Less is Enough: Training-Free Video Diffusion Acceleration via Runtime-Adaptive Caching

Xin Zhou^1*, Dingkang Liang^1*, Kaijin Chen¹, Tianrui Feng¹, Xiwu Chen², Hongkai Lin¹,
Yikang Ding², Feiyang Tan², Hengshuang Zhao³, Xiang Bai^1†

¹ Huazhong University of Science and Technology, ² MEGVII Technology, ³ University of Hong Kong

(*) Equal contribution. (†) Corresponding author.

---

This document provides the implementation for accelerating the HunyuanVideo model using EasyCache.

✨ Visual Comparison

EasyCache significantly accelerates inference speed while maintaining high visual fidelity.

Prompt: "A cat walks on the grass, realistic style." (Base Acceleration)

HunyuanVideo (Baseline, 544p, H20)	EasyCache (Ours)
Inference Time: ~2327s	Inference Time: ~1025s (2.3x Speedup)

Prompt: "A young man at his 20s is sitting on a piece of cloud in the sky, reading a book." (SVG with EasyCache)

HunyuanVideo (Baseline, 720p, H20)	SVG with EasyCache (Ours)
Inference Time: ~6572s	Inference Time: ~1773s (3.71x Speedup)

---

🚀 Usage Instructions

This section provides instructions for two settings: base acceleration with EasyCache alone and combined acceleration using EasyCache with SVG.

1. Base Acceleration (EasyCache Only)

a. Prerequisites ⚙️

Before you begin, please follow the instructions in the official HunyuanVideo repository to configure the required environment and download the pretrained model weights.

b. Copy Files 📂

Copy easycache_sample_video.py into the root directory of your local HunyuanVideo project.

c. Run Inference ▶️

Execute the following command from the root of the HunyuanVideo project to generate a video. To generate videos in 720p resolution, set the --video-size argument to 720 1280. You can also specify your own custom prompts.

python3 easycache_sample_video.py \
    --video-size 544 960 \
    --video-length 129 \
    --infer-steps 50 \
    --prompt "A cat walks on the grass, realistic style." \
    --flow-reverse \
    --use-cpu-offload \
    --save-path ./results \
    --seed 42

2. Combined Acceleration (SVG with EasyCache)

a. Prerequisites ⚙️

Ensure you have set up the environments for both HunyuanVideo and SVG.

b. Copy Files 📂

Copy hyvideo_svg_easycache.py into the root directory of your local HunyuanVideo project.

c. Run Inference ▶️

Execute the following command to generate a 720p video using both SVG and EasyCache for maximum acceleration. You can also specify your own custom prompts.

python3 hyvideo_svg_easycache.py \
        --video-size 720 1280 \
        --video-length 129 \
        --infer-steps 50 \
        --prompt "A young man at his 20s is sitting on a piece of cloud in the sky, reading a book." \
        --embedded-cfg-scale 6.0 \
        --flow-shift 7.0 \
        --flow-reverse \
        --use-cpu-offload \
        --save-path ./results \
        --output_path ./results \
        --pattern "SVG" \
        --num_sampled_rows 64 \
        --sparsity 0.2 \
        --first_times_fp 0.055 \
        --first_layers_fp 0.025 \
        --record_attention \
        --seed 42

📊 Evaluating Video Similarity

We provide a simple script to quickly evaluate the similarity between two videos (e.g., the baseline result and your generated result) using common metrics.

Usage

# install required packages.
pip install lpips numpy tqdm torchmetrics

python tools/video_metrics.py --original_video video1.mp4 --generated_video video2.mp4

• --original_video: Path to the first video (e.g., the baseline).
• --generated_video: Path to the second video (e.g., the one generated with EasyCache).

🌹 Acknowledgements

We would like to thank the contributors to the HunyuanVideo, and SVG repositories, for their open research and exploration.

📖 Citation

If you find this repository useful in your research, please consider giving a star ⭐ and a citation.

@article{zhou2025easycache,
  title={Less is Enough: Training-Free Video Diffusion Acceleration via Runtime-Adaptive Caching},
  author={Zhou, Xin and Liang, Dingkang and Chen, Kaijin and and Feng, Tianrui and Chen, Xiwu and Lin, Hongkai and Ding, Yikang and Tan, Feiyang and Zhao, Hengshuang and Bai, Xiang},
  journal={arXiv preprint arXiv:2507.02860},
  year={2025}
}