Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Welcome to Waifu Labs v2: How Do AIs Create?

waifulabs.com

Post reply on HN
Print view View original

81–90 of 236 posts

Re: Welcome to Waifu Labs v2: How Do AIs Create?

#81

Post by wodenokoto »

liuru wrote: post #2

Hey HN, one of the team members here! I hope you all enjoy playing with the new and improved generator! We've been hard at work improving the model quality since the last time the site was posted[1] As both a professional fantasy illustrator & software engineer, I find the concept of AI creativity so fascinating. On one hand, I know that mathematically AI only can hallucinate images that fit within the distribution o…

Naïvely I thought Waifu generator was just “some guy having a laugh” fine-tuning a model off of hugging face, but reading through the comments here, it is obviously a much, much bigger enterprise.

Can you talk a little about team size, work process, funding and revenue stream? I think the effort required for such an undertaking is vastly underestimated by readers.

Re: Welcome to Waifu Labs v2: How Do AIs Create?

#82

Post by ridaj »

liuru wrote: post #2

Hey HN, one of the team members here! I hope you all enjoy playing with the new and improved generator! We've been hard at work improving the model quality since the last time the site was posted[1] As both a professional fantasy illustrator & software engineer, I find the concept of AI creativity so fascinating. On one hand, I know that mathematically AI only can hallucinate images that fit within the distribution o…

Fascinating... Thanks for sharing

A couple questions:

1) I didn't really understand how you went about identifying what vectors of the latent space stand for various things, like pose or color. Did you train one of the AIs to that effect, or did you manually inspect a bunch of vectors, twiddling through them one by one, did to the outcome?

2) If one were to train an AI to the same level using commodity cloud services, what's the order of magnitude cost that you would pay for the training? More like $100, $1,000, $10,000 or $100,000?

Re: Welcome to Waifu Labs v2: How Do AIs Create?

#83

Post by xcambar »

zaik wrote: post #80
bmitc wrote: post #74

This isn't really creativity is it? I like to call stuff like this statistical copying, and indeed, the linked Wikipedia article on GANs says: > Given a training set, this technique learns to generate new data with the same statistics as the training set. There isn't a creative process here nor any creative introspection going on. While the technical results are impressive, this article does not address creativity ev…

> There isn't any AI either. It's machine learning, i.e., statistical models and algorithms. I always thaught those were synonyms.

Both the A and the I in AI are debatable terms.

The sole definition of intelligence is under heavy load of reconsideration the last decades with the emergence of a better knowledge of animal cognition, for example.

Re: Welcome to Waifu Labs v2: How Do AIs Create?

#84

Post by isaacimagine »

zaik wrote: post #80
bmitc wrote: post #74

This isn't really creativity is it? I like to call stuff like this statistical copying, and indeed, the linked Wikipedia article on GANs says: > Given a training set, this technique learns to generate new data with the same statistics as the training set. There isn't a creative process here nor any creative introspection going on. While the technical results are impressive, this article does not address creativity ev…

> There isn't any AI either. It's machine learning, i.e., statistical models and algorithms. I always thaught those were synonyms.

There isn't any I either. It's biological learning, i.e. neuronal models and feedback loops.

/s?

Re: Welcome to Waifu Labs v2: How Do AIs Create?

#85

Post by numpad0 »

bmitc wrote: post #74

This isn't really creativity is it? I like to call stuff like this statistical copying, and indeed, the linked Wikipedia article on GANs says: > Given a training set, this technique learns to generate new data with the same statistics as the training set. There isn't a creative process here nor any creative introspection going on. While the technical results are impressive, this article does not address creativity ev…

I find odd that no one is tracing back portions of generative arts back to publicly posted sources. Internet horde of bored people were used to be good at finding where an unattributed pieces come from. Are these really generalized and stored in highly abstract forms? I doubt it...

Re: Welcome to Waifu Labs v2: How Do AIs Create?

