Is SynthID reliable? I tested Google's AI image check on screenshots, stripped files and real photos

Description and use

Tests show SynthID reliably detects AI images after metadata removal and in screenshots, but missed a small pasted AI segment and the Detector site was down during the test.

Problem

Determine whether Google’s SynthID watermark detection works on screenshots, metadata‑removed files, and real photos.

Outcome

SynthID reliably flags AI‑generated images even after metadata removal and when captured as a screenshot, but it missed a small pasted AI segment and the public Detector site was down during the test. Real photos without AI content were correctly identified as non‑AI.

Workflow

• Generate two Gemini images (2400×1792)
• Create test variants: original, metadata‑removed, screenshot PNG, NASA photo, NASA photo with sky edited, AI image pasted into photo
• Attempt uploads to SynthID Detector (failed with error)
• Use Gemini app’s AI‑generation check for each file
• Record Gemini’s responses and content credentials presence

Evidence

Google's SynthID check in the Gemini app flagged a Gemini-generated image after its metadata was removed and again after a screenshot, and found nothing on a real photo.

Source

The Detector was the plan, but on October 12 all four of my uploads, between 4:36 and 5:33 pm UTC, ended in "Something went wrong! Please wait and try again later."

Source

Gemini said about each file: ... "Most or all of this image was edited or generated with Google AI." (2 of 2 runs)

Source