Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

ChatGPT Images 2.5

openai.com

Post reply on HN
Print view View original

81–90 of 472 posts

Re: ChatGPT Images 2.5

#81

Post by AyanamiKaine »

cobolcomesback wrote: post #59
AyanamiKaine wrote: post #12

Omg, I love how the first examples just show how easy you can fake things. Fake being at a party with your friends. Didn't make your bed, no problem, just fake it. The sad part is my mother would love "remixing" my old child photos of me.

Seriously, are these really the best examples they could come up with? What is the point of having a fake picture of your dog in a costume? What’s the point of having a fake picture about being at a party? The only use case I can think of for this is for someone who likes to make up stories and lie about what they’ve done. Is that really the target market?

That was exactly my first reaction!

Mark where you at the party today? Yes of course look at these pictures I took (╥﹏╥)

Re: ChatGPT Images 2.5

#82

Post by yoz-y »

minimaxir wrote: post #56

A big wtf at the LM Arena scores: https://arena.ai/leaderboard/text-to-image gpt-image-2.5-sunburst: 1421 gpt-image-2.5-flare: 1399 gpt-image-2 (medium): 1381 mai-image-2.6: 1331 Even with LM Arena being flawed, this is significant. I was planning to do a writeup on the original gpt-image-2 as it crushed every complex image comprehension benchmark I had...I'm glad I procrastinated since ChatGPT Images 2.5 seems like…

> Many people still think AI images output the wrong number of fingers on a regular basis.

Last time I used a frontier image model it made me a seal with three hands so…

Re: ChatGPT Images 2.5

#83

Post by handbanana_ »

vlyan wrote: post #48
recitedropper wrote: post #38

Earlier quoted context omitted.

The sheer waste that accompanies image and video genAI is actually terrifying. I challenge OpenAI to put a "your carbon footprint" field next to each generation. If you have nothing to hide, more information is surely better, right?

would you like a carbon receipt for your video gaems, netflix, and reddit sessions?

Absolutely, I'd love to see how tiny they are in comparison.

Re: ChatGPT Images 2.5

#84

Post by vunderba »

bufbupa wrote: post #23

Still can't make sprite sheets :(

You can get semi-decent sprite work out of GenAI models, but you still have to put in some manual work (scale normalization, palette reduction, grid alignments, etc). It's definitely not "out-of-the-box" yet.

https://mordenstar.com/other/hobbes-animation

Re: ChatGPT Images 2.5

#85

Post by kfarr »

pelzatessa wrote: post #6

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.

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

#88

Post by applfanboysbgon »

paimapi wrote: post #49
ksd482 wrote: post #44

Earlier quoted context omitted.

I think @conradludgate was alluding to that we have so much of AI slop these days. But I see your point too. I too like to generate cute and funny images and share it with my friends and family. But this being done at scale can lead to overall degradation of the online experience.

plus the energy cost - https://www.technologyreview.com/2023/12/01/1084189/making-a... >Generating 1,000 images with a powerful AI model, such as Stable Diffusion XL, is responsible for roughly as much carbon dioxide as driving the equivalent of 4.1 miles in an average gasoline-powered car. In contrast, the least carbon-intensive text generation model they examined was responsible for as much CO2 as driving 0.0006 mi…

This is an embarrassingly idiotic article. They're comparing against the least intensive text model, ie some million parameter model nobody uses. Stable Diffusion is a 3.5 billion parameter model, while the GPT models a billion people are using for text generation are over 10 trillion parameters. To say nothing of the differences in average context size usage. The actual ratio of SDXL to text generation pollution is probably literally reversed from what the article claims by lying with statistics.

(Note, however, that OpenAI's image model is much much larger than SDXL; however, we don't have precise numbers for it. Nonetheless, misinformation is misinformation.)

Re: ChatGPT Images 2.5

#89

Post by embedding-shape »

sanid wrote: post #69
mattbettinson wrote: post #3

I hope my local kebab shops switch to this for their signs.

I can't disagree more. A lot of shops in our city have done this and I actually miss the shitty photoshopped images that did not even look like the real life dish anyway. AI signs are way worse.

I thought that was parents point, that currently they're using so shitty AI generated logos, that even if we despise that as a concept, at least better models output slightly less sloppy shit. But re-reading it, I'm not sure that was the right initial reading.
Post reply on HN