Posted on November 5, 2022 Posted by John Scalzi 20 Comments
I’ve been playing with the AI art generator Stable Diffusion, and one of the things it lets you do is use a photo for reference. So I went ahead and popped in a couple of pictures of Krissy to see what would come up, and as a text prompt used various iterations of “queen” and other references. The pictures that came out don’t look like Krissy (nor did I expect them to, I had the setting for fidelity to the originating photo set to low), but they do look cool. What they mostly do is make me wonder which artists the AI was trained on, and if I can hire them to do some work for me.
Because, of course, that is the thing: These AI generators have been trained on various artists, many if not most still alive and producing work. I’m happy, for the purposes of my own amusement, to play with these art robots and see what comes out, and show off the results on my non-commercial outlets. But when it comes time to commission art for paid work, or for the house, or wherever, it’ll also be time for me to pay up for actual living humans making art. Support actual humans, folks! You might be one yourself, after all.
This is something that I’m not sure I fully understand, although I do support the idea of supporting artists. At the same time, I’m an amateur musician. I’ve never made any money off my music, but suppose one day I do. My own personal neural net (i.e., my brain) has been trained, so to speak, on all the music I’ve heard over my lifetime. I certainly have been influenced by certain creators. If I make money from my own original compositions, but realize that they were influenced by earlier artists, do I own those artists a cut of what I make? I might be standing on the shoulders of giants, but how much do I own them financially just for providing me with inspiration. How is this different than if an AI produces something original, but influenced by other artists, which we know to be the case? This isn’t meant to be a troll post, I honestly am not sure how to separate an AI learning from a body of work vs me learning from a body of work, why one is different from the other. How much does John Scalzi owe the estate of Robert Heinlein?
John, I appreciate your sentiments towards your fellow humans. But I’m not afraid of these apps.
I’m too old to care.
Self publishing writers need nice covers, and they aren’t getting them. Maybe this or one of the many self-generating-image (don’t know what they’re called) programs can help.
Art going truly from the soul into one’s hand and onto a canvas will always have intimacy.
Children really should be taught how to draw as a basic skill in school. It improves the mind, makes you a better observer, and improves the imagination. You’re really missing out on the creative process if you need an app to make a decent picture.
The joy of starting from nothing and making something from it with your own hands will never go out of fashion.
Seeing Scott’s comment: Copyright and trademark law has addressed your question. A gray area remains. Not a law on the books, is a requirement to credit someone’s work you use, post, or make a pastiche from. I think you should, even when the art is in the public domain. It’s the least we can do for the long dead who continue to enrich our lives.
That last picture looks how I image Princess Irulan looked when I first read Dune, years ago.
Neil Clarke says many of the AIs use Clarkesworld artists.
The first pic looks like it was trained on Angelina Jolie…. Cool images otherwise.
I know it’s AI, but they are beautiful.
They are all great images. I like the second one most as a picture of a queen.
What does Krissy think of these images?
I grew up in NYC. I saw the title of the post and wondered what you had to say about the borough.
Lovely pics, but if you want an actual living breathing artist, I can highly recommend Dean Williams https://www.facebook.com/TheArtOfDean/
He lives in Inglewood CA, and has done several commissions for us for mystery novel covers, he’s a great deal of fun to work with.
That first one will probably be a cover on one of your next books.
Wow! These AI generated art is amazing, but I’m pleased that you will still be supporting real human artists for your book art.
There was a case years ago where a … gorilla? I tbink? Painted some art and someone tried to copyright it. And the supreme court said, nope, copyright is for human-created works only, no copyright for non homosapiens.
Personally, i think its a good ruling because it also prevents soneone from writing a script that just cranks out every chord progression and trying to copyright everything.
Dont know if that ruling still applies or not, but i would think it should apply to AI art. No doubt, deep pockets behind ai might be able to get that law/ruling changed, and that would be scary. The whole point of ai is automation, and decentralization. Ai is kinda expensive right now and if one company gats to have ai make all the art and copuright it, that would suck.
If ai does decentralize this, then the idea of art and artist is going to be radically different in 10 years.
The transistion that puts people out of work will suck, but the eventual result is that anyone could type in a few words and get free, bespoke art. Thats an improvement.
That first picture is clearly Kiva Lagos!
One of the big challenges with the view of “AI as democratization” is the risk of stagnation of the public domain. And, eventually, of the very art used to train these neural nets.
By automating the “easy” classes of art, we deprive beginning artists of the commisions they could get (because, being beginners, nobody will turn to them for the “complex” pieces). After a generation, we would have almost no professional artists – as all the could-be-artists have turned to some other pursuit to bring bread to the table (and, at most, dabble in art as a hobby).
Alon: “One of the big challenges with the view of “AI as democratization” is the risk of stagnation of the public domain.”
Corporate bribery of politicians extending copyright from a reasonable 40 years to the current insanity that wont be happy until terms are at least five centuries has been devastating for the public domain.
The idea that ai will rob would be artists of basic art practice sounds a lot like older people complaining that drum machines and samplers arent real music. But daft punk is in a genre of music impossible to create just playing drums, piano, and guitar.
Johannes Vermeer used camera obscura technology to create lifelike art. Should he have practiced painting freehand-only instead?
Punk was created in part in response to dinosaur rock and other genres based on people spending a lifetime practicing and learning music, something punk artists couldnt do, but still had something to say that wasnt being expressed anywhere else in any other musical genre, so punk created its own genre of stripped down music.
AI could be the camera obscura for the next artist, AI could enable the Daft Punk of visual artists, AI could enable a punk visual art genre where people have a message they want to convey but dont have the visual arts training to draw it themselves.
There are a lot of things about AI that are problematic: putting unconscious bias into the data reinforces that bias, putting ai developed by libertarian fools into the driver seat of cars could get a lot of people killed, and so on. But that doesnt mean we need to fear all AI.
Its kinda mind boggling that sci fi, as a genre, who prides itself in being open to new ideas, exploring new frontiers, has, since Mary Shelley’s Frankenstein, settled into a cave-man science fiction approach to AI, where the moral of the story is “me go to far!”, ai always violates the 3 laws, terminators hunt mankind, and the singularity is always feared.
This is ai making art, ffs. It could be a new tool that enables the creation of many new venres of art that never existed before, people creating works who couldnt before. And the standard refrain in response to this echos the South Park episode line of “they took our jobs!”
Kiva Lagos’ ancestry is West African (the last name is a hint), so, not really, no.
I would really like to know more about the text strings you used for these. Specifically I wonder if you explicitly asked for “shadow” or “twilight” or “chiaroscuro” — because they all have that quality.
Oh… Ah… Um… I’ll get my coat.
(Not even in a general sense of badassery suggested by the picture?)
The alternative, of course, is to split the revenue from the new tool 50/50 with artists — stick the artist’s 50% portion in a sovereign wealth fund invested in the broader economy, get the machines to cite their damn sources as we demand of every college freshman, and then pay out proportionate royalties to every artist cited from the earnings of said fund every quarter.
Elinor Ostrom said a thing or two about common pool resources as applied to non-natural resources. Alaska’s a good model.
Petition to this effect.