An honest 2026 review of ChatGPT Sketch, OpenAI's draw-to-image feature: how well it controls composition, its limits, and where it stops for creators.
ChatGPT Sketch is the most intuitive composition-control feature OpenAI has shipped for image generation — draw a rough guide with @Sketch and the model honors your layout in a way a text prompt rarely can. For nailing framing and placement in a single image, it is a genuine upgrade and it is free inside ChatGPT. But it is one input to one image, not a workflow: it does not brand, caption, keep a face consistent across posts, or publish, so judge it as a better steering wheel, not a content engine.
ChatGPT Sketch launched on September 8, 2026 as part of ChatGPT Images 2.5, OpenAI's new image model. It lets you draw a rough guide directly in the chat — you type @Sketch, doodle the layout you want, and add a text description — and the model renders a detailed image that follows your drawing as a compositional reference. This review judges Sketch specifically, the draw-to-image feature, not the whole Images 2.5 release (which we cover separately).
I run a content engine and ship visual posts every week, so the question I care about is narrow and practical: does drawing a guide actually get me the framing I wanted faster than prompting for it? The honest answer is yes, clearly — composition is the one thing text prompts are worst at, and a scribble communicates it in a single gesture. Where Sketch earns a mixed score is everything around that: it controls one image, and the model still decides the final render, and nothing about it touches the steps after the image exists.
The ratings below reflect that. High marks for composition control, ease, and the fact that it is bundled free; middling marks for how literally it honors a drawing and for the ceiling on complex layouts; and low marks for the things it was never meant to do — brand governance, consistency across a calendar, and publishing. Everything here is reconciled against OpenAI's launch announcement and ChatGPT as of 2026-09-08; where a detail could shift (plan-level image caps), I say so rather than inventing precision.
ChatGPT Sketch is an input feature inside ChatGPT Images 2.5. You invoke it by typing @Sketch, which opens a drawing canvas in the chat; you rough out the layout of what you want — the shape of a room, the contour of an outfit, the arrangement of objects — and pair it with a text description of the style and details. The model treats the drawing as a guide for composition and placement, not as a literal image to trace, so a crude sketch is enough to steer where elements land. OpenAI's pitch is that you do not need to draw well; the sketch is just a faster way to communicate framing than words. Sketch is one of several additions in the Images 2.5 release, alongside Templates (prebuilt starting points for formats like posters and merch) and the ability to comment directly on a generated image for focused edits, with generation up to 50% faster than Images 2.0. It rolled out to all ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web. What Sketch is not is a standalone product or a content tool: it produces one image in a chat, and it does not caption for social, turn that image into a branded set, hold a persona identity across many posts, or publish anywhere.
Sketch fits anyone who knows what they want a frame to look like and is tired of describing it: creators laying out a thumbnail, marketers arranging a product shot or flyer, designers roughing a concept, and non-artists who can gesture at a composition faster than they can prompt one. If your images live or die on placement — where the subject sits, how the scene is arranged, what the framing is — this is the fastest control ChatGPT offers. It is a weak fit as the engine of a content operation: a solo creator or brand whose real bottleneck is producing a week of on-brand posts across many feeds, with the same face and frame every time, gets one better image from Sketch and still faces the fan-out, branding, and publishing work on their own.
| Dimension | Score | Why |
|---|---|---|
| Composition & layout control | 4.5 / 5 | Drawing a guide is a real gain over prompt-only steering — you communicate framing and placement in one gesture. |
| Ease of use | 4.4 / 5 | Type @Sketch, doodle, describe — no drawing skill or prompt tricks needed, and it is available across desktop, mobile, and web. |
| How faithfully it honors the drawing | 3.8 / 5 | It uses the sketch as a reference, not a literal trace, so the model still reinterprets — great for intent, less so for exact placement. |
| Output image quality | 4.4 / 5 | It rides Images 2.5's quality — natural lighting and rich texture — so a rough guide can still produce a polished result. |
| Handling complex or dense layouts | 3.5 / 5 | Simple compositions translate cleanly; busy, multi-element scenes drift more and need text to disambiguate. |
| Value / accessibility | 4.3 / 5 | Bundled free into every ChatGPT tier for in-app use, though some plan-level image caps are not officially published. |
| Brand & persona consistency across a calendar | 2.4 / 5 | Sketch controls one image; there is no locked persona face or brand frame reused across a feed of posts. |
| Multi-platform publishing | 1.5 / 5 | None — it is an input to a single image; captioning, fan-out, and posting happen entirely elsewhere. |
Sketch is not priced on its own — it is a feature of ChatGPT Images 2.5, which is bundled into every ChatGPT tier. It works on the free plan, on Plus (about $20/mo), and on the Pro plans (up to about $200/mo), with higher tiers getting faster and higher-volume image creation. There is no separate purchase to try Sketch; if you have ChatGPT, you have it. That makes it one of the better-value control features in image generation this year — you are not paying extra for a capability that meaningfully improves your hit rate on composition.
