OpenAI launches GPT Image 1.5, up to 4x faster and more precise, lifelike

This lifelike image of Sam Altman as Hugh Hefner was generated in one shot.
One-shotted: “Generate a picture of Sam Altman in the style of Hugh Hefner, in a smoking jacket with a cigar and a whiskey.” (Picture: GPT Image 1.5)
The new image model is faster by a wide margin, and way better at composing detailed, lifelike images.

It also «adheres to your intent more reliably» and keeps elements of the picture, like lighting, composition, and appearance across the editing phase.

The outputs are also much more lifelike, making it ever more difficult to discern AI images from reality.

Does it beat Nano Banana? Yes, it does, according to the almost gold-standard LMArena benchmark, where it instantly climbed to the top spot on «Text-to-image» and «Image Edit.»

At the same time, OpenAI is launching a new feature called ChatGPT Image on the web, which is a handy way to access predefined styles for your pictures and share those you like.

They are even seasonally aware enough to let you turn pictures into Christmas cards.

Read more: OpenAI’s announcement (includes lots of examples), ChatGPT Image. Writeups on Engadget, TechCrunch. Discussion on r/Singularity.