Free tool from Liz on the Web
Remove hidden AI watermarks, invisible characters and metadata from your text, images and PDFs before you post them. No sign up, no upload, no cost.
Pick one. Text, PNG, JPG, WebP, GIF, PDF or SVG.
Most tools are vague here. This one is not. Find the tool you used and see exactly what comes off and what stays on.
| Made with | What comes off | What stays |
|---|---|---|
| ChatGPT text | Invisible characters, em dashes, curly quotes, narrow spaces | Nothing. OpenAI does not watermark text. |
| Claude, Gemini, Grok text | Invisible characters and formatting quirks | Nothing. None of them watermark text. |
| ChatGPT and DALL-E images | C2PA Content Credentials, EXIF, XMP | Nothing known. |
| Google Gemini, Imagen, Nano Banana | C2PA Content Credentials, EXIF, XMP | SynthID. It is in the pixels. It stays. |
| Meta AI images | Metadata tags | Their pixel watermark. It stays. |
| Adobe Firefly | C2PA Content Credentials, EXIF, XMP | Any pixel-level mark Adobe adds. |
| Midjourney | EXIF, XMP, prompt data saved in the file | Nothing publicly known. |
| Your camera or phone | GPS location, device name, date, serial number | Nothing. |
| Word, Google Docs, Canva PDFs | Author name, producer, title, dates, XMP | Text hidden in compressed streams. |
| Any visible logo or stamp | Nothing | It stays. That is a photo editing job. |
Read this part. There are two very different kinds of AI watermark, and only one of them can be removed without wrecking your file.
Metadata watermarks are hidden text fields bolted onto the file. Things like "made with Imagen 4". This tool removes those, and your file comes out looking exactly the same.
Pixel watermarks like Google SynthID are woven into the picture itself. Every pixel is nudged by a tiny amount in a pattern only Google can read. Cropping it, resizing it, screenshotting it or saving it again does not shift it. The only way to remove one is to have an AI redraw the whole image, which changes what your picture looks like. This tool does not do that, and no free tool does it well.
This will not beat an AI writing detector. Detectors like GPTZero and Turnitin read your writing style, not hidden characters. Cleaning a file changes nothing they look at. Anyone selling you a watermark remover as a detector bypass is lying to you.
Three steps, and none of them involve a server.
When you drop a file in, your browser reads it straight from your hard drive into memory. There is no upload, no server and no internet request. You can switch off your wifi after the page loads and it still works.
Every file type keeps its hidden notes in a known place. A PNG stores them in labelled blocks. A JPEG keeps them in segments near the front. A PDF lists them in an info section. The tool walks through your file, finds those spots, and drops them. Your text, your pixels and your page content are copied across untouched.
You get a download button and a list of exactly what was taken out. No guessing. If nothing was hidden in there, it tells you that too.
Pictures are never re-compressed. For a PNG the tool reads the chunk list and drops tEXt, iTXt, zTXt, eXIf and iCCP. For a JPEG it walks the segment markers and drops every APPn block, which is where EXIF, XMP and C2PA live, plus any comment block. The compressed image data after the start-of-scan marker is copied across byte for byte. That is why your picture comes out looking identical and the file just gets smaller.
The tool reads your text one character at a time and drops anything from the known invisible ranges. Zero-width spaces and joiners, word joiners, byte order marks, soft hyphens, direction overrides, variation selectors, and the Unicode Tags block, which is the trick people use to smuggle a whole hidden message inside a normal-looking sentence. Odd spaces get swapped for a normal one.
A PDF keeps a table of byte positions, so changing the file size breaks it. The tool gets around that by replacing each metadata value with the same number of spaces. The file stays exactly the same size, the position table stays correct, and the document still opens anywhere. No PDF software required.
The whole thing is a few small JavaScript files with no libraries and no tracking. There is no account system, no database and no logging, because there is no server doing any of the work. Open your browser's network tab and watch it: after the page loads, cleaning a file makes zero requests.
No. SynthID is hidden inside the pixels of the picture itself, not in the file's text fields. It survives cropping, resizing, screenshots and re-saving. This tool never changes a pixel, so it cannot remove SynthID. No free metadata tool can. Anything claiming otherwise is either re-drawing your image or is not telling the truth.
OpenAI does not publicly watermark ChatGPT text. What it does leave behind are invisible characters and formatting habits like em dashes, curly quotes and narrow no-break spaces. This tool removes all of those.
No, and be careful of any tool that promises it will. AI detectors look at writing style, sentence rhythm and word choice, not hidden characters. Cleaning the file changes none of that. The only thing that changes a detector's result is rewriting the words yourself.
No. Every bit of the work happens in your own browser using JavaScript on your device. Nothing is sent to a server, so there is no file to store, log or leak. You can turn off your internet after the page loads and it still works.
Not at all. The image is never re-compressed. The hidden text blocks are cut out and the pixel data is copied across untouched, so the picture that comes out is identical to the one that went in. The file just gets a little smaller.
C2PA Content Credentials are a signed label tucked inside an image file that records which tool made it and when. ChatGPT, DALL-E and Adobe Firefly all add them. This tool removes them.
Photos from your phone quietly carry the exact spot they were taken, the date, and your device name. If you post that picture, you post all of that too. Cleaning it first is a good habit.
Yes. No sign up, no account, no credit card, no limit. It costs nothing to run because your own browser does the work.
55 free guides, plus the tools I build. All of it written for people who want AI to do the boring parts of their business, with no tech background needed.
Join 106,000 people getting the free newsletter. Real AI workflows you can copy, explained in plain English.
Get the free newsletter