That stained glass whale thing makes me especially grumpy. When it scrolled onto the page, my first thought was "whoah, what a neat idea" and then it was _immediately_ followed by "what a terribly execution."
Like, I'm not even an artist, so I'm sure there's a million things I'm missing, but it's just so upsetting to look at. There's weird random reflections on things that shouldn't have reflections, the coloured light absolutely doesn't actually track through the stained glass onto any surface that makes any sense, light is coming from every which way, it's just... gross to look at for anything more than a quarter of a second.
And this is what they're using to hype people up for it. /shrug
Yeah. I think that the pelican SVG benchmark is actually a much better reflection of the state of AI art than the visual quality of the images.
Based on current trends I would not be at all surprised if LLMs working with tools, SVGs and looking at their own work get there before diffusion models. I've stopped keeping up with the image models so maybe I'm underestimating the progress but I dropped out because they were so frustratingly bad at composition and integrating details together.
That whale image makes me think that the situation hasn't improved as much as I'd expected. It is a really cool idea and a pretty picture, but the whale looks a lot like it is flying over the top of some sort of huge glass dome on the ground.
Using the grid trick for animations to generate the individual frames has been around for awhile and not specific to ChatGPT: I'm surprised it took this long for one of the official providers to make it explicit. However, adherence between frames is generally tricky, and even in the examples there is a lot of noise between frames although that could be due to compression.
SVG support is interesting but I suspect shenanigans are afoot since that's not how image generation models work, and it would likely strip any SynthID watermarks. It's possible OpenAI has pipelined a raster-to-vector tool that does the conversion automatically. (I developed a similar pipeline and it works extremely well; I'll open source it at some point)
EDIT: Per documentation, there is an explicit `output_format: "svg"`, and I am indeed correct there is raster-to-vector in play:
> The API creates a reference image and uses it to produce the SVG. The quality setting controls that reference image. Decode data[0].b64_json as base64 and save the resulting UTF-8 XML bytes with a .svg extension.
OpenAI claims 50% faster generation with Images 2.6 but curiously price is the same as Image 2.5.
I tried it with Sunburst with both gpt-5.5 and gpt-5.6-sol, and the results were nothing spectacular. It’s something you could already do in the ChatGPT interface by asking it to convert an existing image to an equivalent SVG.
Recraft is probably the only group I know of that had a native text-to-SVG generator pretty early on, but since it’s a closed model, I’m not familiar with what they were doing under the hood technically.
Yeh using sprites to make frames consistent (kinda tilting latent space towards a character consistency, zero-shot ~LoRA hack?) is old-school. I did this 3 years ago. https://j11y.io/dance/
Coding related have had massive step-up improvements over time.
The same cannot be said for image generation. If anything, I swear it's gone backwards.
It peaked with very early midjourney (like 2022), maybe follow ups with stable diffusion, and then with some of the nano banana models for actual editing. But overall, very lackluster in comparison to other areas.
> Feels like ai image generation has plateaud.
> Coding related have had massive step-up improvements over time.
> The same cannot be said for image generation. If anything, I swear it's gone backwards.
I would argue both are exactly the same. The images are getting nicer, the AI can alter pictures better. It's the same as coding.
The difference is our perception. Looking at image takes 2s, and we can spot what we don't like. Looking at a code change takes hours to understand. It's very difficult to tell how much code generation is improving because almost nobody is reading it.
Image and code generation don't correlate exactly, but I would be surprised to see a big jump of capabilities in either without the other quickly following.
There have been a lot of strides in open-weight models, well beyond SDXL. Flux, back in August of 2024 was a game changer, finally allowing us to move past SDXL’s nuclear pile of comma-delimited keywords for prompting and allowing more natural language.
And this summer we got Ideogram V4 and Krea2, both of which are highly capable.
You can see how the open/closed models have advanced since SDXL in the Progress tab.
MJ is pretty out-of-date by this point though it can still be fun from an image space exploration perspective. It scored a rather pitiful 2 out of 15 on my bench back in the summer of 2025.
You could argue that progress has been relatively flat since Nano Banana Pro, but that's still less than a full year old (just barely), and that was the first model that could really fully maintain visual consistency across complex edits. The GPT ones are a little stylistically disappointing in ways, but they've still been making steady progress too. I don't think we can claim a peak just yet, even if it's been slower.
Your feeling is generally right. There have been no seminal moments in image generation anymore in the last couple of months. Most attention from the big labs has gone to video since that‘s where the money is.
We have a page to track the history of it all here https://lumenfall.ai/arena/history
(Right now missing some models which I‘ll add today.)
Coding flattened out like 9 months ago. In reality, people are getting dumber and lazier from over using them, so they feel like they're improving when they aren't really, outside some areas like RE.
In reality RE has only gotten better because they RLd it into the models because they NEED data, so now every kid is decomping proprietary softwate and sending them back great training data.
I miss the outpainting feature they used to have (3 years ago?) in the dall-e days. It feels like tooling overall for image generation has been standing still or almost regressed the past few years, while the quality has improved
I wish it was properly adjusted to design work. I can get some work done with Gemini over multiple prompts, center this, adjust this, keep stylized properties of this, etc.
I would caution against using the webapp to test the quality of this new model because OpenAI does not indicate which image model and effort is being used to generate the image, to my continued frustration.
I think it's more impressive that it managed so much coherence in a single image than anything else. Very few problems with each of the pokemon anatomically speaking. Mew's kind of a mess but the rest are pretty accurate.
That being said, the ukiyo-e artwork looks more like a standard colored pencil illustration, and it sort of tried to cheat by adding some vaguely Japanese natural scenery.
See this Kickstarter all the way back in 2013 for a great example of what Pokémon might actually look like as ukiyo-e woodblock prints.
I remember that Kickstarter. I follow the woodblock carver who worked with the artist. His name is David Bull and he has some great videos on woodblock carving on YouTube: https://m.youtube.com/@seseragistudio
ChatGPT Images (and some others like Nano Banana) are autoregressive models, and the text model generates the tokens in the same way which are then passed in the correct form for an image model to decode. That is how it can reason the prime numbers + Pokemon + art style conditional.
It is different from earlier image models which just use a simple text encoder and struggle with conditional logic if not explicitly trained for it.
If you look at the thinking steps it clearly includes:
> * Mapped Pokémon primes
> * Organized the grid
before generating the image.
So really it is first figuring out the exact specification of the layout, then prompting the image generation tool with this specification - perhaps unless you do it with an "instant" model.
I think that particular one would be a prompting problem. Instagrammers and anime artists do left-right two-tone dyes all the time, so left-right two-tone rather than meshed gray blonde would be a reasonable interpretation of "half blond and grey hair" based on representations on Internet collected data.
I think the super-AGI should be able to figure this one out. This is the kind of things I would expect to have to think about when prompting a local 4B model, not the latest Sol or whatever is driving the image model here.
I don't think this ChatGPT's interpretation was necessarily wrong, given the prompt "Image of a middle aged man with half blond and grey hair". Maybe it should have asked back for clarification but that's it.
This to me as if someone is complaining for a basket full of extremely thin sliced fried potatoes for an order of "chips", rather than like wood and metalworking scraps.
That prompt seems almost deliberately obtuse. The natural phrasing would be “greying blonde hair”. If you described someone as having half-X, half-Y hair, I’d expect to see something like that image, too.
whats incredible is when in their video they're talking about complexity, the cat's images have extra limbs in low and medium complexity. They could've easily generated more versions to get the clean versions. But it just doesn't seem like they care. Maybe this is what someone who is so used to using AI to move fast becomes: someone who drops the quality bar to 80%?
They mentioned multiple times in previous blog posts that they do not cherry pick examples.
Would you appreciate more if their model generated cats with 5 legs in 99% of cases, and they cherry picked that 1 in 100 image where it has 4 legs? Do you think doing that would make them more caring and professional?
I doubt their models are that bad for simple and common things like cats, probably takes just a handful of tries at most for something like getting the correct number of limbs. They could just mention the number of tries on the side like a little bubble showing up to show “Rendered in 5 attempts” or something along those lines. I wouldn’t think less of the tech personally.
