Remove an object from a photo

One of the most wanted photo edits, and usually behind a sign-up and an upload. Here the model is downloaded once and runs on your own device: paint over what should go, press Remove, and keep going until the photo is clean.

Loading the tool…

How it works

  1. Drop a photo — JPEG, PNG, WebP or HEIC. It opens at full resolution in this tab.
  2. Paint over what you want gone, with a little margin around it, using a mouse, a pen or a finger. Pick MI-GAN (fast, 28 MB) or LaMa (better on large areas, 208 MB).
  3. Press Remove and compare with the original. Undo, or keep painting and removing, then download the photo as JPEG or PNG.

Why nothing is uploaded

Every operation on this page is done by code running inside your browser tab, using the same engine that renders web pages. The file is read from disk into your tab’s memory, transformed there, and written back out as a download. It is never sent anywhere — not to us, not to a third party.

Verify it yourself

  1. Open your browser’s developer tools (F12) and select the Network tab.
  2. Load your file and run the tool.
  3. The only requests you will see fetch the tool’s own code — and, for a few heavy tools, their open-source engine (also from this site) or an AI model from Hugging Face — plus one small page-view ping to loreatec.jp (page address and title, nothing more). None of them carry your file.

Proof it stays local →

Frequently asked questions

Is my photo uploaded anywhere?

No. The photo is opened and edited inside this browser tab. Only the model travels: it is downloaded once from Hugging Face, kept by your browser, and after that the tool works without a connection. The downloaded result is re-encoded by the browser, so it does not carry the original EXIF data such as location or camera.

How good is the result, honestly?

It fills the gap with plausible background — the model invents what might be there; it does not know what was really behind. Small things on simple backgrounds (a person on a beach, a sign in a field, a cable against the sky) usually disappear cleanly. Large areas come out softer and more invented, and faces or text inside the gap are not reconstructed: they turn into a blur or into something made up. Always compare with the original, which the tool shows at the press of a button.

MI-GAN or LaMa?

Start with MI-GAN: a 28 MB download and a few seconds per removal. LaMa is a 208 MB download and handles larger areas and repeating textures (brick, grass, water) better. When the browser offers WebGPU, LaMa runs on the graphics card and takes a few seconds too (the first removal longer, while the model is prepared); without it, LaMa runs on the processor and needs 30 seconds to a minute per removal on a computer, longer on a phone — the tool tells you when that is the case.

Does it keep the full resolution of my photo?

Yes. Only a square around what you paint is processed — brought to the 512-pixel working size of the models when it is larger — and pasted back with a soft edge that stays inside the painted area. Every other pixel is left exactly as it was. The flip side: when you paint a large part of a high-resolution photo, the fill is generated at lower resolution and looks softer than the rest.

Does it work on a phone?

Yes: paint with your finger and set the brush size with the slider. On a phone, use MI-GAN. LaMa needs a large download and a lot of memory, so on a phone without WebGPU it is better left for a computer.

Which models are these, and under what licence?

MI-GAN by Picsart AI Research (MIT licence) and LaMa by Samsung AI Center, converted to ONNX by Carve (Apache-2.0). Both are fetched from Hugging Face at a fixed version and run with ONNX Runtime Web; they are listed on the licences page.