Article8 min readUpdated
Sketch to UI With AI: What tldraw Make Real Shows Us
Sketch to UI with AI turns a rough drawing into a clickable screen. See how tldraw make real does it, its current status, limits and safe uses.
By Tithi Mondal, UI/UX Designer
Imagine a founder in Kochi standing at a whiteboard. She draws a box for a phone number, a button that says "Send OTP", and an arrow to a second box named "Home". Her team nods. Someone takes a photo and posts the photo on WhatsApp. Then nothing happens for a week, because a photo of a whiteboard cannot be clicked.
Sketch to UI with AI is an answer to that gap. Sketch to UI with AI means you draw a rough interface by hand and an AI model turns the drawing into a working screen that you can click.
The clearest demonstration of the idea is make real, an experiment from the team behind tldraw. The tldraw website describes tldraw as an infinite canvas SDK for React. SDK means software development kit: a set of building blocks that developers use to add a whiteboard to their own product.
Checked in November 2026. Tools change quickly, so confirm details on the project's own page.
A note on status before anything else. The make real repository on GitHub carries a notice that the owner archived the repository in February 2026 and that the repository is now read-only. I could not confirm from primary sources whether the public demo website still works. So this article treats make real as a demonstration of an idea, not as a product to depend on.
What is tldraw make real?
Make real is an experiment in which you draw an interface on a tldraw canvas, press a button, and receive a working web page on the same canvas. The repository describes the project in one line: "Draw a ui and make it real."
tldraw published a blog post about the experiment in November 2023, written by Steve Ruiz of tldraw. The post describes the flow as "sketch an interface, press a button, and get a working website".
The post also explains the origin. An engineer named Sawyer Hood shared a small project named draw-a-ui. The project sent a drawing from a tldraw canvas to an AI model that can read images, and received a working website back. The tldraw team built make real from that project.
How does a drawing become a working interface?
The app takes a picture of your drawing, sends the picture to an AI model with instructions, and places the returned web page on the canvas. tldraw also published a starter template so that developers can build similar apps. According to the project's documentation in that template, the steps are:
- You draw a mockup of an interface on the canvas.
- You select the drawing and press the Make Real button.
- The app captures an image of the selection.
- The app sends the image to an OpenAI model that can read images, with instructions to turn the image into an HTML file.
- The app places the returned HTML on the canvas inside an iframe. An iframe is a small window that shows one web page inside another page.
- You can interact with the result on the canvas.
The interesting part is the next round. The template's README says you can annotate the result, select the result, and press the button again. To annotate means to draw notes and arrows on top. The blog post adds a technical detail. The embedded page appears as an empty rectangle in the exported picture. So the app also sends the earlier HTML back to the model, with a note that the earlier HTML belongs in that rectangle.
The blog post also says the result can be resized on the canvas to test how the page behaves at different widths. The template's README says the template needs an OpenAI API key. An API key is a secret code that lets an app use a paid AI model on your account.
Why can a drawing instruct a model better than a typed prompt?
A drawing shows position, size and order at once, and those are the things that words describe badly. This section is my opinion as a designer, not a claim from the tldraw documents.
Try to type the layout of a fee payment screen. "Put the student name at the top. Put the month below the name, but smaller. Put the amount on the right." After 3 sentences the reader is lost. A sketch of 4 boxes says the same thing in a few seconds.
A drawing also keeps the conversation at the right level. A rough sketch looks unfinished, so people talk about structure. A polished mockup invites comments about colour too early. I explain the difference between these stages in wireframe vs mockup vs prototype.
The blog post describes the canvas as a "conversation space" between the person and the model. I think that description is the real lesson of make real. The drawing is the prompt. The correction is also a drawing. This is why people describe the idea as whiteboard to working interface.
How could a founder use this idea this week?
You can use the idea without depending on any one website: draw first, and then ask a model to build exactly what is drawn. The preparation is the valuable part, and the preparation needs only paper.
