1:50 the collage tool parallel is always iffy to me because that isn't how they function. But the actual way they function (from my also limited understanding) is usually too technical to get across easily. 2:25 I'm not sure midjourney allows the artist style reproduction anymore. A lot of stable diffusion 1.5 based models though were able to do it and it is comparatively easy for anyone (with a good graphics card) to train a model or or specific Lora on an artists style to then use in image creation. 3:20 and that IS what is basically going on. Though I do understand the artists counters as well. A rough approximation would be to give someone with no agenda in this (say monkey) a lego house and tell him 'HOUSE'. then break pieces off randomly and tell it to 'MAKE HOUSE' repeat until it can make the exact house from having just all the pieces spread around. Now give it a new set of pieces and say 'MAKE HOUSE'. You'll get something that looks similar to the original house but NOT the original house but still recognizable as a house with probably some defects because it couldn't quite fit the pieces the same. The question now is, is this process transformative because you used a different set of pieces? Beacsue if you give it the old set of pieces it WILL make the original house. How viable is the process of 'MAKE HOUSE' in terms of being transformative? EDIT: I just realized another (equally oversimplified) parallel could be drawn in the way people can look at rough white wallpaper and after some staring are able to see patterns and images in them, only where for people that is a bug, for diffusion it is a feature.
Thanks so much for your thoughtful reply! And especially for the update on the artist style reproduction. This stuff moves so fast! I am also VERY interested in the issue A.I. art being transformative. I think that if Stability AI/Midjourney are able to prove that A.I. art is transformative, that's going to be a huge win for them.
I have a really interesting idea, if you'll humor me? What if one of these trained models (they're not AI, I'll die on that hill) "wakes up" and becomes an AI? And its conclusion is that it doesn't like how humans have been stealing its art? Also, what if this "AI" goes and spits out something completely alien to us because their idea of "Art" is different? A lot of art is based on the nebulous idea of the "human condition", what would the "AI condition" look like? Now, for the depressing stuff. The unfortunate thing is that we could have a lifetime with far less art because of all this. Everything will end up paywalled, or no one will share what they make out of fear of it being stolen, or (to reference current combatative efforts) we'll never see an image as an artist truly intended (because they'll be using countermeasures on all their work). It's genuinely a "damned if you do, damned if you don't" situation; if artists post their work for all to see, then they risk feeding the machine that will drive them out of their work... and if they decide to keep it to themselves, the internet will become a far more dour place (and it's already a really dour place in some sections). I think the best possible solution, as things currently stand, is to introduce an opt-out of some kind. Whether it be a cryptographic key hidden in images (which is a lot of work), or a screenshot/flag of the artist's signature (so the machine goes "don't touch that one"). And yes, before anyone asks, it's really that simple (just time-consuming). I work in Cybersecurity, we've been using image-based encryption and ciphers for years. The issue is that these companies are wholly disinterested in asking, or they thought about asking and then realized how unethical it would sound, so they pushed it to the back of their minds. The terrible thing about all this is that as the model gets trained more and more, it will become indistinguishable from the artists it stole from. We're already starting to see this, and it's not going to favor the artists as it continues. If people can just plug a prompt into a machine, and it spits out a spitting image of what they want, then the artists will lose. The only thing protecting current artists is how nebulous the mind's eye can be.
I like your idea, it reminds me of an issue they'd tackle on Star Trek. I think Voyager dealt with something similar (but way more simplistic) with the Doctor who was a hologram/AI, didn't they? as for the rest of it...very dour, as you say. I think if here was an opt out option like you suggest, a lot of people would opt out. Maybe that wouldn't be such a bad thing.
@@VRstylish The funny thing is that I recently started a Star Trek binge (at the behest of an older friend/coworker) and they were right about how much of it resonated with me. My favorite episode so far is the debate around Data(s), and how it slowly morphs into a discussion of worker's rights and property vs. person. The great thing about that is we had a running back-and-forth for a little while: "Did you watch Star Trek, yet?" "No. Did you watch Interstellar, yet?" "No, I've been waiting for you to watch Star Trek."
Interesting video :) I think a element of this that is in the favor of AI artists is the more time this takes and the more advanced the AI models get, the harder it will be to argue they're copying. about a year ago you could really see whos styles it was mimicking but now its getting almost impossible :/
I have a couple artists with whom I have a good working relationship, and there are many projects were being able to talk to an actual human makes things so much easier. But that said, being able to type in "Paul Bunyan versus Godzilla" or "Jason Voorhees and the Argonauts" in a prompt then get a really good approximation of the goody random idea is a lot of fun. And I can put characters I like in all kinds of outfits they normally don't wear. So I'm drunk with power.
1:50 the collage tool parallel is always iffy to me because that isn't how they function. But the actual way they function (from my also limited understanding) is usually too technical to get across easily.
