OpenAI's newest image model now makes your image ads. Sharper detail, faster renders, and the same credit cost.
Image ads in the Ad Studio are now made on GPT Image 2.5, which OpenAI released on September 8, 2026 and calls its new state-of-the-art image model. It ships in two variants and we run Flare, the one built for speed and for generating at volume, because that is what making ads in batches actually is.
Lighting looks more natural and textures hold more detail, which is most of the difference between an image that reads as a real photo and one that reads as generated. It also holds on to your subject more faithfully when it is working from a reference, so your actual product and your actual people survive the render. That matters most for the two ways in that start from something of yours: "From a link" and your brand kit. And it follows editing instructions more reliably over several rounds, so a second and third pass in the image editor stay on target instead of drifting away from what you asked for.
OpenAI puts the new generation at up to 50% lower latency than the one before it, so there is less waiting between a prompt and a batch you can judge. The part we control has not changed: an image preview is still 1 credit, unlocking an approved image is still 30, and ad copy, brand research and your own uploads are still free. You get the newer model at the price you were already paying.
There is no setting and no migration. Every new image generation in the Ad Studio already runs on it. Anything already sitting in your library is untouched, so nothing you have approved or launched changes under you.
GPT Image 2.5 is live in the Ad Studio today. Every new image generation already uses it.