Runcrate Logoruncrate

Kijai/WanVideo_comfy_fp8_scaled

diffusion-single-filediffusion-single-filecomfyuibase_model:Wan-AI/Wan2.1-VACE-1.3Bbase_model:finetune:Wan-AI/Wan2.1-VACE-1.3Blicense:apache-2.0region:usapache-2.0
474
HuggingFaceHuggingFace
825.8K

Better fp8 scaled models (when measured against fp16) based on quantization code from https://github.com/Tencent-Hunyuan/HunyuanVideo/blob/main/hyvideo/modules/fp8_optimization.py

Can be used with: https://github.com/kijai/ComfyUI-WanVideoWrapper (latest version) and ComfyUI native WanVideo nodes.

14B-T2V comparison test without LoRAs, 25 steps, 832x480x81

<video controls autoplay width=50% src=https://cdn-uploads.huggingface.co/production/uploads/63297908f0b2fc94904a65b8/DwlAGbj20it1unZW54NDC.mp4>

2.2 A14B-T2V test

<video controls autoplay width=50% src=https://cdn-uploads.huggingface.co/production/uploads/63297908f0b2fc94904a65b8/6A_AZ7GN_uxeRH0vwsWkH.mp4>

<video controls autoplay width=50% src=https://cdn-uploads.huggingface.co/production/uploads/63297908f0b2fc94904a65b8/GpuqQ4YwoR3kjxkhuvP8P.mp4>

The e5m2 marked as v2 is the one uploaded here and these are all scaled even if I forgot to label properly.

DEPLOY IN 60 SECONDS

Run WanVideo_comfy_fp8_scaled on Runcrate

Deploy on H100, A100, or RTX GPUs. Pay only for what you use. No setup required.