That's an interesting observation, and that's also my feeling: Everything is now expected to move so fast, dropping quality bar to 80% or below is not uncommon.
Not like this is the fault of the AI progress, we could use AI to go beyond the quality what we used to deliver, too. It's just the pace expectations that are now "in the room" ?
People who make things often have an unrealistic expectation of how much users care about quality (and don't realize how much users actually care about price).
We've seen this play out over and over again, in many different sectors. The whole shtick of the old (Bellhead) telecom world was quality at the (literal) expense of price, which didn't work out well for them. Same for CD versus pirated mp3s, DVD versus pirated videos (and, to a lesser extend, theaters versus bootleg cam releases), native texting / GSM calls versus WhatsApp and similar, mainstream newspapers and media versus blogging and social media, traditional electrical appliances versus Chinese-made, and so on.
Because of the innovator's dilemma, In many cases, the cheaper, lower-quality product later surpasses the quality of its more expensive counterpart.
AI is just one more example of this happening. It lets you make more things which otherwise would be far too expensive to make, even if that often means a lower quality product.
Yeah, I've been mystified for years why they can't just train the "AI look" out of images. Is it, like, something undetectable by machine? But then how come we can often detect it so easily?
I think this is fundamental. People will probably dislike whatever it is that AI images look like, because whatever they look like is going to be associated with cheap. They could have a totally different aesthetic and I think most people's conception of cheap would adapt to whatever that is.
Some of it has to do with Classifier Free Guidance (CFG), a math thing that makes it more likely for a diffusion model to follow what you ask for in the prompt. However, turning up CFG inevitably causes high contrast and certain other artifacts.
With that being said, my understanding of that issue is a few years old, people have always said CFG is a havk until models get better at following prompts on their own. So thee “look” that tools like Nano Banana ends up with could be something else. It may or may not be a necessary artifact of some part of the process.
The problem isn't the classifier free guidance itself, but the choice of parameters. When a parameter above 1 is chosen you get images which very stronger conform to theme, but which there's not a good mathematical justification for.
Yeah, it must be some algorithmic thing like CFG, or some other technique, that causes the unrealistic AI look. Because Dall-E 2 didn't have it, but Dall-E 3 did to an extreme degree. Google's Imagen <= 3 also didn't have it, but Imagen 4 (and Nano Banana, which seems to use Imagen 4 in the background) suddenly had significant AI look, apart from much better prompt following and fewer visual mistakes (like too many fingers).
It's the opposite: the average person likes the AI look of images (the more "aesthetic" look) and RLHF keep converging toward that.
It is entirely possible to prompt engineer away from that style, but that requires effort and knowledge that said average person does not have, so it's stuck in a feedback loop.
This is not true. Think of the problems AI has with creating appetizing food imagery, for example. Images that every human annotator marks as unappealing with a high degree of consensus.
The problem is in the rewards, but it's more how they average across domains and contexts, so AI misapplies them. Actually, in human preference evaluations, most of Gen Z in particular extremely adverse to the common look of AI images.
So it's not the annotators involved in RLHF, it's the models failing to map to human preferences at the task level. The failure rate is also uneven across domains.
Where there are challenges with human annotation, it's not the ability to gauge what's good but rather how to correct what's wrong. Most people can tell when something is off but struggle to articulate why. That's where more domain expertise is required.
>This is not true. Think of the problems AI has with creating appetizing food imagery, for example. Images that every human annotator marks as unappealing with a high degree of consensus.
I think the way they'll solve it in the end is to create an ultra-high-fidelity simulation of the world including modelling taste and other senses on the level of physics, and then incorporate all of this into the training rewards.
I think this is how the machines know what tasty wheat tasted like.
I see restaurants post AI flyers and ads to random local Facebook groups and most of the replies are about them being AI slop. I think it’s an uncanny valley problem: People don’t mind if you post a real photo of your food or a cartoon drawing of a pizza or a pig on a spit, but when it’s a weird fake photo with texture issues, it looks gross, especially when it’s food.
I think a lot of people who open restaurants probably aren’t extremely online, partly because it would be physically impossible with the amount of work they have to do, but they definitely pay attention -sometimes too much attention — to public online feedback.
I get why restaurants use the cartoon-looking slop. If you don't particularly care that it signals that you put no effort into your menu or ad or that your restaurant's menu has the same shitty aesthetic as all other slop, fine. It's a cartoon and I could understand someone thinking it looks ok.
I absolutely do not get the uncanny-valley images. Those all look gross. Besides not actually being representative of the dish, the images have all sorts of weird texture issues, look too glossy, and are just unappetizing. I cannot understand why a restaurant owner would want to put up an ad that has food that looks straight up gross.
Professional food photography is expensive. This is a case of the free mediocre offer crowding out the much better but expensive option. Same way that free mail with ads crowded out email without ads.
The thing with those "AI Slop" comments are generally from people whom are opinionated about the image. However it's abundantly clear that the target demographic which is the customers generally don't mind. If you think about it, We have been conditioned by the "for illustrative purposes only" label on most food images that at this point most people just want a reference so they can decide what to eat. I'd take an AI image over just words any day.
I've been seeing "AI slop" stickers posted on things
> I see just everywhere including food menus
The places I have seen it in use are the places that would have a handwritten scribble that was misspelt and would be something mocked by people because it was written by ESOL (or vice versa, English speakers trying, but failing, to write in a different language)
Honestly, I think the AI slop is a step up - I can read the signs clearly, and they are only mildly offensive
Edit: I guess I take the view that - like a pirated movie filmed in a cinema - the only people that use/watch it are people that were never going to pay for the real thing anyway
You are actually a very skilled pattern recognizing machine. There's likely always going to be some bias and purple are going to be able to detect it and the human style will move away from it.
I doubt it, same with people that use ChatGPT for everything and don’t see the slip anymore and are confused when you point it out. LLM blindness is a real thing.
Someone posted an AI picture on Twitter, that was done in the style of a great classic painter of renown. Everyone tore it apart, discussing all the details that betrayed it as a pale imitation of the original. It was roundly panned as an example of how AI will never be good enough to replace humans. After this went on for some time, the initial poster revealed it wasn't actually AI at all, it was a genuine original from the artist in question.
People just hate AI and will find a post hoc justification for their hatred in every situation. The link you gave could have just as easily been someone expressing surprise and delight at the artists choice to not follow traditional physics... but instead be playful in unconstrained ways. Shrug.
Artists aren't machines that churn out one masterpiece after another, they do create bad works (most artists would acknowledge that) and there are one hit wonders. I look at a lot of art and sometimes I like only one painting out of a series, or even out of a lifetime of work. Also, you often need the context of the artist's broader work to judge the quality of a single piece, which might otherwise appear like a bad picture in vacuum.
that kind of experiment was done multiple times, someone put a very famous photo from a very famous photographer on some photography critique forum and it was torn apart: badly framed, blurry, no composition, ....
the vast majority of people have no idea what they are doing, going on vibes and copying the opinions of their peers
It does look terrible. But keep in mind these decisions aren't made to please you or me on HN. The numbers clearly show the average user does not care about the AI look because they can't really tell the difference between human-gen and AI-gen. Whatever saves them money is what they will go with.
> The numbers clearly show the average user does not care about the AI look because they can't really tell the difference between human-gen and AI-gen.
Do you have a link to the numbers that clearly show this?
Do you need statistical numbers for every information you receive? I don’t have any numbers to show you besides the fact that everyone I have spoken to in the last 6 months knows how to generate an image on ChatGPT. I work in the construction industry, and 100% of the companies I deal with have at some point posted an AI-generated poster on their social media, or created their flyers and business cards. YMMV.
So nothing at all. Most people would notice the difference between no ads everywhere, and ads everywhere, would prefer no ads, and still there are ads. That there are ads tells us nothing about people.
I'm deeply interested in this subject. In conversation, most people I talk to say they hate the look of AI generated "art," or find it uncanny or uncomfortable. However, whenever I'm on the bus or train, if I peek over the shoulder of someone, it's 50% chance they're endlessly scrolling AI generated short form video, either as unrelated clips or Chinese dramas (this being in Taiwan).
