From Ars Technica:
On Friday, the online art community DeviantArt announced DreamUp, an AI-powered text-to-image generator service powered by Stable Diffusion. Simultaneously, DeviantArt launched an initiative that ostensibly lets artists opt out of AI image training but also made everyone’s art opt in by default, which angered many members.
DreamUp creates novel AI-generated art based on text prompts. Due to its Stable Diffusion roots, DreamUp learned how to generate images by analyzing hundreds of millions of images scraped off sites like DeviantArt and collected into LAION datasets without artists’ permission, a potential irony that some DeviantArt members find problematic.
As we’ve reported frequently on Ars in the past, Stable Diffusion’s web-scraping nature ignited a huge debate earlier this year among artists that challenge the ethics of AI-generated artwork. Some art communities have taken hard stances against any AI-generated images, banning them completely.
Perhaps anticipating a backlash, DeviantArt is making overtures to pacify artists who might be upset about their work being used to train AI image generators. The site is providing a special “noai” flag that artists can check in their image settings to opt out of third-party image datasets. (Whether third-party image scrapers will honor this flag, however, remains to be seen.)
. . . .
Also, DeviantArt will let artists opt out of letting their images train DreamUp in the future, but each artist must fill out a form that requires human review first. This policy has led to significant pushback among DeviantArt members, some of whom have threatened to delete all of their work and deactivate their accounts.
DeviantArt’s DreamUp information page also takes a defensive tone, stating that DeviantArt did not consent to third-party AI image models (such as Stable Diffusion) that scraped their site to make their models work. And further down the page, the site attempts to debunk common misconceptions about how AI image synthesis works.
Link to the rest at Ars Technica
For those not familiar with DeviantArt, it’s an artists’ social network. It’s also a place for those who might might want to commission an artist to produce something like a cover design for a fantasy or science fiction novel.
PG’s understanding is that a great many artists who create images with computers/tablets+computers, etc., show some of their work in order to attract visitors to their websites. Of course, everything an artist puts up on Deviant Art is a digital image.
Based on the Ars Technica article, it sounds like DeviantArt really screwed up the launch of its AI Art tool.
Of course, standard legal advice is that a creator should read all contracts, terms and conditions, terms of use, terms of service, etc., prior to uploading any creation to a website or other online destination.
One of the things a creator may find in the T’s&C’s is a provision that says the site owner can change the T’s&C’s at any time without notifying the creator in advance.
Here are a few sample provisions from Facebook’s Privacy Policy which is referenced in Facebook’s Terms of Service for your edification and enjoyment:
On our Products, you can send messages, take photos and videos, buy or sell things and much more. We call all of the things you can do on our Products “activity.” We collect your activity across our Products and information you provide, such as:
- Content you create, like posts, comments or audio
- Content you provide through our camera feature or your camera roll settings, or through our voice-enabled features. Learn more about what we collect from these features, and how we use information from the camera for masks, filters, avatars and effects.
- Messages you send and receive, including their content, subject to applicable law. We can’t see the content of end-to-end encrypted messages unless users report them to us for review. Learn more.
- Metadata about content and messages, subject to applicable law
- Types of content you view or interact with, and how you interact with it
- Apps and features you use, and what actions you take in them. See examples.
- Purchases or other transactions you make, including credit card information. Learn more.
- Hashtags you use
- The time, frequency and duration of your activities on our Products
Information with special protections
You might choose to provide information about your religious views, political views, who you are “interested in” (which could reveal your sexual orientation) or your health in your Facebook profile fields or life events. This and other information (such as racial or ethnic origin, philosophical beliefs or trade union membership) could have special protections under the laws of your country.
(PG Note: You have to click to a separate page to continue reading the Terms of Service)
(The Facebook Docs continue.)
Friends, followers and other connections
Information we collect about your friends, followers and other connections
We collect information about friends, followers, groups, accounts, Facebook Pages and other users and communities you’re connected to and interact with. This includes how you interact with them across our Products and which ones you interact with the most.
Information we collect about contacts
We also collect your contacts’ information, such as their name and email address or phone number, if you choose to upload or import it from a device, like by syncing an address book.
