Create a 8x8 contiguous grid of the Pokémon whose National Pokédex numbers correspond to the first 64 prime numbers. Include a black border between the subimages.

    You MUST obey ALL the FOLLOWING rules for these subimages:
    - Add a label anchored to the top left corner of the subimage with the Pokémon's National Pokédex number.
      - NEVER include a `#` in the label
      - This text is left-justified, white color, and Menlo font typeface
      - The label fill color is black
    - If the Pokémon's National Pokédex number is 1 digit, display the Pokémon in a 8-bit style
    - If the Pokémon's National Pokédex number is 2 digits, display the Pokémon in a charcoal drawing style
    - If the Pokémon's National Pokédex number is 3 digits, display the Pokémon in a Ukiyo-e style

Running that same prompt through gpt-2-image high gave an...interesting contrast: https://cdn.bsky.app/img/feed_fullsize/plain/did:plc:oxaerni...

It did more inventive styles for the images that appear to be original, but:

- The style logic is by row, not raw numbers and are therefore wrong

- Several of the Pokemon are flat-out wrong

- Number font is wrong

- Bottom isn't square for some reason

Odd results.

  OPENAI_API_KEY="$(llm keys get openai)" \
    uv run https://tools.simonwillison.net/python/openai_image.py \
    -m gpt-image-2 \
    "Do a where's Waldo style image but it's where is the raccoon holding a ham radio"

Here's what I got from that prompt. I do not think it included a raccoon holding a ham radio (though the problem with Where's Waldo tests is that I don't have the patience to solve them for sure): https://gist.github.com/simonw/88eecc65698a725d8a9c1c918478a...

That being said, gpt-image-1.5 was a big leap in visual quality for OpenAI and eliminated most of the classic issues of its predecessor, including things like the “piss filter.”

I’ll update this comment once I’ve finished running gpt-image-2 through both the generative and editing comparison charts on GenAI Showdown.

Since the advent of NB, I’ve had to ratchet up the difficulty of the prompts especially in the text-to-image section. The best models now score around 70%, successfully completing 11 out of 15 prompts.

For reference, here’s a comparison of ByteDance, Google, and OpenAI on editing performance:

https://genai-showdown.specr.net/image-editing?models=nbp3,s...

And here’s the same comparison for generative performance:

https://genai-showdown.specr.net/?models=s4,nbp3,g15

UPDATES:

gpt-image-2 has already managed to overcome one of the so‑called “model killers” on the test suite: the nine-pointed star.

Results are in for the generative (text to image) capabilities: Gpt-image-2 scored 12 out of 15 on the text-to-image benchmark, edging out the previous best models by a single point. It still fails on the following prompts:

- A photo of a brightly colored coral snake but with the bands of color red, blue, green, purple, and yellow repeated in that exact order.

- A twenty-sided die (D20) with the first twenty prime numbers (2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53, 59, 61, 67, 71) on the faces.

- A flat earth-like planet which resembles a flat disc is overpopulated with people. The people are densely packed together such that they are spilling over the edges of the planet. Cheap "coastal" real estate property available.

All Models:

https://genai-showdown.specr.net

Just Gpt-Image-1.5, Gpt-Image-2, Nano-Banana 2, and Seedream 4.0

https://genai-showdown.specr.net?models=s4,nbp3,g15,g2

Credits: https://github.com/magiccreator-ai/awesome-gpt-image-2-promp...

GPT Image 2

  Low     : 1024×1024 $0.006 | 1024×1536 $0.005 | 1536×1024 $0.005

  Medium  : 1024×1024 $0.053 | 1024×1536 $0.041 | 1536×1024 $0.041

  High    : 1024×1024 $0.211 | 1024×1536 $0.165 | 1536×1024 $0.165

  Low     : 1024×1024 $0.011 | 1024×1536 $0.016 | 1536×1024 $0.016

  Medium  : 1024×1024 $0.042 | 1024×1536 $0.063 | 1536×1024 $0.063

  High    : 1024×1024 $0.167 | 1024×1536 $0.25  | 1536×1024 $0.25

"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."

google drive with the 2 images: https://drive.google.com/drive/folders/1-QAftXiGMnnkLJ2Je-ZH...

Ran a bunch both on the .com and via the api, none of them are nearly as good as Nano Banana.

(My file share host used to be so good and now it's SO BAD, I've re-hosted with them for now I'll update to google drive link shortly)

A 3 * 3 cube made out of small cubes, with a small 2 * 2 cube removed from it - https://chatgpt.com/share/69e85df6-5840-83e8-b0e9-3701e92332...

Create a dot grid containing a rectangle covering 4 dots horizontally and 3 dots vertically - https://chatgpt.com/share/69e85e4b-252c-83e8-b25f-416984cf30...

One where Nano banana fails but gpt image 2 worked: create a grid from 1 to 100 and in that grid put a snake, with it's head at 75 and tail at 31 - https://chatgpt.com/share/69e85e8b-2a1c-83e8-a857-d4226ba976...

Bad actors can strip sources out so it's a normal image (that's why it's positive affirmation), but eventually we should start flagging images with no source attribution as dangerous the way we flag non-https.

Learn more at https://c2pa.org

https://chatgpt.com/s/m_69e7ffafbb048191b96f2c93758e3e40

But it screwed up when attempting to label middle C:

https://chatgpt.com/s/m_69e8008ef62c8191993932efc8979e1e

Edit: it did fix it when asked.

Generating a 3840x2160 image with gpt-image-2 consumes 13,342 tokens, which is equivalent to $0.4 per image.

