Can you remove SynthID? What works and what does not
SynthID remover sites promise to strip Google's invisible watermark. What it survives, how to test any claim yourself, and which AI marks you can really remove.
4 min read
On this page7 sections
Short answer. No method has been shown to remove SynthID while leaving the image intact. Google built it to survive cropping, compression, filters and screenshots, and the sites that call themselves SynthID removers do not publish detector results. What you can remove reliably is the metadata layer (C2PA Content Credentials and AI tags), which is what platforms read to label posts. HideSlop does that and gives the file a new fingerprint; it makes no claim about SynthID.
What SynthID is, briefly
SynthID is Google DeepMind's invisible watermark. A neural network adjusts pixel values across the whole image at generation time; a matching detector recognizes the pattern. It is in every image from Gemini and Imagen, every Veo video and its audio, and, since 2026, images from ChatGPT. What is SynthID? covers how it works.
The important property is that it is not stored anywhere you can delete. There is no field, no block and no corner. The mark is the pixels.
What the "remover" sites do
Searching for a SynthID remover turns up a mix of pages. They fall into four groups:
| Type | What it does | Effect on SynthID |
|---|---|---|
| Text humanizers that rank for the keyword | Rewrite text | None; the page is about text, not images |
| Metadata strippers | Remove C2PA and AI tags | None; SynthID is not metadata |
| Noise, blur, resize and re-save tools | Degrade the image slightly | Little; the watermark is designed for this |
| Regeneration tools (image-to-image, upscalers, diffusion passes) | Redraw the pixels | Degrades the mark in proportion to how much the image changes |
Only the last group does anything to the watermark, and it does it by producing a different picture. Faces shift, text breaks, textures change. Research on removing this family of watermarks reaches the same place: detection confidence can be pushed down by regenerating enough of the image, at a visible cost, and Google updates the detector as attacks are published.
How to test any claim
You do not have to take anyone's word, including ours.
- Generate an image in Gemini and save it.
- Run it through Google's SynthID Detector. Note the result.
- Run it through the tool you are evaluating.
- Run the output through the detector again.
A tool that removes SynthID shows "not detected" on step 4 with an image that still looks like step 1. If a site cannot show that pair, it has not removed anything. The same test works for ChatGPT images now that Google's detector reads OpenAI's marks.
What you can remove
The metadata
Every Gemini and ChatGPT image also carries C2PA Content Credentials and an IPTC "AI generated" field. Those are what Instagram, Facebook, Threads and TikTok read when they label a post, and they are plain metadata. The image remover rebuilds the file so none of it survives, writes camera metadata from a real iPhone in its place, and saves a JPEG at phone quality with an invisible variation. The result has no automatic label trigger and a fingerprint that does not match the original. First image free, 20 per batch. Video goes through the video remover.
The visible logo
Gemini draws a sparkle in the corner on most plans; Veo draws its logo in the frames. Cropping or an object-removal tool handles those, with the usual cost of changing the framing or repainting pixels. The logo and SynthID are independent: removing one does nothing to the other.
The fingerprint
Platforms match reposts by hashing frames, audio and files. HideSlop's small zoom, color, grain and timing changes defeat that matching. This is a different problem from watermarking and should not be confused with it.
What we do not claim
HideSlop changes every pixel slightly, and that disrupts simple fingerprinting. It is not designed to remove SynthID, and we do not say it does. The product FAQ has said so since launch: pixel watermarks are built to survive edits, so no tool can honestly promise to remove them. What we can promise is that the embedded AI metadata is removed and the file is new.
What to do instead
- Assume SynthID is there on anything from Google or ChatGPT, and plan as if it can be detected.
- Remove the metadata so platforms do not label the post automatically.
- Generate elsewhere if a watermark-free pixel layer matters to you. Models such as Midjourney, Stable Diffusion and Flux do not embed SynthID.
- Disclose realistic content. Platform rules apply regardless of what is or is not in the file.
FAQ
Is there a free SynthID remover?
Free or paid, none has demonstrated removal with before-and-after detector results. Free metadata removal exists; HideSlop's first removal is free.
Does cropping remove SynthID?
No. The mark is spread across the image, and Google says it survives cropping as long as enough of the picture remains.
Does converting to JPEG or resizing remove it?
No. Compression and resizing are exactly the edits it was designed to survive.
Does a screenshot remove SynthID?
No. A screenshot copies the pixels and the pattern with them. It does remove the metadata.
Is it illegal to remove a watermark?
Editing images you own or have the rights to is legal in general. Platform terms and AI labeling rules in some jurisdictions may require disclosure of realistic AI content, and those apply to you regardless of marks.