AI Video Object Removal — Local-first, Fast, and Simple

VideoVanish brings professional-grade video inpainting to your desktop with a friendly UI. Under the hood it combines DiffuEraser for diffusion-based inpainting and SAM2 for robust masking — no cloud uploads, no account, no fuss.

⬇️ Download .zip ⭐ View on GitHub
Local-first DiffuEraser + SAM2 GUI & CLI Windows & Linux

Why VideoVanish?

  • Local-first workflow — keep your footage private and skip slow uploads and time limits.
  • Simple installer + GUI — get started without touching the command line.
  • State-of-the-art models — DiffuEraser for inpainting and SAM2 for precise masks.
  • Automation ready — command-line options for batch jobs and pipelines.

Quick Facts

  • Works best on a GPU with generous VRAM (Windows or Linux).
  • Mask inside the app (SAM2) or import a mask video you made elsewhere.
  • Adjust inference resolution; results are upscaled & blended back into the original.
  • Weights and model dependencies download automatically on first run.

Install — Windows

  1. Install Miniconda with latest installer (choose all defaults).
  2. Download VideoVanish and extract anywhere.
  3. Double-click windows_install.bat.
  4. Launch with start_videovanish.bat.

Install — Linux

git clone https://github.com/calledit/VideoVanish
cd VideoVanish
./install_videovanish.sh
conda activate videovanish
python videovanish.py

Usage — GUI

  • Load your color video and optional mask video.
  • No mask? Create one in the GUI with SAM2:
    • Left click = add point
    • Right click = remove point/box
    • Drag = define area
    • Click Generate Mask to build a mask video
  • Press Vanish to run DiffuEraser and remove objects.

Usage — CLI

Here are the main CLI entry points:

# Run main VideoVanish pipeline
python videovanish.py \
  --color_video input.mp4 \
  --mask_video mask.mp4 \
  --infilled_video output.mp4

# Create a colored mask video with SAM2
python sam2_masker.py \
  --color_video input.mp4 \
  --annotations ann.json \
  --start_frame 0 \
  --max_frames -1 \
  --out mask.mkv

# Run DiffuEraser to remove objects
python diffuerase.py \
  --color_video input.mp4 \
  --mask_video mask.mkv \
  --prior_video prior.mp4 \
  --start_frame 0 \
  --max_frames -1 \
  --out vanished.mkv

Video Tutorial

Walkthrough of the workflow from mask creation to export.

FAQ

Does VideoVanish upload my footage?

No. VideoVanish is local-first. Models run on your machine and assets stay with you.

What models are used?

DiffuEraser powers diffusion-based inpainting; SAM2 provides high-quality interactive segmentation for mask creation.

Where can I report issues or request features?

Open an issue on GitHub Issues.

Contributing

Pull requests are welcome. If you plan a larger change, please open an issue first to discuss what you would like to change.