A tool that sketches app screens and writes AI prompts is racing up GitHub — but is the hype real?
M3E Canvas has picked up over 6,000 stars in under a fortnight, promising to turn rough app sketches into ready-made prompts for AI coding tools — here's what's actually been verified.
What it actually is
M3E Canvas is a free, open-source web app that lets you sketch out app screens in the style of Google’s Material 3 Expressive design language, then wire those screens together and export the result as a written prompt. That prompt is meant to be handed to an AI coding assistant — the project’s own materials name Claude Code, Codex, Gemini CLI and Cursor — which then attempts to build the actual app from your rough layout.
It’s built in TypeScript, released under the MIT licence, and hosted on GitHub by a user going by lnkiai. According to the repository, you can draw a screen, tap through a simple prototype of it in the browser, retheme the colours, and copy a prompt describing what you’ve made. The idea sits squarely in the “vibe coding” trend — describing software in natural language rather than writing it line by line — but with a visual sketchpad bolted on the front end.
The numbers behind the buzz
The headline figure is growth: 6,191 stars and 625 forks in under two weeks, which is a genuinely fast climb by GitHub standards. Stars and forks are a rough proxy for developer interest, and a jump like this normally means either a project has struck a nerve, or it’s being amplified — through social media pushes, newsletter mentions, or in less charitable cases, coordinated star campaigns that are not uncommon on GitHub’s trending page.
The source material available doesn’t tell us which. There’s no changelog history, contributor breakdown, or independent commentary to confirm the growth is organic rather than promoted. The repo does include documentation translated into Japanese, Chinese and Korean, which suggests either a genuinely international user base or a deliberate effort to broaden reach — again, not something we can distinguish from the listing alone.
Does it actually work?
That’s the harder question, and the one that matters most. A tool that turns a sketch into a prompt is only as good as the AI coding assistant reading that prompt, and prompt-to-app pipelines are notoriously hit-or-miss — they can produce a working prototype for something simple, and something unusable for anything with real logic behind it. The project’s own demo shows a recipes app being sketched, re-themed and then built by an AI tool and run on Android, but a vendor-supplied demo video is not independent verification that the workflow holds up for anything more ambitious than a showcase example.
Who this actually affects
Right now, this is a tool for developers and hobbyists already comfortable using AI coding assistants — not something aimed at, or usable by, the general public. It requires you to bring your own AI coding tool (Claude Code, Cursor, and so on), most of which have their own costs and limitations. Nobody’s data is at risk here, nothing is being installed system-wide, and there’s no security concern to speak of based on what’s published.
The takeaway
M3E Canvas is a neat, low-stakes idea — sketch a screen, get a prompt, let an AI have a go at building it — and its star count shows plenty of developers are curious. But a fast-rising GitHub repo is a signal of attention, not proof of quality or of how many people are actually using it in anger. Worth a look if you’re already knee-deep in AI coding tools; not yet something to take on faith.