LTX-2.5 distilled NVFP4 fails in ComfyUI
Is anyone else able to run the officialltx-2.5-22b-distilled-transformer-nvfp4.safetensorscheckpoint in ComfyUI?
I used ComfyUI’s built-in LTX 2.5 Image to Video template. Sampling fails immediately with:
RuntimeError: mat1 and mat2 shapes cannot be multiplied (3520x4096 and 2048x4096)
My environment:
- ComfyUI 0.32.0
- Ubuntu 24.04.4 LTS
- NVIDIA RTX 5090
- PyTorch 2.13.0+cu130
The official ltx-2.5-22b-distilled-transformer-comfy-int8-convrot.safetensors
checkpoint works with the same template.
I’m not sure whether this is related to the checkpoint, ComfyUI’s loading path,
or something specific to my setup. Can anyone confirm whether the NVFP4
checkpoint works?
NVFP4 doesn't work for me.
It’s not working for me either. I’m getting an error too.
Thank you for reporting about this issue.
We are on it!
This one works for me, or at least it was the only one i could find for ComfyUI :
https://huggingface.co/BennyDaBall/LTX-2.5-22b-distilled-nvfp4-comfy/tree/main
@YieumYoon , @hackeys , @Killerrr2008
We fixed the issue.
NVFP4 checkpoint should be working both in ComfyUI and in the LTX-2 inference pipeline