Nodes used in the process: https://github.com/TTPlanetPig/TTP_Comfyui_FramePack_SE
First of all, thanks to lvmin Zhang for the FramePack application—it provides a very interesting method that makes video generation simpler and more accessible.
Original repository:
https://github.com/lllyasviel/FramePack
Changes in this project
Based on the original repository, we injected a simple end_image to enable starting and ending frame references
View my PR for the full differences:
https://github.com/lllyasviel/FramePack/pull/167
This PR addresses the criticism of "background freezing"
Issues encountered
When the starting and ending frames differ too greatly, the model struggles to process them, often resulting in "slideshow-like" transitions.
Although some attribute this to "the need for further training," I believe:
The Hunyuan model handles static human poses well but lacks smooth dynamic transitions
With lvmin Zhang's improvements to the Hunyuan base model, we can unlock more possibilities
My optimizations
Adjusted the generation pipeline code to balance variability and frame consistency
Tested and fine-tuned multiple parameters to ensure smooth transitions in most scenarios
The ComfyUI code is referenced from HM RunningHub/ComfyUI_RH_FramePack and modified
Model download and storage location
You can manually download the models from Hugging Face or use the bundled model package.
1. Manual download
HunyuanVideo
https://huggingface.co/hunyuanvideo community/HunyuanVideo/tree/main
Flux Redux BFL
https://huggingface.co/lllyasviel/flux_redux_bfl/tree/main
FramePackI2V
https://huggingface.co/lllyasviel/FramePackI2V_HY/tree/main
Baidu Netdisk: https://pan.baidu.com/s/17h23yvJXa6AczGLcybsd_A?pwd=mbqa
Quark: https://pan.quark.cn/s/80ff4f39c15b
2. Model storage location
Copy the content to the models/ folder, referring to HM RunningHub/ComfyUI_RH_FramePack
text
Copy
Edit
comfyui/models/
├── flux_redux_bfl
│ ├── feature_extractor/
│ │ └── preprocessor_config.json
│ ├── image_embedder/
│ │ ├── config.json
│ │ └── diffusion_pytorch_model.safetensors
│ ├── image_encoder/
│ │ ├── config.json
│ │ └── model.safetensors
│ ├── model_index.json
│ └── README.md
├── FramePackI2V_HY
│ ├── config.json
│ ├── diffusion_pytorch_model 00001 of 00003.safetensors
│ ├── diffusion_pytorch_model 00002 of 00003.safetensors
│ ├── diffusion_pytorch_model 00003 of 00003.safetensors
│ ├── diffusion_pytorch_model.safetensors.index.json
│ └── README.md
└── HunyuanVideo
├── config.json
├── model_index.json
├── README.md
├── scheduler/
│ └── scheduler_config.json
├── text_encoder/
│ ├── config.json
│ ├── model 00001 of 00004.safetensors
│ ├── model 00002 of 00004.safetensors
│ ├── model 00003 of 00004.safetensors
│ ├── model 00004 of 00004.safetensors
│ └── model.safetensors.index.json
├── text_encoder_2/
│ ├── config.json
│ └── model.safetensors
├── tokenizer/
│ ├── special_tokens_map.json
│ ├── tokenizer_config.json
│ └── tokenizer.json
├── tokenizer_2/
│ ├── merges.txt
│ ├── special_tokens_map.json
│ ├── tokenizer_config.json
│ └── vocab.json
└── vae/
├── config.json
└── diffusion_pytorch_model.safetensors
Parameter guide
padding_mode
Still in the experimental stage—currently using optimized
end_condition_strength & enable_feature_fusion can be combined; lower strength reduces tail frame matching
Lower end_condition_strength increases freedom but reduces similarity to the ending frame
history_weight
Controls historical influence, default 100%
history_decay
Linearly decays historical weight; for more variation, increase decay
Feel free to share feedback or suggestions in Issues or PRs!
