Desynthetic
Remove AI artifacts from images
Re-encode to JPEG and inject fabricated camera EXIF.
Drop your images here
or
What is AI image laundering?
"Laundering" an AI image means stripping the signals that reveal its origin — the C2PA provenance and tell-tale metadata — and dressing it up with fabricated camera data so it passes as a real photo. Desynthetic reproduces the technique as an educational red-team demo, so red and blue teams can study it and learn to detect it.
How it works
Drop an image, pick a device profile, and Desynthetic re-encodes it to JPEG and injects a fabricated camera EXIF block (optionally renaming the file to that device's convention). It is intentionally shallow — paired with a detector, it teaches that metadata laundering leaves forensic traces.
Frequently asked questions
What is Desynthetic for?
It is an educational security demo that shows, for red/blue-team training, how AI-image "laundering" works so it can be recognized and defended against.
What does "laundering" an AI image mean?
Re-encoding it and replacing its metadata so a synthetic image superficially resembles a genuine camera photo.
Why build a tool that does this?
To teach that metadata laundering is shallow and detectable; understanding the attack is how you learn to defend against it.
Does it remove invisible or in-pixel watermarks?
No. It only changes metadata and re-encodes; durable in-pixel watermarks are not reliably removed.
Is the output meant to deceive?
No. It is a demonstration artifact for education and research, not a tool for passing AI images off as authentic.