6 ms·

I do work in tech. Almost 25 years. As a traditionally trained artist, my advantage over the years always was the knowledge of composition, color and my drawin

by nbzso 4y ago

I do work in tech. Almost 25 years. As a traditionally trained artist, my advantage over the years always was the knowledge of composition, color and my drawing skills when producing the visual part of the products. I focused early in my company (2005) on UX and UCD processes, and with the time realized that people don't care about craftsmanship or maximizing the quality. We are approaching a time of generative design. Personalization of UX by data based generative interfaces. So even the flow and journey mapping will be automated soon (UX researchers, hello). For me, this is the end of an era of innovation driven by human creativity. You will have a data scientist and secretary, which will prompt all day the "boss" vision. A.I. art is not art. I will give you a simple example: Who will be the creator of the famous David sculpture? Michelangelo or the rich patron who gave the description of the task? According to the definition of A.I. "art" the patron is the artist. So in this scenario, the art becomes a commodity generated without human touch and personal expression. Just a synthetic membrane of average results trained on stolen data and genius intellectual property laundering machine. Personally, my decision is to capitalize the remains of an industry and detach myself from any form of metaverse interactions. I started to paint recently again and don't give any F* about commercial viability of the process. If people of the future want to live in A.I. induced corporate coma, so be it.

jstummbillig 4y ago
> I started to paint recently again and don't give any F* about commercial viability of the process. That is an affordance a lot of people have to be granted now.
fintechie 4y ago
> A.I. art is not art. IMO you can certainly output art, code, etc from AIs. AI as a new kind of productivity tool, a very powerful one.
- nbzso 4y ago
  Just yesterday, I gave the ChatGPT prompts to generate Midjourney prompts. It transformed a simple sentence of "Portrait of beautiful woman on the beach with yellow dress." to the specific output of prompts with different lenses and lighting setups generated with astonishing results. My input in the process was minimal. No compositional skills or artistic taste. Powerful tool? Yes. Productivity is not an Artistic measure. Transitional value and personal expression is. But hey, be productive while you can.
raincole 4y ago
> A.I. art is not art. I will give you a simple example: Who will be the creator of the famous David sculpture? Michelangelo or the rich patron who gave the description of the task? According to the definition of A.I. "art" the patron is the artist. No? What are you even talking about? AI is not a person. In every single legal and ethicical discussion, "human" and "non-human" are treated completely differently. AI is a tool, while Michelangelo isn't. Who's the creator of David sculpture is completely unrelated to whether AI art is art or not.
precompute 4y ago
> A.I. art is not art. It isn't human-generated art. It would also be difficult to really label the generating process as a process that generates "art". I'd rather only label the end result as "art" for people that think it qualifies. It's the same as not considering photography as art. How many photographers actively contribute to camera technology? Or how many people that have worked on camera tech can profess to be actual photographers? Very, very few for either case. It's similar for prompting for an image. It's a "skill", the same way photography is a "skill". You'd be hard-pressed 20 years ago to take pictures of a quality consistent with a high-end DSLR today. Yet are people today "better" photographers just because they own better cameras? Or is an "AI Artist" a "better" AI artist because image generation got better, and / or he got a much better model (seeing how the quality of diffusion-based image generation wholly depends upon the model being used, and not so much upon prompt tuning)? > I will give you a simple example: Who will be the creator of the famous David sculpture? Michelangelo or the rich patron who gave the description of the task? According to the definition of A.I. "art" the patron is the artist. Yes, and I hope that changes, but it won't as long as there is monetary incentive for companies to shell out models. Genuine human contribution to the models will also decrease exponentially, because everything that could be mined has already been. They can make small gains with labelling but the rest is a hard ceiling. Attribution is very finicky right now because there's some effort involved in generating those images. If it was completely effortless (with a UI that labeled it as such) and gave sources / similar images, you could stop calling it "art", and instead just plain old machine-based image generation.
falsaberN1 4y ago
"Trained on stolen data" This is emotionally charged wording with a clear confrontational bias and a dark intent to discredit. As someone with over 30 years of experience with art and artists, do you realize how many influences are present in any given artist? Do you know how many artists use "reference images" heavily? Do you know how many artists trace images (specially in 3D modeling)? By your logic, every piece of art is stolen, if not from other artists, from reality itself. You are insulting every artist present in this forum with your bias. Thanks for that. Also, art made by commission or contract is technically property of the one who paid. Only the top recognized artists get to retain ownership of a specific piece or design done for a company, and even that is rare. How many cases are there where artists jumped ship off a property and needed to make lawyer-friendly pastiches of the characters, because they didn't own the rights? And, the Unet does not store images. This is not a photobasher nor a extreme compression algorithm. You'd know this if you had bothered to try. Which I have. I have trained it on my own artwork and in fact I maintain one of the largest guides about how to train Stable Diffusion using Low Rank Adaptation. You can't just "take out" an image in any exact capacity, and a lot of "proofs" are fabricated using img2img or models trained so aggressively on a single image that it's going to resemble it no matter what (and yet it'll never be reproduced 1:1). People fueled by strong emotions are not incapable of deceit. In fact, it's the stuff you MUST double-check, precisely because fueled by emotions. I'm so disappointed in "tech people" lately.
- nbzso 4y ago
  Sorry, the idea that artists are synthesizers of the past art is ludicrous. The human is not the machine. There are a billion factors for interpretation in creating a personal style. The A.I. image generators are just machines which produce average from all previous artists. They have no personality or agency (yet). > Stable Diffusion trained its algorithm on data sets collected by the German nonprofit LAION, which has collected billions of captioned images from art shopping sites and websites such as Pinterest. And it has done so without consent, causing artists and advocates to raise copyright concerns. https://bit.ly/3FU4OSa https://bit.ly/3FU4OSa
  - lowbloodsugar 4y ago

