AI feature
AI Plan Recognition: Turn any floor plan into an editable 3D model
Upload a floor plan image, Claude Vision detects walls, doors, windows, and furniture, then drops them into Home 3D Plan as an editable 3D project. Bring your own Anthropic API key; ~$0.005 per scan.
Most home-design projects start with something already on paper: a PNG export from the architect, a real-estate listing photo, a quick sketch of "where the couch goes." Home 3D Plan’s AI Plan Recognition turns any of those into a live, editable 3D model in under a minute, no manual tracing, no specialist software, no learning curve.
It reads the geometry the way an architect would: walls become editable line segments, doors and windows snap to their host walls, room boundaries auto-close, and dimensions stay true to scale. You stay in control, every detection is editable, draggable, and easy to refine on the canvas.
Getting started
AI Plan Recognition lives on your projects dashboard. From `/projects`, click **New plan from image** (the secondary button next to *New plan*). Drop your floor-plan image, hit **Generate**, and Claude returns the parsed geometry as a preview. Click **Create project** and you land directly in the editor on the new 3D plan.
The first time you open the modal you'll be asked to sign in and to add your personal **Anthropic API key** on your profile. Once it's saved, every future recognition runs with one click. Your key is encrypted on our server with a per-user secret and only ever decrypted in memory for the duration of a single call.
How AI Plan Recognition works
Drop a PNG, JPEG, or WebP image onto the modal. We downscale very large uploads to the resolution Claude can actually read (≤ 2400 px on the long edge), then send the image to the Claude Vision API alongside a structured-output schema that forces the model to return a valid floor-plan JSON.
The model identifies room polygons, doors, windows, and furniture, normalises everything to metres (using any dimension annotations, a scale bar, or the door-width heuristic), and submits the result via a tool call so you never see free-form prose. The frontend converts that JSON into a working scene, rooms with walls, openings snapped to their host walls, furniture matched against our catalog, and seeds it into a fresh project named "AI - Untitled plan".
What you can upload
PNG, JPEG, and WebP images today. Architect-issued PDFs, multi-page PDFs, and high-res phone photos of printed plans are on the roadmap. For now, export your PDF to PNG/JPEG first.
Best results come from top-down orthographic plans with clear wall lines and visible dimensions or room labels. Isometric drawings, perspective renders, and very densely annotated plans still work but may need a few minutes of manual cleanup once the AI hands the geometry off to you.
Pricing and your API key
Home 3D Plan is free to use today. There's no subscription, no premium tier, no usage caps. The AI feature is one place where pure-zero is impossible: each recognition runs through Anthropic's API, which costs money. So we put that bill where it makes sense, directly on you, via your own Anthropic account.
Typical cost is **around $0.005 per recognition** (slightly more in *High* mode using Opus, slightly less in *Fast* mode using Sonnet). You add your key once on your profile; it's end-to-end encrypted on our server and used only when you trigger a scan. You can revoke it any time from the same screen, and at Anthropic, and you keep full control of your spend.
Don't have an Anthropic account yet? Sign up at console.anthropic.com, add a payment method, generate an API key, paste it on your profile. The modal walks you through it the first time.
Editable from the first second
AI Plan Recognition is a starting point, not a black box. Every wall the model finds is a draggable segment with the standard properties panel attached, change thickness, swap a wall for a hidden opening, retag a door as a window. Materials, furniture, and lighting layer on top exactly as they would on a from-scratch plan.
Once the geometry looks right, switch to 3D to walk through the space, snapshot a photo render (PNG), or share a public view link. The same project file works across every Home 3D Plan surface: desktop, mobile PWA, and the public share URL.
Who it's for
Renters and homeowners who want to redesign without learning CAD. Interior designers who get listing photos from clients and need to mock up a layout the same day. Contractors and architects who want a quick "as-built" digital twin from a paper drawing. Real-estate agents who want a 3D walkthrough from a listing JPEG.
Because Home 3D Plan is free and runs in a browser tab, and because the AI bill goes straight to your Anthropic account at cost, there's no licence to renew, no plug-in to install, and nothing stopping you from re-running the recogniser on a fresh upload tomorrow. Bring the plan, get a model, ship the design.
Frequently asked questions
Is Home 3D Plan free?
Yes. Home 3D Plan runs in your browser at no cost, with no sign-up requirement and no paywalled features. The PWA install is also free.
Do I need to install anything?
No. Home 3D Plan is a Progressive Web App, so you can open it in any modern browser and start designing. You can optionally install it to your home screen for offline access.
Can I export my plans?
Yes. Snapshot your lit 3D scene to an image (PNG) with one click, unlimited and with no watermark, and share a read-only or editable live link that opens in any browser without an account.
Do I need an Anthropic API key?
Yes. AI Plan Recognition runs on Claude Vision, billed per call to your own Anthropic account. Add your key once on your profile (Profile → AI features), then every future recognition uses it automatically. Recognitions cost about $0.005 each.
Where is my Anthropic API key stored?
Encrypted on our server with AES-256-GCM using a per-user secret derived from your account ID. The plaintext key never returns to your browser, the server decrypts it in memory for the duration of one recognition call. You can remove the key any time from Profile → AI features.
What file types can the AI read?
PNG, JPEG, and WebP images today. PDF and multi-page support is on the roadmap. For now, export your PDF to a single PNG or JPEG first. Hand-drawn sketches, phone photos of printed plans, and architect PNG exports all work.
High vs Fast: which model should I pick?
High runs Claude Opus, which is much better at complex / dense plans (L-shaped rooms, multiple openings, lots of furniture). Fast runs Claude Sonnet, about 3× quicker and cheaper, fine for clean architect exports. Default is High; toggle it on the recognition modal.
How accurate is AI Plan Recognition?
On clean architect exports the model typically captures every wall and opening correctly. On rougher inputs (phone photos, sketches) you usually get an 80-95% accurate starting point that takes a couple of minutes to refine, far faster than redrawing from scratch.
Does the AI set the right scale?
If your plan has a labelled dimension, scale bar, or visible door, the AI calibrates against it. Otherwise you can drag any wall in the editor afterwards and every other dimension follows proportionally.
Can I redraw or fix the AI output?
Yes. Everything is editable. Walls are draggable segments, doors and windows snap to walls, and the standard Properties panel applies. The AI just gives you a head start; the final plan is yours to shape.
Will it touch my existing 3D projects?
No. AI Plan Recognition always creates a brand-new project, it never edits or replaces something you've already drawn. The new project is named "AI - Untitled plan"; rename it like any other from the dashboard card menu.