I suspect people are somewhat hypnotized by the content, in the same way a toddler is hypnotized by Cocomelon. Something about it is enthralling and keeps the attention. I wonder if it's related to the phone game icon / YouTube thumbnail phenomenon of "person's face making an extreme expression with the mouth open." AI generated videos for example still tend to have very exaggerated facial expressions and the mouth is always unusually wide when talking. I also wonder if something about the uncanny smoothness of movement (I assume caused by any form of transition defaulting to "linear") has a hypnotic effect.
I'm not sure how to look into this further. Elsagate is an interesting past example, and what I've learned so far I threw in a blog post a couple months back https://blog.calebjay.com/posts/psychic-attack/
Right now I'm looking into "Lula's army" and the AI generated content they used. I feel like there's something significant and perhaps even dangerous that could be developed bere.
The input and output pricing per token remains the same between 2.5 and 2.6, although using their image pricing calculator shows a quadrupling of token usage at low and medium quality settings for 2.6. For e.g. at 1024x1024:
2.5 low, 196 tokens, $0.006
2.6 low, 823 tokens, $0.025
2.5 med, 439 tokens, $0.013
2.6 med, 1449 tokens, $0.043
Interestingly, higher quality settings show no change in token usage between the models.
Similar results for my workflow. And the Round art comes out significantly less round without prompt changes (more input tokens)
Setting (1024x1024) Tokens Cost/image Time
2.5 medium 439 $0.014 12.0 s
2.6 low 824 $0.025 6.7 s
2.6 medium 1,449 $0.044 7.9 s
2.6 high 1,758 $0.053 8.9 s
Decided to try out one of the new features in GPT Image 2.6: the ability to natively generate animated GIF files.
I didn't have high expectations, and I still think you can probably get better results by doing a First-Last-Frame using something like Minimax H3 but I was still rather surprised that it could do it at all.
Here are some of the results. The last one shows its ability to take an existing image and convert it into an animated GIF. It's a picture of my husky inside a snow globe.
That's a bit hyperbolic. The windmill one is honestly not terrible (slight flicker on the door).
Honestly, you have to think about the use case. GIFs are going to be little throwaway things that you're pasting into Discord or meme-ing.
Nobody's reaching for a GIF if they're looking for professional results.
It's a shame that the SVG output seems to be limited to static vector graphics only though - or rather I couldn't get it to produce anything with the <animate> attribute.
Everyone is highly sensitive to slop, and it still looks like slop.
If you actually want something that looks decent, it feels like tool use would at least help cover the obvious blunders. Building layers to composite would go a long way to removing some slop here.
I love images 2, my only issue is the iterative changes. I want to change a tiny thing and it changes too or not exactly what I want, end up being a long conversation to get what I desire. I see this new model promises more control in editing, has anybody tried and experienced a decent improvement?
Did a considerably worse job than Image Gen 2.0 on my standard test prompt.
"A macro close-up photograph of an old watchmaker's hands carefully replacing a tiny gear inside a vintage pocket watch. The watch mechanism is partially submerged in a shallow dish of clear water, causing visible refraction and light caustics across the brass gears. A single drop of water is falling from a pair of steel tweezers, captured mid-splash on the water's surface. Reflect the watchmaker's face, slightly distorted, in the curved glass of the watch face. Sharp focus throughout, natural window lighting from the left, shot on 100mm macro lens."
I've been playing with it for the last hour or so. I immediately updated my platform to use both since 2.5 is my primary model (i run a prompt->design site).
After playing with it for an hour, I'm rolling back 2.6-flare to 2.5-flare. I've found 2.6-flare specifically to either be too simplistic in its approaches, or create outputs that are inherently broken in other cases.
I wonder if there's some weird accounting going on, or if this is real. It would mean around 24M images an hour and that's in the area of "more images than anyone will see". It means there's lots of wasted automated effort in this process somewhere, even if we assume some creators will request 8x variants of their ideas.
I don't use AI image generation anymore, but when I was fiddling with it I would generate dozens of images per idea, limited only by my patience. "The same thing but a little different" is the easiest feedback you can give anything.
Same, when I messed around with it to learn how the models worked I would generate batches of hundreds or thousands with slightly tweaked prompts and negative prompts to get a sense of how the inputs changed the outputs.
What's the state of the art for image gen? I'm far removed and pretty happy with the free Gemini tier but its notable to me that I rarely see boasting about being "the best image generator yet" and I wonder if that's because they're all just good enough?
It has to be Google, since OpenAI always compares to other companies when it feels its winning at some benchmarks. They didn't even compare on cost, speed, or any benchmarks at all.
I've been running a comparative image benchmark site for years now (see my profile for the link), ever since SDXL back in 2023, and the answer really comes down to it depends. My benchmark focuses primarily on prompt adherence, but that's just one metric you can use. (fidelity, censorship, explorative, speed, features, etc are others)
In terms of overall state of the art, the two major contenders are probably Nano Banana Pro (Google) and the GPT Image 2 series (OpenAI).
For pure prompt adherency, OpenAI has consistently lead the pack since the original gpt-image-1 released back in March of 2025.... but its OpenAI's fidelity has always left something to be desired. Version 2 of the series has definitely closed the gap a bit.
It's not a stupid question. It is multi-modal but it's about efficiency. A model that is trained purely on generating images is smaller and faster and has its own set of optimisation techniques, compared to a "generalist" multi modal LLM.
I... doubt that's true? I don't know all the details how these recent image generation models work, but last time I checked people still used diffusion models for it. Diffusion pipelines are _slower_ and _more expensive_ than the language model, not cheaper.
"generalist" multi-modal LLM mostly uses images (and videos, sounds, etc.) as input. I'm quite sure they have an output head for it, as well, used during training, but their outputs are probably not as good. They generate images much faster (hence probably cheaper), too: typically a token is 16x16 pixels, so to generate an image of modest 512x512 resolution would be the same as generating a 1024 token length string, that nowadays is pretty fast and cheap (e.g. with Sol it would be about $0.01). But the result won't be great.
Diffusion models instead are trained to generate a "denoising" step of the image. Typically they have some inputs (e.g. the owl sketch in the blog post and an embedding of your textual request), start either from pure noise or from a noisy version of an input image, and then repeatedly slightly modify it in anywhere between 4 and 100 (or more) steps (hyperparameter to be chosen, and also depends if you start from noise or from something non-pure-noise), each time making a version that the model believes to be a bit closer to where you want to get. The underlying model _could_ be similar to the language model, but I think it isn't, because (1) if you want to call it 100 times in 4 seconds it must be smaller and dedicated, and (2) it's trained to do _exactly_ that, i.e. remove 1/100th of the noise, no more, no less. So it must be a dedicated one.
A few years ago people did research on "one shot" methods (earliest were GANs of course, but also some autoregressive model like a multimodal LLM directly), but as far as I know, quality has always stayed well behind diffusion models. It was faster, though, not more expensive.
If you know how an LLM could have an output image head, that is purely generative without using diffusion, and not a classification/selection head, I’m really interested in learning more. I think you are onto something very new and novel here
On a phone, if you share a gif, it creates a link to ChatGPT.com with a static image. You can also save the image to photos and share directly from there which feels like an unnecessary step.
It feels like a sleazy pattern and defeats the purpose of why I’d want to generate a gif in the first place, which to make clear is likely going to be to share with friends in a chat. I’m not interested in driving traffic to ChatGPT.com to impress my friends.
As an aside, there was a recent post about building in an emulator in an open model. I think it was Qwen 3.6. Actually, I think this type of hard-coded calculation would be very well suited to a small image model as seemingly trivial tasks like symmetry or moving text or placing a vector are quite difficult to do for models currently.
I also take it for what it is, incredible none determinstic engineering.
That being said, i always feels a bit of sympathy for "the machine" when they say things like:
- Oh it horrible to use
- It changed everything instead of the one thing i asked (probably biggest misconceptions and misunderstood technical part among my none-techie friends).
