this post was submitted on 15 Jul 2026
127 points (97.7% liked)

Technology

87426 readers
3818 users here now

This is a most excellent place for technology news and articles.

Our Rules

  1. Follow the lemmy.world rules.
  2. Only tech related news or articles.
  3. Be excellent to each other!
  4. Mod approved content bots can post up to 10 articles per day.
  5. Threads asking for personal tech support may be deleted.
  6. Politics threads may be removed.
  7. No memes allowed as posts, OK to post as comments.
  8. Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
  9. Check for duplicates before posting, duplicates may be removed
  10. Accounts 7 days and younger will have their posts automatically removed.

Approved Bots

founded 3 years ago
MODERATORS
L3s@lemmy.world
enu@lemmy.world
technopagan@lemmy.world
L4s@lemmy.world
L3s@hackingne.ws
127
Meta AI image detector fails to identify some of its own cropped AI images, Reuters analysis finds (www.reuters.com)
submitted 1 month ago by AcidiclyBasicGlitch@sh.itjust.works to c/technology@lemmy.world
4 comments fedilink hide all child comments
 

A new AI detection tool ​from Meta which the tech company previewed this week alongside the launch of its image-generation model, Muse ‌Image, failed to identify some of its own AI-generated images once they were cropped, according to a Reuters analysis.

In an analysis of 40 images generated using Muse Image, Reuters found the detection tool verified all of the original AI-generated images but failed to verify 55% of ⁠the same images after they were cropped to approximately one-third to one-half of their original size.

When asked about the results of the Reuters analysis of the detection tool, Meta noted that the tool was a preview. The company said the watermark is designed to remain intact after common edits, but that the signal may be lost if an image is heavily cropped.

Rival tech companies Google and OpenAI have cautioned that their own detection ‌tools ⁠are not foolproof against image-alteration techniques.

top 4 comments
sorted by: hot top controversial new old
[–] goatinspace@feddit.org 6 points 1 month ago

[–] lyralycan@sh.itjust.works 6 points 1 month ago* (last edited 1 month ago)

Last year I generated 8 images during a day of experimenting with Google's Gemini Image model. Running this same test on my own images:

  • 5/8 were verified using Google's own embedded SynthID;
  • After running the failures through once again, one verified, making the successful total 6/8. Two remained ambiguous according to the AI.

Of these images:

  • 5 were generated with just a prompt (4*/5), including the image that was not recognised at first
  • 2 were generated using personal images alongside a prompt (1/2), the recognised one of a human, the other of a rabbit
  • 1 was generated using sourced imagery alongside a prompt (0/1)

Conclusion: Hilariously, it seems the in-house watermark method is absolutely not accurate.

[–] FaceDeer@fedia.io 2 points 1 month ago

Just use one of these.

[–] Zarobi@aussie.zone 1 points 1 month ago

Cache miss