TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Show HN: This Food Does Not Exist

248 pointsby MasterScratalmost 3 years ago

32 comments

JadoJodoalmost 3 years ago
OP: Forgive me if this is out of place. Also, please know that my question is genuine, not at all a reflection on the author&#x2F;their project, and most certainly born out of my own ignorance:<p>Why are these kinds of things impressive?<p>I think part of my issue is that I don&#x27;t really &quot;get&quot; these ML projects (&quot;This X does not exist&quot; or perhaps ML in general).<p>My understanding is that, in layman&#x27;s terms, workers are shown many, many examples of X and then are asked to &quot;draw&quot;&#x2F;create X, which they then do. The corollary I can think of is if I were to draw over and over for a billion, billion years and each time a drawing &quot;failed&quot; to capture the essence of a prompt (as deemed by some outside entity), both my drawing, and my memory of it was erased. At the end of that time, my skill in drawing X would be amazing.<p>_If_ that understanding is correct, it would seem unimpressive? It&#x27;s not as though I can pass a prompt of &quot;cookie&quot; to an untrained generator and, it pops out a drawing of one. And likewise, any cookie &quot;drawing&quot; generated by a trained model is simply an amalgam of every example cookie.<p>What am I missing?
评论 #32169975 未加载
评论 #32174274 未加载
评论 #32170056 未加载
评论 #32178854 未加载
评论 #32176091 未加载
评论 #32170081 未加载
forgotusername6almost 3 years ago
I wonder how close the nearest match from the training data is. Was there a cheesecake that looked almost like these generated images?
评论 #32169513 未加载
评论 #32169649 未加载
评论 #32170041 未加载
jdthedisciplealmost 3 years ago
Looks impressive but I can&#x27;t escape the notion that surely some of the generated images will be very close to the some of the training images?<p>How am I to assess how original the generated results really are?
评论 #32174783 未加载
评论 #32173735 未加载
评论 #32173251 未加载
xg15almost 3 years ago
At least with DALL-E you can be sure the food has a name. For a moment I was worried this would produce vaguely food-like images where on closer look you realise you have no idea what you&#x27;re looking at - like a lot of other &quot;this X does not exist&quot; projects seem to do.<p>Also a bit of cultural bias in the training is shown I think. The &quot;pile of cookies&quot; prompt seems to mostly generate American cookies, while e.g. a German user might be disappointed they didn&#x27;t get this: <a href="https:&#x2F;&#x2F;groceryeshop.us&#x2F;image&#x2F;cache&#x2F;data&#x2F;new_image_2019&#x2F;ABSB0005XOJBS_0-600x600.jpg" rel="nofollow">https:&#x2F;&#x2F;groceryeshop.us&#x2F;image&#x2F;cache&#x2F;data&#x2F;new_image_2019&#x2F;ABSB...</a> :)
评论 #32175077 未加载
评论 #32171395 未加载
评论 #32170534 未加载
herpderperatoralmost 3 years ago
Is there a way to trigger a fresh image on demand? That&#x27;s kind of what I expect when I see a does-not-exist site.
评论 #32173244 未加载
notamyalmost 3 years ago
I was really hoping that this would be never-before-seen, AI-generated recipes or something similar ):
评论 #32170419 未加载
MasterScratalmost 3 years ago
We have trained four StyleGAN2 image generation models and are releasing checkpoints and training code. We are exploring how to improve&#x2F;scale up StyleGAN training, particularly when leveraging TPUs.<p>While everyone is excited about DALL·E&#x2F;diffusion models, training those is currently out of reach for most practitioners. Craiyon (formerly DALL·E mega) has been training for months on a huge TPU 256 machine. In comparison our models were each trained in less than 10h on a machine 32x smaller. StyleGAN models also still offer unrivaled photorealism when trained on narrow domains (eg thispersondoesnotexist.com), even though diffusion models are catching up due to massive cash investments in that direction.
评论 #32174243 未加载
评论 #32173524 未加载
gus_massaalmost 3 years ago
Are you using the same model for cookies and cheesecakes? Do you get sometimes a cookiecake?
评论 #32167788 未加载
评论 #32168383 未加载
andrewmcwattersalmost 3 years ago
Darn! I was hoping for other-worldly foods that don&#x27;t actually exist being generated from real food attributes. I suppose I should have known better.
derbOacalmost 3 years ago
The Cake is a Lie meme was never so relevant.
