OpenAI's livestream of GPT-4o Image Generation shows that it is slowwwwwwwwww (maybe 30 seconds per image, which Sam Altman had to spin "it's slow but the generated images are worth it"). Instead of using a diffusion approach, it appears to be generating the image tokens and decoding them akin to the original DALL-E (https://openai.com/index/dall-e/), which allows for streaming partial generations from top to bottom. In contrast, Google's Gemini can generate images and make edits in seconds.
No API yet, and given the slowness I imagine it will cost much more than the $0.03+/image of competitors.
LLMs are autoregressive, so they can't be (multi-modality) integrated with diffusion image models, only with autoregressive image models (which generate an image via image tokens). Historically those had lower image fidelity than diffusion models. OpenAI now seems to have solved this problem somehow. More than that, they appear far ahead of any available diffusion model, including Midjourney and Imagen 3.
Gemini "integrates" Imagen 3 (a diffusion model) only via a tool that Gemini calls internally with the relevant prompt. So it's not a true multimodal integration, as it doesn't benefit from the advanced prompt understanding of the LLM.
Edit: Apparently Gemini also has an experimental native image generation ability.
No that seems to be indeed a native part of the multimodal Gemini model. I didn't know this existed, it's not available in the normal Gemini interface.
This is a pretty good example of the current state of Google LLMs:
The (no longer, I guess) industry-leading features people actually want are hidden away in some obscure “AI studio” with horrible usability, while the headline Gemini app still often refuses to do anything useful for me. (Disclaimer: I last checked a couple of months ago, after several more of mild amusement/great frustration.)
That's pretty disappointing, it has been out for a while, and we still get top comments like (https://news.ycombinator.com/item?id=43475043) where people clearly think native image generation capability is new. Where do you usually get your updates from for this kind of thing?
ByteDance has been working on autoregressive image generation for a while (see VAR, NeurIPS 2024 best paper). Traditionally they weren't in the open-source gang though.
The VAR paper is very impressive. I wonder if OpenAI did something similar. But the main contribution in the new GPT-4o feature doesn't seem to be just image quality (which VAR seems to focus on), but also massively enhanced prompt understanding.
Gemini added their multimodal Flash model to Google AI Studio some time ago. It does not use Imagen via tool, it's uses native capabilities to manipulate images, and it's free to try.
Meta has experimented with a hybrid mode, where the LLM uses autoregressive mode for text, but within a set of delimiters will switch to diffusion mode to generate images. In principle it's the best of both worlds.
That's overly pessimistic. Diffusion models take an input and produce an output. It's perfectly possible to auto-regressively analyze everything up to the image, use that context to produce a diffusion image, and incorporate the image into subsequent auto-regressive shenanigans. You'll preserve all the conditional probability factorizations the LLM needs while dropping a diffusion model in the middle.
As a user, images feel slightly slower but comparable to the previous generation. Given the significant quality improvement, it's a fair trade-off. Overall, it feels snappy, and the value justifies a higher price.
> it appears to be generating the image tokens and decoding them akin to the original DALL-E
The animation is a lie. The new 4o with "native" image generating capabilities is a multi-modal model that is connected to a diffusion model. It's not generating images one token at a time, it's calling out to a multi-stage diffusion model that has upscalers.
You can ask 4o about this yourself, it seems to have a strong understanding of how the process works.
There are many clues to indicate that the animation is a lie. For example, it clearly upscales the image using an external tool after the first image renders. As another example, if you ask the model about the tokens inside of its own context, it can't see any pixel tokens.
A model may not have many facts about itself, but it can definitely see what is inside of its own context, and what it sees is a call to an image generation tool.
Finally, and most convincingly, I can't find a single official source where OpenAI claims that the image is being generated pixel-by-pixel inside of the context window.
Sorry but I think you may be mistaken if your only source is ChatGPT. It's not aware of its own creation processes beyond what is included in its system prompt.
i find this “slow” complaint (/observation— i dont view this comment as a complaint, to be clear) to be quite confusing. slow… compared to what, exactly? you know what is slow? having to prompt and reprompt 15 times to get the stupid model to spell a word correctly and it not only refuses, but is also insistent that it has corrected the error this time. and afaict this is the exact kind of issue this change should address substantially.
im not going to get super hyperbolic and histrionic about “entitlement” and stuff like that, but… literally this technology did not exist until like two years ago, and yet i hear this all the time. “oh this codegen is pretty accurate but it’s slow”, “oh this model is faster and cheaper (oh yeah by the way the results are bad, but hey it’s the cheapest so it’s better)”. like, are we collectively forgetting that the whole point of any of this is correctness and accuracy? am i off-base here?
