VideoMaMa++

VideoMaMa++

An interactive demo of video matting that stays consistent over time. SAM3 tracks an object you click or name, and VideoMaMa++ turns the track into an alpha matte. Non-commercial license.

Made by SammyLim on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: This Space is the interactive demo for VideoMaMa++, a video matting method from a NeurIPS 2026 paper. Matting cuts a subject out of a video as a soft alpha matte, which is a per-pixel see-through mask.

SAM3 tracks the object you click on or name. VideoMaMa++ then turns that mask track into a matte that stays consistent across chunk boundaries of the video.

Who it suits: It suits video and computer vision researchers who want to try consistent matting, and effects artists who want to see a research-grade cutout in the browser.

Industry | Who uses it | How they use it | Example
Computer vision | Researchers | Click an object and inspect the matte | Checking consistency at chunk boundaries
Visual effects | Compositors | Try the matte on a sample clip | Judging edges around hair or fine detail
Video editing | Editors | Test cutting out a subject for non-commercial work | Previewing a background swap
AI development | Developers | Read app.py, sam3_tracker.py and painter.py | Learning how a tracker feeds a matting model

Skip it if your use is commercial, since the license limits the work to non-commercial use.

Where to use it: The app runs in these places.

• A web browser, on the Space's page at Hugging Face
• A local copy on Windows, macOS or Linux, running Python 3.12
• Hugging Face's zero-a10g hardware, where the online Space runs

Use case | Who | What they get back
Matte a tracked subject | Researchers | An alpha matte that stays consistent across the clip
Try the sample clips | Anyone curious | Results on the clips in the samples folder
Study the pipeline | Developers | A tracker module and a painter module to read

How to set it up: Online you only need a browser. A local copy needs these.

• Python 3.12 on Windows, macOS or Linux
• The pip command
• Room on disk for the SAM3 and VideoMaMa++ models

1. Install the huggingface_hub package.
2. Download the Space's files into a folder with the hf command.
3. Move into that folder.
4. Install gradio at version 6.29.1.
5. Install the packages listed in requirements.txt.
6. Start the app with python app.py.

pip install -U huggingface_hub
hf download SammyLim/VideoMaMa-PlusPlus --repo-type space --local-dir VideoMaMa-PlusPlus
cd VideoMaMa-PlusPlus
pip install gradio==6.29.1
pip install -r requirements.txt
python app.py

How to use it: Open the Space's page, or run the command below in your copy and open the page the app serves. Then add a video, click the object you want or name it, and run it.

python app.py

Example: You open the Space, load a sample clip and click on the person in the frame. SAM3 tracks that person, and the app returns a matte of them that holds steady from one chunk of the clip to the next.

For newcomers and experts: If you are new, open the Space's page and try one of the sample clips first. Click a clear, single subject, because it makes the matte easier to judge.

If you know the field, SAM3 supplies a mask track and VideoMaMa++ refines it into an alpha matte using preserve-and-refine tokens to keep consecutive chunks consistent. The repository splits the work into sam3_tracker.py, painter.py and app.py. Online it runs on zero-a10g hardware with Gradio 6.29.1.

Open the Space and add a video
Click or name the object to track
SAM3 tracks it across frames
VideoMaMa++ makes a consistent matte

Good to know: The Space names two models, facebook/sam3 and SammyLim/VideoMaMa-PlusPlus. The repository has 12 files in all, including a samples folder.

• License: cc-by-nc-4. The NC part means non-commercial use only
• Cost to run: none named
• Data it reaches: the SAM3 and VideoMaMa++ model repositories, and the video you give it
• Not for: commercial use, since the license is non-commercial

On Hugging Face, 13 November 2026

• Likes: 3
• Built with: gradio
• License: cc-by-nc-4.0

Links

• Space: https://huggingface.co/spaces/SammyLim/VideoMaMa-PlusPlus
• Files: https://huggingface.co/spaces/SammyLim/VideoMaMa-PlusPlus/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.