I've heard the "reason" from my non-techie friends "oh 'they' do that on purpose ! So you have to upgrade.
Ive once or twice had the non-deterministic discussion with them.. but dont think explained it very well.
Either way for all its warts etc i still find Ai-images extremely useful, as someone that before couldnt really produce any visual-assets.
I really don't like this generative AI. I may look at it like an artist but completed generated software, writing, music and imagery is a robbery of human worth. Absolutely.
No, it's created by data scanned from the internet, books and fed to the models. How is that a good thing for anyone? I'd love your take. People spend years giving themselves value to others to live a life of worth and value.
The data just scanned itself and organized itself into an neural network? Would be an awesome display of bootstrapped intelligence if that happened! Sadly, it was the result of decades of human research.
Decades of human research and mountains of work used without permission. Not that they legally did or didn't need that permission, but ethically, kind of a dick move.
If you define pixel art as just a vague overall look, then yes, it can do pixel art. If you define pixel art as being limited to a fixed color palette, like 16 or 256 colors and aligned to a square grid, then no, not without some manual post-processing.
It's not helped by the fact that a lot of people who are not artistically inclined will generate pixel art on these image generators at 1MP resolutions (far higher than traditional pixel art), such that even the casual observer can typically see the bad smearing and fringing in the final result.
That depends on your definition of pixel art. (people are exceptionally pedantic about this)
With the right prompt engineering, it (and similar quality models) can indeed do relevant pixel art, but due to the existence of blotchiness/SynthID watermarks you'll need to do some sort of postprocessing regardless.
There are SDXL pixel art models out there that do actual pixel art. They generate images that are 8x too large, you need to downscale them and reduce colors externally.
yeah, the original Retro Diffusion models were an early example of that - IIRC it had models separately fine-tuned for 32x32, 64x64, etc. so you'd nearest neighbor squash'em down to that size after the initial gen to kill the high frequency noise.
You may notice that luxury brands are making stuff by hand now. It's already painfully uncool to use AI images, regardless of how "real" they look – because they're cheap
It looks like the images in there are 'based' on the "hot springs symbol" - a Japanese map symbol encoded into Unicode that seems to have become synonymous with "piping hot bread" in the West!
The images in the GPs link show a common pictogram for artisanal bread. The image itself is actually a close representation of the real thing. An image search for sourdough will return many images showing bread with a similar appearance.
Anecdotally, I’ve never seen that hot springs symbol used for bread, and I’ve lived in the west my whole life.
Not without irony, but connecting the bread symbol with the hot springs symbol feels like the sort of mistake an llm would make.
The only animation format it supports right now is GIF so you can only have a transparency channel that's either on or off. It doesn't support animated WebP.
Not sure what that says about grok, Claude, and runway flagging that and being unable to generate that, or me not having the prompting skills to do that without help.
That stained glass whale thing makes me especially grumpy. When it scrolled onto the page, my first thought was "whoah, what a neat idea" and then it was _immediately_ followed by "what a terribly execution."
Like, I'm not even an artist, so I'm sure there's a million things I'm missing, but it's just so upsetting to look at. There's weird random reflections on things that shouldn't have reflections, the coloured light absolutely doesn't actually track through the stained glass onto any surface that makes any sense, light is coming from every which way, it's just... gross to look at for anything more than a quarter of a second.
And this is what they're using to hype people up for it. /shrug
Yeah. I think that the pelican SVG benchmark is actually a much better reflection of the state of AI art than the visual quality of the images.
Based on current trends I would not be at all surprised if LLMs working with tools, SVGs and looking at their own work get there before diffusion models. I've stopped keeping up with the image models so maybe I'm underestimating the progress but I dropped out because they were so frustratingly bad at composition and integrating details together.
That whale image makes me think that the situation hasn't improved as much as I'd expected. It is a really cool idea and a pretty picture, but the whale looks a lot like it is flying over the top of some sort of huge glass dome on the ground.
Using the grid trick for animations to generate the individual frames has been around for awhile and not specific to ChatGPT: I'm surprised it took this long for one of the official providers to make it explicit. However, adherence between frames is generally tricky, and even in the examples there is a lot of noise between frames although that could be due to compression.
SVG support is interesting but I suspect shenanigans are afoot since that's not how image generation models work, and it would likely strip any SynthID watermarks. It's possible OpenAI has pipelined a raster-to-vector tool that does the conversion automatically. (I developed a similar pipeline and it works extremely well; I'll open source it at some point)
EDIT: Per documentation, there is an explicit `output_format: "svg"`, and I am indeed correct there is raster-to-vector in play:
> The API creates a reference image and uses it to produce the SVG. The quality setting controls that reference image. Decode data[0].b64_json as base64 and save the resulting UTF-8 XML bytes with a .svg extension.
OpenAI claims 50% faster generation with Images 2.6 but curiously price is the same as Image 2.5.
From what I can tell after some preliminary testing, you have to pass an additional large language model to the OpenAI API when generating SVGs.
I couldn't find this requirement in the documentation, but if you don't pass one in, you get this error:
I tried it with Sunburst with both gpt-5.5 and gpt-5.6-sol, and the results were nothing spectacular. It’s something you could already do in the ChatGPT interface by asking it to convert an existing image to an equivalent SVG.https://imgpb.com/bkUrc
Recraft is probably the only group I know of that had a native text-to-SVG generator pretty early on, but since it’s a closed model, I’m not familiar with what they were doing under the hood technically.
https://fal.ai/models/fal-ai/recraft/v4/pro/text-to-vector
Image 2.6 costs more tokens for generating low and medium quality images, so it ends up being a bit more expensive. Could refer to https://developers.openai.com/api/docs/guides/image-generati...
Yeh using sprites to make frames consistent (kinda tilting latent space towards a character consistency, zero-shot ~LoRA hack?) is old-school. I did this 3 years ago. https://j11y.io/dance/
Feels like ai image generation has plateaud.
Coding related have had massive step-up improvements over time.
The same cannot be said for image generation. If anything, I swear it's gone backwards.
It peaked with very early midjourney (like 2022), maybe follow ups with stable diffusion, and then with some of the nano banana models for actual editing. But overall, very lackluster in comparison to other areas.
> Feels like ai image generation has plateaud. > Coding related have had massive step-up improvements over time. > The same cannot be said for image generation. If anything, I swear it's gone backwards.
I would argue both are exactly the same. The images are getting nicer, the AI can alter pictures better. It's the same as coding.
The difference is our perception. Looking at image takes 2s, and we can spot what we don't like. Looking at a code change takes hours to understand. It's very difficult to tell how much code generation is improving because almost nobody is reading it.
Image and code generation don't correlate exactly, but I would be surprised to see a big jump of capabilities in either without the other quickly following.
There have been a lot of strides in open-weight models, well beyond SDXL. Flux, back in August of 2024 was a game changer, finally allowing us to move past SDXL’s nuclear pile of comma-delimited keywords for prompting and allowing more natural language.
And this summer we got Ideogram V4 and Krea2, both of which are highly capable. You can see how the open/closed models have advanced since SDXL in the Progress tab.
https://genai-showdown.specr.net
MJ is pretty out-of-date by this point though it can still be fun from an image space exploration perspective. It scored a rather pitiful 2 out of 15 on my bench back in the summer of 2025.
You could argue that progress has been relatively flat since Nano Banana Pro, but that's still less than a full year old (just barely), and that was the first model that could really fully maintain visual consistency across complex edits. The GPT ones are a little stylistically disappointing in ways, but they've still been making steady progress too. I don't think we can claim a peak just yet, even if it's been slower.
Your feeling is generally right. There have been no seminal moments in image generation anymore in the last couple of months. Most attention from the big labs has gone to video since that‘s where the money is. We have a page to track the history of it all here https://lumenfall.ai/arena/history
(Right now missing some models which I‘ll add today.)
There’s less money to be made with images, than code gen.
Investment in new models follows the money.
Seedream 4 continues to be the best model by far at character consistency. I havnt done my tests on this model yet, however.