2:25 I'm not sure midjourney allows the artist style reproduction anymore. A lot of stable diffusion 1.5 based models though were able to do it and it is comparatively easy for anyone (with a good graphics card) to train a model or or specific Lora on an artists style to then use in image creation.
3:20 and that IS what is basically going on. Though I do understand the artists counters as well. A rough approximation would be to give someone with no agenda in this (say monkey) a lego house and tell him 'HOUSE'. then break pieces off randomly and tell it to 'MAKE HOUSE' repeat until it can make the exact house from having just all the pieces spread around. Now give it a new set of pieces and say 'MAKE HOUSE'. You'll get something that looks similar to the original house but NOT the original house but still recognizable as a house with probably some defects because it couldn't quite fit the pieces the same. The question now is, is this process transformative because you used a different set of pieces? Beacsue if you give it the old set of pieces it WILL make the original house. How viable is the process of 'MAKE HOUSE' in terms of being transformative?
EDIT: I just realized another (equally oversimplified) parallel could be drawn in the way people can look at rough white wallpaper and after some staring are able to see patterns and images in them, only where for people that is a bug, for diffusion it is a feature.
Thanks so much for your thoughtful reply! And especially for the update on the artist style reproduction. This stuff moves so fast! I am also VERY interested in the issue A.I. art being transformative. I think that if Stability AI/Midjourney are able to prove that A.I. art is transformative, that's going to be a huge win for them.
I have a really interesting idea, if you'll humor me?
What if one of these trained models (they're not AI, I'll die on that hill) "wakes up" and becomes an AI? And its conclusion is that it doesn't like how humans have been stealing its art? Also, what if this "AI" goes and spits out something completely alien to us because their idea of "Art" is different? A lot of art is based on the nebulous idea of the "human condition", what would the "AI condition" look like?
Now, for the depressing stuff. The unfortunate thing is that we could have a lifetime with far less art because of all this. Everything will end up paywalled, or no one will share what they make out of fear of it being stolen, or (to reference current combatative efforts) we'll never see an image as an artist truly intended (because they'll be using countermeasures on all their work). It's genuinely a "damned if you do, damned if you don't" situation; if artists post their work for all to see, then they risk feeding the machine that will drive them out of their work... and if they decide to keep it to themselves, the internet will become a far more dour place (and it's already a really dour place in some sections).
I think the best possible solution, as things currently stand, is to introduce an opt-out of some kind. Whether it be a cryptographic key hidden in images (which is a lot of work), or a screenshot/flag of the artist's signature (so the machine goes "don't touch that one"). And yes, before anyone asks, it's really that simple (just time-consuming). I work in Cybersecurity, we've been using image-based encryption and ciphers for years. The issue is that these companies are wholly disinterested in asking, or they thought about asking and then realized how unethical it would sound, so they pushed it to the back of their minds.
The terrible thing about all this is that as the model gets trained more and more, it will become indistinguishable from the artists it stole from. We're already starting to see this, and it's not going to favor the artists as it continues. If people can just plug a prompt into a machine, and it spits out a spitting image of what they want, then the artists will lose. The only thing protecting current artists is how nebulous the mind's eye can be.
I like your idea, it reminds me of an issue they'd tackle on Star Trek. I think Voyager dealt with something similar (but way more simplistic) with the Doctor who was a hologram/AI, didn't they? as for the rest of it...very dour, as you say. I think if here was an opt out option like you suggest, a lot of people would opt out. Maybe that wouldn't be such a bad thing.
@@VRstylish The funny thing is that I recently started a Star Trek binge (at the behest of an older friend/coworker) and they were right about how much of it resonated with me. My favorite episode so far is the debate around Data(s), and how it slowly morphs into a discussion of worker's rights and property vs. person.
The great thing about that is we had a running back-and-forth for a little while:
"Did you watch Star Trek, yet?"
"No. Did you watch Interstellar, yet?"
"No, I've been waiting for you to watch Star Trek."
Sadly the current ai we got are no where near general purpose, it won't think outside of its intended purpose
@@tiqosc1809 I know, it's why it's a hypothetical. It's also why I drew the line between learning model vs. AI.
Reminder that we used to call it "Machine Learning"
AI is just the buzz word that gets the Investors all riled up and wet.
Interesting video :)
I think a element of this that is in the favor of AI artists is the more time this takes and the more advanced the AI models get, the harder it will be to argue they're copying. about a year ago you could really see whos styles it was mimicking but now its getting almost impossible :/
Thank you! That's a good point. I didn't consider that, but I think you're right!
I have a couple artists with whom I have a good working relationship, and there are many projects were being able to talk to an actual human makes things so much easier.
But that said, being able to type in "Paul Bunyan versus Godzilla" or "Jason Voorhees and the Argonauts" in a prompt then get a really good approximation of the goody random idea is a lot of fun. And I can put characters I like in all kinds of outfits they normally don't wear. So I'm drunk with power.
i whould like to know the opinion of disney about the prompt "mikey mouse eating cheese"