_{If this project helps you, a coffee helps me keep working on it.}

Remove a background. Keep control of the mask.

AlphaCut is a desktop app for local image and video cutouts. Compare the original with the result, paint corrections into the mask and export for the editor you already use. There's also a command line for batch work and watched folders.

Download for Windows · Portable download · User guide

Windows downloads use CPU inference. Python isn't required. Install FFmpeg separately for video; image cutouts work without it. The downloads are unsigned, so Windows may show an unknown-publisher warning. Compare your download with the release checksums.

Actual AlphaCut window with a generated sample photograph. U2Net produced the cutout on CPU; the result wasn't retouched.

• Check the cutout before exporting. Drag the before/after divider or switch to a grayscale mask to spot holes and rough edges.
• Use a region selection and keep/remove brushes when the first mask needs work. Edge softness and spill controls are available for further adjustment.
• Export transparent PNGs or alpha video. Solid-color and image backgrounds let you prepare an opaque export too.
• Process a queue of images and videos, save presets or watch a folder from the command line.

The mask summary describes this preview frame. Its jitter-risk indicator is a heuristic, not a quality score for the whole video.

1. Open a photo with Browse.
2. Choose u2net for a general subject, or u2net_human_seg for a person.
3. Click Preview and inspect the mask. The processing buttons stay below the scrollable settings.
4. Use Start Processing and choose where to save the PNG.

The first use downloads the selected model. Download sizes vary, and larger models can take time to load. Once cached, that model can process media without a network connection.

Alpha support varies between players and editors. Test a short clip in your destination app. Animated GIF and WebP are available for short clips, with a memory check before export.

Use Python 3.12 or newer. Windows, Linux and macOS source profiles are available; this release was exercised on Windows with CPU inference.

git clone https://github.com/SysAdminDoc/AlphaCut.git
cd AlphaCut
python -m venv .venv
.venv\Scripts\python -m pip install -r requirements.txt
.venv\Scripts\python AlphaCut.py

On Linux/macOS, use .venv/bin/python. See the guide for CUDA, DirectML, CoreML and headless environments. Startup checks dependencies but never installs or removes packages.

A command-line image cutout:

.venv\Scripts\python AlphaCut.py -i photo.jpg -m u2net -o cutout.png

The complete CLI reference covers batch jobs, watch folders and video exports. Exact model names take priority; ambiguous names are rejected.

Your images and video are processed locally. AlphaCut checks GitHub for updates at startup and downloads uncached models from their configured sources. Settings, recent paths, logs and cached models are stored in ~/.alphacut.

Results depend on the subject, lighting and model. Fine hair, motion blur and low-contrast edges can need manual corrections. The video pipeline uses per-frame segmentation, not a dedicated temporal matting model. There's no blanket speed or quality guarantee.

.venv\Scripts\python -m pip install -r requirements-dev.txt
.venv\Scripts\python -m pytest -q
.\packaging\windows\build-release.ps1

The local build creates the Windows executable and installer, AlphaCut-windows.exe.sha256, AlphaCut-Setup-<version>.exe.sha256 and dist/SHA256SUMS.txt. Without a configured signing certificate these are produced unsigned.

User guide · Release notes · Original artwork and review archive · Report a problem

AlphaCut's source is MIT licensed. Windows downloads also contain PyQt6 (GPLv3) and other dependencies with their own terms. The combined Windows distribution is provided under GPLv3-compatible terms, with notices and source-rebuild information. Model files are separate downloads governed by their respective licenses.