SnackSnap
Delivery-app food photos that look appetising, from the snap you already took.
The question I was chasing
Does the approach behind RoomSnap (multi-provider AI imaging wrapped in a real, paid product) travel, if I point it at a completely different trade?
RoomSnap proved a solo build could carry commercial image-editing for estate agents. SnackSnap was the test of whether that playbook travels. Same problem shape, a phone photo that has to look professional. Different industry, and a different idea of what 'good' looks like.
Independent restaurants can't afford menu photography, so their dishes turn up on delivery apps as dim phone snaps that do the food no favours. The fix has to be effortless: upload a camera-roll photo, get a clean, appetising, menu-ready image, and get straight back to running the kitchen.
The constraints
Food is its own problem. A property photo run through the wrong model looks fake. A dish looks inedible, which is worse. So quality control couldn't be an afterthought, and the output had to be export-ready for the exact places it's used: social, and the major delivery apps. And like any product with money moving through it, the accounts, payments and limits had to genuinely work, not demo-work.
The decisions that mattered
Reuse RoomSnap's commercial playbook (Supabase for accounts, Stripe for payments) so the new bet was about the imaging, not the plumbing.
Route across several imaging providers instead of betting on one, because the best model for appetising food isn't the best for anything else, and food is unforgiving.
Offer a few clean style options rather than infinite knobs: a restaurant owner wants 'make this look good', not a parameter panel.
Built with: Codex, Claude, Gemini, multiple AI imaging APIs, Cursor, Supabase, Stripe
Where it landed
Restaurants get multi-provider imaging and real commercial scaffolding, live at snacksnap.ai. Two products still carry their own copy of the provider-routing logic. I have not consolidated them yet.
Part of the Rolling Waves work archive.