Adobe Will Sell AI-made Stock Images (axios.com) 46
Adobe is opening its stock images service to creations made with the help of generative AI programs like Dall-E and Stable Diffusion, the company said. From the report: While some see the emerging AI creation tools as a threat to jobs or a legal minefield (or both), Adobe is embracing them. At its Max conference in October, Adobe outlined a broad role it sees generative AI playing in the future of content generation, saying it sees AI as a complement to, not a replacement for, human artists. Adobe says it is now accepting images submitted from artists who have made use of generative AI on the same terms as other works, but requires that they be labeled as such. It quietly started testing such images before officially announcing the move today. "We were pleasantly surprised," Adobe senior director Sarah Casillas told Axios. "It meets our quality standards and it has been performing well," she said.
well no doubt (Score:2)
While some see the emerging AI creation tools as a threat to jobs or a legal minefield (or both), Adobe is embracing them.
Adobe makes art tools more than art, they can make their own AI tools, in fact they already have for a long time. For the foreseeable future people will still need Photoshop, to fix hands :)
Re: (Score:2)
Re: (Score:2)
The key to making good AI art on things that it performs poorly with (such as hands) continues to be... ... to not be lazy.
Extremely lazy: Just hitting "generate", grabbing something that comes up, and posting it online
Semilazy: Running through one or more rounds of img2img on whatever's mangled until you get something "good enough"
Not lazy: Manually retouching (even if poorly) the things that got mangled between img2img rounds, aka using img2img only to clean up your hasty retouching work.
Throw it a friggi
Re:They Are ALL Based from Other Images (Score:5, Insightful)
Yeah, mate, that's not even remotely how these tools work. They're not compostors. They don't have some database of images.
These models are trained on billions of images and a typical checkpoint is a couuple gigs. Aka, there's about one byte of neural net weightings per image. If you think you have some magical algorithm to compress an image in one byte, by all means share it with the class.
SD learns motifs and styles. SD ***CANNOT*** reproduce single images from its dataset. Here's an easy test: search LAION-400M [github.io], turn on captions and aesthetic scoring, and pick an image with a fairly unique caption. Now go to StableDiffusion's Dreamstudio, enter in that exact caption, and generate a bunch of images. Result? You don't get anything like the original image. Thematic, stylistic, motifs? Yes. The same image? Not at all. And it cannot, because the data just isn't there.
Now, I specified *single* for a reason. If you have something like the Mona Lisa or the Earthrise photo or whatnot, it's going to exist thousands of times in the dataset in different forms, and it will be able to reproduce those things to varying degrees. But such things are pretty much the definition of a motif - a common stylistic artistic element.
Again, you don't have to take my word for it - do the above test for yourself.
Again, to repeat: AI art tools are NOT compositors. They're NOT samplers. They're denoisers [fosstodon.org]. They're given latent static dotproducted with a textual latent, and try to clean it up so that the latent image and text together make sense. You can easily picture doing this manually: you see a staticky image and can picture, "well, it'd look more like a house if I pushed this here and tweaked that there". But what sort of pushing and tweaking you'd do would vary much depend on whether someone said it was, say, a picture of a horse vs., say, a picture of a go-kart. The training teaches it how to do just that: how to push and tweak static to look a tiny bit more "houselike" or "go-kart-like" or whatnot. And that's run again and again and again. In a way, it's sort of reverse image recognition - it's trying to *make* the static recognizable.
Re: (Score:2)
That doesn't mean that ANY AI tool will be like this, mind you. If you have a huge neural net and a small dataset and train it enough, you'll overfit to the dataset, and it'll start producing originals, with good fidelity - indeed, in extreme overfit cases, to near perfection.
But that simply cannot happen with one byte per image. Sorry, data compression just isn't that good.
Re: (Score:2)
grey goo
stahp
You don't understand the words you're using, because if you did, you'd realize why that's a stupid name for this.
and dedicated it to looking into why it also has massive theft issues
The two houses are not the same. The claim is that it was spit right out, but it was false, because you can see that the two houses are not the same even from the scaled down sample images.
Got any claims that aren't false?
