NaviCache for HunyuanVideo

NaviCache is a training-free test-time self-calibration caching method for accelerating video generation. This directory provides the NaviCache sampling script for HunyuanVideo.

Usage

Follow HunyuanVideo to clone the repo, finish the installation, and download the required model weights. Then copy navicache_sample_video.py in this repo to the HunyuanVideo repo.

git clone https://github.com/Tencent-Hunyuan/HunyuanVideo.git
git clone https://github.com/HelloZicky/NaviCache.git

cp NaviCache/NaviCache4HunyuanVideo/navicache_sample_video.py HunyuanVideo/
cd HunyuanVideo

For 544 x 960 generation, set --video-size to 544 960:

python3 navicache_sample_video.py \
    --video-size 544 960 \
    --video-length 129 \
    --infer-steps 50 \
    --prompt "A cat walks on the grass, realistic style." \
    --use-cpu-offload \
    --navicache_thresh 0.025 \
    --navicache_align_steps 5 \
    --save-path ./navicache_results

For 720 x 1280 generation, set --video-size to 720 1280:

python3 navicache_sample_video.py \
    --video-size 720 1280 \
    --video-length 129 \
    --infer-steps 50 \
    --prompt "A cat walks on the grass, realistic style." \
    --use-cpu-offload \
    --navicache_thresh 0.025 \
    --navicache_align_steps 5 \
    --save-path ./navicache_results

The generated video and time_cost.json are saved under --save-path.

Acknowledgements

We would like to thank the contributors to HunyuanVideo.