ChatGPT users can now draw a rough image inside a conversation and ask the system to turn it into a finished picture. Typing @Sketch opens a canvas, and the drawing becomes part of the prompt alongside written instructions. The feature gives people a spatial way to describe composition that can be difficult to express in text alone.
Users can also place comments directly on parts of an existing image to request targeted edits. The Verge’s brief test converted a mouse-drawn cat into a realistic image and then changed its eye color through a localized instruction. Results will still depend on the model interpreting both the sketch and accompanying description correctly.
The tools arrive with ChatGPT Images 2.5. OpenAI claims the model follows multi-step edits more reliably, produces more natural lighting and textures, and cuts generation latency by as much as 50 percent compared with Images 2.0. The update is available now to ChatGPT, ChatGPT Work and Codex users across desktop, mobile and web.