the value to me of a demonstrably wrong chat completion is essentially zero, and the value of a correct one that anticipates things i hadn’t considered myself is nearly infinite. or, at least, worth much, much more than they are charging, and even _could_ reasonably charge. it’s like people collectively grouse about low quality ai-generated junk out of one side of their mouths, and then complain about how expensive the slop is out of the other side.
hand this tech to someone from 2020 and i guarantee you the last thing you’d hear is that it’s too slow. and how could it be? yeah, everyone should find the best deals / price-value frontier tradeoff for their use case, but, like… what? we are all collectively devaluing that which we lament is being devalued by ai by setting such low standards: ourselves. the crazy thing is that the quickly-generated slop is so bad as to be practically useless, and yet it serves as the basis of comparison for… anything at all. it feels like that “web-scale /dev/null” meme all over again, but for all of human cognition.
If you look at the examples given, this is the first time I've felt like AI generated images have passed the uncanny valley.
The results are ground breaking in my opinion. How much longer until an AI can generate 30 successive images together and make an ultra realistic movie?
> ChatGPT’s new image generation in GPT‑4o rolls out starting today to Plus, Pro, Team, and Free users as the default image generator in ChatGPT, with access coming soon to Enterprise and Edu. For those who hold a special place in their hearts for DALL·E, it can still be accessed through a dedicated DALL·E GPT.
> Developers will soon be able to generate images with GPT‑4o via the API, with access rolling out in the next few weeks.
That's it folks. Tens of thousands of so-called "AI" image generator startups have been obliterated and taking digital artists with them all reduced to near zero.
Now you have a widely accessible meme generator with the name "ChatGPT".
The last task is for an open weight model that competes against this and is faster and all for free.
Yep. The coherence and text quality is insanely good. Keen to play with it to find it's "mangled hands" style deficiencies, because of course they cherry picked the best examples.
> Tens of thousands of so-called "AI" image generator startups have been obliterated and taking digital artists with them all reduced to near zero. Now you have a widely accessible meme generator with the name "ChatGPT".
ChatGPT has already had a that via Dall-E. If it didn't kill those startups when that happened this doesn't fundamentally change anything. Now its got a new image gen model, which — like Dall-E 3 when it came out — is competitive or ahead of other SotA base models using just text prompts, the simplest generation workflow, but both more expensive and less adaptable to more involved workflows than the tools anyone more than a casual user (whether using local tools or hosted services) is using. This is station-keeping for OpenAI, not a meaningful change in the landscape.
There are several examples here, especially in the videos that no existing image gen model can do and would require tedious workflows and/or training regimens to replicate, maybe.
It's not 'just' a new model ala Imagen 3. This is 'what if GPT could transform images nearly as well as text?' and that opens up a lot of possibilities. It's definitely a meaningful change.
OpenAI was started with the express goal of undermining Google's potential lead in AI. The fact that they time launches to Google launches to me indicates they still see this as a meaningful risk. And with this launch in particular I find their fears more well-founded than ever.
Exactly. OpenAI isn't going to win image and video.
Sora is one of the worst video generators. The Chinese have really taken the lead in video with Kling, Hailuo, and the open source Wan and Hunyuan.
Wan with LoRAs will enable real creative work. Motion control, character consistency. There's no place for an OpenAI Sora type product other than as a cheap LLM add-in.
Flux 1.1 Pro has good prompt adherence, but some of these (admittingly cherry-picked) GPT-4o generated image demos are beyond what you would get with Flux without a lot of iteration, particularly the large paragraphs of text.
I'm excited to see what a Flux 2 can do if it can actually use a modern text encoder.
Structural editing and control nets are much more powerful than text prompting alone.
The image generators used by creatives will not be text-first.
"Dragon with brown leathery scales with an elephant texture and 10% reflectivity positioned three degrees under the mountain, which is approximately 250 meters taller than the next peak, ..." is not how you design.
Creative work is not 100% dice rolling in a crude and inadequate language. Encoding spatial and qualitative details is impossible. "A picture is worth a thousand words" is an understatement.
We're not trying to replace human artists. We're trying to make them more efficient.
We might find that the entire "studio system" is a gross inefficiency and that individual artists and directors can self-publish like on Steam or YouTube.
> Yeah, but then it no longer replaces human artists.
Automation tools are always more powerful as a force multiplier for skilled users than a complete replacement. (Which is still a replacement on any given task scope, since it reduces the number of human labor hours — and, given any elapsed time constraints, human laborers — needed.)
