Meet GPT-Image-2.5 Flare and Sunburst. Introducing new image models in the API, with sharper detail, stronger style adherence, and more control over edits.

Flare delivers higher-quality images at 50% lower latency than GPT-Image-2. Sunburst adds even more precision for detailed editing. More natural lighting, richer textures, and better handling of complex layouts and transparent backgrounds.

Edits you control: Change a jacket while keeping the subject in the same position. Update a background while preserving the product image. Revise copy without reworking the composition. GPT-Image-2.5 follows editing instructions more precisely, helping preserve the subject and composition around each change.

Build creative tools with more precise editing: • Image generators with greater realism and creative control • AI photo editors that change what users ask for and preserve the rest • Image-to-video workflows with more consistent characters and scenes • Design platforms that turn briefs into complete sets of on-brand assets • Ad platforms that generate, personalize, and localize campaign creative at scale • Ecommerce tools that turn product photos into studio shots, lifestyle imagery, and catalog variations

Start with GPT-Image-2.5 Flare for most applications. Choose GPT-Image-2.5 Sunburst when you need extra precision and control across edits.

Wake up babe, new triple T generator just dropped.

it works well. Left old model. Right new model. Iron Bike

You couldn't entirely remove the image noise still could you? ... that is quite disappointing.

Nice. When is it coming to codex via sub?

You guys had one job. And that was to remove the noise filter. It's still there :(

Thanks for the image model updates. They are great for character consistency...

I really hope you guys fixed the checkerboard fake PNGs for transparent images.

its so confusing - how do I know which model I am using - flare or Sunburst? Or do I even have access to Image-2.5?

what about this pixel checkerboard pattern, probably because of synthID

The image model progress is so impressive to see. Great video @codestantine!

You guys are rolling newmodels like nothing... this is crazyyy, I love OpenAI!!

I wonder if the noise filter is gone 👀

flare for speed. Sunburst for precision. @OpenAI really looked at people already incapable of choosing a text model and said “let’s do it to images too.” Which one are we stress-testing into the ground first?

damn OpenAI taking no breaks

beats Nano Banana Pro btw 👀

@ajambrosino could’ve been you

What is this trained on? Copyrighted work?

Is this actually going to be in ChatGPT? Right now the results are nowhere near the results your blog post is demonstrating, so I think it's still using image-2.

A consistent naming scheme? Internal AGI at OpenAI confirmed!

Sunburst numbers will be out in an hour or so. But with 50k independent respondents, flare is crushing your old best.

"Keep the new background, bring back the old logo." A clickable revision history would let the user point to exactly which "old logo" they mean.

Nano banana is cooked

you tested whether the image models can do bio/cyber reasoning by making them write a scratchpad and a final answer inside the picture? wild

Awesome, can it draw anime girls with swastikas on them?

first

Congrats on the launch. Sharper detail, better style adherence and more precise edits are welcome improvements Looking forward to seeing how Flare and Sunburst perform in the wild especially on multi-turn editing consistency and text rendering. Competition keeps everyone honest🚀

you guys coooooooked so good, it taste wonderful. Congrats to the team!

Please do something with this, at least admit that this is a problem

@OpenAIDevs Curious how new image detail changes branding for AI products. Ever hit the wall with style consistency myself!

how do you chose Sunburst or Flare in chat ?

Buy Seedance Company and integrate it on GPT as Imagine 2.5!

flare and sunburst — they finally stopped naming models after weather events and started naming them after what they actually do to your gpu