floormind-sd-floorplans

FloorMind SD v2 — Architectural Floor Plan Generator

Fine-tuned Stable Diffusion v1-5 on 5,000 real floor plans from the CubiCasa5K dataset.

Model Description

• Base model: runwayml/stable-diffusion-v1-5
• Task: Text-to-image (floor plan generation)
• Best val loss: 0.0522
• Epochs: 10
• Augmentation: rotation ±5° only (no color jitter)
• Resolution: 512×512

Usage

from diffusers import StableDiffusionPipeline
import torch

pipe = StableDiffusionPipeline.from_pretrained(
    "Youssefx64/floormind-sd-floorplans",
    torch_dtype=torch.float16,
    safety_checker=None,
).to("cuda")

image = pipe(
    "floor plan 2 bedrooms living room kitchen bathroom",
    num_inference_steps=30,
    guidance_scale=7.5,
).images[0]

image.save("floor_plan.png")

Training Data

• Dataset: CubiCasa5K (5,000 architectural floor plans)
• Split: 80% train / 10% val / 10% test
• Languages: English & Arabic prompts

Part of CadArena

This model powers CadArena — an AI-powered architectural floor plan generator.