metadata
license: other
license_name: stabilityai-community-license
license_link: >-
https://huggingface.co/stabilityai/stable-video-diffusion-img2vid/blob/main/LICENSE.md
library_name: diffusers
pipeline_tag: image-to-image
VideoMaMa: Mask-Guided Video Matting via Generative Prior
Sangbeom Lim 路 Seoung Wug Oh 路 Jiahui Huang 路 Heeji Yoon 路 Seungryong Kim 路 Joon-Young Lee
[Paper] [Project Page] [GitHub] [Gradio Demo]
VideoMaMa (Video Mask-to-Matte Model) is a framework that converts coarse segmentation masks into pixel-accurate alpha mattes by leveraging pretrained video diffusion models. It demonstrates strong zero-shot generalization to real-world footage, even though it is trained solely on synthetic data.
Inference
To use VideoMaMa for inference, you can use the script provided in the official repository:
python inference_onestep_folder.py \
--base_model_path "stabilityai/stable-video-diffusion-img2vid-xt" \
--unet_checkpoint_path "SammyLim/VideoMaMa" \
--image_root_path "/path/to/your/images" \
--mask_root_path "/path/to/your/masks" \
--output_dir "./output" \
--keep_aspect_ratio
License
The VideoMaMa model checkpoints (specifically unet/* and dino_projection_mlp.pth) are subject to the Stability AI Community License. By using this model, you agree to the terms outlined in the license agreement.
Citation
@article{lim2026videomama,
title={VideoMaMa: Mask-Guided Video Matting via Generative Prior},
author={Lim, Sangbeom and Oh, Seoung Wug and Huang, Jiahui and Yoon, Heeji and Kim, Seungryong and Lee, Joon-Young},
journal={arXiv preprint arXiv:2601.14255},
year={2026}
}