Good questions.
Clear answers.

Yes. The AI models are bundled in the installer, and image processing runs on your computer. The Windows installer may need internet access to install Microsoft WebView2 if it is missing. GitHub and community links also open online pages.

Open PNG, JPG, WebP, BMP, TIFF, and GIF images by browsing, dragging them in, or pasting from the clipboard. Export to PNG, JPG, or WebP. Use PNG for a transparent result; JPG does not support transparency. GIF input is supported as an image, not animated background removal.

The current release has Windows x64, macOS Apple Silicon, and Linux x86_64 installers. Windows 10 or 11 and an AVX2-capable processor are required on Windows. Linux requires Ubuntu 24.04 or similarly recent system libraries, plus AVX2. Intel Macs are not supported. macOS and Linux builds have had less testing than Windows.

No GPU is required. CutOut runs on the CPU. The repository recommends at least 8 GB of RAM; Best and Best+ can briefly use several gigabytes during processing. Runtime varies with your computer, image, and chosen model.

Fast uses U²-Net-P for simple subjects and quick checks. Balanced uses IS-Net general use for everyday work. Best uses BiRefNet Lite for finer details such as hair. Best+ uses the full BiRefNet fp16 model and takes longer. Results depend on the image; model input resolution is separate from export resolution.

CutOut does not connect to the internet to process images. Your photos stay on your computer. This website contacts the public GitHub API to check release information; it has no image upload feature and includes no analytics or tracking.

All four AI models are included so they are available offline after installation. The download page shows the actual file size reported by GitHub for each installer.

CutOut is not code-signed on Windows or notarized by Apple. Windows may show a SmartScreen warning, and macOS may block the first launch. Read the repository’s installation steps and verify that you downloaded from the official GitHub release. On Linux, the AppImage needs executable permission. Links to the exact instructions are on the download page.

CutOut’s application code is available under the MIT license, which permits commercial use subject to its terms. Bundled models and libraries have separate licenses listed in Third-party notices. You remain responsible for the rights to the images you process.

Yes. Batch mode accepts multiple selected images or an entire folder, including subfolders. Process and export the batch together. Format and background options also work in batch mode.

Exports keep the original image dimensions. Each AI model works at its own internal input resolution, then the mask is mapped back to the original image. An original-size export does not guarantee that every fine detail will be preserved.

Updates are manual: get the newer installer from GitHub Releases. Report problems through the repository’s Issues page or join the Discord community. If troubleshooting, the repository asks for the cutout-perf.log file from your temporary directory.