OpenAI's New ChatGPT Image Tool Lets You Draw to Guide What It Creates
OpenAI announced ChatGPT Images 2.5 on September 8, 2026. The update adds a new Sketch feature, lets users leave comments directly on generated images, and cuts the time it takes to create an image by up to half compared with the previous version. It is available now for ChatGPT, ChatGPT Work, and Codex users on desktop, mobile, and web. The Verge
The biggest addition is called Sketch. When you type @Sketch into the chat box, a small drawing canvas opens. You can make a quick doodle, and ChatGPT uses that drawing as a visual reference alongside whatever text you have written. This is a shift from the usual approach, where you describe everything in words. Now you can show roughly where things should go instead of trying to explain it.
The update also lets you leave comments directly on specific parts of a generated image. Instead of starting over or describing a change for the whole image, you can point at one area and say what should be different there. For people who go through several rounds of edits, this could cut down on the back-and-forth that has made precise image editing in chat tools frustrating.
OpenAI says Images 2.5 produces images with more natural lighting and richer textures. The company also says the new version is better at following editing instructions across multiple turns, which addresses a known problem: earlier versions would lose track of what you originally wanted as a conversation went on. OpenAI's release notes describe ChatGPT Images 2.5 as bringing "sharper details, more precise editing, and faster image generation." OpenAI Help Center
The speed improvement is the easiest claim to measure: up to 50% faster image generation compared with Images 2.0. When you are generating, reviewing, editing, and regenerating, cutting the wait time in half changes how the tool feels. It moves closer to the point where you treat the result as something you can quickly adjust rather than something you have to wait for.
The broader context here is that OpenAI is working on image generation as more than just a quality race. The previous version, announced in December 2025, was described as powered by OpenAI's flagship image model, delivering more precise edits and consistent details. OpenAI Images 2.5 builds on that with changes to how you interact with the tool (Sketch, inline commenting) and performance gains (speed, better multi-turn editing) that are about workflow as much as output quality.
The timing places this release amid a busy news cycle for OpenAI. On September 8, 2026, the company's newsroom also listed a proposed solution to a complex math problem called Navier–Stokes, quantum computing experiments run via GPT-5.6 Sol, and a safety post titled "An Alien Mind" published two days earlier. OpenAI Newsroom That breadth signals how OpenAI is spreading its efforts across research, applied AI, safety, and product. Images 2.5 is the consumer-facing release in the batch.
For people who use these tools, Sketch and inline commenting together suggest OpenAI is building something that feels more like a lightweight creative app than a chatbot that happens to produce images. Sketch gives the model information about layout and composition that text alone struggles to convey. Inline commenting gives you control over specific areas across multiple edits. Combined with the speed boost, the result is that editing images inside ChatGPT becomes faster and more precise.
In my view, the Sketch feature is the more interesting of the two additions. Text-to-image tools have been good at producing a decent image from a prompt, but getting them to place things exactly where you want has been hard. A sketch is a simple, direct way to show where things should go, and you do not need to learn special prompt-writing techniques to use it. Whether Sketch works reliably across many different prompts and styles will be the real test.
The multi-turn editing improvement, if it holds up, addresses what has been the single most frustrating part of chat-based image editing: the model's tendency to "fix" one thing while changing something you wanted left alone. OpenAI says Images 2.5 is better here. That claim will be tested quickly by the community.