评论 #32171157 未加载
gffrdalmost 3 years ago
I like the thought that, years from now, we&#x27;re all drinking eating weirdly-presented food &#x2F; drinking weird cocktails because AI synthesized the images of drinks around the web and decided `cocktails always include fruit` and `all food must be piled high on plate`
Animatsalmost 3 years ago
Coming soon to the restaurant site generator of some large delivery service.<p>(&quot;Picture is only for illustration purposes&quot;)
croesalmost 3 years ago
You are killing Instagram influencers
评论 #32168914 未加载
评论 #32169401 未加载
评论 #32169392 未加载
munificentalmost 3 years ago
I love cheesecake with strawraspcherries on top.
wyldfirealmost 3 years ago
Are there any analysis techniques that can easily distinguish between these and real photographs? Do simple things like edge detections or histograms reveal any anomalies?
评论 #32170489 未加载
dylan604almost 3 years ago
When will we see this as a contestant on Is It Cake?
n4bz0ralmost 3 years ago
The food looks great! I suppose these models could use some extra training with dishes, though. The plates and glasses look wobbly, which is an instant giveaway. Otherwise, I can see this being used by food posters! Maybe not as a primary source, but as a &quot;filler&quot; — for sure.
abidlabsalmost 3 years ago
You can try out the model with this interactive Gradio demo: <a href="https:&#x2F;&#x2F;huggingface.co&#x2F;spaces&#x2F;nyx-ai&#x2F;stylegan2-flax-tpu" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;spaces&#x2F;nyx-ai&#x2F;stylegan2-flax-tpu</a>
hammycheesyalmost 3 years ago
I tried to use the linked Colab notebook to generate my own, and it appears to have been successful, but I don&#x27;t see any way to view the generated images via the notebook interface. I&#x27;m not familiar with the notebook tool - have I missed something?
评论 #32170885 未加载
评论 #32177831 未加载
fxtentaclealmost 3 years ago
I&#x27;m honestly surprised that they trained a StyleGAN. Recently, the Imagen architecture has been show to be both easier in structure, easier to train, and even faster to produce good results. Combined with the &quot;Elucidating&quot; paper by NVIDIA&#x27;s Tero Karras you can train a 256px Imagen* to tolerable quality within an hour on a RTX 3090.<p>Here&#x27;s a PyTorch implementation by the LAION people:<p><a href="https:&#x2F;&#x2F;github.com&#x2F;lucidrains&#x2F;imagen-pytorch" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;lucidrains&#x2F;imagen-pytorch</a><p>And here&#x27;s 2 images I sampled after training it for some hours, like 2 hours base model + 4 hours upscaler:<p><a href="https:&#x2F;&#x2F;imgur.com&#x2F;a&#x2F;46EZsJo" rel="nofollow">https:&#x2F;&#x2F;imgur.com&#x2F;a&#x2F;46EZsJo</a><p>* = Only the unconditional Imagen variant, meaning what they show off here. The variant with a T5 text embedding takes longer to train.
评论 #32171497 未加载
评论 #32240270 未加载
georgeburdellalmost 3 years ago
This is the most disturbing “does not exist” yet. A food blog could write itself
评论 #32171142 未加载
spacemanmattalmost 3 years ago
Somewhere in that data set is found an Eigencookie. I want the recipe.
tmountainalmost 3 years ago
Aggregate &quot;does not exist&quot; website for anyone who&#x27;s interested.<p><a href="https:&#x2F;&#x2F;thisxdoesnotexist.com&#x2F;" rel="nofollow">https:&#x2F;&#x2F;thisxdoesnotexist.com&#x2F;</a>
rkagereralmost 3 years ago
My partner is very impressionable when she see&#x27;s food in a TV show. Immediately has a craving for it. This thing is like, limitless porn for her gluttony.
golergkaalmost 3 years ago
And here I was, hoping for new, never seen before dishes.
Sebbeckingalmost 3 years ago
How big was your training dataset?
WalterSearalmost 3 years ago
No hot dogs?<p>Nice work.
ComputerCatalmost 3 years ago
Everything looks delish!
waynesonfirealmost 3 years ago
and what was the licensing for the training data that you used?
twicalmost 3 years ago
This computer has pretty poor taste in cocktails.
immmmmmalmost 3 years ago
should call that DeepCakes
beej71almost 3 years ago
We&#x27;re looking at the complete collapse of the stock photography market.
评论 #32169917 未加载
评论 #32169849 未加载
评论 #32170120 未加载
评论 #32169880 未加载
评论 #32172849 未加载
评论 #32169866 未加载
评论 #32173345 未加载
评论 #32174209 未加载
评论 #32174179 未加载
评论 #32171488 未加载
评论 #32175361 未加载
评论 #32174575 未加载
评论 #32175422 未加载