The 'Nano Banana Pro' experiment: AI detection tools prove useless

[Redaktion]

Can software save us from deepfakes? We put six detection tools to the test with a 'Nano Banana Pro' image. The results show that current tech is easily outsmarted. Simple edits using standard software were enough to drop detection rates to almost zero.

https://www.notebookcheck.net/The-Nano-Banana-Pro-experiment-AI-detection-tools-prove-useless.1201789.0.html

[Dvorak]

Google's SynthID (which is built specifically for Google-generated images) seems to work even for the final edit. Try sending it to Gemini with the query "Is this image ai generated?"

See here for more info: blog[dot]google/innovation-and-ai/products/ai-image-verification-gemini-app/

[toto1234]

Fakes were easier to fake when Google's image generation was heavily woke/DEI biased, with African Roman Emperors or African WW2 SS soldiers :-)