SpaceSolver

output image
imput image
any special requests
asking to input messy image
plans

Inspiration

A messy room can be overwhelming. Often, the hardest part of cleaning is simply knowing where to start. SpaceSolver was inspired by the need to break down the daunting task of room organization into manageable, actionable steps while providing a clear visual goal to stay motivated.

What it does

SpaceSolver is an AI-powered room organization assistant. Users upload a photo of their cluttered space. The app processes the image to generate a customized, step-by-step cleaning checklist based on the exact items visible in the room. Simultaneously, it generates a pristine "After" preview image that maintains the exact architectural layout and furniture placement of the original photo but renders it completely clean. Users can also add custom constraints (e.g., "Don't move the heavy desk").

How we built it

The application is built using a modern full-stack architecture: Frontend: Next.js (App Router) with React and TailwindCSS for a sleek, responsive dark-mode UI. Backend: Next.js Server Actions securely handle direct REST API requests without exposing credentials to the client. AI Vision & Logic: Google's gemini-3.8-flash model analyzes the user's uploaded Base64 image to identify clutter, generate the actionable checklist, and craft a highly specific architectural prompt.

Image Generation: Pollinations AI renders the "After" image based on Gemini's hyper-specific spatial descriptions.

Challenges we ran into

Building the backend infrastructure presented several hurdles. Initial iterations using Python FastAPI faced CORS blocks and deployment execution errors within the sandbox environment. I pivoted to Next.js Server Actions to ensure a stable, secure connection to the Gemini REST API. Another major challenge was prompt engineering the image generator; early attempts yielded completely random room layouts. This was solved by forcing Gemini to output a hyper-specific architectural blueprint (detailing lighting, camera angle, and exact furniture placement) to lock in the room's precise geometry.

Accomplishments that we're proud of

What we learned

We gained deep experience in prompt engineering for multimodal AI, specifically learning how to chain models together by using a vision model (Gemini) to act as a precise prompt-writer for an image generation model (Pollinations). I also strengthened my skills in handling Base64 image data parsing and structuring secure Server Actions in Next.js.

What's next for SpaceSolver

Future updates will include user authentication to save room histories across multiple days, a gamified point system for completing checklist items, and exportable cleaning schedules.