About MagicRemover
MagicRemover is a free, browser-based tool for erasing things out of photos — a passer-by, a bin lorry, a date stamp, a caption, a logo. You paint over what you want gone and a generative AI model rebuilds the surface behind it. This page explains how that actually works, where the limits are, and who is on the other end of it.
What it is
MagicRemover runs entirely in a browser tab. There is nothing to install, no account needed to start, and nothing is added to the picture you get back — no badge, no watermark, no resolution downgrade. You upload a JPG or PNG, brush a red mask over the region you want removed, and press Remove. Fifteen to thirty seconds later you get a PNG at the dimensions you uploaded.
The site is organised as a set of task-specific tools rather than one generic editor, because the right brush size, the right amount of margin, and the realistic success rate are genuinely different for a diagonal stock watermark than for a person standing in front of a hedge. Each tool page carries the guidance for that specific job.
How it works
The technique is called generative inpainting. Your image and your mask are sent to a diffusion model that has been trained to predict what plausibly belongs in a hidden region given everything around it. It is not copying pixels from elsewhere in the photo and it is not revealing anything that was behind the object — it is generating a new, plausible surface and blending it into the existing one.
That distinction matters more than it sounds. On a wall, a sky, sand, water, or a smooth gradient, the prediction is close to perfect and the edit is invisible. Over fine repeating detail — foliage, brickwork, a crowd, patterned fabric — the model is making an educated guess, and a careful viewer may spot it. Nothing here recovers hidden information; it invents a convincing replacement.
Image jobs run on hosted inference providers (currently RunningHub, with Replicate as the fallback path). The video subtitle tool runs on apimodels.app. Results are stored on Cloudflare R2 and served back to you over HTTPS. The full list of processors, and what each one receives, is in the Privacy Policy.
What it cannot do
Being straight about the failure cases saves everyone time:
- It does not recover what was behind an object. The rebuilt area is generated, not uncovered. If a sign covered someone's face, you get a plausible face-shaped guess, not their actual face.
- It is not OCR. The text tools erase writing; they never read or translate it. Run OCR first if you need the words.
- Busy texture is hard. Removals over foliage, crowds, brick, or text-on-text often leave soft residue that needs a second pass or a manual touch-up.
- It works one image at a time. There is no batch mode in the web UI, deliberately, so the free tier stays usable for everyone.
- It does not defeat invisible watermarking. Provenance signals such as SynthID or C2PA metadata are not what the brush touches, and we make no attempt to strip them.
- Very large images are downscaled. Anything over 1536 pixels on the long edge is resized before processing.
What the free tier really is
Guests get 2 removes per rolling 24 hours per device. Signing in with Google adds a separate allowance of 2 per day per account, so a signed-in visitor can use up to 4 in a day. Both counters reset at midnight UTC and are shared across every tool on the site.
There is no credit card, no trial that quietly expires, and no watermark on the output. The image tools are free, full stop. Credits exist only for the heavier credit-based work — currently the video subtitle remover, which is billed per second of video because that is how it is billed to us. If you never touch the video tool you will never need a credit.
There is no welcome bonus. New accounts get the daily free allowance and nothing else. If you find copy anywhere on this site that says otherwise, it is out of date — please tell us.
What happens to your photos
Short version: your upload is used to perform the edit you asked for and nothing else. It is never sold, never used to train a model, and never shown to anyone else.
Longer version, because the honest answer has an asterisk: anonymous results are deleted automatically within 24 hours. If you are signed in, your results are kept so they can appear in your gallery, and they stay there until you delete them or ask us to close your account. The Privacy Policy sets out exactly which service holds what, and for how long.
Where we draw the line
An eraser is a neutral tool, but some uses are not. We ask you to use MagicRemover only on images you own or have the right to edit, and the Terms of Service make that binding. Two cases are worth spelling out because people genuinely do not realise they are different from ordinary editing:
- Stock and agency watermarks. Removing one from a comp preview you have not licensed is copyright infringement, and stripping copyright management information can be a separate violation in its own right (in the US, 17 U.S.C. § 1202). Licence the file instead — it is cheaper than the alternative.
- Trademarks. A logo is usually a registered mark. Cleaning your own branding is fine; removing a mark so that a product's origin is misrepresented, or an endorsement implied, can breach trademark law even where no copyright is involved.
We also refuse the obvious: nothing involving sexual content of minors, nothing designed to fabricate evidence, and nothing intended to harass a private individual.
Every tool on the site
- Remove Object from Photo
- Remove People from Photo
- Remove Passers-by from Photo
- Remove Watermark from Photo
- Remove Text from Image
- Remove Sticker from Image
- Remove Date Stamp from Photo
- Remove Gemini Watermark
- Logo Remover
- Magic Eraser
- AI Image Generator
- Edit Text in Image
- Remove Subtitles from Video
- Remove Text from Video
Step-by-step walkthroughs live on the blog.
Getting in touch
Email hello@magicremover.org, or use the contact page to see which address to use for support, copyright, privacy, or business enquiries. A real person reads it.
Powered by apimodels.app — a unified AI image generation API