Coding flattened out like 9 months ago. In reality, people are getting dumber and lazier from over using them, so they feel like they're improving when they aren't really, outside some areas like RE.
In reality RE has only gotten better because they RLd it into the models because they NEED data, so now every kid is decomping proprietary softwate and sending them back great training data.
>Coding flattened out like 9 months ago.
It absolutely has not. You do have a nice point with data though, somehow haven't thought about that
I miss the outpainting feature they used to have (3 years ago?) in the dall-e days. It feels like tooling overall for image generation has been standing still or almost regressed the past few years, while the quality has improved
I wish it was properly adjusted to design work. I can get some work done with Gemini over multiple prompts, center this, adjust this, keep stylized properties of this, etc.
https://chatgpt.com/share/6aebc6f8-d660-83e8-9cb3-49bdf8caa7...
I guess I should have been careful about “half”?
These models have a hard time imagining men with naturally blond hair getting grey.
I would caution against using the webapp to test the quality of this new model because OpenAI does not indicate which image model and effort is being used to generate the image, to my continued frustration.
That said, I gave it one of my more difficult image prompts and the model now surprisingly passes (without plagiaristic cheating): https://chatgpt.com/share/6aebcbcd-6b8c-83e8-b52b-dbedf20ce4...
I think it's more impressive that it managed so much coherence in a single image than anything else. Very few problems with each of the pokemon anatomically speaking. Mew's kind of a mess but the rest are pretty accurate.
That being said, the ukiyo-e artwork looks more like a standard colored pencil illustration, and it sort of tried to cheat by adding some vaguely Japanese natural scenery.
See this Kickstarter all the way back in 2013 for a great example of what Pokémon might actually look like as ukiyo-e woodblock prints.
https://www.kickstarter.com/projects/1499165518/ukiyo-e-hero...
https://ricedigital.co.uk/this-japanese-ukiyo-e-pokemon-art-...
I remember that Kickstarter. I follow the woodblock carver who worked with the artist. His name is David Bull and he has some great videos on woodblock carving on YouTube: https://m.youtube.com/@seseragistudio
Oh, thanks for showing me these videos! The way that he presents is so calm - he's like a woodprint Bob Ross.
Wow. Are those the correct Pokemon and everything? Quite cool.
Yes, the Pokemon are correct. Compare to my attempt a year ago with Nano Banana Pro, which plagiarized the images, didn't do interesting things with the background, and got 271 wrong: https://minimaxir.com/2025/12/nano-banana-pro/pokemongrid_hu...
When I did a similar prompt with ChatGPT 2.0 it was worse, so interesting to see improvement.
On the back end a language model is effectively writing a long detailed prompt describing this image exactly right? The image model can't know this.
ChatGPT Images (and some others like Nano Banana) are autoregressive models, and the text model generates the tokens in the same way which are then passed in the correct form for an image model to decode. That is how it can reason the prime numbers + Pokemon + art style conditional.
It is different from earlier image models which just use a simple text encoder and struggle with conditional logic if not explicitly trained for it.
If you look at the thinking steps it clearly includes:
> * Mapped Pokémon primes > * Organized the grid
before generating the image.
So really it is first figuring out the exact specification of the layout, then prompting the image generation tool with this specification - perhaps unless you do it with an "instant" model.
Yes it thinks about it and then calls the image tool with the detailed prompt with exactly what to include in the image
I think that particular one would be a prompting problem. Instagrammers and anime artists do left-right two-tone dyes all the time, so left-right two-tone rather than meshed gray blonde would be a reasonable interpretation of "half blond and grey hair" based on representations on Internet collected data.
I think the super-AGI should be able to figure this one out. This is the kind of things I would expect to have to think about when prompting a local 4B model, not the latest Sol or whatever is driving the image model here.
I don't think this ChatGPT's interpretation was necessarily wrong, given the prompt "Image of a middle aged man with half blond and grey hair". Maybe it should have asked back for clarification but that's it.
This to me as if someone is complaining for a basket full of extremely thin sliced fried potatoes for an order of "chips", rather than like wood and metalworking scraps.
It precisely matches your prompt, more accurately I think than your expectation which it meets if described accurately:
https://chatgpt.com/share/6aec2ba7-6edc-83ea-9a48-b41b53874a...
That prompt seems almost deliberately obtuse. The natural phrasing would be “greying blonde hair”. If you described someone as having half-X, half-Y hair, I’d expect to see something like that image, too.
Still looks like an eyesore, I genuinely want to know if the engineers who work on the product don't notice it
whats incredible is when in their video they're talking about complexity, the cat's images have extra limbs in low and medium complexity. They could've easily generated more versions to get the clean versions. But it just doesn't seem like they care. Maybe this is what someone who is so used to using AI to move fast becomes: someone who drops the quality bar to 80%?
They mentioned multiple times in previous blog posts that they do not cherry pick examples.
Would you appreciate more if their model generated cats with 5 legs in 99% of cases, and they cherry picked that 1 in 100 image where it has 4 legs? Do you think doing that would make them more caring and professional?
I doubt their models are that bad for simple and common things like cats, probably takes just a handful of tries at most for something like getting the correct number of limbs. They could just mention the number of tries on the side like a little bubble showing up to show “Rendered in 5 attempts” or something along those lines. I wouldn’t think less of the tech personally.
That's an interesting observation, and that's also my feeling: Everything is now expected to move so fast, dropping quality bar to 80% or below is not uncommon.
Not like this is the fault of the AI progress, we could use AI to go beyond the quality what we used to deliver, too. It's just the pace expectations that are now "in the room" ?
People who make things often have an unrealistic expectation of how much users care about quality (and don't realize how much users actually care about price).
We've seen this play out over and over again, in many different sectors. The whole shtick of the old (Bellhead) telecom world was quality at the (literal) expense of price, which didn't work out well for them. Same for CD versus pirated mp3s, DVD versus pirated videos (and, to a lesser extend, theaters versus bootleg cam releases), native texting / GSM calls versus WhatsApp and similar, mainstream newspapers and media versus blogging and social media, traditional electrical appliances versus Chinese-made, and so on.
Because of the innovator's dilemma, In many cases, the cheaper, lower-quality product later surpasses the quality of its more expensive counterpart.
AI is just one more example of this happening. It lets you make more things which otherwise would be far too expensive to make, even if that often means a lower quality product.
It could be that they've been using AI so long that they're unaware of the problem.
Yeah, I've been mystified for years why they can't just train the "AI look" out of images. Is it, like, something undetectable by machine? But then how come we can often detect it so easily?
I think this is fundamental. People will probably dislike whatever it is that AI images look like, because whatever they look like is going to be associated with cheap. They could have a totally different aesthetic and I think most people's conception of cheap would adapt to whatever that is.
Some of it has to do with Classifier Free Guidance (CFG), a math thing that makes it more likely for a diffusion model to follow what you ask for in the prompt. However, turning up CFG inevitably causes high contrast and certain other artifacts.
With that being said, my understanding of that issue is a few years old, people have always said CFG is a havk until models get better at following prompts on their own. So thee “look” that tools like Nano Banana ends up with could be something else. It may or may not be a necessary artifact of some part of the process.
The problem isn't the classifier free guidance itself, but the choice of parameters. When a parameter above 1 is chosen you get images which very stronger conform to theme, but which there's not a good mathematical justification for.
Yeah, it must be some algorithmic thing like CFG, or some other technique, that causes the unrealistic AI look. Because Dall-E 2 didn't have it, but Dall-E 3 did to an extreme degree. Google's Imagen <= 3 also didn't have it, but Imagen 4 (and Nano Banana, which seems to use Imagen 4 in the background) suddenly had significant AI look, apart from much better prompt following and fewer visual mistakes (like too many fingers).
It's the opposite: the average person likes the AI look of images (the more "aesthetic" look) and RLHF keep converging toward that.
It is entirely possible to prompt engineer away from that style, but that requires effort and knowledge that said average person does not have, so it's stuck in a feedback loop.
This is not true. Think of the problems AI has with creating appetizing food imagery, for example. Images that every human annotator marks as unappealing with a high degree of consensus.
