wan2.1-t2v-14b-controlnet-canny-v1

Dilated Controlnet for Wan2.1 (canny)

<video controls autoplay src="https://cdn-uploads.huggingface.co/production/uploads/63fde49f6315a264aba6a7ed/bA5U06VI9Bs6DjmXJ7Z5S.mp4"></video>

This repo contains the code for dilated controlnet module for Wan2.1 model.
Dilated controlnet has less basic blocks and also has stride parameter. For Wan14B model controlnet blocks count = 6 and stride = 4.
See <a href="https://github.com/TheDenk/wan2.1-dilated-controlnet">Github code</a>.

How to

Clone repo

git clone https://github.com/TheDenk/wan2.1-dilated-controlnet.git
cd wan2.1-dilated-controlnet

Create venv

python -m venv venv
source venv/bin/activate

Install requirements

pip install -r requirements.txt

Inference examples

Inference with cli

python -m inference.cli_demo \
    --video_path "resources/physical-4.mp4" \
    --prompt "A balloon filled with water was thrown to the ground, exploding and splashing water in all directions. There were graffiti on the wall, studio lighting, and commercial movie shooting." \
    --controlnet_type "canny" \
    --controlnet_stride 4 \
    --base_model_path Wan-AI/Wan2.1-T2V-14B-Diffusers \
    --controlnet_model_path TheDenk/wan2.1-t2v-14b-controlnet-canny-v1

Inference with Gradio

python -m inference.gradio_web_demo \
    --controlnet_type "canny" \
    --base_model_path Wan-AI/Wan2.1-T2V-14B-Diffusers \
    --controlnet_model_path TheDenk/wan2.1-t2v-14b-controlnet-canny-v1

Detailed Inference

python -m inference.cli_demo \
    --video_path "resources/physical-4.mp4" \
    --prompt "A balloon filled with water was thrown to the ground, exploding and splashing water in all directions. There were graffiti on the wall, studio lighting, and commercial movie shooting." \
    --controlnet_type "canny" \
    --base_model_path Wan-AI/Wan2.1-T2V-14B-Diffusers \
    --controlnet_model_path TheDenk/wan2.1-t2v-14b-controlnet-canny-v1 \
    --controlnet_weight 0.8 \
    --controlnet_guidance_start 0.0 \
    --controlnet_guidance_end 0.8 \
    --controlnet_stride 4 \
    --num_inference_steps 50 \
    --guidance_scale 5.0 \
    --video_height 480 \
    --video_width 832 \
    --num_frames 81 \
    --negative_prompt "Bright tones, overexposed, static, blurred details, subtitles, style, works, paintings, images, static, overall gray, worst quality, low quality, JPEG compression residue, ugly, incomplete, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, misshapen limbs, fused fingers, still picture, messy background, three legs, many people in the background, walking backwards" \
    --seed 42 \
    --out_fps 16 \
    --output_path "result.mp4"

Acknowledgements

Original code and models Wan2.1.

Citations

@misc{TheDenk,
    title={Dilated Controlnet},
    author={Karachev Denis},
    url={https://github.com/TheDenk/wan2.1-dilated-controlnet},
    publisher={Github},
    year={2025}
}

Contacts

<p>Issues should be raised directly in the repository. For professional support and recommendations please <a>[email protected]</a>.</p>