ChatGPT Sketch lets you doodle prompts for AI images

ChatGPT Images 2.5 release date and features

OpenAI unveiled ChatGPT Images 2.5 on Tuesday, September 8, 2026, introducing a fresh way to guide the model’s picture-making abilities. The new Sketch feature lets users draw a rough doodle directly inside the chat interface and then describe how they want the AI to turn that scribble into a polished image. To start, you type @Sketch in the message box, which opens a small canvas where you can sketch with a mouse, touchpad or finger. After you finish the drawing, you add a natural-language prompt describing the desired style, lighting or details, and the model generates an image that blends your doodle with your instructions.

How to doodle prompts for AI images with ChatGPT Sketch

In a quick test, the reporter drew a clumsy cat outline with a computer mouse and asked the system to produce a realistic photograph. The result was a convincing cat image that matched the requested pose and fur texture. The tool also allows you to click on specific parts of the generated picture and leave a comment requesting a change. In the example, the reporter first asked for green eyes, then later edited the comment to ask for blue eyes, and the model updated the eyes accordingly without needing to regenerate the whole scene.

OpenAI claims that the underlying image model, ChatGPT Images 2.5, delivers several upgrades over the previous version. Lighting appears more natural, surfaces show richer texture, and the model follows multi-step editing requests more reliably. Perhaps most noticeable for developers is the latency improvement; the company says image generation is up to fifty percent faster than with Images 2.0. These changes aim to make the workflow smoother for anyone who needs rapid visual feedback, from designers prototyping concepts to educators illustrating ideas.

ChatGPT Images 2.5 availability on major platforms

The rollout covers all major platforms where ChatGPT is available. Users of the standard ChatGPT service, the professional ChatGPT Work tier, and the Codex coding assistant can access Sketch and the upgraded image model on desktop browsers, mobile apps and the web interface. No additional software installation is required; the feature appears automatically once the update is applied to your account.

Benefits of hand-drawn cues for AI-generated visuals

For builders who rely on AI-generated visuals, the ability to start with a hand-drawn cue lowers the barrier to expressing vague ideas. Instead of struggling to articulate a complex scene in words alone, you can lay down a quick shape and let the model fill in the details. This can speed up brainstorming sessions, reduce the number of revision cycles, and make it easier to explore variations on a concept by simply tweaking the sketch or the accompanying comment.

The introduction of Sketch also hints at a broader trend toward multimodal interaction within conversational AI. By accepting visual input directly in the chat, OpenAI is moving the model closer to acting as a collaborative design partner rather than a pure text-to-image tool. As the feature settles into everyday use, developers may want to experiment with how doodle-based prompts affect output quality, test the limits of the model’s interpretive ability, and consider integrating similar sketch-first workflows into their own applications. Keeping an eye on latency reports and user feedback will help determine whether the promised speed gains hold up under real-world loads.

Frequently asked questions

When was ChatGPT Images 2.5 unveiled?

OpenAI unveiled ChatGPT Images 2.5 on Tuesday, September 8, 2026, introducing the new Sketch feature that lets users draw a doodle in the chat and then describe how the AI should turn it into a polished image.

How do you start using the Sketch feature in ChatGPT?

You type @Sketch in the message box, which opens a small canvas where you can sketch with a mouse, touchpad or finger, then add a natural-language prompt describing the desired style, lighting or details.

What improvements does ChatGPT Images 2.5 offer over the previous version?

ChatGPT Images 2.5 delivers more natural lighting, richer surface texture, better reliability with multi-step editing requests, and up to fifty percent faster image generation compared with Images 2.0.

Can you edit specific parts of an image generated with Sketch without regenerating the whole image?

Yes-you can click on any part of the generated picture, leave a comment requesting a change, and the model will update only that area (e.g., eye color) without needing to regenerate the whole scene.

Source: The Verge