Skip to content

A hybrid AI workflow for modeling a house from a floorplan

Conceptual model only. Not a survey or construction document.

Reconstruct a floorplan, review it through MCP, export GLB, then use GPT-6 Astra through Blender Python or MCP for bounded scene work.

A floorplan-to-house workflow has two different problems: recovering spatial structure from a drawing and developing that structure into a useful scene. Use a floorplan system for the first, then use an AI agent inside the authoring tool for the second. The handoff between them is a file you can inspect.

Conceptual reference study: verify the modeled stack against the source plan before an agent adds scene detail.

1. Reconstruct and check the source

Upload the complete floorplan to INK TO BRICK. Compare the exterior outline, room names, walls and openings with the drawing. Check scale against a known dimension and verify critical measurements on site. Fixing an uncertain source after adding furniture and materials multiplies the cleanup.

2. Use MCP for project context

Connect Codex or another compatible MCP client with read access to https://mcp.inktobrick.com/mcp. Copy the project ID from its dashboard URL, then ask the client to read that workspace, summarize rooms, compare revisions or search the fixture catalog. The MCP server does not list projects. Self-serve Codex connections are read-only. Contact us to register an editing client that supports preview and confirmation.

Worth checkingINK TO BRICK MCP works with projects that already exist. It does not upload the source, start reconstruction, generate a photoreal image, export the scene or drive Blender. Keep those boundaries in the prompt so the agent does not describe an operation the server cannot perform.

3. Export for the work ahead

  • GLB for the simplest Blender handoff: one file with geometry, materials and textures.
  • Packaged glTF when you want separate, inspectable texture and buffer files.
  • DXF or DWG for a layered 2D CAD plan, with no claim that it recreates the original CAD file.
  • IFC4 for a semantic BIM reference on Business, without claiming native Revit fidelity.

4. Give Astra a narrow Blender brief

Give Astra access to Blender through the Blender Python API or a separate Blender MCP server, import the GLB into an empty scene and save a working copy. Ask for one bounded change at a time: rename collections, clean selected mesh topology, replace one material family, place cameras, or prepare lighting. Require the agent to describe which objects it changed and inspect the scene after each step.

Keep each system responsible for the part it can expose and verify.
StageSystemReview before continuing
Source reconstructionINK TO BRICK web or APIOutline, scale, rooms, walls and openings
Project questions and supported option previewsINK TO BRICK MCPTarget project, revision, preview and confirmation
Scene authoringAstra with Blender Python or MCPObject changes, dimensions, transforms, normals and materials
DeliveryBlender or the receiving CAD/BIM toolReopen the file and test it in the destination application

This hybrid keeps a reviewable floorplan-derived starting point while leaving bespoke modeling to the authoring tool. It also makes failure easier to locate: a room-topology error belongs before export, while a lighting or subdivision problem belongs in Blender. Read the cost and time comparison before treating model-token arithmetic as a project quote.