Prompt adherence and additional tricks such as ControlNet/ComfyUI pipelines are not mutually exclusive. Both are very important to get good image generation results.
It is when it's kept behind an API. You cannot use Controlnet/ComfyUI and especially not the best stuff like regional prompting with this model. You can't do it with Gemini, and that's by design because otherwise coomers are going to generate 999999 anime waifus like they do on Civit.ai.
That's a fun idea—but generating an image with 999,999 anime waifus in it isn't technically possible due to visual and processing limits. But we can get creative.
Want me to generate:
1. A massive crowd of anime waifus (like a big collage or crowd scene)?
2. A stylized representation of “999999 anime waifus” (maybe with a few in focus and the rest as silhouettes or a sea of colors)?
3. A single waifu with a visual reference to the number 999999 (like a title, emblem, or digital counter in the background)?
Let me know your vibe—epic, funny, serious, chaotic?
It can do in-context learning from images you upload. So you can just upload a depth map or mark up an image with the locations of edits you want and it should be able to handle that. I guess my point is that since its the same model that understands how to see images and how to generate them you aren't restricted from interacting with it via text only.
EDIT: Seems not, "The smallest image size I can generate is 1024x1024. Would you like me to proceed with that, or would you like a different approach?"
OpenAI themselves discourages using GPT-4 outside of legacy applications, in favor of GPT-4o instead (they are shutting down the large output gpt-4-32k variants in a few months). GPT-4 is also an order of magnitude more expensive/slower.
I think both of these points are what sow doubt in some people in the first place because both could be true if GPT-4 was just less profitable to run, not if it was worse in quality. Of course it is actually worse in quality than 4o by any reasonable metric... but I guess not everyone sees it that way.
I work on a product for generating interactive fanfiction using an LLM, and I've put a lot of work into post-training to improve writing quality to match or exceed typical human levels.
I'm excited about this for adding images to those interactive stories.
It has nothing to do with circumventing the cost of artists or writers: regardless of cost, no one can put out a story and then rewrite it based on whatever idea pops into every reader's mind for their own personal main character.
It's a novel experience that only a "writer" that scales by paying for an inanimate object to crunch numbers can enable.
Similarly no artist can put out a piece of art for that story and then go and put out new art bespoke to every reader's newly written story.
-
I think there's this weird obsession with framing these tools about being built to just replace current people doing similar things. Just speaking objectively: the market for replacing "cheeky expensive artists" would not justify building these tools.
The most interesting applications of this technology being able to do things that are simply not possible today even if you have all the money in the world.
And for the record, I'll be ecstatic for the day an AI can reach my level of competency in building software. I've been doing it since I was a child because I love it, it's the one skill I've ever been paid for, and I'd still be over the moon because it'd let me explore so many more ideas than I alone can ever hope to build.
> That is a great right, as long as it's not programmers.
You realize that almost weekly we have new AI models coming out that are better and better at programming? It just happened that the image generation is an easier problem than programming. But make no mistake, AI is coming for us too.
This is really impressive, but the "Best of 8" tag on a lot of them really makes me want to see how cherry-picked they are. My three free images had two impressive outputs and one failure.
While drawing hands is difficult (because the surface morphs in a variety of ways), the shapes and relative proportions are quite simple. That’s how you can have tools like Metahuman[0]
SD extensions like rembg are post-processing effects - with their video transparency demo I'd be curious if 4o actually did training with an alpha channel.
am I dumb or every time they release something I can never find out how to actually use it and forget about it. take this for instance I wanted to try out their newton "an infographic explaining newton's prism experiment in great detail" example, but it generated a very bad result but maybe it's because I'm not using the right model? every release of theirs is not really a release, it's like a trailer. right?
You're not dumb. They do this for nearly every single major release. I can't really understand why considering it generates negative sentiment about the release, but it's something to be expected from OpenAI at this point.
This is what's so wild about Anthropic. When they release it seems like it's rolled out to all users, and API customers immediately. OpenAI has MONTHS between annoucement and roll out, or if they do it's usually just influencers who get an "early look". It's pretty frustrating.
One area where it does not work well at all is modifying photographs of people's faces.* Completely fumbles if you take a selfie and ask it to modify your shirt, for example.
The Americas are quite a bit larger than the USA, so I disagree with 'american' being a word for people and things from mainland USA. Usian seems like a reasonable derivative of USA and US, similar to how mexican follows from Mexico and Estados Unidos Mexicanos.
> We’re aware of a bug where the model struggles with maintaining consistency of edits to faces from user uploads but expect this to be fixed within the week.
