Hacker News Reader

Comment by SamBam | Hacker News Reader

That last image is bizarre. The quiche, cream and even the salad look like they've been given the trypophobia treatment.

Which might even make sense, because there were always (still are?) those horrible ads in the chumbox area of news sites that used trypophobia and other creepy body-horror stuff to get you to click. [1] So maybe the hope is that you don't really look closely at the quiche, but some reptilian party of the brain gets oddly activated and drives you towards the restaurant?

1. https://medium.com/the-awl/a-complete-taxonomy-of-internet-c...

View on HN

The food “photography” I’ve noticed in our local area - and many have started putting up these AI images - all have a weird distribution of shapes to them, a strangely uniform rhythm of same-sized features with almost blue-noise spacing. Every texture looks unnatural in the shapes it presents as, similar to this picture.

This is a quirk of the last gpt image model (gpt-image-2). It put this sort of high frequency noise on all of the image especially if it's in a "drawn" style. There is often lots of other tells that this model in particular generated it.

Image models somewhat watermarking the image in a way that's very easily identifiable by a human seems present in all the image models of the big labs, since DALL-E 3 on OpenAI's side and the first nano banana on Google's side. I have no idea what they did to reach this and why they don't try to fix it.

Just use Ideogram. Locally.

It doesn't look like a quiche but more like a cake to me, and the top would be torched meringue, not mold. Although it's probably some weird ai mix of quiche and cake.

Even with the label absent - you wouldn't serve cake with "tomatoes" and "salad", so that would be another hint of something being horribly wrong

It was labeled as a quiche, I just cut the photo in an unfortunate way :-)

There's a similar "horror burrito" circulating on bluesky and my first thought was "now we know how to produce images for the https://en.wikipedia.org/wiki/Long-term_nuclear_waste_warnin... "

(Actually that was my second thought; my first was "just how much H.R. Giger is in the training data?")

Two week old cake can pass as a quiche i guess!

> probably some weird ai mix of quiche and cake

Quike? Cache?

Cache sounds like a decent meal

Well, it's hard to invalidate.

cAIche

I never understand why I can get better results, with less thinking about it, than some of the Tier1 companies.. Modern diffusion models can easily render the image onto a white background, center it, crop it, relight it, but keep the same item, at no cost, or low cost.

example: https://ordermerchants.com/static/img/clearshot/product-1.jp...

These kinds of images are disingenuous. Modern image generating models have way higher quality of output, and in most cases you wouldn't even know it is generated.

That's not what we're exposed to, though. I've seen AI-generated postcards, t-shirt decals, logos and wall murals in shops and restaurants. They don't use the "good" models, because why would they? If they AI-generated it, it's because they didn't care about the quality anyway.

> If they AI-generated it, it's because they didn't care about the quality anyway.

No? There are far far more AI generated images you see daily, than the awful stuff from 2024 that you immediately discern as AI-generated. You could as well pass by somethimg like this (https://ideogram.ai/g/d6YzZ5XQQ56TU3Tc8eZw6w/0) believing it is a genuine photo or human made art.

It made me recoil honestly!

Reply on news.ycombinator.com