Suppose a coaching centre with 300 students wants parents to pay the monthly fee from a phone. A plan for one week:
- Write the one task in a sentence: "A parent pays this month's fee and gets a receipt." My guide on what a user flow is helps with this step.
- Draw 3 or 4 screens as boxes: Login with phone number and OTP, Fee due, Pay with UPI, Receipt.
- Write the real field labels inside each box: "Student name", "Month", "Amount due", "Pay now".
- Draw arrows between the screens, and write beside each arrow what the button does.
- Ask your developer to turn the drawing into a clickable draft. The public starter template is one route, and the template needs your own model provider key.
- Give the draft to one colleague and one parent. Watch where each person stops.
- Keep the notes. Be ready to discard the generated code.
This kind of drawing to prototype work is useful inside the team. For a meeting with investors, a designed prototype is the better choice. I explain the reason in prototype for an investor demo.
What are the limits and risks?
The idea is good for exploring a layout, and the idea is weak at everything a real product needs after the layout. Here are the limits I would expect, and the risks that the tldraw documents name.
- An experiment can stop. The archived notice on the repository is the reminder. Do not plan a business process around a demonstration.
- API keys on a public site. The tldraw blog post says the original demo asked visitors to paste their own key, and calls that approach "usually a terrible idea and security risk". The template's README says users should be reluctant to add API keys to public sites. If your developer runs a copy, the key should stay with your developer.
- Cost of running models. Every press of the button uses a paid model. Many rounds of drawing and correcting add up.
- Privacy of your data. The picture of your drawing is sent to an outside AI provider. Do not write real customer names or account numbers in a sketch.
- Made-up details. A sketch leaves gaps, and the model fills the gaps with invented text, fields and menu items. Check that every element is one you actually wanted.
- One page, not an app. According to the template, the model returns an HTML file. A button named "Pay with UPI" in that file does not take a payment. There is no login, no database and no receipt.
- A simulation is not evidence. A clickable page proves that the layout can be drawn. A clickable page does not prove that parents want to pay this way.
- Licence. The starter template's repository shows the AGPL-3.0 licence. Your developer should read the licence before building a product on the template.
What does a designer still do after the sketch works?
A designer turns one working sketch into a product that stays consistent and clear on every screen, for every kind of user. The sketch answers "where do things go". The sketch does not answer the harder questions.
Those questions are practical. What does the Fee due screen show when no fee is due? What does the parent see when the UPI payment fails? Does the screen still work in Bengali or Hindi, where words are often longer? Can a grandparent with weak eyesight read the amount? On the edtech product I designed, with live classes and progress reports, questions of this kind took more thought than any single layout.
Real users still must be asked. A few parents trying the draft on their own phones will tell you more than any generated page. My guide to usability testing on a small budget describes a simple way to run those sessions. If you plan to use AI drafts alongside a designer, read how to use AI with your designer first.
My position: sketch first, always. Whether a model or a person builds from the sketch, a clear drawing with real labels saves days. If you want help turning rough sketches into a complete set of app screens, my mobile app design service starts from exactly that kind of drawing.
Sources
Frequently asked questions
Is tldraw make real still available?
The make real repository on GitHub is marked as archived and read-only, as checked in November 2026. I could not confirm the status of the public demo website, so check the tldraw project pages directly. The starter template repository is still public.
Do I need to know coding to turn a sketch into a UI with AI?
You do not need coding to draw the sketch and write the labels, and that is the most valuable part. Running a tool such as the make real starter template needs a developer and a model provider key.
Can a sketch-to-UI tool replace wireframing?
No. The sketch is the wireframe. The tool only turns that wireframe into something clickable faster, and the quality of the result depends on how clearly the flow and labels were drawn.
Is the generated page ready to launch?
No. The result is a single web page without login, payments, saved data, error states or accessibility checks. Treat the page as a draft for discussion inside the team.
Working on something like this?
Tell Tithi what you are planning. You get a clear scope, a fixed price and a timeline before any work begins. I reply to every enquiry within 24 hours.