FloorGen AI
Natural-language to floor-plan generation with CAD-ready vector exports.
Text-to-CAD pipeline using fine-tuned FLUX.1-dev with raster-to-vector export to DXF, JSON, and annotated PNG.
Stack
6 technologies
Gallery
6 screenshots
Source
Open on GitHub
Problem
Architects and designers spend significant time translating layout requirements into draft floor plans, and manual tracing from generated images slows handoff into CAD tools.
Solution
Built a two-stage pipeline: structured prompt engineering feeds a fine-tuned FLUX.1-dev model for layout generation, then a custom OpenCV raster-to-vector module exports layered DXF geometry, JSON room metadata, and annotated PNGs.
Highlights
- Natural-language prompts to floor-plan layouts
- Fine-tuned FLUX.1-dev on 80,000+ professional floor plans
- Raster-to-vector conversion with DXF, JSON, and PNG exports
Outcomes
- Prompt-to-CAD workflow with layered DXF and structured JSON metadata
- Exports compatible with AutoCAD, Revit, Blender, and Unreal Engine
- Gradio demo for interactive layout generation and export
Repository
FloorGen AI repository
Turn natural language into production-ready floor plans with fine-tuned FLUX.1-dev and full raster-to-vector automation.
Training data
80,000+ professional floor plans
Model
FLUX.1-dev (LoRA fine-tuned)
Exports
DXF, JSON, annotated PNG
CAD targets
AutoCAD, Revit, Blender, Unreal Engine
- Natural language to floor plan (e.g., 3BHK with L-shaped kitchen and balcony)
- Automatic raster-to-vector conversion from PNG to clean DXF
- Layered DXF output with walls, doors, furniture, and text
- JSON schema with rooms, areas, labels, and dimensions
- LLM-powered prompt structuring layer
- Gradio demo via python app.py
Stack