The problem is in the rewards, but it's more how they average across domains and contexts, so AI misapplies them. Actually, in human preference evaluations, most of Gen Z in particular extremely adverse to the common look of AI images.
So it's not the annotators involved in RLHF, it's the models failing to map to human preferences at the task level. The failure rate is also uneven across domains.
Where there are challenges with human annotation, it's not the ability to gauge what's good but rather how to correct what's wrong. Most people can tell when something is off but struggle to articulate why. That's where more domain expertise is required.
> Think of the problems AI has with creating appetizing food imagery, for example.
For posterity, the memetic ugly food can be controlled and mostly filtered out with prompt engineering. Here's a quick ad hoc test example which (hopefully) uses the new model: https://chatgpt.com/share/6aebf030-9d38-83e8-a103-3509883905...
First image is fine, but attempts to make it more wacky turn it to the AI food slop. So then followup prompts deslopify it, mostly successfully.
Eh, looks like slop still in my opinion. The evenness and density of the teeth especially stands out to me as uncanny.
tbh it got worse every time, the first one is the best lol
>This is not true. Think of the problems AI has with creating appetizing food imagery, for example. Images that every human annotator marks as unappealing with a high degree of consensus.
I think the way they'll solve it in the end is to create an ultra-high-fidelity simulation of the world including modelling taste and other senses on the level of physics, and then incorporate all of this into the training rewards.
I think this is how the machines know what tasty wheat tasted like.
The saddest words of tongue or pen: “you’re not the target demographic”
Isn't this the sum of dating apps?
No, you are confusing a model with reality.
I still can’t believe the amount of deepfakes and slop I see just everywhere including food menus and people don’t even seem to notice.
This kind of content is here to stay and we’re going to see more and more of it. Unfortunately…
People definitely seem to notice.
I see restaurants post AI flyers and ads to random local Facebook groups and most of the replies are about them being AI slop. I think it’s an uncanny valley problem: People don’t mind if you post a real photo of your food or a cartoon drawing of a pizza or a pig on a spit, but when it’s a weird fake photo with texture issues, it looks gross, especially when it’s food.
I think a lot of people who open restaurants probably aren’t extremely online, partly because it would be physically impossible with the amount of work they have to do, but they definitely pay attention -sometimes too much attention — to public online feedback.
I get why restaurants use the cartoon-looking slop. If you don't particularly care that it signals that you put no effort into your menu or ad or that your restaurant's menu has the same shitty aesthetic as all other slop, fine. It's a cartoon and I could understand someone thinking it looks ok.
I absolutely do not get the uncanny-valley images. Those all look gross. Besides not actually being representative of the dish, the images have all sorts of weird texture issues, look too glossy, and are just unappetizing. I cannot understand why a restaurant owner would want to put up an ad that has food that looks straight up gross.
Professional food photography is expensive. This is a case of the free mediocre offer crowding out the much better but expensive option. Same way that free mail with ads crowded out email without ads.
The thing with those "AI Slop" comments are generally from people whom are opinionated about the image. However it's abundantly clear that the target demographic which is the customers generally don't mind. If you think about it, We have been conditioned by the "for illustrative purposes only" label on most food images that at this point most people just want a reference so they can decide what to eat. I'd take an AI image over just words any day.
> people don’t even seem to notice.
I've been seeing "AI slop" stickers posted on things
> I see just everywhere including food menus
The places I have seen it in use are the places that would have a handwritten scribble that was misspelt and would be something mocked by people because it was written by ESOL (or vice versa, English speakers trying, but failing, to write in a different language)
Honestly, I think the AI slop is a step up - I can read the signs clearly, and they are only mildly offensive
Edit: I guess I take the view that - like a pirated movie filmed in a cinema - the only people that use/watch it are people that were never going to pay for the real thing anyway
for the same reason they can't train llmisms out of text.
You are actually a very skilled pattern recognizing machine. There's likely always going to be some bias and purple are going to be able to detect it and the human style will move away from it.
I doubt it, same with people that use ChatGPT for everything and don’t see the slip anymore and are confused when you point it out. LLM blindness is a real thing.
This. You shoulda seen OpenAI’s own cookbook for GPT-Image-2 - literally their own prompting guide - and the dog comic was a complete mess.
Generative AI (coding, images, etc.) has only accelerated the atrophy of basic observational skills.
That stained glass sky-whale, at the bottom of the page, is pretty nice.
and yet https://news.ycombinator.com/item?id=50372945#50373586
Someone posted an AI picture on Twitter, that was done in the style of a great classic painter of renown. Everyone tore it apart, discussing all the details that betrayed it as a pale imitation of the original. It was roundly panned as an example of how AI will never be good enough to replace humans. After this went on for some time, the initial poster revealed it wasn't actually AI at all, it was a genuine original from the artist in question.
People just hate AI and will find a post hoc justification for their hatred in every situation. The link you gave could have just as easily been someone expressing surprise and delight at the artists choice to not follow traditional physics... but instead be playful in unconstrained ways. Shrug.
Artists aren't machines that churn out one masterpiece after another, they do create bad works (most artists would acknowledge that) and there are one hit wonders. I look at a lot of art and sometimes I like only one painting out of a series, or even out of a lifetime of work. Also, you often need the context of the artist's broader work to judge the quality of a single piece, which might otherwise appear like a bad picture in vacuum.
that kind of experiment was done multiple times, someone put a very famous photo from a very famous photographer on some photography critique forum and it was torn apart: badly framed, blurry, no composition, ....
the vast majority of people have no idea what they are doing, going on vibes and copying the opinions of their peers
I created some great designs and illustrations with imagen 2.5. I'm sure 2.6 will be even better.
It does look terrible. But keep in mind these decisions aren't made to please you or me on HN. The numbers clearly show the average user does not care about the AI look because they can't really tell the difference between human-gen and AI-gen. Whatever saves them money is what they will go with.
The average user is very unlikely, if preferences are high dimensional, and they are.
It's not like it's saving people money. Instead it's letting them create things they never could have before
> The numbers clearly show the average user does not care about the AI look because they can't really tell the difference between human-gen and AI-gen.
Do you have a link to the numbers that clearly show this?
Do you need statistical numbers for every information you receive? I don’t have any numbers to show you besides the fact that everyone I have spoken to in the last 6 months knows how to generate an image on ChatGPT. I work in the construction industry, and 100% of the companies I deal with have at some point posted an AI-generated poster on their social media, or created their flyers and business cards. YMMV.
The numbers of AI posters/signs/ads I see on the street…
So nothing at all. Most people would notice the difference between no ads everywhere, and ads everywhere, would prefer no ads, and still there are ads. That there are ads tells us nothing about people.
I'm deeply interested in this subject. In conversation, most people I talk to say they hate the look of AI generated "art," or find it uncanny or uncomfortable. However, whenever I'm on the bus or train, if I peek over the shoulder of someone, it's 50% chance they're endlessly scrolling AI generated short form video, either as unrelated clips or Chinese dramas (this being in Taiwan).
I suspect people are somewhat hypnotized by the content, in the same way a toddler is hypnotized by Cocomelon. Something about it is enthralling and keeps the attention. I wonder if it's related to the phone game icon / YouTube thumbnail phenomenon of "person's face making an extreme expression with the mouth open." AI generated videos for example still tend to have very exaggerated facial expressions and the mouth is always unusually wide when talking. I also wonder if something about the uncanny smoothness of movement (I assume caused by any form of transition defaulting to "linear") has a hypnotic effect.
I'm not sure how to look into this further. Elsagate is an interesting past example, and what I've learned so far I threw in a blog post a couple months back https://blog.calebjay.com/posts/psychic-attack/
Right now I'm looking into "Lula's army" and the AI generated content they used. I feel like there's something significant and perhaps even dangerous that could be developed bere.
Lula's army seems obviously inspired from K-pop armies
The model used by codex-cli is still gpt-image-2. https://github.com/openai/codex/blob/f3074e35c079bd8a5e89b03...
