OpenAI just gave ChatGPT a new way to understand what you actually want an image to look like: let you draw it. The company announced ChatGPT Images 2.5 on Tuesday, and according to The Verge AI, the headline addition is a feature called Sketch that turns a rough doodle into a detailed AI image. The Verge AI reports that a reporter tested it live, drawing a cat with a computer mouse and watching ChatGPT convert that messy scribble into a realistic photo.
What stands out here is how OpenAI is rethinking the prompt itself. Text is precise, but it’s a clumsy way to describe layout, shape, or composition. A quick sketch fills that gap. You show the model roughly what you mean, then use words to fill in the details.
What’s new in the update
- Sketch, a draw-to-image tool. Type @Sketch into the chat box and a drawing window pops up. You sketch something, then tell ChatGPT how to render it. In The Verge AI’s test, a mouse-drawn cat became a realistic photo based on the written instructions that followed.
- Comments directly on images. You can now leave notes on specific parts of an image you want changed. The Verge AI’s reporter used it to switch a cat’s eyes from blue to green, no full re-prompt required. That’s targeted editing instead of regenerating the whole picture and hoping.
- Better image quality. OpenAI says Images 2.5 produces “more natural lighting and richer textures” than the previous version. Small phrasing, but lighting and texture are usually where AI images give themselves away.
- Stronger multi-turn editing. The model is better at following your editing instructions “across multiple turns,” per OpenAI. In plain terms, it should hold context as you refine an image step by step instead of losing the thread.
- Faster generation. OpenAI claims image generation latency is down by “up to 50% compared with Images 2.0.” Speed matters a lot when you’re iterating, and cutting the wait in half changes how much you’re willing to experiment.
Why this matters
The combination is the real story. Sketch handles the “what should it look like” problem, inline comments handle the “fix this one thing” problem, and the speed bump makes running through several versions less painful. Together they push image generation closer to a real editing loop rather than a one-shot slot machine.
This also lines up with where the industry is heading. Rivals like Google’s Gemini image tools and Midjourney have been racing on quality and editing control. OpenAI’s answer leans into interaction: give people more ways to communicate intent, not just a better final render. Drawing is something almost anyone can do badly, and “badly” is now good enough to be useful.
Who can use it and when
ChatGPT Images 2.5 is out now. OpenAI says it’s available to ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web. The article doesn’t break out separate pricing for the feature, so access appears tied to those existing product tiers rather than a new paid add-on.
The caveats
A few things worth keeping in mind. The performance claims, natural lighting, richer textures, and the latency cut, come straight from OpenAI, and The Verge AI presents them as company statements rather than independently benchmarked results. The hands-on test covered a single simple subject, a cat, so how Sketch handles complex scenes, multiple objects, or precise layouts is still an open question. And the “up to 50%” speed figure is a ceiling, not a guarantee.
For anyone who creates visuals, though, the direction is clear. The prompt is no longer just a text box. You can draw, point, comment, and refine, and the tool is fast enough to keep up. Expect competitors to match the draw-to-image idea quickly, because it removes one of the biggest friction points in AI image tools: describing a picture with words alone.
More details are available in the original report from The Verge AI.