2. 1. The Four Categories
Every tool in this space falls into one of these buckets. Knowing which bucket a tool is in tells you almost everything about who it is for and what it cannot do.
| Category | What it does | Who uses it | Key tools |
|---|---|---|---|
| A: Vibe Design Canvas | Prompt (or sketch or voice) in; editable visual design out. No code. | Designers, PMs, founders doing early exploration | Google Stitch, Banani, Flowstep, Moonchild, UXPilot, Visily, Uizard, UXCanvas |
| B: Design-to-Code Hybrid | Generates both a visual design and working frontend code simultaneously | Developers who want to design, designers who want to ship | Figma Make, Magic Patterns, v0, Framer |
| C: Website-Specific Builder | Narrow focus: sitemaps, landing pages, marketing sites. Not app UI. | Agencies, Webflow/Framer designers, marketing teams | Relume, Framer AI |
| D: AI on Incumbents | AI features inside tools people already use (Figma, Canva, Adobe) | Everyone who already uses those tools | Figma AI, Canva Magic Studio, Adobe Firefly Boards |
The most interesting category is A. These are the "Lovable equivalents" for designers. You describe what you want, an infinite canvas materializes multiple connected screens, and you iterate through natural language or manual edits. The best ones can then export to Figma with full layers, or pipe into a coding agent via MCP.
3. 2. Category A: Vibe Design Canvases
These are the tools where a designer (or non-designer) types a prompt and gets an editable, multi-screen design. No code knowledge required. The quality gap between these tools is enormous right now.
Google Stitch
URL: stitch.withgoogle.com | Pricing: Free | Status: Closed source (Google Labs)
The most ambitious tool in this list. Google acquired Galileo AI in mid-2025 and relaunched it as Stitch inside Google Labs with a massive feature expansion. You describe an app in natural language (or speak it, or upload a sketch or screenshot) and it generates up to five interconnected screens at once on an infinite canvas. Click Play and you get an interactive prototype: the AI auto-generates logical next screens when you click through the flow.
The DESIGN.md concept is the genuinely novel idea here. It is an agent-friendly markdown file that exports your design system rules in a format that Claude Code, Cursor, and other coding agents can read directly. Import a URL, extract its visual style, turn it into a DESIGN.md, hand it to your dev agent. The pipeline from "describe it" to "it exists as a codebase" is shorter than anywhere else.
Powered by Gemini. Manual editing shipped in March 2026 (finally). Copy-paste to Figma works without a plugin. Free.
What makes it special: Multi-screen generation, voice input, DESIGN.md export, interactive prototype playback, and it is free. The Galileo AI acquisition was a big deal: Google bought the best pure AI design tool and then made it free.
Banani
URL: banani.co | Pricing: Free (20 credits/month) to unlimited Pro | Status: Closed source; Ukrainian-founded; Berlin-based; raised €850K pre-seed
"If Lovable and Figma had a baby." Prompts, reference images, PRDs, or Figma links as input. High-fidelity editable designs out. Multi-screen flows. Copy-paste to Figma. Export to HTML/CSS.
The Banani MCP server is the reason this tool is on my radar. It connects designs directly to Claude Code, Cursor, Windsurf, and Codex so the coding agent builds your screens directly in your stack. You design in Banani, the AI codes it. No screenshot-to-code translation needed. This is the most direct implementation of "vibe design to vibe coding" pipeline I have found.
Also does style extraction from any URL or screenshot, which is useful for "make it look like this app" prompts.
What makes it special: The MCP bridge to coding agents. Also the only tool in this list founded by a Ukrainian team (building in Berlin), which I find worth noting given the context of 2026.
Flowstep
URL: flowstep.ai | Pricing: Free; Professional $29/month | Status: Closed source; raised $2.6M seed June 2025
AI-native design platform for design teams. Generates complete user flows on an infinite canvas from a single prompt. The differentiator is the Figma integration: Cmd+C in Flowstep, Cmd+V in Figma, full layer fidelity, no plugin required. Just the native clipboard. This is the most frictionless Figma handoff I have seen anywhere.
Real-time collaboration. React/TypeScript/Tailwind code export. Design system management. $2.6M seed is small but enough to ship fast; they launched officially in June 2025 off the back of that raise.
