OpenAI adds doodle-based image creation to ChatGPT

ChatGPT Images 2.5 release September 2026

OpenAI unveiled a new version of its image generation model on Tuesday, September 8, 2026, called ChatGPT Images 2.5. Alongside the update, the company introduced a feature named Sketch that lets users turn rough drawings into polished AI pictures. Instead of typing a description, you can now sketch an idea directly inside the chat window and then tell the model how to turn that sketch into a final image.

To start Sketch, you type @Sketch in the message box. A small canvas pops up where you can draw with a mouse, trackpad or touchscreen. The drawing does not need to be detailed; even a clumsy doodle works as a starting point. After you finish the sketch, you add a prompt describing the style, lighting, or any other details you want the model to apply. The system then processes both the visual input and the text instruction to produce an image that matches your request.

In a quick test, a rough mouse-drawn outline of a cat was turned into a realistic photograph. The user asked for a cat with green eyes, and the model obliged. Later, by clicking on the cat’s eyes in the generated picture and adding a comment, the user changed the eye color from green to blue, showing that the tool supports localized feedback directly on the output.

Technical upgrades in ChatGPT Images 2.5

Beyond the sketching ability, Images 2.5 brings several technical upgrades. OpenAI says the model now renders scenes with more natural lighting and richer surface textures, making results look less artificial. It also follows multi-step editing instructions more reliably, so you can ask for a series of changes without losing track of earlier requests. Latency has dropped noticeably, with the company claiming up to a 50 % reduction compared to the previous Images 2.0 version.

The updated model is available right away for anyone using ChatGPT, ChatGPT Work, or Codex. Access works across desktop browsers, mobile apps and the web interface, so you can sketch and generate images whether you are at a desk or on a phone. No extra subscription tier is required beyond the existing plans that already include image generation.

For developers and creators who rely on AI-generated visuals, the Sketch feature lowers the barrier to getting a concept across. Instead of struggling to find the right words for a complex scene, you can scribble a rough idea and let the model fill in the details. This can speed up brainstorming sessions, help prototype UI elements, or produce quick illustrations for presentations and social media posts. The ability to comment on specific parts of an image also means you can iterate quickly without having to rewrite the whole prompt each time.

Who benefits from latency improvements in ChatGPT Images 2.5

The latency improvements matter for anyone who needs rapid feedback, such as game designers testing concept art or marketers producing ad variations on the fly. Faster generation reduces waiting time and keeps creative flow uninterrupted. The improvements in lighting and texture also make the output more suitable for professional use, where visual fidelity can affect client perception.

If you want to try the new capability, open a ChatGPT conversation, type @Sketch, draw your idea, and add a brief description of the style or details you want. Experiment with adding comments on the generated image to tweak particular areas. Keep in mind that while the model handles rough sketches well, very abstract or ambiguous drawings may still need clearer textual guidance to achieve the desired result.

Overall, the addition of Sketch to ChatGPT Images 2.5 gives builders a more intuitive way to communicate visual ideas to an AI. It blends the immediacy of drawing with the precision of language-based prompting, offering a flexible tool for rapid prototyping, design exploration, and content creation. As the feature rolls out, watching how users integrate it into their workflows will reveal new patterns for AI-assisted creativity.

Frequently asked questions

What is the new Sketch feature in ChatGPT Images 2.5 and how do you start using it?

To use Sketch, type @Sketch in the message box; a small canvas appears where you can draw with a mouse, trackpad or touchscreen. After finishing the sketch, add a text prompt describing style, lighting or other details, and the model creates an image from both inputs.

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

Images 2.5 renders scenes with more natural lighting and richer surface textures, follows multi-step editing instructions more reliably, and reduces latency by up to 50 % compared to Images 2.0.

Who can access the Sketch feature and is there any extra cost?

The Sketch feature is available immediately for anyone using ChatGPT, ChatGPT Work, or Codex, and works on desktop browsers, mobile apps and the web interface. No extra subscription tier is needed beyond the existing plans that already include image generation.

How can users refine images after they are generated with Sketch?

After an image is generated, you can click on a specific area-such as a cat’s eyes-and add a comment to change attributes like color. The model applies the localized feedback, letting you iterate without rewriting the whole prompt.

Source: The Verge