Reply
  • AI art isn't inherently "stealing" art

    1st of all I find it funny people jump to hating it with that knowing nothing about how AI models are trained other than some theyve seen complain about it stemming from irrational fear
    The fact is it's trained on millions of data from real world sunsets or buildings to selfies to clothing to photos of cats and dogs. There are THOUSANDS of different methods using different data. Some don't even use ANY images from a human artist, some might use SYNTHETIC DATA, I've seen scenes created in UE5 game engine to generate new ARTIFICIAL data. Some may look at VIDEO and model the pixels as changes over time to learn consistency and shape. Some might focus more on building representations of TEXT data, some that may learn using parameters of coordinates/position of objects in 3d modeling software like Blender rather than looking at the pixels of 2d images, and so on.
    There are new papers getting published EVERYDAY for new methods that have nothing to do with the conspiracy of scraping your portfolio to make copies of it.
    Even beyond that, all these representations of the data are stored in vector spaces as numbers that have ABSTRACT relationships with eachother in a completely different way than we perceive as a 2d image. This is called a latent space and it's different from the concept of a person visually looking at an image to trace something from it.
    If you ask it for an astronaut as an oil painting, it doesn't just search for a preexisting drawing of that from an artist's work and then maybe photoshop a few differences and done. It extracts the concept of a real astronaut (not from any 1 particular image) and then extracts the concept of oil paint (not from any 1 particular image) and then infers things to create something that hasn't existed before.
    But let's just say every model is a copy and paste of non-consenting artists' hard work as you think it is. How many artists have you seen on social media posting anime versions of preexisting characters? How many times have you seen an artist that looks like a Picasso or a Rothko in the modern day? Haven't you heard of Picasso himself taking inspiration from African cubism that he saw?
    The fact is humans do this process you describe as stealing EVERYDAY. Artists don't magically come up with the art styles and concepts they want to draw when they were born, they learn through EMULATION of what they've seen and add onto it.
    Good artists use new mediums to find ways to do more with their art. Traditional artists trained on artists like Rembrandt complained about cartoonists, traditional animators complained when computers started to take away the art of drawing every frame manually, painters complained about people using photoshop to draw and edit digitally, and so on. I find that some people who generate art have a lot more agency or push boundaries more than some "traditional" artists complaining who repeat boring styles or concepts we've seen 100 times.
    An artist can even train one on their own early wireframes of body shapes or something to speed up a tedious part of their flow and then draw over it in the same way they previously worked. And in fact to get some of the GREAT ai art out there, you have to constantly rework things like this and do complicated iterations that is a skill in itself. Fun fact: EVERY person that makes CGI or 3d games started with free models you can download online, EVERY person that codes has used open-sourced software as a template to start, EVERY artist borrows from things that came from someone else's mind even if it's subconscious. But according to your logic if you tell a 5 year old to paint a forest with blue leaves, he's stealing the concept of leaves and trees and the color blue from God.

    Getting good art isn't just typing words and selling the 1st thing you get from it, at least for the more creative people not just using Midjourney to do minimal work and if it was that easy I'd love to see everyone in this thread do it and see the quality of their output bc there are AI artists that captured a unique style are selling thousands in NFTs daily and gaining audiences of hundreds of thousands so I'm sure everyone in this thread would quit to make money if it was so easy.

    For example the workflow for Stable Diffusion involves so many steps: LoRAs, using your GPU to finetune the model you're using, experimenting with CLIP Skip and different samplers and number of sample steps, hours of semantics to explore the representational vector space and tokenizing phrases to maybe get 1-2 great images in a FULL day of repeating this (I'm not talking about a s***ty generic fantasy girl with elf ears, I mean actual great award-winning photography that would otherwise not exist), using controlNET, downscaling and cropping the image then inpainting or outpainting whatever you got, inputting that into RealEsrgan or another post-processor, then working in photoshop to do what you need to do

  • 🤔🤔🤔

  • Rt

  • Nort 💫
    Mar 25
    ·
    1 reply

    ai art is poop bro

  • "I'm not talking about a s***ty generic fantasy girl with elf ears, I mean actual great award-winning photography that would otherwise not exist"

    bruh it's all the same s***

  • Mar 25
    ·
    3 replies

    The models are all trained on giant pools of data they don't own. It's pretty straight forward