by nbzso 4y ago

jstummbillig 4y ago

> I started to paint recently again and don't give any F* about commercial viability of the process. That is an affordance a lot of people have to be granted now.

fintechie 4y ago

> A.I. art is not art. IMO you can certainly output art, code, etc from AIs. AI as a new kind of productivity tool, a very powerful one.

nbzso 4y ago
Just yesterday, I gave the ChatGPT prompts to generate Midjourney prompts. It transformed a simple sentence of "Portrait of beautiful woman on the beach with yellow dress." to the specific output of prompts with different lenses and lighting setups generated with astonishing results. My input in the process was minimal. No compositional skills or artistic taste. Powerful tool? Yes. Productivity is not an Artistic measure. Transitional value and personal expression is. But hey, be productive while you can.

raincole 4y ago

> A.I. art is not art. I will give you a simple example: Who will be the creator of the famous David sculpture? Michelangelo or the rich patron who gave the description of the task? According to the definition of A.I. "art" the patron is the artist. No? What are you even talking about? AI is not a person. In every single legal and ethicical discussion, "human" and "non-human" are treated completely differently. AI is a tool, while Michelangelo isn't. Who's the creator of David sculpture is completely unrelated to whether AI art is art or not.

precompute 4y ago

> A.I. art is not art. It isn't human-generated art. It would also be difficult to really label the generating process as a process that generates "art". I'd rather only label the end result as "art" for people that think it qualifies. It's the same as not considering photography as art. How many photographers actively contribute to camera technology? Or how many people that have worked on camera tech can profess to be actual photographers? Very, very few for either case. It's similar for prompting for an image. It's a "skill", the same way photography is a "skill". You'd be hard-pressed 20 years ago to take pictures of a quality consistent with a high-end DSLR today. Yet are people today "better" photographers just because they own better cameras? Or is an "AI Artist" a "better" AI artist because image generation got better, and / or he got a much better model (seeing how the quality of diffusion-based image generation wholly depends upon the model being used, and not so much upon prompt tuning)? > I will give you a simple example: Who will be the creator of the famous David sculpture? Michelangelo or the rich patron who gave the description of the task? According to the definition of A.I. "art" the patron is the artist. Yes, and I hope that changes, but it won't as long as there is monetary incentive for companies to shell out models. Genuine human contribution to the models will also decrease exponentially, because everything that could be mined has already been. They can make small gains with labelling but the rest is a hard ceiling. Attribution is very finicky right now because there's some effort involved in generating those images. If it was completely effortless (with a UI that labeled it as such) and gave sources / similar images, you could stop calling it "art", and instead just plain old machine-based image generation.

falsaberN1 4y ago

"Trained on stolen data" This is emotionally charged wording with a clear confrontational bias and a dark intent to discredit. As someone with over 30 years of experience with art and artists, do you realize how many influences are present in any given artist? Do you know how many artists use "reference images" heavily? Do you know how many artists trace images (specially in 3D modeling)? By your logic, every piece of art is stolen, if not from other artists, from reality itself. You are insulting every artist present in this forum with your bias. Thanks for that. Also, art made by commission or contract is technically property of the one who paid. Only the top recognized artists get to retain ownership of a specific piece or design done for a company, and even that is rare. How many cases are there where artists jumped ship off a property and needed to make lawyer-friendly pastiches of the characters, because they didn't own the rights? And, the Unet does not store images. This is not a photobasher nor a extreme compression algorithm. You'd know this if you had bothered to try. Which I have. I have trained it on my own artwork and in fact I maintain one of the largest guides about how to train Stable Diffusion using Low Rank Adaptation. You can't just "take out" an image in any exact capacity, and a lot of "proofs" are fabricated using img2img or models trained so aggressively on a single image that it's going to resemble it no matter what (and yet it'll never be reproduced 1:1). People fueled by strong emotions are not incapable of deceit. In fact, it's the stuff you MUST double-check, precisely because fueled by emotions. I'm so disappointed in "tech people" lately.

nbzso 4y ago
Sorry, the idea that artists are synthesizers of the past art is ludicrous. The human is not the machine. There are a billion factors for interpretation in creating a personal style. The A.I. image generators are just machines which produce average from all previous artists. They have no personality or agency (yet). > Stable Diffusion trained its algorithm on data sets collected by the German nonprofit LAION, which has collected billions of captioned images from art shopping sites and websites such as Pinterest. And it has done so without consent, causing artists and advocates to raise copyright concerns. https://bit.ly/3FU4OSa https://bit.ly/3FU4OSa
- lowbloodsugar 4y ago