The input and output pricing per token remains the same between 2.5 and 2.6, although using their image pricing calculator shows a quadrupling of token usage at low and medium quality settings for 2.6. For e.g. at 1024x1024:
2.5 low, 196 tokens, $0.006
2.6 low, 823 tokens, $0.025
2.5 med, 439 tokens, $0.013
2.6 med, 1449 tokens, $0.043
Interestingly, higher quality settings show no change in token usage between the models.
Same results here, seems like a hidden price increase, hopefully they will still keep 2.5 running for some time.
Similar results for my workflow. And the Round art comes out significantly less round without prompt changes (more input tokens)
same in my testing; super deceptive marketing here. it’s 3-4x the true cost of 2.5.
Output image tokens per image (2.5 Flare → 2.6 Flare / Sunburst)
low, 1024x1024: 196 → 824 (4.2x) medium, 1024x1024: 439 → 1,449 (3.3x) low, 1024x1536: 158 → 664 (4.2x) medium, 1024x1536: 343 → 1,132 (3.3x)
Same $30/1M output tokens on all three, so per image: low: $0.006 → $0.025 medium: $0.013 → $0.043
from 800 to 1500 is barely 2.
from 200 to 450 is just a little over 2.
Nothing near "quadrupling"
Decided to try out one of the new features in GPT Image 2.6: the ability to natively generate animated GIF files.
I didn't have high expectations, and I still think you can probably get better results by doing a First-Last-Frame using something like Minimax H3 but I was still rather surprised that it could do it at all.
Here are some of the results. The last one shows its ability to take an existing image and convert it into an animated GIF. It's a picture of my husky inside a snow globe.
https://imgpb.com/OzSMKJ
The splotchiness on that simple chess animation makes this GIF feature a nonstarter.
Yeah, what's up with those splotches? Are those watermarks? I am sure they weren't there a few months ago.
That's a bit hyperbolic. The windmill one is honestly not terrible (slight flicker on the door).
Honestly, you have to think about the use case. GIFs are going to be little throwaway things that you're pasting into Discord or meme-ing.
Nobody's reaching for a GIF if they're looking for professional results.
It's a shame that the SVG output seems to be limited to static vector graphics only though - or rather I couldn't get it to produce anything with the <animate> attribute.
Everyone is highly sensitive to slop, and it still looks like slop.
If you actually want something that looks decent, it feels like tool use would at least help cover the obvious blunders. Building layers to composite would go a long way to removing some slop here.
Slight flicker on the door? The axle jumps to the right.
Yeahhhh but that’s like saying Dall-E isn’t good enough
Ain’t wrong buuuuut…
Just wasted 4hrs on the model by trying to do consistent sprite sheet art.
Nothing to see here.
How well does this handle SVG generation? Does it perform better than existing LLMs when asked to create SVGs?
Ask HN: Does anyone know whether a Codex subscription can generate images specifying gpt-image-2.6-sunburst without relying on an api_key?
I love images 2, my only issue is the iterative changes. I want to change a tiny thing and it changes too or not exactly what I want, end up being a long conversation to get what I desire. I see this new model promises more control in editing, has anybody tried and experienced a decent improvement?
Based on stolen artwork. F*ck this.
Off-topic, but the looping GIFs of stock prices going up / going down remind me of the currently airing anime https://en.wikipedia.org/wiki/FX_Fighter_Kurumi-chan .
Did a considerably worse job than Image Gen 2.0 on my standard test prompt.
"A macro close-up photograph of an old watchmaker's hands carefully replacing a tiny gear inside a vintage pocket watch. The watch mechanism is partially submerged in a shallow dish of clear water, causing visible refraction and light caustics across the brass gears. A single drop of water is falling from a pair of steel tweezers, captured mid-splash on the water's surface. Reflect the watchmaker's face, slightly distorted, in the curved glass of the watch face. Sharp focus throughout, natural window lighting from the left, shot on 100mm macro lens."
2.6: https://gov.f9.is/JRDpHJGu
2.0: https://gov.f9.is/cqyCwtq9
Whatever site you are hosting images seems to trap my mobile browser and require 3-4 back buttons to get out of.
Hmm, it's jumpshare one of those drag and drop file share companies (that I pay for). Thanks for the heads up I'll open a support ticket.
tweezers not holding the gear, and THOSE FINGERS, yick. What else?
2.5 is better than 2.6
I've been playing with it for the last hour or so. I immediately updated my platform to use both since 2.5 is my primary model (i run a prompt->design site).
After playing with it for an hour, I'm rolling back 2.6-flare to 2.5-flare. I've found 2.6-flare specifically to either be too simplistic in its approaches, or create outputs that are inherently broken in other cases.
Comparison examples: https://diffui.ai/app/canvas/178db5d5-ff87-4b43-aa98-2897773...
2.6-sunburst seems solid however.
Update: seems to be confirmed on arena.ai - 2.5 scores higher than 2.6 https://arena.ai/leaderboard/image/text-to-image
Update: based on info from design arena, it looks like gpt-image-2.6 was post-trained to output less text: https://x.com/DesignArena/status/2118124381155303918
This may be why text elements in particular appear more "zoomed in", and why it's providing less detail for UI design.
Oh thank god, the awful inspirational text shoved needlessly into everything has been excised.
I preferred 2.6 in all examples supplied!
I firmly disagree. There's subjective comparison, and there's objective comparison.
In two of the examples the images from 2.6 are inherently broken, ie: https://image.non.io/161d8f02-fc2a-40e0-91b2-9d27ceb8fef1.we...
Objectively there are tests here where 2.5 are better. Subjectively prefering the layout of 2.6 is fine, but there are blocking flaws.
FOUR BILLION IMAGES PER WEEK???
I wonder if there's some weird accounting going on, or if this is real. It would mean around 24M images an hour and that's in the area of "more images than anyone will see". It means there's lots of wasted automated effort in this process somewhere, even if we assume some creators will request 8x variants of their ideas.
I don't use AI image generation anymore, but when I was fiddling with it I would generate dozens of images per idea, limited only by my patience. "The same thing but a little different" is the easiest feedback you can give anything.
Same, when I messed around with it to learn how the models worked I would generate batches of hundreds or thousands with slightly tweaked prompts and negative prompts to get a sense of how the inputs changed the outputs.
How depressing.
What's the state of the art for image gen? I'm far removed and pretty happy with the free Gemini tier but its notable to me that I rarely see boasting about being "the best image generator yet" and I wonder if that's because they're all just good enough?
It has to be Google, since OpenAI always compares to other companies when it feels its winning at some benchmarks. They didn't even compare on cost, speed, or any benchmarks at all.
How do you sensibly benchmark this product?
https://openrouter.ai/benchmarks/media/images
Depends on the use. For pics, nano banana still feels like the best. For designs/posters, gpt-image is king by a very large margin.
Generally depends on use case and style. Midjourney, Seedream, GPT Image are commonly regarded contenders IMO.
I've been running a comparative image benchmark site for years now (see my profile for the link), ever since SDXL back in 2023, and the answer really comes down to it depends. My benchmark focuses primarily on prompt adherence, but that's just one metric you can use. (fidelity, censorship, explorative, speed, features, etc are others)
In terms of overall state of the art, the two major contenders are probably Nano Banana Pro (Google) and the GPT Image 2 series (OpenAI).
For pure prompt adherency, OpenAI has consistently lead the pack since the original gpt-image-1 released back in March of 2025.... but its OpenAI's fidelity has always left something to be desired. Version 2 of the series has definitely closed the gap a bit.
The blog post says
> Images 2.6 is available globally today in ChatGPT and Codex across all subscription tiers on desktop, mobile, and web.
But I just updated my codex-cli to the latest version (still 0.166.0) and it is still unable to generate gifs. Very disappointing
Stupid question: why is there a separate image model? Shouldn't GPT-6.x be multimodal?
It's not a stupid question. It is multi-modal but it's about efficiency. A model that is trained purely on generating images is smaller and faster and has its own set of optimisation techniques, compared to a "generalist" multi modal LLM.
Same applies for speech models.