The caveat is the same one that applies to all in-app image generation: per-plan image caps are not all officially published, and community usage suggests free and Plus users can hit rolling limits during heavy sessions, while the top Pro tier is effectively unlimited within OpenAI's abuse guardrails. If you sketch and regenerate heavily to dial in a layout, you burn through your allowance faster, so confirm current caps at OpenAI's pricing page.
The honest framing on cost is that Sketch improves the making of one image, not the running of a content operation. Your ChatGPT subscription buys you better single images; it does not buy you the branding, fan-out, captioning, and publishing that turn an image into posted content. If you are pricing a full visual workflow, budget for the separate tools or hours those steps still take.
| Use case | Fit | Why |
|---|---|---|
| Nailing the framing of a single image | Strong | Drawing the layout is exactly what Sketch is built for and its clearest advantage over prompting. |
| Roughing a thumbnail or product-shot composition | Strong | Placement matters most here, and a sketch communicates it faster and more reliably than text. |
| Directing an image without design or drawing skill | Strong | A crude scribble is enough to steer the model, so non-artists get compositional control. |
| Precisely reproducing an exact, dense layout | OK | It reads the sketch as a reference, not a trace, so complex placements still drift and need text. |
| Keeping one persona's face and frame identical across a month of posts | Weak | Sketch controls one image; there is no locked identity or brand template reused across a calendar. |
| Turning one sketched image into a branded carousel and per-platform set | Weak | It produces a single image; it has no fan-out into multiple branded formats. |
| Scheduling and publishing images across many platforms | Weak | There is no scheduler or publishing layer — every post is placed by hand elsewhere. |
Sketch and Kompozy solve control at different scales, and it is worth being precise about the difference. Sketch gives you composition control over one image — you draw where things go, once, per generation. Kompozy is not a better draw-to-image tool and does not try to be; its answer to layout control is a repeatable brand system. HyperFrames renders pixel-exact templated layouts in your own brand style across many outputs, so the framing is consistent by design rather than re-sketched each time, and Gemini face-lock keeps a persona's face identical across every render. One is a per-image steering wheel; the other is a brand template that stamps the same structure onto a whole calendar.
So they are complementary, not competitive. Use Sketch to nail the composition of a hero image or thumbnail, then bring that image into Kompozy to fan it into a branded carousel, quote card, and Photo Posts, write per-platform copy in your voice through the Persona Brief, reframe per destination, and publish across nine platforms plus blog and email. Sketch controls the frame of one image; Kompozy controls the brand across many and ships them. For the fuller comparison of the model Sketch lives in, see our ChatGPT Images 2.5 review.
For controlling the composition of a single image, yes. Drawing a rough guide with @Sketch communicates framing and placement far better than a text prompt, it needs no drawing skill, and it is free inside ChatGPT. Its limits are workflow, not control: it steers one image and does not caption, brand, or publish.
You type @Sketch to open a drawing canvas in the chat, doodle the rough layout you want, and add a text description of the style and details. ChatGPT uses the drawing as a compositional reference rather than a literal image and generates a detailed result. A crude scribble is enough to steer where elements land.
No. It treats your sketch as a guide for composition and placement, not as a literal image to trace, so the model reinterprets it into a detailed render. That is great for communicating intent, but exact, dense layouts can drift and usually need text to disambiguate.
It shipped 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.
No. Sketch controls one generated image; it does not caption for social, turn that image into a branded multi-format set, hold a consistent persona face, or publish. Fanning one image into a captioned, reframed, per-platform set and scheduling it is what a content engine like Kompozy does.
Sketch lets you draw a compositional guide for a fully custom image. Templates are prebuilt starting points for common formats like posters and merch, so you start from a preset frame instead of a blank one. They are two different ways to reduce the guesswork in a single image; neither handles distribution.
For layout and placement, usually yes — a drawing communicates where things go in one gesture, which text struggles to do. For style, mood, and content detail, text is still the better channel, so Sketch works best paired with a clear description rather than replacing it.