If you don’t use Meta Products, or use them without an account, your information might still be collected. Learn more about how Meta uses contact information uploaded by account holders.
App, browser and device information
We collect and receive information from and about the different devices you use and how you use them.
Device information we collect and receive includes:
- The device and software you’re using, and other device characteristics. See examples.
- What you’re doing on your device, like whether our app is in the foreground or if your mouse is moving (which can help tell humans from bots)
- Identifiers that tell your device apart from other users’, including Family Device IDs. See examples.
- Signals from your device. See examples.
- Information you’ve shared with us through device settings, like GPS location, camera access, photos and related metadata
- Information about the network you connect your device to, including your IP address. See more examples.
- Information about our Products’ performance on your device. Learn more.
- Information from cookies and similar technologies.
Learn how to upload and delete contacts on Facebook and Messenger, or how to connect your device’s contact list on Instagram.
Information we collect or infer about you based on others’ activity
We collect information about you based on others’ activity. See some examples.
We also infer things about you based on others’ activity. For example:
- We may suggest a friend to you through Facebook’s People You May Know feature if you both appear on a contact list that someone uploads.
- We take into account whether your friends belong to a group when we suggest you join it.
Information from Partners, vendors and third parties
What kinds of information do we collect or receive?
We collect and receive information from Partners, measurement vendors and third parties about a variety of your information and activities on and off our Products.
Here are some examples of information we receive about you:
- Your device information
- Websites you visit and cookie data, like through Social Plugins or the Meta Pixel
- Apps you use
- Games you play
- Purchases and transactions you make
- Your demographics, like your education level
- The ads you see and how you interact with them
- How you use our Partners’ products and services, online or in person
Partners also share information like your email address, cookies and advertising device ID with us. This helps us match your activities with your account, if you have one.
We receive this information whether or not you’re logged in or have an account on our Products. Learn more about how we connect information from Partners to your account.
Partners also share with us their communications with you if they instruct us to provide services to their business, like helping them manage their communications. To learn how a business processes or shares your information, read their privacy policy or contact them directly.
How do we collect or receive this information from partners?
Partners use our Business Tools, integrations and Meta Audience Network technologies to share information with us.
These Partners collect your information when you visit their site or app or use their services, or through other businesses or organizations they work with. We require Partners to have the right to collect, use and share your information before giving it to us.
Lawyer PG notes all the links to other places sprinkled through Facebook’s TOS. Each of the links includes yet more information that is part of the TOS. While PG didn’t click and read what was to be found in each of the links in the Mother TOS, PG will note that the links can include information and definitions that changes the meanings Mother TOS substantially.
PG doesn’t know whether Facebook’s Terms of Service as a whole are great, sorta-great, somewhere-in-the-middle, sorta terrible or terrible because he swore off of Facebook a long time ago. He has more than one Facebook account that contains information that has nothing to do with PG-in-the-flesh if he finds out about something Facebook is doing that may be of interest to PG or visitors to TPV.
PG hasn’t read Deviant Arts’ TOU, TOS, etc., but these are some of the concerns that artists who use Deviant Arts as a marketing platform are likely thinking about at the moment.
This reminds PG that he hasn’t taken a look at KDP’s Terms of Use for awhile. He thinks he has copies of such documents from some earlier exploration of them, so he may check out the latest and see what Zon’s lawyers have altered, likely as a result of some disaster, minor or major, that transpired under an earlier TOU.
This will take PG awhile to finish, so don’t hold your breath waiting for it.
Are there any replacement sites for former DeviantArt users? Preferrably paid, so I am the customer, not product.
Anticipating backlash? They didn’t anticipate backlash, they expected us to roll over and ignore it. Twitter erupted, and with good reason. Not only would a person have to go through each of their pieces one by one and check the box, there was no protection in place for those who may have died since starting their account. A flurry of people responded. Many deleted all their art and deactivated their accounts too. I’m one of the ones who rid my gallery of everything.
They tried to backpedal, by making the ‘no ai’ thing a default setting, but the bridge has burned as far as I’m concerned.
