OpenAI Releases ChatGPT Images 2.5 With 50% Faster Generation, Sketch Feature, and Two New API Models

My take: Speed has always been one of the biggest bottlenecks in AI image generation workflows: waiting 10 to 30 seconds per image does not seem like much until you multiply it by a hundred revision cycles. ChatGPT Images 2.5 cuts that latency by up to 50% compared to the previous version, which in practice means more iterations in less time.

The new Sketch feature changes how visual intent is communicated. Instead of trying to describe an image composition in text, you can draw directly in ChatGPT as a guide. For design teams, content creators, or freelancers working with clients, that closes the gap between what you have in mind and what the model generates.

For developers, OpenAI added two separate API models: Flare for most applications, and Sunburst for workflows that need tighter control across edits. How much time do you spend today iterating on images in your workflow, and how much of that could be resolved with tools that generate faster and follow your instructions more precisely?