  • its just one of those generic talking point liberals can blurt out to seem ethical and concerned when they really don't give a s*** at all

    social media is all optics

  • Rock Mudson

    The models are all trained on giant pools of data they don't own. It's pretty straight forward

    Yea but I think op is trying to say that
    AI art isn't inherently "stealing" art

    1st of all I find it funny people jump to hating it with that knowing nothing about how AI models are trained other than some theyve seen complain about it stemming from irrational fear
    The fact is it's trained on millions of data from real world sunsets or buildings to selfies to clothing to photos of cats and dogs. There are THOUSANDS of different methods using different data. Some don't even use ANY images from a human artist, some might use SYNTHETIC DATA, I've seen scenes created in UE5 game engine to generate new ARTIFICIAL data. Some may look at VIDEO and model the pixels as changes over time to learn consistency and shape. Some might focus more on building representations of TEXT data, some that may learn using parameters of coordinates/position of objects in 3d modeling software like Blender rather than looking at the pixels of 2d images, and so on.
    There are new papers getting published EVERYDAY for new methods that have nothing to do with the conspiracy of scraping your portfolio to make copies of it.
    Even beyond that, all these representations of the data are stored in vector spaces as numbers that have ABSTRACT relationships with eachother in a completely different way than we perceive as a 2d image. This is called a latent space and it's different from the concept of a person visually looking at an image to trace something from it.
    If you ask it for an astronaut as an oil painting, it doesn't just search for a preexisting drawing of that from an artist's work and then maybe photoshop a few differences and done. It extracts the concept of a real astronaut (not from any 1 particular image) and then extracts the concept of oil paint (not from any 1 particular image) and then infers things to create something that hasn't existed before.
    But let's just say every model is a copy and paste of non-consenting artists' hard work as you think it is. How many artists have you seen on social media posting anime versions of preexisting characters? How many times have you seen an artist that looks like a Picasso or a Rothko in the modern day? Haven't you heard of Picasso himself taking inspiration from African cubism that he saw?
    The fact is humans do this process you describe as stealing EVERYDAY. Artists don't magically come up with the art styles and concepts they want to draw when they were born, they learn through EMULATION of what they've seen and add onto it.
    Good artists use new mediums to find ways to do more with their art. Traditional artists trained on artists like Rembrandt complained about cartoonists, traditional animators complained when computers started to take away the art of drawing every frame manually, painters complained about people using photoshop to draw and edit digitally, and so on. I find that some people who generate art have a lot more agency or push boundaries more than some "traditional" artists complaining who repeat boring styles or concepts we've seen 100 times.
    An artist can even train one on their own early wireframes of body shapes or something to speed up a tedious part of their flow and then draw over it in the same way they previously worked. And in fact to get some of the GREAT ai art out there, you have to constantly rework things like this and do complicated iterations that is a skill in itself. Fun fact: EVERY person that makes CGI or 3d games started with free models you can download online, EVERY person that codes has used open-sourced software as a template to start, EVERY artist borrows from things that came from someone else's mind even if it's subconscious. But according to your logic if you tell a 5 year old to paint a forest with blue leaves, he's stealing the concept of leaves and trees and the color blue from God.

    Getting good art isn't just typing words and selling the 1st thing you get from it, at least for the more creative people not just using Midjourney to do minimal work and if it was that easy I'd love to see everyone in this thread do it and see the quality of their output bc there are AI artists that captured a unique style are selling thousands in NFTs daily and gaining audiences of hundreds of thousands so I'm sure everyone in this thread would quit to make money if it was so easy.

    For example the workflow for Stable Diffusion involves so many steps: LoRAs, using your GPU to finetune the model you're using, experimenting with CLIP Skip and different samplers and number of sample steps, hours of semantics to explore the representational vector space and tokenizing phrases to maybe get 1-2 great images in a FULL day of repeating this (I'm not talking about a s***ty generic fantasy girl with elf ears, I mean actual great award-winning photography that would otherwise not exist), using controlNET, downscaling and cropping the image then inpainting or outpainting whatever you got, inputting that into RealEsrgan or another post-processor, then working in photoshop to do what you need to do

  • Mar 25
    Rock Mudson

    The models are all trained on giant pools of data they don't own. It's pretty straight forward

    Everyone is ignoring this part. None of this would be an issue if it was trained on public domain s***. There's substantial evidence it is trained on heavy copyrighted work lol.