AI art is horrible, and steals work from artists. When someone can type “Art in so-and-so’s” style into the generator (https://www.technologyreview.com/2022/09/16/1059598/this-artist-is-dominating-ai-generated-art-and-hes-not-happy-about-it/) and get art that’s made based on scraping that person’s art off the net, it’s bad. When companies look to generators instead of real artists, it’s bad. When umpty-thousands of thousands of people who’ve kept the small-time commission’s market going for artists trying to make ends meet start saying “When they start getting all Air BnB with their prices, I just remind them Midjourney exists’ (yes, I’ve seen this exact comment and I about chucked my phone across the room), it’s BAD. BAD BAD BAD.
That DA, who had previously been one of the few sites protecting artists from NFT rackets, decided to turn to AI and let the art get scraped, it broke a lot of trust, and they can’t get it back.
What AI is doing is hurting people’s income, and making alot of us wary of EVER sharing our art online at all. All the fun anticipation of posting and showing off is tainted now.
(edited to elaborate and tidy paragraphs)
Deviant Art doing this was a definite misstep; I used the site to find artists for hire, and encouraged another artist I knew to use it. They should have had an open discussion about this. The reaction they’re getting was very foreseeable.
How I imagine AI art working out is having known artists license their style. Similar to how you can have stock art or model releases. You can use my image for this and this and this, but not that and definitely not that!
But there are an assortment of legal issues to work out, so I would be wary for now. Just a little while ago I was trying to find out if any of the Pre-Raphaelites ever painted a particular mythical figure, and it appears they did not. It might legally be plausible to to ask a machine to “paint this person in the style of John Waterhouse Williams” because JWW has been dead for over a century. Or, “render me this other subject in the style of a Greek red-figure vase.”
But for a living artist? What if a living artist has an ethical objection to using their art for a particular purpose, or would never want anyone to think they were participating in whatever output their art style is used for? Do they get a say, legally? I picture AI-art being next level to iStock or Getty, but those current artists voluntarily participated — and get paid. I haven’t seen any indication yet that any artists whose styles are being aped by AI are getting any money out of this. Licensing one’s style is potentially an income stream, but again, how will this happen?
Your reaction doesn’t surprise me. And I just don’t see AI-art truly becoming a going concern until someone resolves the rights / payment portions of this issue.
Two problems: a lot of folks put stuff out without considering what terms (if any) they are agreeing to. Also, the internet is forever.
There will be court cases and a likely possibility is that a final ruling will be along the lines of “expectations of privacy”.
First fight is likely to be “can a style be copyrighted/trademarked/patented?”.
Lots of billable hours in that one alone. Especially if the subject hasn’t been.
Three boundary cases:
1- Trade dress patents: https://en.wikipedia.org/wiki/Trade_dress#:~:text=Trade%20dress%20is%20the%20characteristics%20of%20the%20visual,is%20a%20form%20of%20intellectual%20property%20protection%20law.
2- Ideas can’t be copyrighted. Only patented for 17 years.
3- Clean room reverse enginering is legal. This one is going to weigh heavily.
BTW, PENTIMENT is out and it’s everything promised.
Beautiful gtaphics, historically accutate, good mystery, in a narrative driven game with meaningful options.
https://www.windowscentral.com/gaming/xbox/pentiment-review
When pundits dream of interactive books, this is what they aspire to.
But it’s already here and all you need is a browser to play.
I thought it was coming tomorrow. Say no more, I can’t play just yet and I don’t like to face temptation until I can give in to it.
In the meantime, I’m going to look into trade dress because at first glance it makes me think of “signature” looks or designs. For example, here:
https://search.brave.com/images?q=trade%20dress%2Bchanel%20suit
Note that on the Chanel suits in this search result, the pockets and cuffs and hem of the suit ensembles are always in a contrasting color, like white or black. This is a signature look to a Chanel suit, the way Louboutin has the red heel on their stilettos. Yves St. Laurent lost a court case after trying to imitate that look with yellow heels.
If an artist can prove to have a signature style, I’m willing to bet that said artist will sue. A lawsuit will surely come, so this question will be settled sure enough. In the meantime, this will surely prompt artists and other creatives to care more about the TOS of sites where they post. So there’s an upside there, at least.
