Four key rooms, twelve vibes. Each variant is an AI restyle of the actual listing photo that keeps the architecture and only changes furniture, rugs, art, lighting, plants & decor — sized for a household with a 2- and 4-year-old. Pick a profile, go one image at a time: 💚 / 😐 / 🚫, tap why, jot a note. The page summarizes you as you go; export the JSON and hand it to Claude to learn your taste, generate a refined batch, and (once you've both rated) find your overlap.
Tip: use ← → (or the big Next button) to move through images, and Y / M / N to rate the one you're on. Anything not generated yet shows its prompt — re-run GEMINI_API_KEY=… python3 StyleStudy/generate.py to add more.
How this works — the feedback loop
Pick your profile (Earl / Natasha) — top-left. Ratings are kept separately for each of you, in this browser.
Go one at a time: the big reviewer shows a single room+vibe. Hit 💚 Yes / 😐 Meh / 🚫 No (or Y/M/N), tap the chips for why, add a note. Then → or Next. (With "auto-advance" on, a verdict moves you along automatically — turn it off if you want to dwell.)
Compare: the ⇄ before button on the image swaps to the current staged listing photo, so you can sanity-check "is this actually better?". Click any image to see it full-screen.
★ My summary ranks the 12 vibes by your votes, shows your recurring likes/dislikes, and gives a rough read of your taste.
Export ("⬇ Export JSON" / "📋 Copy for Claude") and send it to Claude — Claude reads the pattern, tells you what you seem to like, and generates a sharper next round.
Natasha does the same on her profile (same page, "Natasha" tab — or send her this folder and have her import yours).
Once you've both rated, open "⇄ Both of us" — what you agree on, plus the short list where you'll need to negotiate. Claude can take both JSONs and propose the intersection set + next batch. Repeat → converge → expand the winner across every room.
Keys: ←→ move · Y/M/N rate · B before · F full-screen · U next unrated