OpenAI releases ChatGPT Images 2.5 with built-in sketch tool

OpenAI announced Tuesday the 2.5 version of its in-chat image generator alongside a new feature called Sketch that lets users feed a hand-drawn sketch directly as a visual reference. Instead of describing what they want in words, users can type @Sketch in the chat box, open an internal drawing window, scribble with the mouse, and then instruct the model on how to turn the doodle into a detailed image. In a brief test published with the announcement, a rough cat sketch became a realistic photograph following the user's guidance.

Beyond creation from sketch, the new version allows users to leave notes directly on specific areas of a generated image and request local changes. In the example provided, the user highlighted the cat's eyes and asked to change the color from blue to green; the model executed the edit without disturbing other details. The move points toward more natural iterative editing that mimics working with a human designer rather than re-rolling prompts.

According to OpenAI, Images 2.5 delivers "more natural lighting and richer textures," excels at following editing instructions across multiple turns, and cuts image generation time by "up to 50%" compared with Images 2.0 — not the immediate predecessor. The phrasing "up to 50%" and the choice of baseline suggest the figure represents optimal conditions rather than a daily average. No independent benchmark has confirmed the number at this stage.

The update is available immediately to regular ChatGPT users, ChatGPT Work subscribers, and Codex users on desktop, mobile, and web. The broad rollout signals that OpenAI treats Sketch and point-editing as core product capabilities, not a separate experiment. For users, that means the new tools can be put to work right away without waiting for a staged release.