Intro to the AI-image generation debate
We have been doing a bunch of talks about how AI image generators make pictures from text, trying out some different android apps that offered free text to ai image generation and a couple of different articles about this niche. It was just something that interested us and seems to be just the beginning of the questions that are started to get stirred up. If AI takes ques to make art… from other artists… who is the artist? Who had the idea to generate that art? If a computer copies a humans work, should the human get credit for it?
First things first, how does AI generate art?
It is pretty easy honestly, and we were shocked by how it could be easily run on a phone without any type of skill or knowledge more than how to enter text. Once you install the app the user simply enters the search terms they are looking for into a search box, and these prompts can be used to describe what they are looking for. It functions similarly to Google Images search except that the search terms are replaced with brand-new artworks.
But what if you add an artists work to the filter?
This is where it gets weird because now that the AI neural networks are expanding they are adding more and more work. So you can actually go and find the style of your favorite artists and you can add it to the filter. You can use the artist’s name to create something similar to an artist’s style. So at this point, who is making the art? Is it me, because I am telling the program what to generate? Or, because the artists made the base of the image generation, is it their work and inspiration that creates this output?
A boom in imitation of art, or is it copying?
Image generators that users are becoming increasingly popular among consumer, because they are super easy to use and make nice outputs for in seconds. Think about that, you have a random idea and you can type in some words. Years of school to learn theory and how to paint… not needed at all…. Supplies… nope? So what is the barrier to entry? In this example you need the AI system… and the human work to base the imitation off of.
OpenAI, which Elon Musk started in 2015, opened its DALL_E generator to the public in September. OpenAI claimed that it had more than 1.5 million users when the waitlist was lifted.
The Graphic Artist Guild’s president is Liz DiFiore. This group supports illustrators, designers, and photographers across the US. She stated that AI’s ability to copy styles quickly could cause financial problems for artists.
She stated that artists spend a lot of time on their careers and make a lot of money by being able to license their images.
“So, an AI can copy an artist’s look, and a company can generate an image that looks like the artist’s, but without actually paying artists for their art, that could pose a problem. “
The new issue: Artists are prohibited from copying their original artworks.
DALL-E, Mid journey, and others are two of the most popular AI-image generators. Their policies prohibit users from using their products in specific ways, and OpenAI, for example, forbids images of celebrities or politicians.
These programs block users from creating “harmful content” by filtering for things like nudity and gore.
Stable Diffusion representatives stated that they were developing an opt-out program to allow artists who don’t want to use AI programs to learn about their work.
A spokesperson stated that artists’ names are only one component of an AI model’s instructions that creates a unique style for each artist.
Open AI representatives didn’t give any details about protections for artists living in the public domain. Still, they said the company would seek artists’ opinions as it expands DALLE access.
AI data Training
AI-image generators can learn from large quantities of images and captions to feed their neural networks. OpenAI representatives said that DALL-E’s training dataset is a mixture of images from public sources and images licensed by the company.
Stable Diffusion representatives stated that the program uses web crawls for images and information. RJ Palmer, another illustrator and designer, was also “anti-artist” on Twitter. He stated they had been “explicitly trained by current working artists.” “
Artists can verify that their work has been used to train AI programs through a website called “Have You Been Trained,” created by Mat Dryhurst (a German sound artist) and Holly Herndon (an American sound artist).
They’ve been working on tools artists can use to opt-out AI training sets. To locate programs, this website searches approximately 5.8 million images from the Mid journey and Stable Diffusion datasets.
Is consent required by artists?
Some people feel that artists should have given their consent to be scanned to create data for AI generators. But that begs the question, should someone say that they don’t want their stuff copied? What about artists that have passed away? Can Picaso restrict consent, or is it fair game because he is dead? It might almost make the most sense that there has to be consent granted to allow AI to learn from the art they find, versus having artists state they don’t want their art used in the training of neural networks.
Copyright laws around AI images are unclear.
It’s not clear if copyright laws will cover AI programs. Getty Images and other stock-image collections have opposed the inclusion of AI-generated art due to uncertainty over copyright. An insider from the US Copyright Office was told by a spokesperson that works created using artificial intelligence do not have the authorship necessary to support a copyright claim. It is unclear if someone using search prompts for an AI art piece to be used in a program is a human-AI collaboration.
Representatives of Stable Diffusion said that images could be used in commercial offerings. They were not able to say whether the photos were copyrightable. This decision would be made at the national legislative level. Representatives from OpenAI stated that they believed images created by their programs could be copied for commercial use. OpenAI spokesperson stated that images could be copied if DALL-E was used to aid human creativity. DALL-E users have the right to commercialize and distribute ideas they create as long as they follow our content policy.
They stated that copyright laws have been adapted to new technologies in the past and would need to adapt to AI-generated content. Despite their reservations, many artists are excited about this technology’s potential. Giles Christopher is a London commercial photographer who specializes in food and drinks. DALL-E, as well as other AI-image generators, is used to create artificial backgrounds for commercial shots. He also experiments with portraits.
Our final thoughts, for now, on AI generated content and copying artists works to create it
The term “copied” is often used in a negative light when discussing AI-created art, as it implies that the AI is simply imitating the work of human artists rather than creating something new. However, it is important to remember that all art is ultimately derivative, and that even the most original artist is drawing on the work of those who came before them.
When we talk about AI created art, we are talking about art that is created by algorithms or computer programs. These programs are designed to mimic the way humans create art. The programs are fed data from sources such as images, videos, and text. They then use this data to create new artworks.
Some people argue that AI created art is simply a copy of existing artwork.