Most image-model updates lead with a benchmark chart. OpenAI’s ChatGPT Images 2.5, rolled out September 8, leads with two workflow changes that are easier to feel than to graph: you can now draw on the canvas instead of describing it, and you can leave a comment on the part of a photo you want changed instead of writing a paragraph explaining where it is.
Comments as edit instructions, sketches as layout guides
The comment-based edits feature lets you place a comment directly on a specific region of an image — a face, a background element, a piece of text — and have that annotation function as the edit instruction, rather than needing to describe the target location in prose (“the lamp in the back left corner behind the couch”). It’s a small interface change with an outsized effect on iteration speed for anyone doing multi-pass photo editing.
Sketch goes the other direction: instead of starting from text, you draw a rough layout or outline directly in ChatGPT — a floor plan, a composition sketch, a UI wireframe — and pair it with a style description. The model treats your drawing as structural guidance and fills in a complete image around it, rather than ignoring the sketch and generating from the text prompt alone.
What changed under the hood
OpenAI says the underlying model is meaningfully faster and more consistent: roughly a 50% reduction in generation latency, better preservation of a reference photo’s subject across edits, more reliable instruction-following across multi-turn editing sessions, and improved handling of transparent backgrounds and real-world visual accuracy. Early API customers reported the new default model generating images at two to four times the speed of the prior version while holding onto visual identity through repeated edits.
Two new models ship in the API alongside the ChatGPT-facing update: GPT-Image-2.5 Flare, the new default, aimed at creator content, product imagery, and high-volume generation at lower latency; and GPT-Image-2.5 Sunburst, a premium tier for production workflows that need tighter, more precise control across a sequence of edits — pitched at creative campaigns and polished product photography specifically.
Rollout
ChatGPT Images 2.5 is live now across ChatGPT, ChatGPT Work, and Codex, on desktop, mobile, and web, with the two new models available separately through the API.