Updated
Updated · The New York Times · Oct 8
Google Gemini Creates Near-Perfect Fake Photo as 7 Subtle Flaws Still Expose It
Updated
Updated · The New York Times · Oct 8

Google Gemini Creates Near-Perfect Fake Photo as 7 Subtle Flaws Still Expose It

3 articles · Updated · The New York Times · Oct 8

Summary

  • Google’s Gemini produced a fake FedEx Office image realistic enough to erase many old giveaways, including malformed hands, broken logos and obviously garbled text.
  • 7 remaining clues still expose the image: warped floor perspective, overemphasized brand prompts, zoom-level text glitches, impossible object connections, nonsensical wording, nonstandard package symbols and other visual artifacts.
  • Researchers say the shift marks a new stage for AI imagery, where detection depends less on quick visual checks and more on close inspection of physical relationships, semantics and context.
  • The result is a narrower window for spotting fakes online: images can now look convincing at a glance, even if trained viewers can still catch errors for now.

Insights

If AI perfectly mimics reality at first glance, what hidden clues in plain sight can still expose the ultimate digital deception?
When invisible watermarks and digital credentials fail, can human intuition still outsmart the most advanced synthetic image generators?