OpenAI Adds Canvas Sketching and Faster Rendering to ChatGPT Images 2.5

OpenAI has rolled out ChatGPT Images 2.5, updating its visual generation system with an in-app doodling tool and direct canvas annotations designed to make spatial editing more precise.

The centerpiece of the update is a feature called Sketch, according to a report by Jay Peters, senior reporter at The Verge. By entering @Sketch into the prompt field, users can open a dedicated drawing window, sketch a basic layout using a mouse or touchscreen, and pair the visual outline with a written prompt to guide the final output.

Alongside the drawing canvas, the interface now supports pinpoint feedback on generated outputs. Instead of relying purely on descriptive text to request revisions, users can select specific regions of an image to leave targeted adjustment notes, such as changing the color or texture of an isolated object.

OpenAI claims that the underlying Images 2.5 model reduces generation latency by up to 50 percent compared to the previous Images 2.0 release. The company also states that the updated model provides improved handling of natural lighting, richer surface textures, and stronger consistency when following iterative instructions across multiple conversation turns.

Solving the Spatial Prompting Gap

Text-to-image generators have long struggled with precise composition and layout control, forcing users into lengthy trial-and-error cycles to position subjects accurately. By integrating rough spatial sketches and localized annotations directly into the conversational workflow, OpenAI is adopting a hybrid interface approach that bridges conversational prompts with traditional digital drafting controls.

The Images 2.5 rollout covers web, desktop, and mobile versions of ChatGPT, and is accessible to standard users, ChatGPT Work teams, and Codex accounts.