Image watermark remover
AI image tools sign what they make with a pattern hidden in the pixels themselves. It survives format conversion, screenshots and cropping, and no image editor can see it. Drop a picture in and pick what comes out: SynthID, the Stable Diffusion mark, and everything the file was carrying about you.
The tool
What comes out
Several different mechanisms get called by one name, and they are not variants of one thing. One has a public specification, which means it can be measured. The rest cannot be, so the report shows numbers where numbers exist and says nothing where they do not.
| Format | Supported | Notes |
|---|---|---|
| SynthID | Removed | Google’s mark, in Gemini, Imagen and Nano Banana pictures. It lives in the fine detail, so the picture is redrawn on a GPU rather than filtered. Small text can come out slightly malformed and fine detail can shift. |
| Stable Diffusion (DwtDctSvd) | Removed, and measured | The one scheme with a public specification, so the payload is read before and after and both numbers are printed next to the result. |
| Stable Signature | Removed | Meta’s mark, written by a tuned decoder. Same position as SynthID and the same answer: redrawing reaches it, filtering does not. |
| Tree-Ring | Removed | Its signal sits in the structure the model started from rather than in the pixels, so redrawing is the only thing that touches it. |
| Low-bit steganography | Removed | Plain hiding in the least significant bits. A dither at the final step randomises them everywhere. |
| EXIF, XMP, IPTC, ICC, C2PA | Removed | Included with any pixel work, because the output is written fresh. It is also available on its own if the picture should not be touched. |
Why redrawing, and not filtering
A learned watermark is trained to survive exactly the things an editor does to a picture: noise, blurring, sharpening, compression. Turning any of those up does not reach it. It only costs you detail, which is why the strongest local setting is not the answer and is not offered as one.
Redrawing works because it does something different in kind. The picture is run back through a diffusion model at a low strength, which replaces the fine detail rather than smoothing it. The mark lives in that detail, so it goes with it. That is the whole of the difference: anything that keeps the original fine detail keeps the watermark with it.
Two things worth knowing first
Redrawing is not free. Small text is fine, high-frequency detail, which is the kind the model rewrites most, so a screenshot of a paragraph can come back subtly wrong where a photograph comes back clean. Look at the result before you use it.
Working on the pixels also destroys any Content Credentials the file carried, permanently. A signature covers exact bytes, and after a redraw the bytes are new. That is usually the point, but it cannot be undone, so the panel says so before it happens.
Common questions
- Does this remove SynthID?
- Yes. Tick SynthID and the picture is redrawn on a GPU, which replaces the fine detail the mark lives in. Filtering pixels cannot do it at any strength, which is why the local settings alone will not, and why this one takes a minute or two rather than a moment.
- Will my picture still look the same?
- Almost always, and the difference is worth knowing about. Redrawing replaces fine detail rather than smoothing it, so small text can come out slightly malformed and very fine detail can shift a little. How much depends entirely on the picture: a portrait or a landscape usually comes back indistinguishable, a screenshot full of small type is the hard case.
- What is an invisible watermark, and how is it different from metadata?
- Metadata sits beside the picture in the file, in EXIF or a signed C2PA manifest, and any stripper removes it. An invisible watermark is in the pixel values themselves: a pattern of tiny changes spread across the image, too small to see, which a matching detector can still read. That is why it survives being converted, screenshotted, cropped and re-saved, and why no ordinary image editor can find it.
- How do I know it worked?
- For the Stable Diffusion mark, you can see it. That scheme is open source, so the decoder runs twice, once on your file and once on the result, and the report prints both readings. The others have no public detector, so there is no before and after to show for them.
- Does it remove metadata too?
- Yes, automatically, whenever the pixels are worked on. A redrawn picture is a new file, so the old container and everything in it is left behind, and the panel shows that as included rather than as another thing to tick. If you only want the metadata gone and the picture untouched, turn the pixel options off.
- Can I do a whole folder at once?
- Yes, up to thirty images in one go, and they come back as a single ZIP. A file that fails takes only itself down.