OpenAI's ChatGPT Sketch turns rough doodles into polished AI images

OpenAI on Tuesday rolled out ChatGPT Images 2.5, an upgrade to its image generation model, and with it came a new way to direct the AI: Sketch. The feature lets users draw a rough doodle directly inside ChatGPT and then describe what they want the final image to look like. Instead of wrestling with text prompts alone, you can sketch a basic shape or scene and let the model fill in the details.

Sketch is designed to close the gap between what people imagine and what they can articulate in words. Many users struggle to describe a specific composition, angle, or layout, but a quick drawing can communicate that instantly. Once you've scribbled your idea, you can type instructions like "make it a photorealistic sunset" or "turn this into a cartoon character," and ChatGPT Images 2.5 will generate a finished image based on both inputs.

The announcement follows a broader push by OpenAI to make its image tools more interactive and practical. Earlier this year, the company introduced native image generation inside ChatGPT, which allowed the model to render text, logos, and diagrams more accurately. The new Sketch feature builds on that by giving users a visual shorthand for directing the model, rather than relying solely on long, detailed prompts.