Pixl.pet uses BEN2, a state-of-the-art background-matting neural network. Your image is resized to 1024×1024, processed through the model to produce an alpha mask, then the mask is upscaled back to your original resolution using a guided filter for crisp, natural-looking edges.

All of this runs inside a Web Worker in your browser — nothing is sent anywhere.

The output is a transparent PNG. The foreground subject is kept at full quality; the background is replaced with full transparency. You can overlay it on any colour or new background in any design app.

The BEN2 model is approximately 235 MB (fp16 precision). On a fast connection this downloads in about 15–30 seconds. After the first visit the model is stored in your browser's Cache API — subsequent uses are instant.

Input: PNG, JPEG and WebP. Output: transparent PNG. The image is processed at its full original resolution.

Yes. Once the AI model is cached and the site's Service Worker has stored all static assets, the tool works completely offline — no internet connection required.

Pixl.pet tries to use the WebGPU API for hardware-accelerated inference. If WebGPU is unavailable (e.g., in Firefox or some Safari versions) or encounters a runtime error, it automatically falls back to multi-threaded WebAssembly.

Never. The full pipeline — model inference, mask post-processing and PNG export — runs inside a Web Worker locally in your browser. No pixel data leaves your device at any point.