What makes it special: The clipboard-based Figma integration. No plugin friction. Full layer preservation. If your team lives in Figma, Flowstep is the lowest-friction design canvas to add to the workflow.
Moonchild AI
URL: moonchild.ai | Status: Closed source; 2024
"Chat for product design." The story here is design system rigor. You build or import a complete design system (from Figma, Storybook, or code) and every generated screen is bound to your actual components, tokens, and typography. Two generation modes: Gold (polished, precise) and Silver (fast, rough). Full design system editor with version control.
Export to Figma, Claude Code, Lovable, Cursor, or Bolt. Export as PNG, SVG, or HTML. Multi-screen flow generation with interactive preview.
What makes it special: If you care about design system consistency at scale, this is the most disciplined tool in the list. Every generated screen is provably on-brand because it is built from your actual component library.
UX Pilot
URL: uxpilot.ai | Pricing: From $19/month | Status: Closed source; 2023
AI platform trained specifically on UX/UI design. Understands complex multi-layered prompts well. Generates connected screen flows (not just single screens). Has a Figma plugin. The standout feature is predictive heatmaps: after you generate a design, UX Pilot shows you where users will look. This feature exists nowhere else in this space. It bridges the gap between design tool and UX research tool.
What makes it special: Predictive heatmaps. Useful for anyone doing user research alongside design work, or for agencies who need to justify design decisions to clients.
Visily
URL: visily.ai | Pricing: Free; Pro ~$14/editor/month | Status: Closed source; 2021
Built specifically for non-designers: PMs, founders, marketers. The richest multi-modal input of any tool here. Screenshot, text, sketch, or diagram as input. Magic Theme applies a visual style across all screens at once. Design Assistant works via chat; you can mix screenshots, text, and canvas selections in a single prompt. Figma plugin for export.
1,500+ templates. The most polished onboarding for non-designers.
What makes it special: The combined input modes (screenshot + sketch + text + diagram in one workflow) and the explicit "no design skills required" positioning. Best tool for early-stage founders who want to prototype before hiring a designer.
Uizard
URL: uizard.io | Pricing: Free (2 projects, 3 AI gen/month); Pro from $12/seat/month | Status: Closed source; ~2019 (OG in this space)
One of the original AI design tools. Pioneered sketch-to-wireframe: you draw on paper, take a photo, it digitizes it into an editable design. The Autodesigner 2.0 generates full multi-screen UI flows from text prompts. Also does screenshot-to-design (upload any app screenshot, get an editable version).
The sketch digitization feature is still unique five years later. No other tool does it as reliably. Best for workshop settings where you want to whiteboard with a client and immediately have an editable digital version.
What makes it special: The OG. Sketch-to-design from a paper drawing. Still the go-to for design workshops and client sessions where you need to capture physical whiteboard output.
UXCanvas.ai
URL: uxcanvas.ai | Pricing: Free trial; from $49/month | Status: Closed source; 2024/2025
"AI UI/UX Design Agent." Natural language to UI design with chat-based iteration. One-click Figma export with full layers. React/TypeScript/Tailwind code export. AI image and asset generation inline. Shareable collaboration URLs.
Positioned at the premium end ($49/month) with a fine-tuned frontier model. Less well-known than the others but the code export quality is reportedly strong.
4. 3. Category B: Design-to-Code Hybrids
These tools blur the line between design canvas and code generator. You get both a visual output and working frontend code. Developers love them. Designers use them too but the output is optimized for shipping, not for Figma handoff.
Figma Make
URL: figma.com/make | Pricing: Included in Figma paid plans (AI credits) | Status: Closed source; GA July 24, 2025
Figma's own AI app builder. Generates interactive, high-fidelity prototypes from text prompts or directly from existing Figma Design files. Powered by Claude 3.7. Supabase backend integration. Publishes directly from the canvas.
The native Figma integration is the whole point. If your team already has a Figma component library and uses variables and tokens, Make understands all of it. The output respects your design system in a way no third-party tool can match.
Professional plan gets 3,000 AI credits/month. Separate AI credit subscription available since March 2026.