It’ll be a horse race: the way the tech works parallels reverse engineering which done properly is legal. And the art scanned is millions of images. On the internet. And that adds extra complication even for copyrighted images because the output might resemble the general style rather than any specific image.
Where’s the violation in a picture of, say, “shirtless mine worker in the style of Frazetta”? Whelan? Boris? Rowena? Freas?
My guess is it will take new laws, worldwide, and we all know how…fast…that happens. Expressed as a filter on acceptable prompts, time limited. But even that may not be doable. Who’s style is to be protected? Everbody? How is their style to be described for protection? And what if the prompt leads to the same output without triggering the filter? “Hyper realistic muscular thin woman with outsized butt?”
And that assumes style is deemed protectable. And how is the software going to distinguish between a prominent artist from associates working at the same studio? A lot of the big name illustrators have studios with “understudies” who mimic their style and fullfil projects under their name. A lot of those folks are more brands than artists.
It is easy to describe the desired outcome but not so easy to get there from here.
(Also: check out the rules for clean room reverse engineering. That will almost certainly be the third line of defense. The first: “No existing law forbids this.” The second: “No law forbid it when done.)
Finally, “AI” tech is evolving beyond the need for big data training and into “AI” training “AI”.
The laws may come faster if someone slips up and crosses Disney 🙂
Looking into “clean room reverse engineering” — based on how it works, this does look like a viable defense.
And that adds extra complication even for copyrighted images because the output might resemble the general style rather than any specific image.
Oh, similar to how when scientists create the “average” face they end up with something beautiful ( a 10) rather than what would normally be defined as “average looks” (a 5). With millions of sources to pull from…
Things are going to get interesting. I truly do not envy the artists; they appear to be staring into Kurt Vonnegut’s “Player Piano” abyss where machines have even taken over art.
The problem is, the assorted things covered by that theoretical licensing are already happening, without permission and with no sign of stopping. DA having, or not having, a ‘no ai’ flag on their art does nothing to protect anything else out there, and given how many people post to Twitter, IG, Artstation, their own sites, or any of a number of other places, there’s no way to tell these AI companies to stop and have it stick. Artists who share their opinions (like me), get laughed at, yelled at, told ‘it’s going to happen anyway, give up’, told ‘you’re being elitist’, and any number of other things. There’s no good faith here, and DA broke what little they had with this stunt.
Yeah, I don’t really see a protective defense against the scraping as things stand right now. It can take years for a court case to result in a ruling. Use a digital watermark? Paywalls?
I vaguely remember my old paper using an anti-scraping measure for our website. IIRC, it may have involved the honeypot route mentioned here:
https://resources.infosecinstitute.com/topic/how-to-prevent-web-scraping/
But determined humans can circumvent all of them, e.g., someone could take a screenshot of an artist’s portfolio. And the methods suggested to protect against scraping tend to be hindrances to legitimate users, e.g., imagine using CAPTCHA for every photo in a carousel.
So this will be tricky, and I don’t envy the artists caught up in this. But DA shooting themselves in the foot is a prime opportunity for any alternative sites to make their presence known right about now. Even a few weeks ago, when I was doing my art search I kept wondering if DA and Art Station were my only options to find contemporary artists. I have a feeling a lot more people are asking that question.
Four letters should scare the DeviantArt management:
GDPR.
Because DeviantArt never has been compliant, and this fiasco is worse. So congratulations: All it’s going to take is a military spouse in Germany (or elsewhere in the post-Brexit EU) to really ruin your collective day with a privacy complaint to/through the EU authorities. Or an EU citizen. Or…
A comic book series/graphic novel produced through Midjourney is available for free download or paid in dead tree pulp via Amazon. Links here:
https://www.cnet.com/culture/youd-never-know-ai-drew-this-beautiful-comic-series/
A case study of economical uses for the tech.
Looks good enough for creating book covers…
…if you get good at the prompts, which is esentially a programing language of sorts, distantly related to the likes of LOGO and its turtle graphics.
