Hunyuan Video

Note Models marked as "Kijai" have the full vision model and blocks - They only work with the Kijai Nodes

Use Comfy Native models for Comfy native nodes

• Converted to safetensors.

• Comfy Native Node Users do not use Kajai TE use the Scaled Version

• For CLIP-L they recommend using the full vision model. I have merged Zer0Int and SimV4 into the Vision Model

• If you have low VRAM but high system RAM it is possible to use the FP32 CLIP and VAE (USE TILED VAE at 128px)

• GBlue has posted a working FP8 workflow using native COMFY nodes on a 12GB card. (It will work on a 8GB card if the video latent size is around 320px)

Using the Kajai marked models on COMFY native will cause rainbow or black output.

I have posed a FP8 VAE that works with Comfy Native but may take more time then BF16 or even FP32