The Sameness Behind Unappetizing AI Menus

According to a TechCrunch report by Amanda Silberling, generative artificial intelligence menus are entering the restaurant industry. These systems are trained on narrow aesthetic datasets, producing images that look oddly smooth, symmetrical, or bizarrely shaped. Experts like Reality Defender CTO Alex Lisle explained that large language models and diffusion models identify patterns from existing commercial corpuses, leading to convergence where outputs mimic and reinforce popular fast-food styles.

Furthermore, when users repeatedly edit images generated by artificial intelligence systems, the visuals tend to become increasingly unnatural and smooth. Lee Rainie from Elon University noted that optimization prioritizes pleasingness, shaving off edges and causing homogenization.

The phenomenon highlights broader systemic issues with artificial intelligence content generation, including convergence and data homogenization. Researchers at the University of Duisburg-Essen found that these computer-generated food images trigger an uncanny valley effect, eliciting disgust and unease. This technological shift also extends far beyond dining tables, challenging traditional notions of visual evidence in societal systems where seeing and hearing were once standard proof.

Alex Lisle noted that models draw from older commercial corpuses, while Lee Rainie explained that optimization for pleasingness leads directly to homogenization.

As artificial intelligence imagery becomes more widespread in commercial advertising, consumer backlash underscores the limits of automated design and our psychological aversion to synthetic visuals.