Re: (Score:2)
The "AI Art" community is absolutely embarrassing and unbearable.
Well, there's definitely elements of that, but the art community has a lot of cringe too. I know full well that making stuff yourself is not easy. My lady is an artist, and we have had lots of conversations about this, so I'm not unversed in the artist's view at all. In fact, I'm much more familiar with it than I am with the actual technology behind these image models! I've been hearing about it since I started playing with these toys.
Though no one asked, I don't consider myself to be "an artist", but I do
Re: (Score:3)
Beyond the simple issue of "Exactly how much diversity do you expect in lowpoly houses anyway?", that is NOT "StableDiffusion". That is a custom trained checkpoint for StableDiffusion. A mod. Created by a user. Most likely trained with Dreambooth. They likely took - tops - several hundred images for the training, maybe no more than a dozen or two, and used that to train a SD's several gigs of weightings. It's very easy to overtrain in such a situation - indeed, it will overtrain if you run it for too
Re: (Score:2)
Where's the theft? None of the generated images are the same as the 'original'. They bear some resemblance because the AI was asked to produce images similar to the original. None are close enough to trigger any sort of copyright concern.
Or are you some IP extremist who thinks nobody can build a house without stealing because someone already built a house with walls and a roof? Can nobody ever do a professional portrait photo again because all of the hand poses are already taken?
You've only supported your opposition. (Score:2)
Re: (Score:2)
As if a human artist would make great paintings if they were blind from birth and had never seen the world?
Weird Business Model (Score:2, Interesting)
How is this going to work? Computer generated images are not copyrightable. A human has to do something to the image to make it a work you can copyright which, I think, defeats the purpose of having a computer pump out tons of images.
Re: (Score:2)
Re: (Score:1)
Computer generated images are not copyrightable.
False. Computer generated images have already been copyrighted (formally, with registration.)
A human has to do something to the image to make it a work you can copyright
Wrong again! What the USPTO said was that a piece of software can't hold a copyright.
Re:Weird Business Model (Score:4, Informative)
False. Computer generated images have already been copyrighted (formally, with registration.)
Not per the US copyright office.
https://www.natlawreview.com/a... [natlawreview.com]
Wrong again! What the USPTO said was that a piece of software can't hold a copyright.
That sentence has nothing to do with the previous sentence. You cannot copyright a computer generated artwork. If a human takes a computer-generated artwork, and does something signifgant to it - crops it, filters it, tiles it, adds text, composites in other images - then it *is* copyrightable.
Here's an in-depth discussion by Leonard French, an IP lawyer:
https://www.youtube.com/watch?... [youtube.com]
Re: (Score:2)
So 3d artists can't get copyright on artwork that they render, because a computer spat it out and they didn't postprocess it? A photographer can't gain a copyright on a photograph because they just hit a button and the camera made the picture? Of course they can.
If it's about human creative endeavour, then creative post OR pre processing steps should apply. Which should include prompting. The question is what is a sufficient level of human creative endeavour. Obviously, if you spent ten seconds to write
Re: (Score:2)
Don't worry, his own citation doesn't actually say what he thinks it does. The USPTO has not said what he thinks they said.
Re: (Score:2)
Not per the US copyright office.
Did you not read that, or not understand it? Did you just read one paragraph? Because that story does not say what you want it to say, as usual. What it says is this:
Computer Generation (Score:2)
I read the article correctly. The USPTO has refused, did refuse, and continues to refuse copyright protection for images created purely by algorithm with no human input beyond feeding it seed images. It is settled doctrine. If a human does not significantly determine the composition of an image it is not copyrightable.
This harkens all the way back to fractal images. Spitting out a Mandelbrot image with fractint is not copyrightable. If you crop the image, sharpen it, and change the color balance, it *is* co
They are all fakes!!! (Score:2)
I've only read a bit about these image generators but they sound impressive. I'm sure they will be able to create some amazing photos and depending on your media needs, this would be a wonderful way to get random awesome stuff.
The difference between this and a human photographer would seem to be the focus of the shot. Human's only shoot stuff that exist in our physical space or being displayed on screen that takes up physical space. IE, we don't create the scene. We capture the scene at hand.
