←back to thread

Gemini 2.5 Flash Image

(developers.googleblog.com)
1092 points meetpateltech | 5 comments | 26 Aug 25 14:01 UTC | HN request time: 0s | source
Show context
skybrian ◴[26 Aug 25 16:55 UTC] No.45029197[source]▶
Like most image generators, it didn’t pass the piano keyboard test. (Black keys are wrong.)

https://aistudio.google.com/app/prompts?state=%7B%22ids%22:%...

replies(9): >>45029266 #>>45029269 #>>45029353 #>>45029404 #>>45029503 #>>45029767 #>>45029823 #>>45029961 #>>45032710 #
1. carimura ◴[26 Aug 25 17:37 UTC] No.45029767[source]▶
or my "hands with palms facing down" test.... no matter how hard I try it just can't get open hands, palms down.
replies(2): >>45030007 #>>45030065 #
2. pbhjpbhj ◴[26 Aug 25 17:56 UTC] No.45030007[source]▶
I guess the vast majority of images have the palms the other way, that this biases the output. It's like how we misinterpret images to generate optical illusions, because we're expecting valid 3D structures (Escher's staircases, say).
replies(1): >>45030084 #
3. vunderba ◴[26 Aug 25 18:00 UTC] No.45030065[source]▶
It's probably just a matter of rerolling a few times. I was able to get it around 25% of the time.

https://imgur.com/a/H9gH3Zy

replies(1): >>45035104 #
4. vunderba ◴[26 Aug 25 18:01 UTC] No.45030084[source]▶
Yes - it's the same reason generating a 5-leaf clover fails - massive amounts of training data that predisposes the model against it.
5. carimura ◴[27 Aug 25 03:17 UTC] No.45035104[source]▶
that's pretty good. I was using a cartoon girl as an example of a dance move for kids.

https://g.co/gemini/share/0e0de0d42029