Space Reimagined

Inspiration

Moving into a new place or redecorating? It's hard to visualize how a space could look in a different style. I wanted a tool where anyone could upload a floorplan or room photo and instantly see it transformed—without hiring a designer or learning 3D software.

What it does

Space Reimagined lets you upload any floorplan or room photo and choose from 8 interior styles (Scandinavian, Mid-Century Modern, Industrial Loft, Minimalist Japanese, Contemporary Luxury, Coastal/Hamptons, Art Deco, Rustic Farmhouse). In seconds, you get a photorealistic render showing your space in that style.

How I built it

Frontend: Next.js with TypeScript and Ant Design, deployed on Vercel
Backend: FastAPI (Python) deployed on Railway
AI: Google Gemini's new image generation model (gemini-2.0-flash)

The key technical approach is a two-stage pipeline:

Spatial analysis – Gemini first analyzes the floorplan to extract room dimensions, wall positions, and layout details
Style-aware rendering – A second call generates the photorealistic render using both the original image and the spatial analysis, ensuring architectural accuracy

Challenges I faced

Getting the AI to respect the original layout instead of hallucinating new room configurations
Balancing prompt specificity—too vague and the style doesn't come through, too detailed and it ignores the source image
Deploying a Python backend that initializes the Gemini client correctly with environment variables

What I learned

Gemini's image generation capabilities are impressive but require careful prompting. The two-stage approach (analyze first, then render) significantly improved output quality compared to a single prompt.