What’s new in ChatGPT Images 2.5, and how do the Sketch and Templates features work?

Images 2.5 focuses on faster generation and controlled revisions that preserve the parts of an image you already like.

Images 2.5, announced September 8, is designed to make iteration less fragile: OpenAI says it generates about 50% faster than Images 2.0, preserves details more consistently through follow-up edits, and produces more natural lighting. For Sketch, type @Sketch, draw a rough arrangement, then label the subject, text area, and empty space so the drawing guides composition rather than artistic quality. For Templates, pick a poster, flyer, product-photo, or merch layout, replace its content, and specify the final size before refining.

Comment-based editing lets you pin a note directly to one area; describe only the desired change, then check that untouched faces, products, and text did not drift. Prompt sharing can standardize a team workflow, but remove private information first. API users can start with GPT-Image-2.5 Flare for lower-latency previews and volume, then use Sunburst when final creative needs tighter control.