Why are my food delivery apps AI generating photos of food? (shub.club)

A food-delivery app (Forkable) has started using AI-generated photos for menu items — and customers are noticing. The synthetic images can be wildly inaccurate and unappetizing (a half Detroit-style pizza rendered as a whole, wrong crust and plating, garish “plastic” textures), which undermines trust and purchase intent. The writer argues that with a limited set of partner restaurants it would be cheaper and more effective to use real photos, and points out an additional UX bug: the app’s AI-driven auto-order feature sometimes selects side dishes by mistake. This is a useful real-world example of generative-model limits and product design tradeoffs. Technical issues include hallucination and domain mismatch from image-generation models (diffusion/GAN-style systems), poor conditioning on metadata (size, style, plating), and lack of human-in-the-loop validation. For the AI/ML community it underscores the importance of grounding generative outputs, improving conditional generation or retrieval-augmented approaches, and measuring fidelity against real-product images. Practical fixes: use actual photos where possible, fine-tune models on branded/restaurant-specific datasets, add strict metadata constraints, and prioritize robustness of the recommendation/auto-ordering system to avoid harming conversion and CX.