  • Mar 25

    this rethard again

  • Mar 25
    ·
    1 reply
    Rock Mudson

    The models are all trained on giant pools of data they don't own. It's pretty straight forward

    so is your brain my guy
    so are the algorithms implemented in all the social media you use and you don't see ppl complaining nearly as much about

    but anyways if you actually read the post, with models like Sora openai.com/sora
    and semi-photorealistic game engines like Unreal Engine 5

    models are now being trained with SYNTHETIC data (human-created images/video that didn't exist before). Every model is different.

    And to your point, most of them are actually "ethical" and use public data provided by Gettyimages etc. People just make up things to complain about. It just stems from artists that can't accept change

  • Idk bruh if you know people are out here hustling creating art and you choose AI over collabing with other humans who can probably make something better, that’s pretty wack

  • nicobarret

    so is your brain my guy
    so are the algorithms implemented in all the social media you use and you don't see ppl complaining nearly as much about

    but anyways if you actually read the post, with models like Sora https://openai.com/sora
    and semi-photorealistic game engines like Unreal Engine 5

    models are now being trained with SYNTHETIC data (human-created images/video that didn't exist before). Every model is different.

    And to your point, most of them are actually "ethical" and use public data provided by Gettyimages etc. People just make up things to complain about. It just stems from artists that can't accept change

    The technology itself was developed on scraped data they didn't own the rights to brother. Sure, now that you have infinite VC you can use other sources lol.

  • Mar 25
    ·
    edited

    Yeah i agree with you. Nothing in art is original, everything is built on the shoulders of giants. It’s no different than math in that way, where we all build and learn from eachother. Everything artists use to create is just a tool, whether its gesture, form, composition etc. AI is just another tool to add to the arsenal that can be used or ignored like any tool. Is Ai generated art really art? It’s debatable, because the definition of art is the expression of people and the expression of a computer isnt from a person. But the programs were written by people. So it’s technically their self expression if that makes sense. I think the problem with AI is going to be the growing pains from the in between when people loose their jobs at major film and game studios, before a group of a handful of people can use it to make a movie or game that competes with modern AAA movies and games. Once it’s used in that way and the major film and game studios are altered permanently or fail then we will see a lot of the controversy die down. The growing pains are going to be tough though. I think other fields outside of art is more threatened by AI, because AI will replace a lot of jobs, but artists will end up in a small group of people who wont ever really be replaced and their manual labour will always be valued (whether they use some AI in their work or not).

  • While I appreciate the detailed explanation of how AI art is created and the effort put into its development, I disagree with the notion that AI art isn't inherently "stealing" art. The process you described involves training AI on vast amounts of existing art and data, which inherently captures and learns from the intellectual and creative efforts of human artists.

    While humans may take inspiration from existing art or styles, there's a difference between inspiration and replication. AI doesn't merely emulate; it replicates and generates new works based on learned patterns. This can undermine the originality and uniqueness of human-created art, potentially devaluing the artistic process and the emotional and intellectual investment that artists put into their work.

    Furthermore, the argument that humans have always borrowed or emulated artistic styles throughout history doesn't justify AI's replication of art. Traditional artists often pay homage or reinterpret existing styles, but they bring their own unique perspective, emotion, and creativity to their work. AI lacks this intrinsic human touch and understanding, which is crucial in creating truly meaningful and authentic art.

    Lastly, while the technical complexity and effort involved in creating AI art are commendable, it doesn't necessarily equate to artistic merit or value. Art is not just about the end result but also the process, intention, and emotional connection it evokes. While AI-generated art may produce visually appealing or novel images, it often lacks the depth, meaning, and soul that human artists infuse into their creations.

  • Nort

    ai art is poop bro

  • written by ChatGPT

  • Literally 99% of AI is the most bland and uninspired s*** ever

  • Nice try ChatGPT