WotC So it seems D&D has picked a side on the AI art debate.

[Charlaquin]

I very much doubt it. These models require mountains of data to train; public domain art is a miniscule set to work from. At least, it is if you're verifying ownership, instead of just harvesting anything on the Internet that some yahoo uploaded and tagged "public."

If someone built a Stable Diffusion-type model trained exclusively -- from the ground up -- on public domain art, or art for which the artist was properly paid, I'd have no objection to it. But I'm not aware of any such thing.

Yeah, as I said on the first page of this thread. If a frog had wings.

 

[Jadeite]

@Charlaquin
"I don't care what the law says, it's unethical to me!" isn't a great offense, either. And lamplighters became a curiosity at best once electrical lighting got introduced. Similar to portrait painters and photography. And maybe mediocre artists and Stable Diffusion. The technology exists and will be used. If not now, certainly in the future. Wait long enough and the art of today will be public domain.
There's no point in paying humans to do a job a machine could do better. Not when there is plenty of work left to be done that humans can do far better than machines, like taking care of the elderly.

 

[Charlaquin]

@Charlaquin
"I don't care what the law says, it's unethical to me!" isn't a great offense, either.

Good thing my argument is more substantive than that, then.

And lamplighters became a curiosity at best once electrical lighting got introduced. Similar to portrait painters and photography. And maybe mediocre artists and Stable Diffusion.

One of these things is not like the others. As I’ve been over a few times now, labor-saving inventions tend to put laborers out of work, which is certainly a flaw in our current economic system. But Image generation algorithms aren’t just saving labor; they actually require artists’ labor to function. They just exploit that labor without paying for it. That is a problem for artists in a similar way to labor-saving inventions like electric lighting, photo cameras, and automated manufacturing is a problem for the people who do the labor those inventions perform. But in addition to that, image generation algorithms necessarily harvest the products of artists’ labor in order to make what they do possible, which makes them superlatively unethical since they don’t compensate the artists for that labor.

The technology exists and will be used. If not now, certainly in the future.

Right, which is why regulation is needed to protect people’s livelihoods.

There's no point in paying humans to do a job a machine could do better.

Except that, for now, humans need jobs to live. If we had some sort of universal basic income, I would be fully onboard with this, but as long as people need to work to live, reducing the demand for work is going to be a bad thing for people.

Not when there is plenty of work left to be done that humans can do far better than machines, like taking care of the elderly.

The work that humans can do better than machines is an ever-shrinking pool; meanwhile the population of humans who need to do work to live is ever-growing. And both are happening at exponential rates. At some point, something has to give.

 

[tetrasodium]

And it does so by taking the fruits of labor those artists did and giving them nothing in return. You’re literally describing theft.

This is factually incorrect

Spoiler: training data

Training data[edit]​

Stable Diffusion was trained on pairs of images and captions taken from LAION-5B, a publicly available dataset derived from Common Crawl data scraped from the web, where 5 billion image-text pairs were classified based on language and filtered into separate datasets by resolution, a predicted likelihood of containing a watermark, and predicted "aesthetic" score (e.g. subjective visual quality).[15] The dataset was created by LAION, a German non-profit which receives funding from Stability AI.[15][16] The Stable Diffusion model was trained on three subsets of LAION-5B: laion2B-en, laion-high-resolution, and laion-aesthetics v2 5+.[15] A third-party analysis of the model's training data identified that out of a smaller subset of 12 million images taken from the original wider dataset used, approximately 47% of the sample size of images came from 100 different domains, with Pinterest taking up 8.5% of the subset, followed by websites such as WordPress, Blogspot, Flickr, DeviantArt and Wikimedia Commons.[17][15]

They get free hosting in most cases. It seems they can even opt out of the dataset while still enjoying the free hosting. Don't like it? pay for your own hosting. Don't want to do that? opt out for your image.

Tetra. How is this 'child-like', which is what I figure you're trying to say here? Is it child like to not want people's things stolen? Childish to not want people mass-generating low effort stories to flood magazines? Childish to assume good faith in people and they won't steal everything about you, even the words you've written, to squeeze a bit of data from?

Plenty of the contempt over it being stealing and theft has gone well, and every time it comes up people are rightfully dunking on AI generated art every time. The discourse is a hot mess, sure, but everyone in the art side of things is against it for a reason and with most of the people supporting it claiming its "democratising" which, aside from being a crypto/NFT-bro buzzword, seems to mean "pushing people out of jobs", is an easy write-off lie.

Think of the children a phrase used to describe a call for reducinga discussion down to a moral panic. Quoting the linked wikipedia entry "You don't have to make a rational argument as long as you can appeal to ...."

