the use of generative AI does not automatically disqualify a piece from being considered art. but why not? i messily type in half word salad, half run-on sentence to a diffusion model, and i get back a high resolution image that might’ve taken a friend weeks. isn’t that cheap? unfair?
maybe. but the realism painters of the 1800s would likely have the same reaction to today’s cameras. completely unfair that, in my pocket, i have a device that can capture reality more clearly than their brushstrokes ever could. and yet we consider photographers to be artists. what are we missing?
one fear is that models obscure the amount of effort put into a given piece. it feels like a betrayal to zoom in on the hands and see the backwards knuckle. what you thought was a human’s handiwork was just matrix multiplications on silicon. is it then the toil that goes into a piece what imbues it with their memories, perspectives, subjective experience? we’re far removed from handprints on cave walls, but the viewer still wants to see the artist’s fingerprint.
but we’re looking in the wrong place. the fingerprint isn’t in the blood, sweat, and tears. someone who sucks at painting and had to put in more effort doesn’t automatically have a more ensouled piece. instead, its in each choice made in the art’s creation. the ones that carry history with them. waking up at 4am to capture the sunrise is just effort, but its the chasing after that specific hex of orange that reminds one of the kumquat tree in their front yard, the kraft mac and cheese, the logo of their alma mater. it has to be that one.
sometimes these connections can be traced. other times artists just know - it has to be cerulean and not cobalt. not being able to explain why doesn’t make the choice any less theirs. they don’t have to remember that it was the color of their childhood bedroom. what matters is matching the vision, even if it couldn’t be described until it was seen.
i’m designing the protagonist of a game i’m making. opening up chatgpt.com - i type in ‘cute anime girl’ and hit enter. what i initially get back is not art. still, even those three words carry faint traces - why cute and not pretty? why anime rather than realism? these preferences formed … somewhere. for now the ensoulment is minimal, generic. and each generation helps bring that subconscious preference into something more real and alive. finally, #257/500 sparks the vision like finding a perfect stick out in the woods. i didn’t carve it - but i recognized its beauty. the choice to take it with me is still a creative act. i refine the prompts further. brown hair, not blue. a backdrop of a cyberpunk city, not a fantasy forest. each prompt brings the idea in my head closer to being on my screen. she’s leaning on a ramen stand that reminds me of Tokyo-3. i open up photoshop to get her smirk to look just like my ex wife (she took the kids). each application of discrimination pulls more of ‘me’ into the final product. a faint fingerprint of preference starts to accumulate the weight of intention and vision.
photographers know that there’s more to a picture than clicking a button. you can see the yoga pose the photographer stood in to get everything in frame. the hours spent in after effects to capture one orange’s vibrance but another’s calmer warmth. why it fits just at the top-right corner of the collage. each revision carries with it more and more of the memory of each choice the artist made - more uniquely identifying than any biometric or signature.
the artist’s soul enters the work over time not through the medium, the tools, or the raw effort involved in generating the piece. but through each decision, each rejection of what could have been into what it must be, makes it uniquely theirs.
I created a vision of David in my mind and simply carved away everything that was not David - Michelangelo
TODOs:
- multiple focused pieces, or have more argument/detail about why model training/image generation is not ‘stealing’. claim here is roughly that if its not stealing to be inspired by a movie i watched as a kid and not cite it, then is it really stealing for a model to have been trained on some data if it can’t or doesn’t 1:1 replicate it. maybe training is more like subconscious inspiration. math is nature something something.
- possibly expand on depth of how diffusion models are not the end piece of the generated art. fine tuning your own model, pose control, all the other layers people add onto the initial generated artifact are similar to the depth of post-processing in other forms of art. curation in what is interesting + control
- strengthen the idea that art is meant to be not a binary measure. something can be more of the quality of art/ensouled than another thing, but there’s no tip between this is vs this isn’t art. emphasize the spectrum
- make more clear the claim that the initial generation does not have the same intentionality or discretion that artists have when applying their taste/discretion in creation, as they collapse the distribution of what the piece could be towards what the piece actually is. the choice of setting up the camera shot/etc is the important part more so than stumbling upon the naturally occurring lake/mountain/whatever
- consider replacing anime girl example entirely for something that might resonate better emotionally. perhaps the silliness of the anime girl example as a device for showing that this ‘unworthy’ use case actually has depth for it is lost in the fact that it turns off readers too easily. also i like the wife took the kids joke but unclear if it messes with overall tone
- point about perceived value with respect to effort? could make the connection about how art is more impressive if it was more difficult to create, as it represents being even closer to the artist’s vision. the big blue painting or whatever being more impressive bc of how much thought was put into making it as perfectly singularly blue as possible, with brushstroke control etc.
- potentially add in a point about iphones and how good image quality from phone cameras has made it easier for people to enter photography and has lowered the difficulty bar, and doesn’t mean they’re any less of an artist for their chosen technique. again there’s an element of spectrum of art here, in that randomly snapping a pic of my food doesn’t make me a food photographer, but more discretion can be applied towards the vision of how the food should look by being intentional about lighting/etc - even if the device is a phone.
- central claim about art/not art being reductive/needlessly binary, rather the application of fine grained control and choice is what makes a piece of work more artful/ensouled over time, as the piece is imbued with the artist’s unique vision and experience and perspective
- i should consider getting better at creating art before writing about creating art