What it does Upload a photo of your room. The app ships it to a ControlNet model hosted on Replicate, then shows you a restyled version. Image storage is handled by Bytescale; rate limiting is optional via Upstash Redis. That’s the whole trick.

The interesting bit This is explicitly the previous open-source release of RoomGPT.io, deliberately gutted of auth, payments, and extra features. The author treats it as a deployable teaching aid rather than a product — a rare honest framing in the “open core” era.

Key highlights

• One-click Vercel deploy with just a REPLICATE_API_KEY
• ControlNet preserves room structure while changing style (the actual ML heavy lifting lives elsewhere)
• Optional Redis-based rate limiting if you’re worried about API costs
• MIT licensed, TypeScript/Next.js stack
• 10.6k stars suggest it hit a nerve as a “first AI app” tutorial

Caveats

• You need paid third-party accounts (Replicate, Bytescale) to run it; the “free” repo is not free to operate
• The README is thin on error handling, testing, or model customization
• No mention of image privacy or retention policies for uploaded photos

Verdict Grab this if you want a minimal, working example of wiring a Next.js frontend to a Replicate model. Skip it if you need a production interior-design tool — this is scaffolding with a cool demo, not a product.