Sounds like it may be a safety thing that's still getting figured out
It just doesn't have that kind of image editing capability. Maybe people just assume it does because Google's similar model has it. But did OpenAI claim it could edit images?
Yes it does, and that's one of the most important parts of it being multi-modal: just like it can make targeted edits at a piece of text, it can now make similarly nuanced edits to an image. The character consistency and restyling they mention are all rooted in the same concepts.
It's incredible that this took 316 days to be released since it was initially announced. I do appreciate the emphasis in the presentation on how this can be useful beyond just being a cool/fun toy, as it seems most image generation tools have functioned.
Was anyone else surprised how slow the images were to generate in the livestream? This seems notably slower than DALLE.
I've never minded that an image might take 10-30 seconds to generate. The fact that people do is crazy to me. A professional artist would take days, and cost $100s for the same asset.
I ran stable diffusion for a couple of years (maybe?, time really hasn't made sense since 2020) on my Dual 3090 rendering server. I built the server originally for crypto heating my office in my 1820s colonial in upstate NY then when I was planning to go back to college (got accepted into a university in England), I switched it's focus to Blender/UE4 (then 5), then eventually to AI image gen. So I've never minded 20 seconds for an image. If I needed dozens of options to pick the best, I was going to click start and grab a cup of coffee, come back and maybe it was done. Even if it took 2 hours, it is still faster than when I used to have to commission art for a project.
I grew out of Stable Diffusion, though, because the learning curve beyond grabbing a decent checkpoint and clicking start was actually really high (especially compared to LLMs that seamed to "just work"), after going through failed training after failed fine-tuning using tutorials that were a couple days out of date, I eventually said, fuck it, I'm paying for this instead.
All that to say - if you are using GenAI commercially, even if an image or a block of code took 30 minutes, it's still WAY cheaper than a human. That said, eventually a professional will be involved, and all the AI slop you generated will be redone, which will still cost a lot, but you get to skip the back and forth figuring out style/etc.
Edit: Please ignore. They hadn't rolled the new model out to my account yet. The announcement blog post is a bit misleading saying you can try it today.
My bad, I was trying the conversational aspect, but that's not an apples to apples conparison. I have put a direct one shot example in the original post as well.
I'm my test a few months ago, I found that just starting a new prompt would not clear GPT's memory about what I had asked for in previous conversations. You might be stuck with 2D animation style for a while. :)
True. I had that conversation before deciding to compare to others. I have updated the post with other fairer examples. Nowhere near Leonardo Phoenix or Flux for this simple image at least.
It's rolling out to everyone starting today but i'm not sure if everyone has it yet.
Does it generate top down for you (picture goes from mostly blurry to clear starting from the top) like in their presentation ?
On mine I tried it "natively" and in DALL-E mode and the results were basically identical, I think they haven't actually rolled it out to everyone yet.
Yeah, its just not good enough. The big labs are way behind what the image focused labs are putting out. Flux and Midjourney are running laps around these guys
It seems like an odd way to name/announce it, there's nothing obvious to distinguish it from what was already there (i.e. 4o making images) so I have no idea if there is a UI change to look for, or just keep trying stuff until it seems better?
If only OpenAI would dogfood their own product and use ChatGPT to make different choices with marketing that are less confusing than whoever's driving that bus now.
Ever since Midjourney popularized it, image generation models are often posttrained on more "aesthetic" subsets of images to give them a more fantasy look. It also help obscure some of the imperfections of the AI.
Wild speculation: video game engines. You want your model to understand what a car looks like from all angles, but it’s expensive to get photos of real cars from all angles, so instead you render a car model in UE5, generating hundreds of pictures of it, from many different angles, in many different colors and styles.
A large part of deviantart.com would fit that description. There are also a lot of cartoony or CG images in communities dedicated to fanart. Another component in there is probably the overly polished and clean look of stock images, like the front page results of shutterstock.
"Typical" AI images are this blend of the popular image styles of the internet. You always have a bit of digital drawing + cartoon image + oversaturated stock image + 3d render mixed in. Models trained on just one of these work quite well, but for a generalist model this blend of styles is an issue
> There are also a lot of cartoony or CG images in communities dedicated to fanart.
Asian artists don't color this way though; those neon oversaturated colors are a Western style.
(This is one of the easiest ways to tell a fake-anime western TV show, the colors are bad. The other way is that action scenes don't have any impact because they aren't any good at planning them.)
I've heard this is downstream of human feedback. If you ask someone which picture is better, they'll tend to pick the more saturated option. If you're doing post-training with humans, you'll bake that bias into your model.
