AI & MCP
Best AI UI generators in 2026 (179 records)
AI UI generators turn a prompt into a screen, not a component a design system can absorb. What to check before you ship the output.
Key points
- 179 components tagged ai and 44 tagged agent in the EneUI catalogue as of 1 November 2026.
- 123 of the 179 records declare a licence; the rest need an upstream check.
- No generator is repeatable: the same prompt returns different markup.
AI UI generators turn a prompt into a screen, not a component a design system can absorb. The EneUI catalogue holds 179 components tagged ai as of 1 November 2026, and that count is the measure this page uses to separate a prototype from a reusable record.
| Tool | What it produces | Licence | Measured figure |
|---|---|---|---|
| v0 (Vercel) | React and Tailwind page from a prompt | Proprietary, SaaS | No public repository |
| Lovable | Complete React app from a prompt | Proprietary, SaaS | No public repository |
| Bolt.new (StackBlitz) | Full-stack project in the browser | Proprietary, SaaS | No public repository |
| Figma Make | UI generated from a Figma file | Proprietary, SaaS | Inside Figma, still in beta |
| Onlook | Open-source visual React editor | Apache-2.0 | 27,070 stars on 1 November 2026 |
What the number means
179 is the number of records the catalogue collected from public libraries tagged ai, not the number of generative tools on the market. It includes chat, loaders, heroes and prompt boxes: 123 of the 179 records declare a licence, so the other 56 need an upstream check before use.
How to judge a generator
- Read the output: JSX with explicit dependencies, or HTML with arbitrary classes.
- Ask for the licence on the result and on the component it started from.
- Test the states: focus, disabled, loading, error.
- Check whether the result enters your design system or replaces it.
- Run the same prompt twice: if the structure changes, it is not repeatable.
The catalogue record
Two records show the difference that matters. ai-prompt-box has 2,826 stars declared upstream on 1 November 2026 and no licence in the record: until the author publishes terms, treat it as unlicensed. animated-ai-chat declares MIT and 1,513 stars.
Where to look in the catalogue
The ai category holds the 179 records. The Magic UI library is an example of a source with a declared licence and a shadcn install path, so it is a sensible destination after the first generated draft.
curl "https://ui.eneawork.it/api/vault/items?cat=ai&limit=5"
The total field in the response is the number quoted here. Swap cat for agent or ai-chat to count the other tags, without opening the data file.
The limitation
No AI UI generator is repeatable: the same prompt returns different markup on every run. Generated code carries no licence from the component the model learned from, so the right to use it stays a manual check, case by case.
Where to continue
The criteria for picking a library are in how to choose a UI library; the handover from the design file to code is in Figma to React without rebuilding. To see what actually lands in the repository read copy-paste or headless.
Frequently asked questions
Which AI UI generator is the best?
There is no absolute first. v0 and Lovable win when you need a React page quickly, Onlook when you want an open editor you can read. Count the dependencies the result adds before you choose.
Does generated code come with a licence?
Rarely an explicit one. The service grants use of the output in its terms, but the component the model learned from may carry a different licence: check the source record.
How many AI components are in the EneUI catalogue?
179 tagged ai, 170 tagged ai-chat and 44 tagged agent, measured on 1 November 2026.
Should I use a generator or a catalogue component?
The generator for the first draft, the catalogue for the code that stays. A component with a declared licence costs less than markup you rewrite.
Will a generator respect my design system?
Only if you pass it the tokens. Without CSS variables and a type scale in the prompt, the result invents colours and spacing, and fixing it costs more than writing the component by hand.