About EraseObject

EraseObject is a photo cleanup workstation for one job: removing things that should not be in a picture — people, objects, text, wires, dust — and rebuilding what was behind them.

Why it runs on your device

Object removal used to need a server with a large GPU, which is why most tools upload your photo, meter you in credits, and keep a copy you did not ask them to keep. Small, capable models and WebGPU have changed that. A modern laptop or phone can run an inpainting model in seconds, so EraseObject sends the model to you instead of asking for your photo. The consequences are the product: nothing uploaded, no credits, no account, and it keeps working offline.

What it is not

It is not a design suite: the few finishing tools are there to tidy up after a removal. It is not a background remover, a face recogniser or a video editor. And it is not for erasing other people's ownership marks or altering documents and evidence; it is a cleanup tool for your own photographs.

Who makes it

EraseObject is made by a small independent team that builds it around one rule: the pixels stay with the person who owns them. Bug reports, hard photos that came out badly and feature requests reach us through the contact page, and they shape what gets built next.

Open models

The AI is built on openly licensed research: MI-GAN (Picsart AI Research, MIT licence) for inpainting, MobileSAM (Apache-2.0) for Smart Select, and MODNet (Apache-2.0) for People mode, all run through ONNX Runtime Web (MIT).