ChatGPT's new Sketch feature transforms rough doodles into polished AI images

OpenAI has unveiled ChatGPT Images 2.5, a refreshed version of its image creation engine that includes a novel input method: the Sketch feature. Rather than relying solely on text descriptions, users can now draw directly within the chat interface to guide the AI's output.

How Sketch works#

The feature is simple to access—simply type @Sketch in the message box and a drawing canvas appears. Users can create rough sketches using their mouse, trackpad, or stylus, then provide text instructions for how they want ChatGPT to render the concept. The system takes the visual template and transforms it into a detailed, polished image. In testing, a hastily drawn cat sketch became a photorealistic image once refined with instructions about lighting, texture, and color.

Enhanced editing capabilities#

Beyond Sketch, the platform now allows pinpoint annotation. Users can comment directly on specific sections of generated images to request alterations—for instance, changing eye color or adjusting composition. This granular feedback system works across multiple iterations, making it easier to arrive at the desired result through conversation rather than restarting from scratch.

Performance improvements#

The updated Images 2.5 engine shows tangible gains under the hood. OpenAI reports that the new model produces images with more naturalistic lighting and textural depth compared to its predecessor. Processing speed has also improved, with generation latency cut by up to 50% relative to Images 2.0. The update is available now for ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web platforms.