gpt-image-2 is here, available today in the API and Codex. The most capable image generation model yet, built for production-grade workflows with stronger text rendering, layout, editing, resolution, and multilingual rendering.

Create assets for the surface you need. gpt-image-2 supports thousands more export ratios and higher-resolution outputs up to 2K, making it easier to create images for apps, ads, social placements, and product workflows.

Text-heavy visuals get more practical. gpt-image-2 improves multilingual text rendering and structured image generation for diagrams, infographics, charts, comics, and multi-panel scenes.

Google already did this 6 months ago. Return GPT-4o, the only model that gave OAI a competitive edge. #keep4o #opensource4o #keep4oAPI

Awesome demos @charlierguo

text rendering looks much better already. are edits in Codex going through the same model path, or is image editing still a separate backend? that's the workflow knob I'd care about most.

Before, I wasn’t so scared of AI-generated images, but now these AI screenshots seem so real with very little if any tells. I won’t believe anything anymore.

API availability from day one signals confidence that this is production-ready, not aspirational Most companies wait months before opening to developers

#keep4o 🙂

wow that's a nice scheduler design can't wait to try it out now 🫣

This is insane! So: I took a screenshot of my favourite game (Monkey Island), gave it to Codex, and told it create a spritesheet and tilemap using gpt-image-2, and wire it all up using Phaser. Ten minutes later: wow! @sama @OpenAIDevs @OpenAI

I haven’t been able to fully use some of Codex’s features, so I’m hoping someone can help explain how they work.

Within Codex it's crazy

it has officially transitioned from a tech revolution to a 24/7 digital billboard they are selling us the same hype in a different wrapper every season just to keep the subscription fees flowing it feels less like innovation and more like a high end furniture store where everything looks pretty but has zero soul we are basically paying a monthly fee to watch a never ending commercial for a product that we are actually building for them with our own data its the ultimate marketing scam disguised as the future of humanity

No, sorry, I was reliably informed by a snarky teenager with an anime avatar that AI images won't, and can't ever get better. He even went "lmaooooo" and everything, so he must be a computer expert. So this must be all in our imaginations and is not really happening.

Integrating this model into Codex allows for a seamless "vibe coding" experience where the AI can generate functional UI components and the visuals to match. This moves us closer to a future where entire frontends are built and populated in a single pass.

🔥

Accurate charts and infographics on demand is a massive win for enterprise. Automating data visualization just got a lot more viable.

finally, a model that can spell words correctly in images. only took us until 2025 to get text rendering that doesn't look like a drunk AI's first attempt at calligraphy

Really happy to see the text rendering fix. If anyone has tested complex layouts yet, let me know if it holds up. Switching my API key now.

This feels like image generation finally going production-ready

API pricing for gpt-image-2 at 1024x1024: Low: $0.006/image Medium: $0.053/image High: $0.211/image High quality is actually more expensive than 1.5 at the same resolution. Also need org verification before the endpoint works. Full breakdown:

😳

Finally! The text rendering alone makes this worth it. Been waiting for something that can actually handle multilingual layouts without breaking.

text rendering in prod image gen has been the biggest pain point for document automation. worth watching. I'd probably drop the OCR pass entirely if this holds up.

#keep4o bring back 4o

It handles text very well

turn the hourglasses over nxt time in background lol

胡散臭い

keep automating!! 🐾💙💚🪄🧙🏾✅🍡 @barkmeta @GodsBurnt @Shieldmetax

gpt-image-2 for workflows sounds prod ready. how's the edit consistency across gens

Stop the madness and listen to your customers!!!!!!! Enough! #keep4o #OpenSource4o

— quick real vs gen check. Fun, actually useful.

digging it so far, wish it could output (or convert) to svg all in one pane of glass 🤞