test_seedvr2_vae_tiled.py
index a2866b609..d64f51918 100644
--- a/tests-unit/comfy_test/test_seedvr2_vae_tiled.py
+++ b/tests-unit/comfy_test/test_seedvr2_vae_tiled.py
@@ -122,6 +122,31 @@ def test_tiled_vae_encode_uses_tensor_return_without_indexing():
     assert tuple(out.shape) == (2, _LATENT_CHANNELS, 1, 8, 8)
 
 
+def test_tiled_vae_preserves_compute_dtype_with_different_parameter_dtype():
+    class DummyVAE(nn.Module):
+        spatial_downsample_factor = 8
+        temporal_downsample_factor = 4
+        slicing_sample_min_size = 8
+
+        def __init__(self):
+            super().__init__()
+            self.device = torch.device("cpu")
+            self._dummy = nn.Parameter(torch.zeros(1, dtype=torch.float16))
+            self.input_dtype = None
+
+        def encode(self, t_chunk):
+            self.input_dtype = t_chunk.dtype
+            b, _, _, h, w = t_chunk.shape
+            return torch.ones((b, _LATENT_CHANNELS, 1, h // 8, w // 8), dtype=t_chunk.dtype)
+
+    vae = DummyVAE()
+    x = torch.zeros((1, 3, 1, 64, 64), dtype=torch.float32)
+
+    tiled_vae(x, vae, tile_size=(64, 64), tile_overlap=(16, 16), encode=True)
+
+    assert vae.input_dtype == torch.float32
+
+
 def test_tiled_vae_preserves_input_dtype_on_single_tile():
     class FloatOutputVAEModel(torch.nn.Module):
         def __init__(self):