From our other sites
The story
AI-generated menu images, apparently used at the Toyota-Kamigo service area in Aichi Prefecture and similar locations, are causing controversy on X (formerly Twitter). Photos of tonkatsu (breaded pork cutlet) and karaage (Japanese fried chicken) came out unnaturally grotesque, prompting a wave of comments like “This kills my appetite.” The controversy began in mid-August, after which YouTubers explained why such images are generated and even released videos recreating the AI-warped dishes with real food. On 5ch, users flooded the thread with comparison images generated by Gemini, ChatGPT, and Grok, sparking a lively discussion pointing out unnatural details.
September 2, 2026
Menu images for restaurants, apparently created using generative AI, are causing controversy on X. Meanwhile, verification videos recreating these distinctly “AI-looking” images using real food have also drawn attention.
The controversy began in mid-August.
Source: pinzuba.news / Original article here
What people said
That reminds me, I was genuinely amazed by Sukiya's "frisbee bowl" lol (a viral incident where a Sukiya beef-bowl chain's promo photo looked nothing like the actual dish)
That's exactly the problem. The original katsu image is, frankly, an appetite-killer that backfires completely, and it's baffling that it got approved as-is. It's also a mystery why it's still being used.
In response to complaints of "this doesn't look like the picture!", turning the sample image itself into the "frisbee" was a genuinely novel solution.
Just like the article says —
if you take a generated food image and keep issuing instructions like "generate a restaurant sign with text added, then add a catchphrase, make the price bigger," deliberately avoiding touching the food itself, it apparently gets more and more unnatural each time.
Other than the tableware placement and the fact there are two cups of tea, it doesn't bother me.
At first glance it doesn't look too off, but there are two cups of tea, the mustard is way too pristine, what's that soy-sauce-looking thing and egg-looking thing behind the pork cutlet?, what's with the green pickles?, and the plate the pickles sit on is transparent and unnatural — plenty of things are off once you look closely.
Makes you wonder if the shops making them aren't even checking, or just haven't noticed.
Grok
Nothing that bad comes out, really.
What an honest, decent-looking diner.
If you look closely at either image, the cabbage is creepy.
What kind of prompt did you even use lol
Okay, surely this one's on purpose, right?
This is just gross at this point.
Even if this comes out, if you just keep refining it — "make it a bit more like this," "no, not that here" — an AI-generated image could turn out decent enough. So why did they use it as-is for a poster? Is whoever's in charge of PR a robot too?
I have trypophobia, so this is rough…
Guess it means everyone's thinking the same thing lol
If you went to that Yayoi Kusama exhibition — the artist who passed away recently — you'd probably collapse foaming at the mouth.
It's become this big a topic precisely because plenty of people are actually looking at it.
I mean, just like there are people who can't tell a creepy AI image from a real photo, there are also people like you who can't even recognize what's happening right in front of them.
Background and Key Points
The controversy began with images of karaage (fried chicken) and tonkatsu displayed on shop signage, reportedly used at the Toyota-Kamigo service area and other locations in Toyota City, Aichi Prefecture. According to the article, AI-generated food photos tend to become more visually distorted the more instructions are piled on top — adding text, catchphrases, enlarging the price, and so on. Opinions in the thread were split on whether this stemmed from simple lack of quality checking on the production side, or whether people critical of generative AI were deliberately creating and spreading ugly images. Many posts also compared images generated with the same composition by Gemini, ChatGPT, and Grok, showing that the degree of distortion varies by model. A point readers can easily misunderstand is that AI-generated images aren’t uniformly bad — they tend to degrade specifically through the process of piling on post-production instructions like adding text, rather than purely from a lack of image-generation capability itself.
*This article is excerpted and summarized from the 5ch (Entertainment & Sports News+) thread “‘It kills my appetite’ — AI-generated ‘extremely creepy menu images’ are surging at restaurants; YouTuber explains why food photos turn grotesque, recreation projects follow [Ailuropoda melanoleuca★].”
Leave a Reply