ComfyUI_FlashVSR

ComfyUI_FlashVSR

FlashVSR: Towards Real-Time Diffusion-Based Streaming Video Super-Resolution,this node ,you can use it in comfyUI FlashVSR_Plus,FlashVSR+ is an efficient one-step diffusion framework for high-quality video super-resolution, enabling spatially untiled 4K inference on a single 24GB RTX 4090 GPU.

Update

• 使用comfyUI的核心封装flashVSR，去掉transformer的版本限制，利用comfyUI的加卸载机制防止OOM（推理超出硬件能力会很慢，但不会OOM）/ Use comfyUI core flashVSR,remove transformer version limit,use comfyUI's add/remove mechanism to prevent OOM (inference will be very slow if it exceeds the hardware capabilities, but it will not OOM)
• Use new loop / 换一个loop切分超长视频
• 支持flashVSR-Plus 模型，plus模式 请在菜单手动选triton-flashvsr_attn，否则还会是 自动选择Block-Sparse-Attention ，support flashVSR-Plus model，plus use triton attention，or auto will choice Block-Sparse-Attention

1.Installation

In the ./ComfyUI/custom_nodes directory, run the following:

git clone https://github.com/smthemex/ComfyUI_FlashVSR

2.requirements

pip install -r requirements.txt

• 2.1 Best quality need 最佳画质需要的依赖
  2.1.1 Block-Sparse-Attention wheel Block-Sparse-Attentiontorch2.8 cu2.8 py311,torch2.12+cu130 py311 py313 wheel or CU128 toch2.7 or Block-Sparse-Attention wheel
  2.1.2 or install flashvsr_attn origin

git clone https://github.com/mit-han-lab/Block-Sparse-Attention  # if 5090 or newest cuda
# git clone https://github.com/smthemex/Block-Sparse-Attention # 无须梯子强制编译 可选
# git clone https://github.com/lihaoyun6/Block-Sparse-Attention # 须梯子 可选
cd Block-Sparse-Attention
pip install packaging
pip install ninja
python setup.py install

• 2.2 optional requirements (flashvsr_attn for plus version) plus版本的可选依赖flashvsr_attn（只用plus且编译麻烦就用这个）
  in node dir 在插件目录下运行

pip install -e operators/flashvsr_attn --no-build-isolation --no-deps

# if python_embeded/ 如果是便携包(也可以复制operators整个文件夹到你的便携python包下，下面的安装命令加上 python -m)

python_embeded/python -m pip install -e you_path/operators/flashvsr_attn --no-build-isolation --no-deps

3.checkpoints

• 3.1 FlashVSRv1.0 ；FlashVSRv1.1 ；FlashVSRv-plus all checkpoints 所有模型，vae 用常规的wan2.1
• 3.2 version 1.0/1.1 emb posi_prompt.pth plus's emb posi_prompt.pth

├── ComfyUI/models/FlashVSR
|     ├── LQ_proj_in.ckpt # v1.1 or v1.0
|     ├── diffusion_pytorch_model_streaming_dmd.safetensors #v1.1 or v1.0，rename 最好重命名
|     ├── posi_prompt.pth
|     ├── lq_proj_in_plus_turbo.safetensors   #rename turbo or base
|     ├── base_dit_plus.safetensors #rename base or turbo
├── ComfyUI/models/vae
|        ├──Wan2.1_VAE.pth # or safetensors  # optional/可选
|        ├──TCDecoder_plus.ckpt/TCDecoder.ckpt   # optional/可选 ,注意旧插件TCD不在vae目录下，需要手动剪切过去

Example

Previous

• 如果觉得项目有用，请给官方项目FlashVSR and FlashVSR_Plus打星；
• Block-Sparse-Attention 源码库已更新，支持5090系显卡及更新的设备。注意，需要cuda>=12.8
• 编译Block-Sparse-Attention window的轮子 可以使用 smthemex 强制编译版 或者 lihaoyun6 要联网 两个fork来
• plus 和 long 模式仅支持tcd解码/plus and long mode only support tcd decoder

Citation

@misc{yao2026flashvsrefficientvideosuperresolution,
      title={FlashVSR+: Efficient Video Super-Resolution via Hybrid Attention and Reward-Yield Distribution Matching}, 
      author={Ruchang Yao and Shi Guo and Junhao Zhuang and Mingde Yao and Xin Cai and Jiahao Zhan and Bin Fu and Xinyuan Chen and Yu Qiao and Tianfan Xue},
      year={2026},
      eprint={2610.17552},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2610.17552}, 
}

@article{zhuang2025flashvsr,
  title={FlashVSR: Towards Real-Time Diffusion-Based Streaming Video Super-Resolution},
  author={Zhuang, Junhao and Guo, Shi and Cai, Xin and Li, Xiaohui and Liu, Yihao and Yuan, Chun and Xue, Tianfan},
  journal={arXiv preprint arXiv:2510.12747},
  year={2025}
}