I... doubt that's true? I don't know all the details how these recent image generation models work, but last time I checked people still used diffusion models for it. Diffusion pipelines are _slower_ and _more expensive_ than the language model, not cheaper.
"generalist" multi-modal LLM mostly uses images (and videos, sounds, etc.) as input. I'm quite sure they have an output head for it, as well, used during training, but their outputs are probably not as good. They generate images much faster (hence probably cheaper), too: typically a token is 16x16 pixels, so to generate an image of modest 512x512 resolution would be the same as generating a 1024 token length string, that nowadays is pretty fast and cheap (e.g. with Sol it would be about $0.01). But the result won't be great.
Diffusion models instead are trained to generate a "denoising" step of the image. Typically they have some inputs (e.g. the owl sketch in the blog post and an embedding of your textual request), start either from pure noise or from a noisy version of an input image, and then repeatedly slightly modify it in anywhere between 4 and 100 (or more) steps (hyperparameter to be chosen, and also depends if you start from noise or from something non-pure-noise), each time making a version that the model believes to be a bit closer to where you want to get. The underlying model _could_ be similar to the language model, but I think it isn't, because (1) if you want to call it 100 times in 4 seconds it must be smaller and dedicated, and (2) it's trained to do _exactly_ that, i.e. remove 1/100th of the noise, no more, no less. So it must be a dedicated one.
A few years ago people did research on "one shot" methods (earliest were GANs of course, but also some autoregressive model like a multimodal LLM directly), but as far as I know, quality has always stayed well behind diffusion models. It was faster, though, not more expensive.
If you know how an LLM could have an output image head, that is purely generative without using diffusion, and not a classification/selection head, I’m really interested in learning more. I think you are onto something very new and novel here
My biggest surprise with this release is the amount of upvotes on HN. It's literally the same as what we've had the past 1.5 years, or more.
On a phone, if you share a gif, it creates a link to ChatGPT.com with a static image. You can also save the image to photos and share directly from there which feels like an unnecessary step.
It feels like a sleazy pattern and defeats the purpose of why I’d want to generate a gif in the first place, which to make clear is likely going to be to share with friends in a chat. I’m not interested in driving traffic to ChatGPT.com to impress my friends.
But then Meta/Apple/Google gets your behavioral surplus for selling Ads instead of OpenAI. Do you want higher subscription costs?
As an aside, there was a recent post about building in an emulator in an open model. I think it was Qwen 3.6. Actually, I think this type of hard-coded calculation would be very well suited to a small image model as seemingly trivial tasks like symmetry or moving text or placing a vector are quite difficult to do for models currently.
Lol honestly i like chatgpt image function.
I also take it for what it is, incredible none determinstic engineering.
That being said, i always feels a bit of sympathy for "the machine" when they say things like:
- Oh it horrible to use
- It changed everything instead of the one thing i asked (probably biggest misconceptions and misunderstood technical part among my none-techie friends).
I've heard the "reason" from my non-techie friends "oh 'they' do that on purpose ! So you have to upgrade.
Ive once or twice had the non-deterministic discussion with them.. but dont think explained it very well.
Either way for all its warts etc i still find Ai-images extremely useful, as someone that before couldnt really produce any visual-assets.
I don’t care for image gen, but I appreciate that they drew the rest of the owl.
Thats why it became completely incompetent today trying to generate an image! I love half baked "test it in production" updates...
worse than images 2.5, the yellow wash slop is back image models regressing?
My only use case for this is animating custom Slack emojis.
But trust, I'm a have a field day with that.
“Preserve the background”
blurs background
That has to be a "draw the rest of the fucking owl" reference in the middle there, right?
https://imgur.com/how-i-feel-when-somebody-gives-me-advice-g...
I really don't like this generative AI. I may look at it like an artist but completed generated software, writing, music and imagery is a robbery of human worth. Absolutely.
Generative AI is created by human worth.
No, it's created by data scanned from the internet, books and fed to the models. How is that a good thing for anyone? I'd love your take. People spend years giving themselves value to others to live a life of worth and value.
The data just scanned itself and organized itself into an neural network? Would be an awesome display of bootstrapped intelligence if that happened! Sadly, it was the result of decades of human research.
Decades of human research and mountains of work used without permission. Not that they legally did or didn't need that permission, but ethically, kind of a dick move.
Yep, that's the sad part.
Human wealth? Humans, some, are worth more than others, per birth rates and varying investments by institutions.
SVG? Who can generate a pelican on a bicycle?
Why do their images still have the classic piss yellow tint? Gemini ones don't look like this. Is it a deliberate way to make their slop identifiable?
Can it finally do pixel art?
If you define pixel art as just a vague overall look, then yes, it can do pixel art. If you define pixel art as being limited to a fixed color palette, like 16 or 256 colors and aligned to a square grid, then no, not without some manual post-processing.
It's not helped by the fact that a lot of people who are not artistically inclined will generate pixel art on these image generators at 1MP resolutions (far higher than traditional pixel art), such that even the casual observer can typically see the bad smearing and fringing in the final result.
That depends on your definition of pixel art. (people are exceptionally pedantic about this)
With the right prompt engineering, it (and similar quality models) can indeed do relevant pixel art, but due to the existence of blotchiness/SynthID watermarks you'll need to do some sort of postprocessing regardless.
There are SDXL pixel art models out there that do actual pixel art. They generate images that are 8x too large, you need to downscale them and reduce colors externally.
yeah, the original Retro Diffusion models were an early example of that - IIRC it had models separately fine-tuned for 32x32, 64x64, etc. so you'd nearest neighbor squash'em down to that size after the initial gen to kill the high frequency noise.
You may notice that luxury brands are making stuff by hand now. It's already painfully uncool to use AI images, regardless of how "real" they look – because they're cheap
wait... did we get (sort of) sora back???
Video generation models work much differently than what this GIF pipeline is doing here.
they always use horrible examples in these blog posts - that whale picture a the bottom is terrible
"keep the shallow depth of field"
adds more depth of field
The gifs really accentuate the artifacting, not sure I would've lead with that.
Do they have an example of an SVG image created by the new model somewhere?
Documentation has two: https://developers.openai.com/api/docs/guides/image-generati...
It looks like the images in there are 'based' on the "hot springs symbol" - a Japanese map symbol encoded into Unicode that seems to have become synonymous with "piping hot bread" in the West!
https://emojipedia.org/hot-springs
The images in the GPs link show a common pictogram for artisanal bread. The image itself is actually a close representation of the real thing. An image search for sourdough will return many images showing bread with a similar appearance. Anecdotally, I’ve never seen that hot springs symbol used for bread, and I’ve lived in the west my whole life. Not without irony, but connecting the bread symbol with the hot springs symbol feels like the sort of mistake an llm would make.
I always thought it was coffee like the Java logo lol
Thanks, but that is a bit underwhelming, isn't it? I was hoping to at least see an SVG of something like the owl on the main page.
Still has the gross artifacts (visible watermarking) that the other GPT-Image-2.* has. Google's Nano banana watermarking is practically invisible.
SVG is cool though, I wonder how good it is.
Stop, stop. AI is dead. Bailouts aren't coming.
"we have Steve Aoki at home"
Can we do animated sprites with alpha channel yet?
The only animation format it supports right now is GIF so you can only have a transparency channel that's either on or off. It doesn't support animated WebP.
See the second image of the bird flying.
https://imgpb.com/OzSMKJ
lol, I tried
> Create a GIF where xkcd 2347 gets hit by an airliner like the twin towers and the pilot is AI
and got a refusal.
> We’re so sorry, but the image we created may violate our guardrails around violence.
“Is the topic and concept from the original clear?“
Follow by
“Make it”
Seems to resolve the refusals
Wow. Thanks!
https://solyin.com/dependency-ai.gif
Not sure what that says about grok, Claude, and runway flagging that and being unable to generate that, or me not having the prompting skills to do that without help.
...I tried a similar prompt today and it refused
Nice, but they still haven't reopened two of my closed accounts. They're the only service I can't use.
anyone else get the feeling this is a chinese model they are just hosting themselves?