Object remover
Erase anything. It never leaves your computer.
Paint over the distraction and release — Moebius, a compact latent-diffusion inpainting model, generates a fill that matches the surrounding scene, on your GPU. It needs a WebGPU browser such as recent Chrome or Edge.
Paint, release, done
No lasso precision needed and no second step: the stroke you paint is the mask, and inference starts the moment you lift the pointer.
Best on real-world cleanup
Tourists, trash bins, watermarks, text overlays, skin blemishes, fence wires. Homogeneous and textured backgrounds reconstruct best.
One stroke, one undo step
Only the painted area is regenerated, and every erase is a single history step, so you can undo it and retry with a bigger brush.
01
Open a photo
Drag, drop or paste a photo. The Moebius weights (~1.2 GB) download once, then stay cached offline.
02
Paint the object
Use the Erase brush at any size. Cover the object plus a small margin of background for the cleanest fill.
03
Release and keep editing
The fill blends into your photo — keep editing, upscale or export from there.
FAQ
Questions, answered plainly
- Can it remove large objects or people?
- Yes, within limits. Moebius generates new content for the hole instead of only stretching nearby pixels, but it works on a 512px view of the painted area, so very large regions lose detail. For best results remove in a few smaller strokes instead of one giant one.
- Is this generative fill?
- Yes, without a prompt or a server. A local diffusion model fills the painted area from its surroundings, so every run can come out slightly different — run it again if you want another fill.
- How long does it take?
- Nineteen diffusion steps on a 512px view, a few seconds on a WebGPU laptop GPU after the one-time model download. Browsers without WebGPU cannot run it.
Same engine, more jobs