It's largely an artifact of classifier-free guidance used in diffusion models. It makes the image generation more closely follow the prompt but also makes everything look more saturated and extreme.
Maybe it’s not useless. 1) it’s only comparing it to their own products and 2) it’s useful to know that the product is the current best in their offering as opposed to a new product that might offer new functionality but isn’t actually their most advanced.
Which is especially relevant when it's not obvious which product is the latest and best just looking at the names. Lots of tech naming fails this test from Xbox (Series X vs S) to OpenAI model names (4o vs o1-pro).
Here they claim 4o is their most capable image generator which is useful info. Especially when multiple models in their dropdown list will generate images for you.
Speaking as someone who'd love to not speak that way in my own marketing - it's an unfortunate necessity in a world where people will give you literal milliseconds of their time. Marketing isn't there to tell you about the thing, it's there to get you to want to know more about the thing.
A term for people giving only milliseconds of their attention is: uninterested people. If I’m not looking for a project planner, or interested in the space, there’s no wording that can make me stay on an announcement for one. If I am, you can be sure I’m going to read the whole feature page.
No, everybody uses marketing because it's a conventional bet. It has proven in many cases to not be effective, but people aren't willing to risk getting fired because they suggested going against the grain.
It kind of is, the iPhone 16e isn’t the best even though it’s the latest, right? Or are we rating best by price/performance, not pure performance (I don’t even know if the 16e would be best there)?
Maybe people also caught up to the fact that the "our most X product" for Apple usually means someone else already did X a long time ago and Apple is merely jumping on the wagon.
Apple isn't really the best software company and though they were early to digital assistants with Siri, it seems like they've let it languish. It's almost comical how bad Siri is given the capabilities of modern AI. That being said, Android doesn't really have a great builtin solution for this either.
Apple is more of a hardware company. Still, Cook does have a few big wins under his belt: M-series ARM chips on Macs, Airpods, Apple watch, Apple pay.
We're in the middle of a massive and unprecedented boom in AI capabilities. It is hard to be upset about this phrasing - it is literally true and extremely accurate.
o3 mini wasn't so much a most advanced model, as it was incredibly affordable for the IQ it was presenting at the time. Sometimes it's about efficiency and not being on the frontier.
(Shrug) It's common for less-than-foundation-level models to be released every so often. This is done in order to provide new options, features, pricing, service levels, APIs or whatever that aren't yet incorporated into the main model, or that are never intended to be.
Just a consequence of how much time and money it takes to train a new foundation model. It's not going to happen every other week. When it does, it is reasonable to announce it with "Announcing our most powerful model yet."
Most things aren't in a massive boom and most people aren't that involved in AI. This is a rare example of great communication in marketing - they're telling people who might not be across this field what is going on.
> Why would they publish a model that is not their most advanced model?
I dunno, I'm not sitting in the OpenAI meetings. That is why they need to tell us what they are doing - it is easy to imagine them releasing something that isn't their best model ever and so they clarify that this is, in fact, the new hotness.
I wanted to use this to generate funny images of myself. Recently I was playing around with Gemini Image Generation to dress myself up as different things. Gemini Image Generation is surprisingly good, although the image quality quickly degrades as you add more changes. Nothing harmful, just silly things like dressing me up as a wizard or other typical RPG roles.
Trying out 4o image generation... It doesn't seem to support this use-case at all? I gave it an image of myself and asked to turn me into a wizard, and it generate something that doesn't look like me in the slightest. A second attempt, I asked to add a wizard hat and it just used python to add a triangle in the middle of my image. I looked at the examples and saw they had a direct image modification where they say "Give this cat a detective hat and a monocle", so I tried that with my own image "Give this human a detective hat and a monocle" and it just gave me this error:
> I wasn't able to generate the modified image because the request didn't follow our content policy. However, I can try another approach—either by applying a filter to stylize the image or guiding you on how to edit it using software like Photoshop or GIMP. Let me know what you'd like to do!
Overall, a very disappointing experience. As another point of comparison, Grok also added image generation capabilities and while the ability to edit existing images is a bit limited and janky, it still manages to overlay the requested transformation on top of the existing image.
It's not actually out for everyone yet. You can tell by the generation style.
4o generates top down (picture goes from mostly blurry to clear starting from the top).
628 comments
[ 3.4 ms ] story [ 361 ms ] threadNo API yet, and given the slowness I imagine it will cost much more than the $0.03+/image of competitors.
When I first read the parent comment, I thought, maybe this is a long-term architecture concern...
But your message reminded me that we've been here before.
