AI background remover that runs on your device
This background remover uses two AI neural networks to separate the subject from the background. Both models run entirely inside your browser, so your photos never touch a server.
How the AI removes backgrounds
- The AI loads once. The first time you use a mode, the model downloads (4.4 MB for Fast, 94 MB for Best quality) and is cached in your browser.
- It finds the subject. The neural network looks at every pixel and predicts which ones belong to the subject and which belong to the background.
- You get a clean cutout. The result is a mask that separates foreground from background. Touch up lets you fix any spot the AI missed.
How the AI models work
The Fast mode runs U²-Net small, a neural network originally designed for salient object detection. It takes an image, runs it through a series of convolutional layers, and produces a grayscale mask where white means foreground and black means background. The model is only 4.4 MB, so it downloads quickly and runs in under two seconds on most devices.
Best quality runs BiRefNet-lite, a newer architecture that uses bilateral reference to keep more detail at object boundaries. This matters most for curly hair, fur, and fine edges, where the boundary between subject and background is not a clean line. The model is 94 MB, downloaded once, and uses your graphics card (via WebGPU) when the browser supports it.
Both models run through ONNX Runtime in a web worker, so the browser stays responsive while the AI processes your photo. The models are verified by SHA-256 hash before they run, so you always get the exact file that was tested. After the mask is generated, a guided filter refines the edges at full resolution, and color decontamination removes color spill from the old background.
Questions about ai background remover
Which AI model should I choose?
Start with Fast. It handles most photos and finishes quickly. Switch to Best quality for curly hair, fur, or fine edges. Use Plain background for logos, signatures, or any subject on a flat color.
Is the AI as good as server-based tools?
Yes. The same types of neural networks run on servers. The difference is where they run: here they run in your browser, so your photos stay private and there is no queue or daily limit.
Does it use my GPU?
Best quality uses WebGPU when your browser supports it (Chrome and Edge on most computers). On the CPU it takes longer but gives the same result. Fast always uses the CPU and is still very quick.
Can the AI handle complex backgrounds?
Yes. The models are trained on millions of photos with all kinds of backgrounds. Busy, colorful, and cluttered backgrounds work fine. The subject just needs to be recognizable as the main object in the photo.