AI menus facing a sameness problem, struggling with unappetising uniformity
AI-generated menu images often look unappetisingly uniform, a problem dubbed "sameness" by industry experts. This occurs because large language models (LLMs) and diffusion models, which create images and text, are trained on vast datasets. When asked to generate menu items, these models identify patterns in their training data and tend to produce images that are "eerily flawless, precisely symmetrical, and oddly smooth".
Alex Lisle, CTO of Reality Defender, compares it to "an alien trying to make a pizza without understanding its core principles". The problem stems from the training data, which often contains images from established chains like McDonald's or Wendy's that already share a similar, idealised style. The AI then mimics and reinforces this style.
This convergence means AI-generated content can lack the imperfections and uniqueness that make real food appealing. There is a risk of "model collapse" if AI models train too much on their own AI-generated content, which can degrade the quality of outputs over time, although the current issue is described as convergence rather than collapse.
This matters because many in the events and hospitality industry are exploring AI for marketing and content creation. While AI offers efficiency, relying solely on its output for visuals risks creating a bland, unrealistic, or unappealing impression. For event organisers, caterers, and venues, authenticity and visual appeal are crucial. If you are using AI to generate marketing materials, particularly food imagery, be aware of this tendency towards uniformity. Human oversight and creative input remain vital to ensure your visuals genuinely tempt and engage your audience.