Of course after
Re: (Score:3)
Stock photographers absolutely create scenes.
Re: (Score:2)
The comparison to photography is indeed far more apt than the comparison with painting.
And I will use them royalty free (Score:1)
According to current law AI generated images are ineligible for copyright.
to access the site you must pay for adobe CC! (Score:2)
to access the site you must pay for adobe CC!
Re: (Score:2)
Until it gets used on another site.
Re:And I will use them royalty free (Score:4, Insightful)
According to current law AI generated images are ineligible for copyright.
Who told you that, and why did you believe it?
Re: (Score:2)
The US copyright Office, and I believed them because they make the rules. https://www.copyright.gov/ruli... [copyright.gov]
Re: (Score:2)
Is it? So your opinion supersedes the US Copyright Offices review board?
https://www.copyright.gov/ruli... [copyright.gov]
First, they came for our stock photography (Score:1)
Then, they came for Hollywood and porn. Some Hollywood movies resemble adult film, but at least AI can't be blamed for lack of character development or a plot. And sometimes adulting videos have production value too.
The obvious billion dollar questions now are: What sort of kinky shit will AI be into, monetize, and give away for free? Will AI make the marketing decisions to steer content into different platforms? Will AI choose to comply with various countries' censors or dance around it? Will the censors b
Re: (Score:3)
Will AI make the marketing decisions to steer content into different platforms?
If it does, it will do it the same way the humans do it, by interpreting market research. Humans already use forecasting tools to make these decisions.
Will the censors be automated with AI like YouTube too?
Of course they will.
Will AI automate paying of bribes to AI censors?
No, the bribes will be paid the old fashioned way, by humans to humans to get them to twiddle the results.
Re: (Score:2)
Porn is the most annoying thing about the AI art space. You're looking for an instruction guide on how to use some random tool, and the demo images are all of barely-covered (if at all) mutantly-large-breasted women in sexy poses :P
Shoppify their efforts (Score:2)
If they make a marketplace for AI art and allow users to upload to sell their assets that would be interesting.
Dead Internet Strikes Again, xref with other story (Score:2)
There's a fairly obvious trend here. Evilcorp gets opensource "AI" program {Stable Diffusion, Gpt3, etc}, gets chumps like you to upload source material and play with it, once it's trained to a minimum level of competence, re-release but now with a cost/price.
No copyright, means no fees to the creator, so the next logical next step is to make the entire practice of (art, music, "coding") irr
Re: (Score:2)
Yes, the same people claiming AI will reduce the need for graphic artists will happily use templates and tools that reduce the need for a web designer and webmaster on a platform that reduces the need for server admins.
The problem isn't reducing the need for work, it's an economic system that makes people depend on demand for their work in order to make a living. That system only works when demand for rewarding work exceeds supply. But unmet demand sparks a search for suitable substitutes.
The good news is t
The prize will go to... (Score:2)
Re: (Score:2)
Just lazily hand-paint them in and then let img2img "prettify" your work.
Re: (Score:2)
Will they require... (Score:2)
Cue the lawyers (Score:2)
Adobe is in the habit of lifting its hind leg and pee... er, 'monetizing' everything in sight. How long will it be before they claim copyright on CC-licensed AI images whose only connection to Adobe is an extended middle finger?
Re: (Score:2)
Unless the Supreme Court changes it's mind, never. Copyright requires a human creator. Courts interpreting the Copyright Act, including the Supreme Court, have uniformly limited copyright protection to creations of human authors.
Cannot be copyrighted. (Score:2)
"Both in its 2019 decision and its decision this February, the USCO found the âoehuman authorshipâ element was lacking and was wholly necessary to obtain a copyright, Engadgetâ(TM)s K. Holt wrote. Current copyright law only provides protections to âoethe fruits of intellectual laborâ that âoeare founded in the creative powers of the [human] mind,â the USCO states."
In order to file for a copyright under current law you have to be the creator of the images in question. Softw
"Sell"? (Score:2)
This is Adobe we're talking about here. Don't they mean "rent", or "subscribe", or some other form of ongoing permanent monetization?