Gemini "integrates" Imagen 3 (a diffusion model) only via a tool that Gemini calls internally with the relevant prompt. So it's not a true multimodal integration, as it doesn't benefit from the advanced prompt understanding of the LLM.
Edit: Apparently Gemini also has an experimental native image generation ability.
The (no longer, I guess) industry-leading features people actually want are hidden away in some obscure “AI studio” with horrible usability, while the headline Gemini app still often refuses to do anything useful for me. (Disclaimer: I last checked a couple of months ago, after several more of mild amusement/great frustration.)
They haven't been focusing attention on images because the most used image models have been open source. Now they might have a target to beat.
That's overly pessimistic. Diffusion models take an input and produce an output. It's perfectly possible to auto-regressively analyze everything up to the image, use that context to produce a diffusion image, and incorporate the image into subsequent auto-regressive shenanigans. You'll preserve all the conditional probability factorizations the LLM needs while dropping a diffusion model in the middle.
If anything, your feedback is of low value.
The animation is a lie. The new 4o with "native" image generating capabilities is a multi-modal model that is connected to a diffusion model. It's not generating images one token at a time, it's calling out to a multi-stage diffusion model that has upscalers.
You can ask 4o about this yourself, it seems to have a strong understanding of how the process works.
A model may not have many facts about itself, but it can definitely see what is inside of its own context, and what it sees is a call to an image generation tool.
Finally, and most convincingly, I can't find a single official source where OpenAI claims that the image is being generated pixel-by-pixel inside of the context window.
im not going to get super hyperbolic and histrionic about “entitlement” and stuff like that, but… literally this technology did not exist until like two years ago, and yet i hear this all the time. “oh this codegen is pretty accurate but it’s slow”, “oh this model is faster and cheaper (oh yeah by the way the results are bad, but hey it’s the cheapest so it’s better)”. like, are we collectively forgetting that the whole point of any of this is correctness and accuracy? am i off-base here?
the value to me of a demonstrably wrong chat completion is essentially zero, and the value of a correct one that anticipates things i hadn’t considered myself is nearly infinite. or, at least, worth much, much more than they are charging, and even _could_ reasonably charge. it’s like people collectively grouse about low quality ai-generated junk out of one side of their mouths, and then complain about how expensive the slop is out of the other side.
hand this tech to someone from 2020 and i guarantee you the last thing you’d hear is that it’s too slow. and how could it be? yeah, everyone should find the best deals / price-value frontier tradeoff for their use case, but, like… what? we are all collectively devaluing that which we lament is being devalued by ai by setting such low standards: ourselves. the crazy thing is that the quickly-generated slop is so bad as to be practically useless, and yet it serves as the basis of comparison for… anything at all. it feels like that “web-scale /dev/null” meme all over again, but for all of human cognition.
The results are ground breaking in my opinion. How much longer until an AI can generate 30 successive images together and make an ultra realistic movie?
> Developers will soon be able to generate images with GPT‑4o via the API, with access rolling out in the next few weeks.
That's it folks. Tens of thousands of so-called "AI" image generator startups have been obliterated and taking digital artists with them all reduced to near zero.
Now you have a widely accessible meme generator with the name "ChatGPT".
The last task is for an open weight model that competes against this and is faster and all for free.
ChatGPT has already had a that via Dall-E. If it didn't kill those startups when that happened this doesn't fundamentally change anything. Now its got a new image gen model, which — like Dall-E 3 when it came out — is competitive or ahead of other SotA base models using just text prompts, the simplest generation workflow, but both more expensive and less adaptable to more involved workflows than the tools anyone more than a casual user (whether using local tools or hosted services) is using. This is station-keeping for OpenAI, not a meaningful change in the landscape.
It's not 'just' a new model ala Imagen 3. This is 'what if GPT could transform images nearly as well as text?' and that opens up a lot of possibilities. It's definitely a meaningful change.
Was it public information when Google was going to launch their new models? Interesting timing.
Sora is one of the worst video generators. The Chinese have really taken the lead in video with Kling, Hailuo, and the open source Wan and Hunyuan.
Wan with LoRAs will enable real creative work. Motion control, character consistency. There's no place for an OpenAI Sora type product other than as a cheap LLM add-in.
I'm excited to see what a Flux 2 can do if it can actually use a modern text encoder.
The image generators used by creatives will not be text-first.
"Dragon with brown leathery scales with an elephant texture and 10% reflectivity positioned three degrees under the mountain, which is approximately 250 meters taller than the next peak, ..." is not how you design.
Creative work is not 100% dice rolling in a crude and inadequate language. Encoding spatial and qualitative details is impossible. "A picture is worth a thousand words" is an understatement.
