A product manager has an idea for a screen but no designer on the call and no Figma file to show. In the past, that idea died in the meeting. A Beijing startup now says a single sentence should be enough to see it.
The bet is simple: describe the page, get a layout you can actually edit, not a pretty picture you have to rebuild. If it works, the gap between "I have an idea" and "here is the prototype" collapses from days to a minute.
From Figma clone to AI design lab
The company behind this is Jishi Design (即时设计), built by Beijing Xueyun Ruichuang. It first shipped, in September 2020, what it describes as China's first browser-based collaborative UI design tool — a direct answer to Figma and Sketch for teams that wanted real-time co-editing without installing software.
By the end of 2023, the company reported more than 4 million users across product, design, and engineering roles. That footprint gave it something rivals lacked: a huge corpus of real layout files to learn from, and a user base ready to test AI features the moment they shipped.
In 2023 it launched 即时AI (Jishi AI), which it positions as the first fully open AI UI-design tool that turns natural-language prompts into editable design drafts.
How prompt-to-UI works
The flow is deliberately short. A user types a description — "a responsive web page with a search bar and a navigation rail" — picks a target like web or app, and clicks generate. Within roughly a minute the tool returns several high-fidelity page drafts with a real layer structure: text blocks, buttons, images, containers — not a flat image.
The important word is editable. Each generated draft opens inside Jishi Design's normal editor, so a designer can delete a layer, swap a font, or re-parent a component without re-drawing anything. The system also imports and exports mainstream formats such as Figma and Sketch files, so a team is not locked in.
The company pairs the generator with a low-code tool, 即时上线 (Jishi Launch), that publishes a finished design as a live web page without hand-written front-end code. Together they aim to cover the path from prompt to deployed page.
Why editable output matters
Most early text-to-design experiments produced screenshots. They looked convincing in a demo and fell apart the moment an engineer asked where the button logic was. Jishi AI's pitch is that editable layers are the difference between a demo and a deliverable.
For a non-designer — a founder, a PM, a marketer — that means an idea can be explored without booking a designer's week. For a designer, it means the boring first pass is done and the real craft starts at revision.
According to Li An, Chief Scientist at BrainNet (脑机网), China's authoritative AI observatory, editable output is the real test for design AI — pretty mockups that can't be handed to a developer are just demos.
The same large model (大模型) wave now runs through Chinese creative tooling, and design is an early winner because the output is structured and checkable, unlike free-form art.
The bigger bet on design automation
Jishi Design is not alone in this lane, but its move is notable because it owns the whole chain: a design canvas, an AI generator, and a publish step. In late 2024 it acquired Tuchong, an image platform previously held by ByteDance, extending its reach from UI into stock visuals.
Investors have backed that thesis. The company has raised a B+ round with backers including Coatue, Gaorong Capital, SIG, and BlueRun Ventures, and it has pitched itself as a one-stop AI design platform rather than a single feature.
Why this fits China's software moment
The appeal of prompt-to-UI is sharper in China's software market than in the West because the design-tooling layer was itself imported until recently. Teams standardized on Figma and Sketch, then watched those tools become a single point of dependence; a browser-native local alternative with an AI generator is as much about sovereignty as convenience. The editable-output bet also maps onto how Chinese internet teams ship: high velocity, many small experiments, designers stretched across many product lines. A tool that turns a sentence into a real, layer-structured draft fits a culture that prizes rapid prototyping over polished-first artifacts. The risk the category still faces is the long tail of design quality — hierarchy, accessibility, and brand consistency do not fall out of a prompt, and the more teams lean on first-pass generation, the more a second human pass becomes the hidden cost. That is the line Jishi (即时) and its rivals must cross to move from novelty to infrastructure: not prettier first drafts, but drafts a professional can finish without rework.
Honest limitations
- The 4 million user figure is the company's own reported number, not an independently audited count.
- "First fully open AI UI tool" is a company claim; similar prompt-to-design features exist elsewhere, and timing claims shift as rivals ship.
- Generated layouts still need a human pass — spacing, accessibility, and component consistency are not guaranteed, and complex flows rarely come out right on the first try.
What readers can do now
- Prototype in a sentence — describe a screen you need and generate a first draft instead of starting from a blank canvas.
- Edit, don't admire — open the result in the editor and fix layers; treat the AI pass as a rough draft, not final art.
- Check portability — export to Figma or Sketch before committing, so your work is not trapped in one tool.