* initial gemma4 support
* parity with reference implementation
outputs can 100% match transformers with same sdpa flags, checkpoint this and then optimize
* Cleanup, video fixes
* cleanup, enable fused rms norm by default
* update comment
* Cleanup
* Update sd.py
* Various fixes
* Add fp8 scaled embedding support
* small fixes
* Translate think tokens
* Fix image encoder attention mask type
So it works with basic attention
* Handle thinking tokens different only for Gemma4
* Code cleanup
* Update nodes_textgen.py
* Use embed scale class instead of buffer
Slight difference to HF, but technically more accurate and simpler code
* Default to fused rms_norm
* Update gemma4.py
SolidMask had a hardcoded device="cpu" while other nodes (e.g.
EmptyImage) follow intermediate_device(). This causes a RuntimeError
when MaskComposite combines masks from different device sources
under --gpu-only.
- SolidMask: use intermediate_device() instead of hardcoded "cpu"
- MaskComposite: align source device to destination before operating
Co-authored-by: Alexis Rolland <alexisrolland@hotmail.com>
Co-authored-by: Jedrzej Kosinski <kosinkadink1@gmail.com>
* Change save 3d model's filename prefix to 3d/ComfyUI
As this node has already changed from `Save GLB` to `Save 3D Model`, using the filename prefix `3d` will be better than `mesh`
* use lowercase
---------
* initial RIFE support
* Also support FILM
* Better RAM usage, reduce FILM VRAM peak
* Add model folder placeholder
* Fix oom fallback frame loss
* Remove torch.compile for now
* Rename model input
* Shorter input type name
---------
Mesh-producing nodes (VoxelToMeshBasic, VoxelToMesh, VaeDecodeShapeTrellis)
previously stacked per-batch vertex/face tensors with torch.stack, which
failed under batch>1 because per-item meshes have variable shapes. Store
per-item tensors as lists when shapes differ; keep stacked-tensor fast
path when shapes match. Update SaveGLB, PostProcessMesh, and
VaeDecodeTextureTrellis consumers to iterate per-item when input is a
list.
Trellis2Conditioning.execute previously collapsed batched image/mask
input to index 0, yielding identical conditioning for every batch item.
Loop over the batch and produce per-image cond_512/cond_1024/neg_cond
tensors stacked along the batch dim, matching the latent batch size.
batch_size=1 behavior is unchanged. batch_size>1 runs now emit one GLB
per batch item per SaveGLB node and carry per-image conditioning
through the structure/shape/texture cascade.
Yousef R. Gamaleldin
drozbay
Alexis Rolland
Alvin Tang
Talmaj
Jukka Seppänen
Jedrzej Kosinski
Gilad Schreiber
Daxiong (Lin)
comfyanonymous
John Pollock
John Pollock
rattus