Friends who sell prints and stock photos keep asking me the same question lately: "If I search for SynthID remove tools, what am I actually getting, and is it okay to use them?" This post is my plain answer, after testing what's inside a typical AI-assisted image and how to clean it up responsibly.
Two very different things hide inside an image
1. Metadata and labels. These are text fields stored alongside the pixels: EXIF (camera, lens, GPS location, timestamps), XMP and IPTC (editing history, creator and copyright fields, "AI generated" flags), C2PA content credentials, and PNG text chunks that some generators fill with the full prompt and settings. They are easy to read with any metadata viewer and easy to remove without touching the picture itself.
2. Invisible watermarks such as SynthID. These are not text fields. They are patterns spread across the pixels of an image, designed to survive cropping, compression and colour changes. Stripping metadata does nothing to them.
Most confusion I see comes from mixing these two up.
Why someone would clean metadata
The most common legitimate reasons I hear:
- Privacy. Phone photos can carry exact GPS coordinates of your home. A prompt stored in a PNG can reveal client details.
- File hygiene. Editing apps pile up long XMP histories that bloat files and confuse stock-site uploaders.
- Consistent labelling. If you composite AI elements into your own photo, you may want to remove stale or wrong tags and then add an accurate disclosure yourself.
For this, I use the free Remove AI Label tool from SynthClear. It runs entirely in the browser, so files are not uploaded anywhere, and it strips EXIF, XMP, IPTC, C2PA and PNG text from JPG, PNG and WEBP files (up to 15 MB, up to 20 at a time). I checked the output with exiftool and the fields were gone while the pixels were untouched.
Where SynthID remove tools fit
SynthClear also offers a paid SynthID watermark remover. My honest take: it only makes sense for images you own or are authorised to edit, for example a background you generated yourself and then heavily repainted, where the watermark no longer reflects how the final image was made. It is not a way to pass off AI images as human-made. Many platforms, marketplaces and clients require you to disclose AI involvement regardless of what is or isn't embedded in the file, and removing a watermark doesn't change that obligation.
A responsible workflow
- Keep your originals with all metadata intact, in case you ever need to show provenance.
- Strip personal or stale metadata (GPS, prompts, old edit history) from the copy you publish.
- Add an accurate disclosure where the platform asks for one, in the description or the platform's own AI toggle.
- Only touch watermarks on images you have the rights to, and never to mislead buyers or readers.
- Check the result with a metadata viewer before uploading.
If you want a quick FAQ-style refresher on the terms, this plain-English FAQ on SynthID, AI labels and image metadata covers the basics well.
Cleaning files is good practice. Being honest about how an image was made is better practice. You can do both.
Owen Harte