Nodes used in the process: https://github.com/TTPlanetPig/TTP_Comfyui_FramePack_SE
First of all, thanks to lvmin Zhang for the FramePack application—it provides a very interesting method that makes video generation simpler and more accessible.
Original repository:
https://github.com/lllyasviel/FramePack
Changes in this project
Based on the original repository, we injected a simple end_image to enable starting and ending frame references
View my PR for the full differences:
https://github.com/lllyasviel/FramePack/pull/167
This PR addresses the criticism of "background freezing"
Issues encountered
When the starting and ending frames differ too greatly, the model struggles to process them, often resulting in "slideshow-like" transitions.
Although some attribute this to "the need for further training," I believe:
The Hunyuan model handles static human poses well but lacks smooth dynamic transitions
With lvmin Zhang's improvements to the Hunyuan base model, we can unlock more possibilities
My optimizations
Adjusted the generation pipeline code to balance variability and frame consistency
Tested and fine-tuned multiple parameters to ensure smooth transitions in most scenarios
The ComfyUI code is referenced from HM RunningHub/ComfyUI_RH_FramePack and modified
Model download and storage location
You can manually download the models from Hugging Face or use the bundled model package.
1. Manual download
HunyuanVideo
https://huggingface.co/hunyuanvideo community/HunyuanVideo/tree/main
Flux Redux BFL
https://huggingface.co/lllyasviel/flux_redux_bfl/tree/main
FramePackI2V
https://huggingface.co/lllyasviel/FramePackI2V_HY/tree/main
Baidu Netdisk: https://pan.baidu.com/s/17h23yvJXa6AczGLcybsd_A?pwd=mbqa
Quark: https://pan.quark.cn/s/80ff4f39c15b
2. Model storage location
Copy the content to the models/ folder, referring to HM RunningHub/ComfyUI_RH_FramePack
text
Copy
Edit
comfyui/models/
├── flux_redux_bfl
│ ├── feature_extractor/
│ │ └── preprocessor_config.json
│ ├── image_embedder/
│ │ ├── config.json
│ │ └── diffusion_pytorch_model.safetensors
│ ├── image_encoder/
│ │ ├── config.json
│ │ └── model.safetensors
│ ├── model_index.json
│ └── README.md
├── FramePackI2V_HY
│ ├── config.json
│ ├── diffusion_pytorch_model 00001 of 00003.safetensors
│ ├── diffusion_pytorch_model 00002 of 00003.safetensors
│ ├── diffusion_pytorch_model 00003 of 00003.safetensors
│ ├── diffusion_pytorch_model.safetensors.index.json
│ └── README.md
└── HunyuanVideo
├── config.json
├── model_index.json
├── README.md
├── scheduler/
│ └── scheduler_config.json
├── text_encoder/
│ ├── config.json
│ ├── model 00001 of 00004.safetensors
│ ├── model 00002 of 00004.safetensors
│ ├── model 00003 of 00004.safetensors
│ ├── model 00004 of 00004.safetensors
│ └── model.safetensors.index.json
├── text_encoder_2/
│ ├── config.json
│ └── model.safetensors
├── tokenizer/
│ ├── special_tokens_map.json
│ ├── tokenizer_config.json
│ └── tokenizer.json
├── tokenizer_2/
│ ├── merges.txt
│ ├── special_tokens_map.json
│ ├── tokenizer_config.json
│ └── vocab.json
└── vae/
├── config.json
└── diffusion_pytorch_model.safetensors
Parameter guide
padding_mode
Still in the experimental stage—currently using optimized
end_condition_strength & enable_feature_fusion can be combined; lower strength reduces tail frame matching
Lower end_condition_strength increases freedom but reduces similarity to the ending frame
history_weight
Controls historical influence, default 100%
history_decay
Linearly decays historical weight; for more variation, increase decay
Feel free to share feedback or suggestions in Issues or PRs!