OpenAI's ChatGPT Images 2.5 adds Sketch, turning doodles into detailed images

OpenAI has rolled out ChatGPT Images 2.5, an updated version of its image generation model, together with a new Sketch feature that turns rough doodles into finished images. According to The Verge, the company announced the update on Tuesday.

Sketch turns doodles into prompts

The headline addition is Sketch, which gives users a second way to tell ChatGPT what to draw. Instead of relying on text alone, you can type @Sketch into the chat box, and a window opens where you can make a quick drawing. You then tell ChatGPT how it should transform that sketch into the final image.

The Verge's Jay Peters briefly tested the feature and reported that it worked as advertised: a deliberately rough drawing of a cat, made with a computer mouse, was converted into a realistic photograph based on the instructions he supplied.

Comments let you mark up images

The update also introduces inline commenting on generated images. Rather than describing a desired change in words and hoping the model finds the right spot, users can now leave a comment pinned to the specific part of the image they want adjusted. Peters used this to change the cat's eyes from blue to green.

What's new in Images 2.5

Beyond the interface changes, OpenAI says the underlying model has improved in three ways. Images are said to come out with more natural lighting and richer textures. The model is better at following editing instructions "across multiple turns", which should make iterative refinement within a single conversation more reliable. And generation latency has been reduced by up to 50% compared with Images 2.0, making the cycle of sketching, generating and editing noticeably faster.

Where you can get it

Images 2.5 is available now to ChatGPT, ChatGPT Work and Codex users on desktop, mobile and web, according to OpenAI.

Why it matters

The Sketch feature points at a broader shift in how people interact with image generators. Text prompts are a blunt instrument for anything spatial, such as composition, layout or the position of one object relative to another, and a few strokes of a doodle can often convey that structure faster and more precisely than a paragraph of description. By letting a crude mouse drawing stand in for an elaborate prompt, OpenAI is lowering the barrier for users who know what they want visually but struggle to articulate it in words.

The combination of inline comments, better multi-turn instruction following and reduced latency suggests the company is betting on iterative editing as the primary workflow, closer to how people work in design or document tools than to one-shot prompt generation. If a user can sketch, generate, point at what is wrong and get a fix within seconds, the model behaves less like a slot machine and more like a collaborator.

The rollout reaching Codex users alongside ChatGPT also indicates that the image model is being positioned as infrastructure across OpenAI's product line rather than as a standalone consumer novelty.