News for hackers. Stuff that matters.
Re:ChatGPT Images 2.5 by pelzatessa on Tuesday September 08, 2026 @02:55PM (#49614999)
The "composite party photo", while impressive, shows that still the miniscule details are being lost, like the teeth structure of the guy in the middle or the fact that the guy on the left is holding the cup with three fingers. Wondering why they chose this edit for the showcase.
Re:ChatGPT Images 2.5 by headz on Tuesday September 08, 2026 @02:59PM (#49615067)
> Wondering why they chose this edit for the showcase.

They probably didn't care to check those images in detail.

Re:ChatGPT Images 2.5 by polytely on Tuesday September 08, 2026 @03:07PM (#49615209)
I'm guessing if you work on these images the whole day you start to lose the ability to judge image quality. like your brain becomes slopified
Re:ChatGPT Images 2.5 by kfarr on Tuesday September 08, 2026 @03:17PM (#49615399)
I've also found the OpenAI image models to lose fine detail on image edits compared to Nano Banana or Flux models which faithfully retain input source image geometry and details. I was hoping this might be different but it sounds similar to previous OpenAI image models where something is lost in translation during image editing.
Re:ChatGPT Images 2.5 by vunderba on Tuesday September 08, 2026 @04:22PM (#49616479)
I know the API (assuming you're using it) lets you set the output size of the edited images - I haven't done a huge amount of testing with > 1mp, so I'd be curious if this might mitigate some of that.

https://developers.openai.com/api/reference/python/resources...

Re:ChatGPT Images 2.5 by vunderba on Tuesday September 08, 2026 @03:14PM (#49615339)
The OpenAI PR/doc team seem to have a history of this kind of attention lapse - their 4-panel comic strip using gpt-image-2 was an absolute mess too.

https://mordenstar.com/blog/gen-failures

Re:ChatGPT Images 2.5 by Caracas288 on Wednesday September 09, 2026 @05:43AM (#49623785)
Another example was the doc for GPT-5, which had all its graphs messed up, but they said it was human error rather than AI.
Re:ChatGPT Images 2.5 by fsloth on Tuesday September 08, 2026 @03:35PM (#49615726)
The dog is even more on the nose. The shadow shape is almost the same despite the dog getting quite a lot of volume around it's original body parts.
Re:ChatGPT Images 2.5 by nzach on Tuesday September 08, 2026 @02:59PM (#49615066)
And don't forget the 'Thing' from Addams Family in the shoulder of the guy on the left.
Re:ChatGPT Images 2.5 by pllbnk on Tuesday September 08, 2026 @03:05PM (#49615156)
The guy on the left in that same photo only has three fingers (and a thumb, I suppose). I thought image generation has already outlived that.

Edit: I feel stupid I didn't see the original OP already mentioning three finger issue. I'll just leave it here.

Re:ChatGPT Images 2.5 by timbaboon on Tuesday September 08, 2026 @03:30PM (#49615628)
Haha yes, I immediately saw the fingers and was surprised 'cos I thought that was something they had "solved" by now
Re:ChatGPT Images 2.5 by hmstx on Tuesday September 08, 2026 @03:52PM (#49616006)
Exactly this. "All this progress and they're back to miscounted fingers again?!"
Re:ChatGPT Images 2.5 by andai on Tuesday September 08, 2026 @03:38PM (#49615778)
Maybe it's like how spammers intentionally make their emails more obvious, because they have a very specific target audience.
Re:ChatGPT Images 2.5 by nilsherzig on Tuesday September 08, 2026 @03:31PM (#49615657)
Dogs shadow didn't change
Re:ChatGPT Images 2.5 by nojs on Tuesday September 08, 2026 @10:09PM (#49619925)
The guy’s arm is also extremely long.