What makes it special: The only AI that fully understands your existing Figma files, components, variables, and tokens. Powered by Claude. If you are already a Figma team, this is the most natural upgrade path.
Magic Patterns
URL: magicpatterns.com | Pricing: ~$20/month; free tier | Status: Closed source; 2023
AI prototyping for product teams that care about brand fidelity. Upload your design tokens (colors, spacing, typography) and every generated screen uses them. Chrome extension to capture UI from any website. Export to layered Figma or production-ready React/Tailwind/Vue code. Real-time multiplayer. Interactive prototypes. MCP server available.
Loved by developers and PMs as much as designers. The design system awareness means the output looks like your actual product, not generic AI-generated UI.
What makes it special: Design system upload for brand-consistent generation. Best for established product teams that have an existing visual language and want AI to extend it, not replace it.
v0 by Vercel
URL: v0.dev | Pricing: Free tier; Pro plans | Status: Closed source; launched September 2023
Text to React/shadcn/ui components. More code-first than canvas, but designers use it heavily for rapid UI exploration. The standout feature is image-to-code: photograph a whiteboard sketch, get working React. Also does Figma design token import for brand-matching output.
Deep Vercel and Next.js integration. Not a design canvas in the traditional sense but too important to omit from this list.
What makes it special: Image-to-code. If you can whiteboard it and photograph it, v0 can build it. The only major tool with this feature. Essential for Vercel/Next.js teams.
5. 4. Category C: Website-Specific Builders
Narrow focus on marketing sites and landing pages, not app UI. Much deeper in their specific domain than general-purpose design tools.
Relume
URL: relume.io | Pricing: Free (30 components); Starter $38/month; Pro $48/month | Status: Closed source; 2022
AI website builder specifically for Webflow and Figma designers. Describe a website, it generates a full sitemap, then one-click converts it into wireframes using a 1,000+ component library (all with mobile variants). React component library also available.
The sitemap-first approach before wireframing is a workflow no other tool has. Agencies use it to go from client brief to full site architecture in under 30 minutes.
What makes it special: Sitemap generation plus sitemap-to-wireframe in one workflow. Category leader for website/marketing-focused design. Not useful for product/app UI.
Framer AI
URL: framer.com | Pricing: Basic $10/month; Pro $30/month; Scale $100/month | Status: Closed source; AI features since 2023
Framer evolved from a code-based prototyping tool to a full design-to-publish platform. AI page generation from prompts. Wireframer generates structural layouts. Workshop builds components (cookie banners, tabs, effects) without code. AI translation for multi-locale sites. And crucially: you can publish to a live production website directly from the canvas.
The only tool in this list that collapses the design-to-live-website pipeline entirely. Draw it, publish it. No handoff, no developer needed (for marketing sites).
What makes it special: Design to live production site with no handoff step. If you are building a marketing website, Framer eliminates the entire frontend development phase.
6. 5. Category D: AI on Incumbents
Existing platforms with hundreds of millions of users adding AI design features. Not as interesting for new builders but impossible to ignore given the user scale.
Figma AI
URL: figma.com/ai | Status: Included in Figma plans
Native AI inside Figma: generate designs from text inside existing files, FigJam AI boards and diagrams, AI wireframe generator (words to interactive layouts), auto-rename layers, generate copy, remove backgrounds. First-party, no plugin needed. Figma AI also powers Figma Make (Claude 3.7 under the hood).
Canva Magic Studio
URL: canva.com | Status: Closed source; 200M+ users
AI image generation (now powered by Leonardo/Phoenix models post-acquisition of Leonardo.ai in 2024; photorealistic quality competitive with Midjourney), Magic Write for text, Brand Kit with Brand Guardrails, Grow Insights for content performance analysis. Infinite canvas via Canva Whiteboards.
Not a serious UI/UX design tool. Used by designers for quick mockups and communication design. The Leonardo acquisition raised image quality significantly.
Adobe Firefly Boards
URL: firefly.adobe.com | Status: Closed source
Adobe's AI image generation trained exclusively on Adobe Stock (commercial safety guarantee). Firefly Boards is an infinite canvas for team ideation launched in 2025. Deep integration into Photoshop, Illustrator, InDesign, XD. Style Kits: fine-tune to your brand by uploading 20 images.
