V3

- HunYuan Video Lora support

- Add resize for clipvision

This lora support... works really well. This is gonna be a problem with deepfakes.... a big problem.

Update your framepack nodes.

Tips

Latent window size of 12 works well

Embed Interpolation disabled unless using 2 images

gpu memory, lower for more. 4 for 12gb, 6 for 8gb, 8 for 6gb

Goes faster the lower this is. (as long as your vram does not max out)

F1 Framepack models

Fp16 and fp8 I have not tested these, i assume they work.

https://huggingface.co/kabachuha/FramePack_F1_I2V_HY_20250503_comfy/tree/main

Updated 2.1 with hotfix, i dont know if it was needed but its there. - changed to crop from resize - Ensure you set model to sdpa/flash attention/sage as you need

Uploading v2.1 to ensure there is no other issues.

V2.1

- updated front end for comfy

- Fixed bucket sizing

V2 - First Last Frames

-Added last frame.

-bucket size has been fixed :)

-fixed things from lastest comfy update.

1.1 - small changes before i try the wan 2.1 version of this.

Added rescale render size

Added vram clear nodes

Added notes on a few more things.

Uses

Model links:

I have not changed much from the original Just a few qol changes

Original is in your custom nodes folder frame pack folder examples.

This has a large model, but the workflow can use as little at 6gb vram

Model is based on Hunyuan, so if you know how to prompt for that, this should do similar

Hopefully we get native nodes for this stuff.

Nodes

https://github.com/kijai/ComfyUI-FramePackWrapper

Models

https://huggingface.co/Kijai/HunyuanVideo_comfy/blob/main/FramePackI2V_HY_fp8_e4m3fn.safetensors

https://huggingface.co/Kijai/HunyuanVideo_comfy/blob/main/FramePackI2V_HY_bf16.safetensors

sigclip:

https://huggingface.co/Comfy-Org/sigclip_vision_384/tree/main

text encoder and VAE:

https://huggingface.co/Comfy-Org/HunyuanVideo_repackaged/tree/main/split_files