Controlnet has been the obvious future of image-generation for a while now.
We might find that the entire "studio system" is a gross inefficiency and that individual artists and directors can self-publish like on Steam or YouTube.
Automation tools are always more powerful as a force multiplier for skilled users than a complete replacement. (Which is still a replacement on any given task scope, since it reduces the number of human labor hours — and, given any elapsed time constraints, human laborers — needed.)
"""
That's a fun idea—but generating an image with 999,999 anime waifus in it isn't technically possible due to visual and processing limits. But we can get creative.
Want me to generate:
1. A massive crowd of anime waifus (like a big collage or crowd scene)?
2. A stylized representation of “999999 anime waifus” (maybe with a few in focus and the rest as silhouettes or a sea of colors)?
3. A single waifu with a visual reference to the number 999999 (like a title, emblem, or digital counter in the background)?
Let me know your vibe—epic, funny, serious, chaotic?
"""
EDIT: Seems not, "The smallest image size I can generate is 1024x1024. Would you like me to proceed with that, or would you like a different approach?"
ofc 4.5 is best, but its slow and I am afraid I'm going to hit limits.
We're largely past the days of 7 fingered hands - text remains one of the tell-tale signs.
Character consistency means that these models could now theoretically illustrate books, as one example.
Generating UIs seems like it would be very helpful for any app design or prototyping.
I'm excited about this for adding images to those interactive stories.
It has nothing to do with circumventing the cost of artists or writers: regardless of cost, no one can put out a story and then rewrite it based on whatever idea pops into every reader's mind for their own personal main character.
It's a novel experience that only a "writer" that scales by paying for an inanimate object to crunch numbers can enable.
Similarly no artist can put out a piece of art for that story and then go and put out new art bespoke to every reader's newly written story.
-
I think there's this weird obsession with framing these tools about being built to just replace current people doing similar things. Just speaking objectively: the market for replacing "cheeky expensive artists" would not justify building these tools.
The most interesting applications of this technology being able to do things that are simply not possible today even if you have all the money in the world.
And for the record, I'll be ecstatic for the day an AI can reach my level of competency in building software. I've been doing it since I was a child because I love it, it's the one skill I've ever been paid for, and I'd still be over the moon because it'd let me explore so many more ideas than I alone can ever hope to build.
You realize that almost weekly we have new AI models coming out that are better and better at programming? It just happened that the image generation is an easier problem than programming. But make no mistake, AI is coming for us too.
That's the price of automating everything.
If that's best of 8, I'd love to see the outtakes.
[0]: https://www.unrealengine.com/en-US/metahuman
It's more pragmatic to pipeline the results to a background removal model.
EDIT: It appears GPT-4o is different as there is a video demo dedicated to transparancy.
I suspect we're getting a flood of comment from people who are using Dall-E.
And that created the isolated image on a transparent background.
Thank-you.
EDIT: Ok it works in Sora, and my jaw dropped
One area where it does not work well at all is modifying photographs of people's faces.* Completely fumbles if you take a selfie and ask it to modify your shirt, for example.
* = unless the people are in the training set
The Americas are quite a bit larger than the USA, so I disagree with 'american' being a word for people and things from mainland USA. Usian seems like a reasonable derivative of USA and US, similar to how mexican follows from Mexico and Estados Unidos Mexicanos.
Sounds like it may be a safety thing that's still getting figured out
Was anyone else surprised how slow the images were to generate in the livestream? This seems notably slower than DALLE.
I ran stable diffusion for a couple of years (maybe?, time really hasn't made sense since 2020) on my Dual 3090 rendering server. I built the server originally for crypto heating my office in my 1820s colonial in upstate NY then when I was planning to go back to college (got accepted into a university in England), I switched it's focus to Blender/UE4 (then 5), then eventually to AI image gen. So I've never minded 20 seconds for an image. If I needed dozens of options to pick the best, I was going to click start and grab a cup of coffee, come back and maybe it was done. Even if it took 2 hours, it is still faster than when I used to have to commission art for a project.
I grew out of Stable Diffusion, though, because the learning curve beyond grabbing a decent checkpoint and clicking start was actually really high (especially compared to LLMs that seamed to "just work"), after going through failed training after failed fine-tuning using tutorials that were a couple days out of date, I eventually said, fuck it, I'm paying for this instead.
All that to say - if you are using GenAI commercially, even if an image or a block of code took 30 minutes, it's still WAY cheaper than a human. That said, eventually a professional will be involved, and all the AI slop you generated will be redone, which will still cost a lot, but you get to skip the back and forth figuring out style/etc.