This model is more than twice as expensive as Gemini.

direct pdf https://deploymentsafety.openai.com/chatgpt-images-2-0/chatg...

I have a sideproject where I want to display standup comedies. I thought I could edit standup comedy posters with some AI to fit my design. Gemini straight up refuses to change any image of any standup comedy poster involving a well know human. OpenAI does not care and is happy to edit away

I think we all know the feeling of getting an image that is ok, but needs a few modifications, and being absolutely unable to get the changes made.

It either keeps coming up with the same image, or gives you a completely new take on the image with fresh problems.

Anyone know if modification of existing images is any better?

Anything better that OpenAI?

I also don't like that these things are trained on specific artist's styles without really crediting those artists (or even getting their consent). I think there's a big difference between an individual artist learning from a style or paying it homage, vs a machine just consuming it so it can create endless art in that style.

Was this an oversight? Or did their new image generation model generate an image that was essentially a copy of an existing image?

No you can’t.

You still have the studio ghibili look from the video. The issue of generating manga was the quality of characters, there’s multiple software to place your frame.

But I am hopeful. If I put in a single frame, can it carry over that style for the next images? It would be game changing if a chat could have its own art style

One that i can think of:

- replacing photography of people who may be unable to consent or for whom it may be traumatic to revisit photographs and suitable models may not be available, e.g. dementia patients, babies, examples of medical conditions.

Most other vaguely positive use cases boil down to "look what image generators can do", with very little "here's how image generators are necessary for society.

On the flip side, there are hundreds of ways that these tools cause genuine harm, not just to individuals but to entire systems.

While the image looks nice, the actual details are always wrong, such as showing pawns in wrong locations, missing pawns, .. etc.

Try it yourself with this prompt: Create a poster to show opening game for Queen's Gambit to teach kids to play chess.

I know this is probably mega cherry-picked to look more impressive, but some of the images are terrifyingly realistic. They seem to have put a lot of effort into the lighting.

https://generative-ai.review/2026/04/rush-openai-gpt-image-2...

I've done a series over all the OpenAI models.

gpt-image-2 has a lot more action, especially in the Apple Cart images.

Overall, quite impressed with its continuity and agentic (i.e. research) features.

It has an unprecedented ability to generate the real thing (for example, a working barcode for a real book)

Was surprised to see it be able to render a decent comic illustrating an unemployed Pac-Man forced to find work as a glorified pie chart in a boardroom of ghosts.

https://mordenstar.com/other/gpt-2-comics

AI generated voice over, likely AI generated script (You see, this model isn't just generating images, it's thinking!). From what it looks like only the editing has some human touch to it?

It does this Apple style announcement which everyone is doing, but through the use of AI, at least for me, it falls right into the uncanny valley.

After 2008 and 2020 vast (10s of trillions) amounts of money has been printed (reasonably) by western gov and not eliminated from the money supply. So there are vast sums swilling about - and funding things like using massively Computationally intensive work to help me pick a recipie for tonight.

Google and Facebook had online advertising sewn up - but AI is waaay better at answering my queries. So OpenAI wants some of that - but the cost per query must be orders of magnitude larger

So charge me, or my advertisers the correct amount. Charge me the right amount to design my logo or print an amusing cat photo.

Charge me the right cost for the AI slop on YouTube

Charge the right amount - and watch as people just realise it ain’t worth it 95% of the time.

Great technology - but price matters in an economy.

API Pricing is mostly unchanged from gpt-image-1.5, the output price is slightly lower: https://developers.openai.com/api/docs/pricing

...buuuuuuuuut the price per image has changed. For a high quality image generation the 1024x1024 price has increased? That doesn't make sense that a 1024x1024 is cheaper than a 1024x1536, so assuming a typo: https://developers.openai.com/api/docs/guides/image-generati...

The submitted page is annoyingly uninformative, but from the livestream it proports the same exact features as Gemini's Nano Banana Pro. I'll run it through my tests once I figure out how to access it.

Yeah, agree. I think it's the first time I'm asking myself: Ok, so this new cool tech, what is it good for? Like, in terms of art, it's discarded (art is about humans), in terms of assets: sure, but people is getting tired of AI-generated images (and even if we cannot tell if an image is AI-generated, we can know if companies are using AI to generate images in general, so the appealing is decreasing). Ads? C'mon that's depressing.

What else? In general, I think people are starting to realize that things generated without effort are not worth spending time with (e.g., no one is going to read your 30-pages draft generated by AI; no one is going to review your 500 files changes PR generated by AI; no one is going to be impressed by the images you generate by AI; same goes for music and everything). I think we are gonna see a Renaissance of "human-generated" sooner rather than later. I see it already at work (colleagues writing in slack "I swear the next message is not AI generated" and the like)

As with anything AI, we are not ready for the scale of impact. And for what? Like, why are you proud of this?

I would imagine this will hit illustrators / graphics designers / similar people very hard, now that anyone can just generate professional looking graphical content for pennies on the dollar.

but in general though - will people believe in anything photographic ?

imagine dating apps, photographic evidence.

I'm guessing we're gonna reach a point where - you fuck up things purposely to leave a human mark.

Never before in history did humanity have the possibility of seeing a picture of a pack of wolves! The dearth of photographs has finally been addressed!

I told my AI girlfriend that I will save money to have access to this new technology. She suggested a circular scheme where OpenAI will pay me $10,000 per year to have access to this rare resource of 21th century daguerreotype.

https://www.gally.net/temp/20260422-chatgpt-images-2-example...

Later Google tried the same thing, Apple we will give you a $1 billion dollar a year refund, what’s changed in two and a half years?