 

[Charlaquin]

This is factually incorrect

Spoiler: training data

Training data[edit]​

Stable Diffusion was trained on pairs of images and captions taken from LAION-5B, a publicly available dataset derived from Common Crawl data scraped from the web, where 5 billion image-text pairs were classified based on language and filtered into separate datasets by resolution, a predicted likelihood of containing a watermark, and predicted "aesthetic" score (e.g. subjective visual quality).[15] The dataset was created by LAION, a German non-profit which receives funding from Stability AI.[15][16] The Stable Diffusion model was trained on three subsets of LAION-5B: laion2B-en, laion-high-resolution, and laion-aesthetics v2 5+.[15] A third-party analysis of the model's training data identified that out of a smaller subset of 12 million images taken from the original wider dataset used, approximately 47% of the sample size of images came from 100 different domains, with Pinterest taking up 8.5% of the subset, followed by websites such as WordPress, Blogspot, Flickr, DeviantArt and Wikimedia Commons.[17][15]

They get free hosting in most cases. It seems they can even opt out of the dataset while still enjoying the free hosting. Don't like it? pay for your own hosting. Don't want to do that? opt out for your image.

That hosting isn’t payment for having their content used to train AI, the AI is just scraping content from the websites where the artists were already hosting their work, and not giving the owners of that content notice that they’re using it, let alone compensation. And the opt-out system doesn’t fix this problem because your art has to already be in the AI’s database, you have to select individual works to opt out, and it doesn’t account for the fact that those image hosting sites are themselves notorious for art theft. All criticisms which the very link you provided gets into.

 

[Jadeite]

@Charlaquin
I don't think the march of progress is slowing down any time soon. And that's the way I like it. As should you, considering how flawed your arguments are. There are still tons of jobs for humans. Not only in the care or social sector but also craftsmen (craftpeople? craftfolk?) are in high demand. And you need technological advances in order to feed a growing populace. You could have, however, become more technologically advanced with a decreasing populace (like Japan, even though that creates other problems). Also, if you want to see what happens when certain nations get a technological edge, take a look at the past 500 years of human history, especially the 19th century.
If you want to support artists that fear for their livelihoods, support them in learning another trade. The technology exists, people will use it, it will get better. You can try to outlaw it, but that's impossible to enforce without extreme collateral damage (which would affect other people's livelihoods).

 

[Charlaquin]

@Charlaquin
I don't think the march of progress is slowing down any time soon. And that's the way I like it. As should you, considering how flawed your arguments are. There are still tons of jobs for humans. Not only in the care or social sector but also craftsmen (craftpeople? craftfolk?) are in high demand. And you need technological advances in order to feed a growing populace. You could have, however, become more technologically advanced with a decreasing populace (like Japan, even though that creates other problems). Also, if you want to see what happens when certain nations get a technological edge, take a look at the past 500 years of human history, especially the 19th century.
If you want to support artists that fear for their livelihoods, support them in learning another trade. The technology exists, people will use it, it will get better. You can try to outlaw it, but that's impossible to enforce without extreme collateral damage (which would affect other people's livelihoods).

Technology is all well and good, but we also need to make sure people are being cared for. This particular piece of technology is unethical because it has to steal labor value in order to function. It is also a problem that our current economic system positions labor-saving technologies at odds with laborers.

 

[Dausuul]

This is factually incorrect

Spoiler: training data

Training data[edit]​

Stable Diffusion was trained on pairs of images and captions taken from LAION-5B, a publicly available dataset derived from Common Crawl data scraped from the web, where 5 billion image-text pairs were classified based on language and filtered into separate datasets by resolution, a predicted likelihood of containing a watermark, and predicted "aesthetic" score (e.g. subjective visual quality).[15] The dataset was created by LAION, a German non-profit which receives funding from Stability AI.[15][16] The Stable Diffusion model was trained on three subsets of LAION-5B: laion2B-en, laion-high-resolution, and laion-aesthetics v2 5+.[15] A third-party analysis of the model's training data identified that out of a smaller subset of 12 million images taken from the original wider dataset used, approximately 47% of the sample size of images came from 100 different domains, with Pinterest taking up 8.5% of the subset, followed by websites such as WordPress, Blogspot, Flickr, DeviantArt and Wikimedia Commons.[17][15]

They get free hosting in most cases. It seems they can even opt out of the dataset while still enjoying the free hosting. Don't like it? pay for your own hosting. Don't want to do that? opt out for your image.

Putting your stuff on Pinterest does NOT mean you are making it public domain. Read the terms of service. You grant rights to Pinterest and its users solely for the purpose of operating and using Pinterest. At no point are you offering up your work to be harvested for AI training or anything else.

And anybody can post an image to Pinterest. Most of the images I've seen there were clearly posted by someone other than the creator. Using Pinterest to "launder" stolen art doesn't make it not stolen.

 

[JiffyPopTart]

I started playing around with the Dall-E Android app this weekend to generate some art from past game sessions.

In everyone's favorite session the druid wildshaped into a giant constricter snake which the paladin then rode up to a pyramid occupied by Yuan-Ti. With a super successful roll the paladin declared himself the Snake-Pope and demanded access to the facility.

A couple iterations of the prompt gave me this.

My question to the thread is....If this art (which I absolutely LOVE) is just someone else's art chopped up and redelivered, how do I go about finding the original artist?

 

[Vaalingrade]

My question to the thread is....If this art (which I absolutely LOVE) is just someone else's art chopped up and redelivered, how do I go about finding the original artist?

Luckily tech bros tend to not understand laws and proudly list what archives they stole from.