Some ideas are easier to draw badly than explain well.
ChatGPT can now understand both.
OpenAI has launched ChatGPT Images 2.5 alongside Sketch, a new feature that lets users draw directly inside the chat and use that doodle as the starting point for an AI-generated image.
Instead of spending ten minutes describing where every object should appear, users can sketch the composition, add a few words about the desired style and let ChatGPT finish the picture.
How ChatGPT Sketch Works
Type “@Sketch” into ChatGPT and a drawing window appears inside the conversation.
Users can roughly map out a room, outline a piece of clothing, position objects or draw something with absolutely no artistic merit. The sketch establishes the visual structure, while the accompanying prompt explains what the finished image should look like.
In a test by The Verge, a crude cat drawn with a computer mouse was successfully transformed into a realistic photograph.
The drawing does not need to be good. It only needs to communicate the idea.
ChatGPT also now allows users to place comments directly on specific parts of an image. That means you can point to the cat’s eyes and request a different color instead of trying to describe their location in a follow-up prompt.
Images Should Survive The Editing Process
Sketch is the most immediately entertaining addition, but Images 2.5 also addresses a more practical problem with AI image creation: making one small edit without accidentally changing everything else.
According to OpenAI, the model is better at preserving faces, subjects, compositions and previous changes across multiple rounds of editing.
Lighting should look more natural, textures should feel richer and reference photos should remain more recognizable after being transformed.
Generation latency has also been reduced by up to 50 percent compared with Images 2.0.
The goal is to make image creation feel less like repeatedly starting over. An earlier adjustment should remain intact while the user continues refining another part of the image.
Templates Remove Another Blank Canvas
OpenAI is also introducing templates for frequently created formats, including posters, merchandise and product photography.
Users can select a format and add the relevant information, visual direction and stylistic details without constructing the entire prompt from scratch.
Prompts can now be shared with images too. Someone opening a shared creation can reuse the original idea with their own photos and details, turning successful prompts into remixable templates.
OpenAI says more than three billion images are created each week across ChatGPT Images and its image models in the API. Making those ideas reusable could give image prompts a second life beyond the person who wrote them.
OpenAI Is Turning The Prompt Into A Workspace
The larger change is how people communicate with an image model.
Text prompts are useful for describing mood, style and content. They are less intuitive when someone already knows where the sofa, headline or coffee cup should sit.
Sketches communicate spatial intent. Comments communicate exactly what needs changing. Templates provide structure before the first prompt is written.
Together, those tools make ChatGPT feel less like a box waiting for instructions and more like a visual workspace where an idea can be drawn, generated, reviewed and refined.
This does not turn a rough doodle into production-ready creative every time. Taste, judgment and finishing still matter.
But it does make the difficult first step much easier: showing the machine what you mean.
ChatGPT Images 2.5 is rolling out across all tiers of ChatGPT, ChatGPT Work and Codex on desktop, mobile and the web.
OpenAI is also releasing two API models. GPT-Image-2.5 Flare prioritizes speed and high-volume creation, while GPT-Image-2.5 Sunburst offers greater precision for more demanding creative work.
The future of prompting may involve considerably less typing.
Sometimes, drawing the world’s worst cat is enough.