The commercial safety story is the differentiator. Enterprise legal teams with IP concerns use Adobe over every other AI image tool because the training data is licensed.
7. 6. Open Source Alternatives
The open source design tooling landscape is more interesting than most people realize. Penpot alone has more GitHub stars than most commercial competitors have users.
Penpot
GitHub: github.com/penpot/penpot | Stars: ~45,000 | License: AGPL-3.0 | Stack: Clojure backend, ClojureScript frontend, SVG-based
The leading open-source Figma alternative. Self-hostable. Built on open web standards: every element is real SVG. Used by enterprises for data sovereignty, universities, and open source projects.
The AI story is developing fast. In December 2025 they shipped an MCP Server:
AI agents can analyze design systems, extract style rules, refactor naming, generate semantic HTML/CSS,
and build Storybook setups. A separate penai research repo at
github.com/penpot/penai
experiments with diffusion models and LLMs for design generation and vector shape variations.
The co-pilot concept: describe changes in text, AI edits SVG elements. Not production-ready yet but the architecture (open SVG, self-hosted, MCP) is the best foundation for a truly open AI design tool.
Why it matters: 45K stars. Real enterprise adoption. SVG-first means every element is a standard web asset. The MCP server means AI coding agents can directly manipulate your designs. If you want to build AI design tooling, Penpot is the platform to build on.
OpenPencil
GitHub: github.com/open-pencil/open-pencil | Stars: ~3,841 | License: MIT | Stack: TypeScript; ~7MB desktop app (also on Homebrew)
AI-native design editor and open source Figma alternative. The remarkable feature: reads .fig files natively with round-trip fidelity. You can open Figma files without Figma. Full copy-paste between Figma and OpenPencil. 90+ tools covering shapes, fills, auto-layout, components, variables, boolean ops, design token analysis.
Press Cmd+J to open the AI assistant. Bring your own API key (OpenRouter, Anthropic, OpenAI, Google AI, more). No backend, no account needed. P2P collaboration via WebRTC. MCP server for Claude Code, Cursor, and Windsurf.
Local-first, privacy-respecting, no cloud required for core features. The MCP server slots it directly into any AI coding workflow.
Why it matters: .fig file compatibility without Figma is remarkable. Truly local, bring-your-own-key, privacy-first. The MCP server means Claude Code can manipulate your designs directly. Best choice for developers who want a design tool that integrates with their coding agent.
Jaaz
GitHub: github.com/11cafe/jaaz | Stars: ~6,054 | License: Open source | Stack: TypeScript/React
"World's first open-source multimodal creative assistant." A Canva and Manus alternative that runs locally. You paint directly on canvas, point with arrows, and the AI generates results in context. A LEGO-style "build by placing elements" approach. Image generation integrated directly into the design canvas. Privacy-first, no cloud required.
6K+ stars and growing fast. Highest-starred open source tool specifically in the AI canvas space. Closest to "Canva but open source and AI-native."
Why it matters: The growth curve is steep. Local image generation directly on canvas is a genuinely different interaction model from anything in the commercial space. Watch this one.
8. 7. Graveyard: Acquired and Defunct
| Tool | What happened | When |
|---|---|---|
| Galileo AI | Acquired by Google; relaunched as Google Stitch inside Google Labs | Mid-2025 |
| Diagram / Genius | Acquired by Figma; team and IP now power Figma AI (Figma Make, etc.) | June 2023 |
| Motiff | Discontinued the design editor; pivoted to a Figma plugin | August 2025 |
The pattern is clear: the best independent AI design tools get acquired by the platforms (Google took Galileo, Figma took Diagram). The ones that survive independently are either deeply specialized (Relume for Webflow/Figma agencies), infrastructure plays (Penpot, OpenPencil), or fast-growing enough to build defensible distribution before the platforms catch up.
