OpenAI's draw-to-image feature — sketch a rough guide in ChatGPT with @Sketch and the model renders it into a detailed, composition-controlled image.
Last verified · 2026-09-08 · by Moe Ameen
ChatGPT Sketch is a feature inside ChatGPT Images 2.5, OpenAI's image model announced September 8, 2026, that lets you draw a rough guide directly in the chat and turn it into a detailed image. You invoke it by typing @Sketch, which opens a drawing canvas; you doodle the layout you want — the shape of a room, the outline of an outfit, how objects sit in a frame — and add a text description of the style and details. The model treats the drawing as a compositional reference, not a literal image to trace, so a crude scribble is enough to steer where things land.
The point of Sketch is control over the one thing text prompts handle worst: composition. Describing content and style in words is easy; pinning down placement and framing is not. A rough drawing communicates layout in a single gesture, so you get the framing you had in mind without prompt tricks or a dozen retries. OpenAI's framing is that you do not need to be an artist — the sketch is just a faster way to say where everything goes.
Sketch shipped alongside the rest of Images 2.5: Templates (prebuilt starting points for formats like posters and merch), the ability to comment directly on a generated image for focused edits, and up to 50% faster generation than Images 2.0. It rolled out to all ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web.
The boundary worth planning around: Sketch is an input to one image, not a content workflow. It steers a single generation and stops there — it does not caption for social, turn one image into a branded multi-format set, keep a persona's face consistent across posts, or publish anywhere.
Sketch removes the guesswork on the input side of one image; [Kompozy](/) removes the grind on the output side of a whole content calendar — and the two chain together cleanly. Say you sketch the composition for a launch graphic, describe the style, and let ChatGPT render it. That is a finished image and an unfinished post. Bring it into Kompozy and it becomes the base asset for a coordinated drop: a brand-exact [Carousel](/glossary/hyperframes) built with HyperFrames, a Quote Graphic pulling the headline, and Photo Posts with per-platform copy written in your own voice through the [Persona Brief](/glossary/persona-brief) — each reframed for its destination and scheduled across nine platforms plus blog and email with autopilot and a review step.
The complementary strength is worth naming precisely. Sketch controls the layout of a single image, once; Kompozy makes layout and identity repeatable across everything. HyperFrames stamps your exact brand template onto every output, and Gemini face-lock keeps the same persona face on every render, so a series looks like one brand rather than a set of one-off generations. So the division of labor is clean: use Sketch to nail the composition of the individual image, and use Kompozy to turn that image into a consistent, published presence across every platform. For the wider model Sketch lives in, see our [ChatGPT Images 2.5 overview](/ai-tools/chatgpt-images-2-5).
It is a feature in ChatGPT Images 2.5 that turns a rough drawing into a detailed image. You type @Sketch, draw a rough guide for the layout, add a text description of the style, and the model renders a finished image using your sketch as a compositional reference. It launched September 8, 2026.
Type @Sketch in ChatGPT to open a drawing canvas, sketch the rough layout you want — a room, an outfit shape, a scene arrangement — and describe the style and details in text. The model uses the drawing as a guide for composition rather than a literal image. You do not need drawing skill; a crude scribble is enough to steer placement.
No. It reads the sketch as a compositional reference, not a literal trace, so the model reinterprets it into a detailed render. That is ideal for communicating framing and placement, but exact, dense layouts can drift and usually need a text description to disambiguate.
Generate the composition-controlled image in ChatGPT, then use a content engine to distribute it. On its own Sketch makes one image and stops. In Kompozy you fan that image into a branded carousel, quote graphic, and Photo Posts, add per-platform captions in your voice, and schedule them across nine platforms — the steps Sketch does not do.
It rolled out with ChatGPT Images 2.5 to all ChatGPT, ChatGPT Work, and Codex users, including the free tier, for in-app use. Paid tiers get faster and higher-volume image creation, and some plan-level image caps are not officially published, so confirm current limits at OpenAI's pricing page.