SnackSnap
Delivery-app food photos that look appetising, from the snap you already took.
Keeping menus photographed
My family has roots in the restaurant industry, and I'd been following how hospitality was responding to AI photography and developments in image models.
One challenge I kept seeing was the cost of keeping a menu fully photographed. Big chains could update their menus regularly, while smaller independents struggled to keep up. I wanted to see whether I could help make better photography accessible without the expense of a shoot every time.
Keeping the actual dish
The AI-generated food images I was seeing often looked poor. They didn't give an accurate sense of the dish, and could end up being more off-putting than appetising.
With SnackSnap, I wanted to start from an everyday photo taken on a phone and make it look menu-ready without making it feel overly AI-processed. The dish itself needed to remain recognisable. Better photography wouldn't help if it misled the customer about what they were ordering.
Fine-tuning the images
Advances in Gemini's image-generation APIs made the idea possible. A huge amount of the work went into fine-tuning prompts so that the results reflected the original dishes without looking overly generated.
SnackSnap offers a few style options and exports sized for social media and delivery apps, so the images can be used in the places restaurants need them.
Making it more accessible
For me, the aim was to help level the playing field for independent restaurants: better photography for their business without breaking the bank or misleading customers. I wanted to explore how much of an expensive photography shoot could be replaced while still retaining the actual dish.
Built with: Codex, Claude, Gemini, multiple AI imaging APIs, Cursor, Supabase, Stripe
Part of the Rolling Waves work archive.