How this works. The photo goes to Claude Sonnet vision which extracts every dish, price, and section into structured data with a confidence score per dish. Then each dish is rendered as an AI Preview via OpenAI gpt-image-1 in the cuisine's plating style. Items with low parse confidence are flagged Needs Review. In production, the restaurant approves each image before it goes public. What you see here is the AI Preview pass.

Cost per scan. One vision call (about $0.01) plus one image gen per dish (about $0.04). A 15-dish menu is about $0.60. Images cache per restaurant. Every diner after the first sees them instantly at zero marginal cost.