OpenAI has released ChatGPT Images 2.5, adding an in-chat drawing tool and direct visual editing controls. The update gives users a way to guide image generation with sketches and targeted notes rather than text alone. Details of the release were reported by www.theverge.com following an OpenAI announcement on September 8, 2026.
The main addition is a feature called Sketch. Users can type @Sketch into the chat box to open a drawing window inside ChatGPT. A user can draw a doodle with a mouse and direct the model to generate a finished image based on that drawing. Users can also leave comments directly on specific sections of an existing image to request changes, such as altering eye color.
According to www.theverge.com, OpenAI claims version 2.5 delivers richer textures, more natural lighting, and better adherence to editing instructions across multiple turns. OpenAI also states that image generation latency has dropped by up to 50% compared with Images 2.0. That latency reduction is a vendor claim and has not been verified by an independent benchmark.
The update is available on desktop, mobile, and web for ChatGPT, ChatGPT Work, and Codex users, according to www.theverge.com.
This update is worth testing if you already use ChatGPT for image generation. Sketches and comments give the model visual guidance rather than relying on text descriptions alone. Before using it for client work, run three checks:
- Draw a basic doodle to see how closely the model follows shapes and object placement.
- Make two consecutive detail edits on one image to test whether the rest of the picture stays consistent.
- Time generation speeds against your current workflow to see whether OpenAI’s claimed latency reduction holds up in practice.