--
Comparison with Leonardo.Ai.
ChatGPT: https://chatgpt.com/share/67e2fb21-a06c-8008-b297-07681dddee...
ChatGPT again (direct one shot): https://chatgpt.com/share/67e2fc44-ecc8-8008-a40f-e1368d306e...
ChatGPT again (using word "photorealistic instead of "photo"): https://chatgpt.com/share/67e2fce4-369c-8008-b69e-c2cbe0dd61...
Leonardo.Ai Phoenix 1.0 model: https://cdn.leonardo.ai/users/1f263899-3b36-4336-b2a5-d8bc25...
I'm curious if you said 2d animation style for both or just for chatgpt.
Edit: Your second version of chatgpt doesn't say photorealistic. Can you share the Leonard.ai prompt?
Leonardo prompt: A golden cocker spaniel with floppy ears and a collar that says "Sunny" on it
Model: Phoenix 1.0 Style: Pro color photography
Midjourney hasn't been SOTA for nearly a year now. It struggles to follow even marginally complex prompts from an adherence perspective.
Might take a day or two before it's available in general.
It seems like an odd way to name/announce it, there's nothing obvious to distinguish it from what was already there (i.e. 4o making images) so I have no idea if there is a UI change to look for, or just keep trying stuff until it seems better?
Truly infuriating, especially when it's something like this that makes it tough to tell if the feature is even enabled.
"Typical" AI images are this blend of the popular image styles of the internet. You always have a bit of digital drawing + cartoon image + oversaturated stock image + 3d render mixed in. Models trained on just one of these work quite well, but for a generalist model this blend of styles is an issue
Asian artists don't color this way though; those neon oversaturated colors are a Western style.
(This is one of the easiest ways to tell a fake-anime western TV show, the colors are bad. The other way is that action scenes don't have any impact because they aren't any good at planning them.)
This option is not exposed in ChatGPT, it only uses vivid.
Sorry, but how are these useful? None of the examples demonstrate any use beyond being cool to look at.
The article vaguely mentions 'providing inspiration' as possible definition of 'useful'. I suppose.
Then google:
> Gemini 2.5: Our most intelligent AI model
> Introducing Gemini 2.0 | Our most capable AI model yet
I could go on forever. I hope this trend dies and apple starts using something effective so all the other companies can start copying a new lexicon.
Which is especially relevant when it's not obvious which product is the latest and best just looking at the names. Lots of tech naming fails this test from Xbox (Series X vs S) to OpenAI model names (4o vs o1-pro).
Here they claim 4o is their most capable image generator which is useful info. Especially when multiple models in their dropdown list will generate images for you.
<Product name>: Our most <superlative> <thing> yet|ever.
This one isn't even my biggest gripe. If I could eliminate any word from the English language forever, it would be "effortlessly".
Obligatory Jobs monologue on marketing people:
https://www.youtube.com/watch?v=P4VBqTViEx4
> Every iPhone is their best iPhone yet
Hotwheels: Fast. Furious. Spectacular.
Apple is more of a hardware company. Still, Cook does have a few big wins under his belt: M-series ARM chips on Macs, Airpods, Apple watch, Apple pay.
Just a consequence of how much time and money it takes to train a new foundation model. It's not going to happen every other week. When it does, it is reasonable to announce it with "Announcing our most powerful model yet."
> Why would they publish a model that is not their most advanced model?
I dunno, I'm not sitting in the OpenAI meetings. That is why they need to tell us what they are doing - it is easy to imagine them releasing something that isn't their best model ever and so they clarify that this is, in fact, the new hotness.
And no, not all models are intended to push the frontier in terms of benchmark performance, some are just fast and cheap.
Trying out 4o image generation... It doesn't seem to support this use-case at all? I gave it an image of myself and asked to turn me into a wizard, and it generate something that doesn't look like me in the slightest. A second attempt, I asked to add a wizard hat and it just used python to add a triangle in the middle of my image. I looked at the examples and saw they had a direct image modification where they say "Give this cat a detective hat and a monocle", so I tried that with my own image "Give this human a detective hat and a monocle" and it just gave me this error:
> I wasn't able to generate the modified image because the request didn't follow our content policy. However, I can try another approach—either by applying a filter to stylize the image or guiding you on how to edit it using software like Photoshop or GIMP. Let me know what you'd like to do!
Overall, a very disappointing experience. As another point of comparison, Grok also added image generation capabilities and while the ability to edit existing images is a bit limited and janky, it still manages to overlay the requested transformation on top of the existing image.