9. 8. Full Comparison Matrix
| Tool | Category | Free tier | Multi-screen flows | Figma export | Code export | MCP support | Open source | Notable |
|---|---|---|---|---|---|---|---|---|
| Google Stitch | A | Yes (free) | Yes | Copy-paste | HTML/CSS | DESIGN.md export | No | Voice input; interactive prototype; DESIGN.md |
| Banani | A | 20 credits/mo | Yes | Copy-paste | HTML/CSS | Yes (MCP server) | No | Best MCP-to-coding-agent pipeline |
| Flowstep | A | Yes | Yes | Clipboard (no plugin) | React/Tailwind | No | No | Most frictionless Figma handoff |
| Moonchild | A | ? | Yes | Yes | HTML | No | No | Design system-bound generation |
| UXPilot | A | No ($19/mo) | Yes | Plugin | Yes | No | No | Predictive heatmaps |
| Visily | A | Yes | Yes | Plugin | No | No | No | Best multi-modal input; non-designer friendly |
| Uizard | A | Yes (limited) | Yes | Yes | No | No | No | Sketch-to-design from paper drawings |
| UXCanvas.ai | A | Trial only | Yes | Yes (layers) | React/Tailwind | No | No | Premium positioning ($49/mo) |
| Figma Make | B | Limited (credits) | Yes | Native (it IS Figma) | Yes | No | No | Claude 3.7; native Figma file awareness |
| Magic Patterns | B | Yes | Yes | Layered | React/Vue/Tailwind | Yes (MCP server) | No | Design token upload for brand fidelity |
| v0 by Vercel | B | Yes | No | Token import | React/shadcn | No | No | Image-to-code (photo a sketch) |
| Relume | C | Yes (limited) | Sitemap flows | Yes | React | No | No | Sitemap-first workflow; Webflow specialist |
| Framer AI | C | Design only (free) | Pages only | No | Publishes live site | No | No | Design to live production website |
| Figma AI | D | Limited | Yes | Native | No | No | No | Native inside Figma; no plugin needed |
| Canva Magic Studio | D | Yes | No | No | No | No | No | 200M users; Leonardo AI image quality |
| Adobe Firefly Boards | D | Limited | No | Via CC | No | No | No | Commercially safe training data |
| Penpot | Open source | Self-hosted | Yes | N/A (is Figma alt) | HTML/CSS | Yes (MCP server, Dec 2025) | Yes (AGPL-3.0) | 45K stars; SVG-based; best Figma alternative |
| OpenPencil | Open source | Self-hosted (Homebrew) | Yes | Reads .fig natively | No | Yes (MCP server) | Yes (MIT) | Local-first; .fig compatible; BYOK AI |
| Jaaz | Open source | Self-hosted | No | No | No | No | Yes | 6K stars; multimodal canvas; local image gen |
10. 9. What Is Still Missing
Despite 20+ tools in this space, there are real gaps. These are the things I could not find a good solution for after this research.
Real design system sync (bi-directional)
Most tools claim "design system support" but mean "import a JSON of tokens." Real bi-directional sync between an AI canvas and a live Storybook or shadcn component library does not exist at scale. Moonchild is closest. Penpot with MCP is the open-source path toward this. But nothing nails it yet.
Mobile-first AI canvas
Every tool in this list is desktop-first. The interaction model for AI-assisted design on iPad (touch, Apple Pencil, voice) is completely unexplored. Uizard's sketch digitization hints at it but nobody has shipped a real mobile-native AI design canvas.
Accessibility-aware generation
None of these tools generate designs that are provably accessible (WCAG 2.1 AA by default, correct ARIA roles, sufficient contrast, keyboard navigable). A tool that ships accessible designs without an accessibility audit step would be genuinely differentiated. UXPilot is closest with heatmaps but it is not the same thing.
Figma plugin that does everything
There is no single Figma plugin that combines multi-screen AI generation, design system awareness, and MCP export in one. After Diagram's acquisition by Figma, the plugin ecosystem for AI is surprisingly thin. Motiff tried to go this direction before pivoting. An indie builder could own this niche.
Open source "Lovable for design" with full MCP
Penpot and OpenPencil have MCP servers. Jaaz is growing fast. But there is no open source tool that combines: multi-screen AI generation from prompts + design system awareness + MCP export to coding agents + self-hostable. This is the most obvious gap in the open source space. The architecture exists (Penpot's SVG base is perfect); the AI generation layer is what is missing.