ChatGPT Sketch AI images are OpenAI’s latest attempt to give people more control over image generators without asking them to become prompt engineers. Released September 8 with ChatGPT Images 2.5, Sketch lets a user draw directly inside a ChatGPT conversation and submit that drawing as a visual reference for an image-generation request.
The practical point is modest but useful: a rough drawing can communicate placement, shape, and composition that words often leave ambiguous. The drawing does not become a finished illustration by itself. Users still tell ChatGPT what sort of image they want it to produce from the sketch.
OpenAI is also adding the ability to put comments on specific parts of a generated image. That creates a three-step workflow: sketch the broad idea, ask for an image, then identify an area that needs revision instead of describing the whole picture again.
How do you use ChatGPT Sketch?
According to The Verge, users can invoke the feature by entering @Sketch in the ChatGPT message box, which opens a drawing area. After making a sketch, the user adds an instruction describing the intended result and asks ChatGPT to generate it.
The Verge reporter Jay Peters said a brief test converted a cat doodle made with a computer mouse into a realistic cat image after he supplied instructions. Peters also used the new image-commenting control to request a change to the cat’s eye color. That is one hands-on example, not a quality benchmark or proof that every rough sketch will be interpreted faithfully.
What else changed in ChatGPT Images 2.5?
Sketch is part of a wider Images 2.5 release. OpenAI says the model produces sharper details, more natural lighting and richer textures; holds onto subjects in reference photos more effectively; and follows editing instructions more reliably over several turns. Those are company performance claims, not independent measurements.
OpenAI also says image-generation latency can be as much as 50% lower than for Images 2.0. The update includes templates for common image formats and a way to share prompts, alongside the sketching and annotation controls.
Images 2.5 is available to ChatGPT, ChatGPT Work, and Codex users on desktop, mobile, and the web, according to OpenAI. For developers, the company has released two related API models: GPT-Image-2.5 Flare, which OpenAI says carries the release’s quality, editing, and speed improvements, and GPT-Image-2.5 Sunburst, which it says offers greater editing precision at the cost of longer generation times.
For ordinary ChatGPT users, Sketch is the more legible change. It gives the model a visual constraint before it starts filling in the pixels, then lets the user point to a mistake afterward. That does not remove the need to check the output. It does make “roughly like this” a usable instruction.
This story draws on original reporting from The Verge.
