Why do AI-generated menu images look so perfect and symmetrical?

AI models are trained on vast amounts of data and optimize for 'pleasingness' and avoiding being offensive. This leads to a homogenized style where food looks too round, smooth, and flawless.

What is 'model collapse' and how is it different from 'convergence'?

Model collapse is when AI trains too much on its own output, leading to degradation and collapse, like 'mad cow disease.' Convergence is less extreme, degrading output quality without making it useless.

Is there scientific evidence that people dislike AI-generated food images?

Yes, researchers at the University of Duisburg-Essen found that AI-generated food images exhibit an 'uncanny valley' effect, where nearly real images cause more disgust than obviously fake ones.

Image GenerationTechCrunch AI8h agoSep 4, 2026, 16:00 JST2 min read

AI-generated restaurant menus look off, and science explains why

Send on LINE

Key takeaway

AI-generated menus look strange because models converge on a narrow, overly polished aesthetic. This 'convergence' can degrade output quality.
It stems from training on similar styles and can worsen with each AI edit.
Researchers found almost-real food images can disgust people.

3 Key Points

What happened
AI-generated menus in restaurants often feature food that looks too perfect, symmetrical, and smooth, triggering unease in customers. This stems from AI models trained on a narrow, 'pleasing' aesthetic.
Why it matters
Researchers at the University of Duisburg-Essen found these images create an 'uncanny valley' effect, where almost-real food looks more disgusting than obviously fake images. The homogenization happens because models optimize for 'pleasingness' and 'not being offensive,' which 'shaves off the edges.'
What to watch
Restaurants may reconsider using AI menus due to customer backlash. However, the issue extends beyond food, affecting trust in visual evidence, as Reality Defender CTO Alex Lisle notes: 'The world has fundamentally shifted, for good or for ill.'

Ask the AI about this article →

Summaries like this, in your inbox every morning.

Get news like this every morning — free

Context & Analysis

The phenomenon of AI-generated menus looking 'off' is rooted in how models are trained and the cultural context of AI. Lisle explains that models are trained on a corpus of work, like menus from popular chains, which already share a similar style. When AI generates new menus, it mimics that style, and if those AI outputs end up back in training data, the homogenization is reinforced. This is a form of 'convergence,' distinct from 'model collapse,' but still degrading output quality. The 'uncanny valley' effect documented by German researchers explains the visceral disgust people feel — a reaction that may drive restaurants to abandon AI menus. However, the implications go beyond food: as Lisle points out, with AI-generated images becoming more prevalent, even court evidence like 'taped confessions and videotaped evidence' may no longer be trusted, marking a fundamental shift in how we perceive reality.

FAQ

Why do AI-generated menu images look so perfect and symmetrical?: AI models are trained on vast amounts of data and optimize for 'pleasingness' and avoiding being offensive. This leads to a homogenized style where food looks too round, smooth, and flawless.
What is 'model collapse' and how is it different from 'convergence'?: Model collapse is when AI trains too much on its own output, leading to degradation and collapse, like 'mad cow disease.' Convergence is less extreme, degrading output quality without making it useless.
Is there scientific evidence that people dislike AI-generated food images?: Yes, researchers at the University of Duisburg-Essen found that AI-generated food images exhibit an 'uncanny valley' effect, where nearly real images cause more disgust than obviously fake ones.

TechCrunch AIRead Original Article

Send on LINE

Get the latest Image Generation news every morning

For example, today's edition would include:

Why AI images feel "creepy": a PR expert explainsTop Companies AI · 18h ago
Jasper Research releases text-to-image build cookbookr/MachineLearning · 23h ago
AWS team cuts dashboard failure detection from 72h to 1hAmazon AI Blog · 1d ago

AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.

Subscribe free →

Free · takes 30 seconds · unsubscribe anytimeWhat is AIToday? →