The key takeaway from that link for me is this:
“He does, however, see Midjourney as his true collaborator in The Bestiary Chronicles, even giving it an author credit. Where a comics artist might conceive of a narrative and then create art to illustrate it, AI-assisted images have the potential to more actively steer the story, or even change its direction, thus dramatically redefining the whole creative workflow. Coulson likens this human-machine duet to improv jazz. ”
I couldn’t agree more.
I find in my initial AI experiments, mostly for blog posts (illustrating metaphors), that this is entirely true. You can (and should) spark off of the imagery that is produced and collaborate on the final product, even though the human and the algorithm are nothing alike. You can’t ask an artist to brain-dump massive quantities of alternatives, but the machine can. And it’s very inspiring if you write fiction “on the fly” rather than heavily pre-planned. All sorts of complications and refinements suggest themselves.
Think of “character images” alone… I’m about to go off and start making character families for my WIPs… I mean, if I’m in the business of building worlds with words, why not really build them (short of painted action figures and clay modelled dwellings)?
Agreed on all counts.
The variations and iteration possibilities are something worth paying (modestly 😉 ) for.
Also I can see a variation of the tech that outputs 3D autocad models ready to be fed to a hobbyist grade 3D printer. Lots of uses there.
I did find amusing that “Hitchcock blonde” rendered Grace Kelly but not Janet Leigh.
This does look useful for “concept art” which is needed for a product a writer may be pitching on a budget. Even useful for cover art and certain illustrations. But I notice in the Bestiary story, the characters have no facial expressions. Some of it lends itself to what I’m going to call the “Ivan Effect.”
The idea comes from something I learned in film class, about a Soviet experiment where they had an actor named Ivan standing still, with just one facial expression. And they’d cut from closeups of Ivan’s face to assorted scenes in a montage. Ivan’s face never changes, but viewers perceived him to be reacting to whatever imagery was included in the montage. Viewers were left with the idea that Ivan was just the greatest actor ever. But his face never changed. His expression was always the same.
This Grace-Kelly stand in has the same situation. People may imbue her with emotion based on what is being said in the scenes she appears in, but her facial expression remains the same. It’s devoid of emotion, which can be creepy if the author isn’t careful to present her in situations where that calm look makes sense. But that does take away from depth.
In acting, you’ll sometimes read of directors planning to kill off one-shot characters, e.g. Boyd Crowder in “Justified.” But the actor, in this case Walton Goggins, will be so good that the director realizes they’ll bring value to the story if they keep the character. The expressiveness of Goggins and the spark he brought to the role allowed the “Justified” show runners to take the series in interesting directions. I think Joss Whedon said similar about casting Sarah Michelle Gellar as Buffy. Buffy was just supposed to make wisecracks, but because Gellar could act the writers were able to add drama and depth to her character and the series.
To me the Bestiary story is a cool proof of concept, but AI art won’t really become a killer app — yet — until it can bring the Goggins-Gellar spark to what is otherwise an Ivan-Effect mannequin. I could picture myself bouncing off a facial expression in an AI-character, where a character looks sly or introspective or determined when I wasn’t expecting it. But an Ivan mannequin means I have to tailor a story to its limitations. I hope AI-assisted art does improve past “Ivan,” so the Grace Kelly stand-in can show a range of emotions. That’s when it will become a true game changer.
Also, TIL that Janet Leigh was not a brunette. Maybe it was the lighting, but I remembered her as a brunette in “Psycho.” But Google image searches confirm she was actually a blonde. However, I’ve seen more Grace Kelly movies, and I have long wished modern women’s fashion returned to her style, so I side with the AI in this one 🙂
Anyway, thanks for the link; I need to keep this tech on my radar.
Hitchcock blonde is an industry term apparently because he not only tended to cast blondes in all his movies (a legacy of the b&w era?) but a specific height and body type.
And yes, the tech can be improved but it is improving fast and is useful by now.
I’m a big fan of MidJourney and have dabbled with Stable Diffusion. There are several artists in both communities and many who relate hating the idea of AI generated art until they started using it as part of their creative process. Now, they’re big fans too.
I took a look at DreamUp’s examples and it’s pretty pathetic for now. I assume they will improve it. MidJourney is definitely my favorite, and the office hours the founder holds are amazing to listen to.