OpenAI rolled out ChatGPT Images 2.5 on Tuesday alongside a feature called Sketch that lets users draw directly inside the chat interface to guide image generation. The combination is aimed at closing the gap between what a user can describe in text and what they actually want to see rendered.
Sketch as a Visual Prompt
The Sketch feature activates with @Sketch in the chat input. A drawing canvas appears, and the user sketches whatever they want, then adds a text prompt telling ChatGPT how to interpret it. A rough mouse-drawn cat doodle becomes a realistic photo with the right instructions. Users can also leave comments directly on generated images to request specific changes, like swapping eye color from blue to green. The interaction model is straightforward: draw, describe, refine.
Images 2.5 Improvements
The updated model produces images with more natural lighting and richer textures, according to OpenAI. Editing accuracy across multiple turns has improved, and generation latency dropped by up to 50 percent compared to Images 2.0. The update is available now for ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web.
For developers building tools on top of ChatGPT's image capabilities, the Sketch integration and iterative commenting mean the API surface for image generation is getting more complex but also more expressive. The latency reduction matters for applications where generation speed is part of the user experience, and multi-turn editing opens up workflows where image refinement happens in a conversation rather than through repeated single-shot generations.