#86

Post by chrisdsaldivar »

liuru wrote: post #16
2bitencryption wrote: post #15

Earlier quoted context omitted.

Firstly, amazing work. My question is, how do you figure out how to parameterize "Same character, different pose" / "Same character, different eyes" / "Same character, different gender" / etc? My (super limited) understanding of GANs is that they slowly discover these features over time simply from observation in the data set, and not from any labels. So how could you make e.x. a slider for head position, style, pose…

Yes, your understanding is correct! You can think of it like coordinates on a many-dimensional vector grid. We craft the functions the functions that will illuminate sets of those points based on a combination of observation, what we know about our model architecture, and how our data is arranged. And yes, when the model is retrained, we have to discover them again!

Can you share any resources for reading on this particular topic?

Re: Welcome to Waifu Labs v2: How Do AIs Create?

#87

Post by lynzrand »

unobatbayar wrote: post #63
liuru wrote: post #2

Hey HN, one of the team members here! I hope you all enjoy playing with the new and improved generator! We've been hard at work improving the model quality since the last time the site was posted[1] As both a professional fantasy illustrator & software engineer, I find the concept of AI creativity so fascinating. On one hand, I know that mathematically AI only can hallucinate images that fit within the distribution o…

The quality and style is mindblowing! What data did you train?

As a rough guess, I think it might be trained on the Danbooru archive dataset, since it's the largest anime picture dataset we can get today.

https://www.gwern.net/Danbooru2020

Re: Welcome to Waifu Labs v2: How Do AIs Create?

#88

Post by numpad0 »

jychang wrote: post #57
Gigachad wrote: post #46

This is extremely impressive. It’s the first GAN I have seen which lets you tweak the result in a meaningful way rather than being just random. I think the speed that GANs have come in to the world has really shaken people up and it’s hard to process what this all means and what it will result in. Especially the ones which generate based on real people. But the feeling this gives me, is what happens for the future of…

Well, digital art already allows you to copy/paste a layer and stuff that traditional art never allowed for. Imagine copy pasting a statue. This is sort of the same thing on steroids. You can copy/remix previous art by feeding them into a ML model in training mode, and it will be massively utilized the same way ctrl-c ctrl-v is used, but it's a part of the toolset of art creation, not replacing it.

> You can copy/remix previous art by feeding them into a ML model in training mode, and it will be massively utilized the same way ctrl-c ctrl-v is used,

Then you need attributions for said previous arts, all of it, at least going by texts of laws.

Re: Welcome to Waifu Labs v2: How Do AIs Create?

#89

Post by aniforprez »

numlock86 wrote: post #79
jhanschoo wrote: post #69

Earlier quoted context omitted.

Ganguro is associated with a certain Japanese subculture and aesthetic, and not meant at all to represent darker-skinned races. There are works that feature a more diverse cast, but perhaps the proportion of representation accurately reflects what you would see in Japanese society.

> Ganguro is associated with a certain Japanese subculture and aesthetic, [...] Yes. What's your point? > [...] and not meant at all to represent darker-skinned races. No one ever claimed that. The question to my post was simply where to possibly get data from for "darker skin tones". Ganguro seems like an option because ... well, darker skin tones? Care to elaborate why you bring up "representation of darker skinned…

... because that was the request of the root comment? And also because Ganguro does not, in fact, represent races with darker skin tones so is a falsely equivalent set of data?

Re: Welcome to Waifu Labs v2: How Do AIs Create?

#90

Post by plutonorm »

Gigachad wrote: post #46

This is extremely impressive. It’s the first GAN I have seen which lets you tweak the result in a meaningful way rather than being just random. I think the speed that GANs have come in to the world has really shaken people up and it’s hard to process what this all means and what it will result in. Especially the ones which generate based on real people. But the feeling this gives me, is what happens for the future of…

hahaha. Now everyone will get a piece of my existential ennui. Welcome to the club!
Post reply on HN