Jeezits

"Love your work - Big Fan" — One of my biggest fans, no doubt

Why am I opposed to Stable Diffusion and AI Art in its current incarnation:

Some people seem to believe AI can “learn” art. Like it learns the concepts of perspective, value, anatomy, colour etc. through images and then recreates art based on this knowledge.

This is a misconception.

An AI doesn’t “know” things. It has no concept for artistic fundamentals. It just learns associations based on the data it’s given in a way that’s completely , vastly different to the way a human brain does. An AI can only recreate based on known image data. Those recreations can be blended in a very complex way, but they will ALWAYS be derived directly from image data it’s trained on.

A human can take a paint-bucket, and throw it at a canvas, and then mush paint around with their fingers. An AI can’t do that;  it can only blend image data that best fits “canvas with messy splashed paint”. It will pull from all the image data it’s categorized with “canvas”, “splash”, “paint”, etc. and then blend them by placing datapoints next to other datapoints that it has “learned” will most suitably go next to each other.

Human learning creates complex conceptual structures. Our concept of an “apple” may contain many elements such as the colour red, how heavy it is, its overall form, how you hold it, and what it tastes like. An AI’s concept of an “apple” is whatever images it associates with the word “apple” based on text cues in its training.

When you tell it to paint “a hand holding an apple”, it will recreate and blend many images of hands with many images of apples in a way that best fit each other depending on weights defined by the data its analyzed.

Any presumption that AI can “learn” art theory and then make art through its knowledge of this is incorrect, and would require a level of general AI intelligence we are nowhere near capable of building yet, and we won’t with our current models because they are not creating actual epistemology, merely datapoint-based imitation without actual integration or understanding.

But the bottom line? All AI art is derivative and, unless it was trained exclusively on works in the public domain, there is definitely a case to argue that the companies creating these algorithms are violating the copyright of the artists whose works they are using without there first being a contract, agreement, or royalties (which is the thing these companies are trying to weasel out of by creating these AIs in the first place.) There is a reason why Clip Studio Paint, the latest of money-chasers jumping aboard the Stable Diffusion pony, issued out a warning that states, verbatim, “we cannot guarantee that images generated by the current model will not infringe on the rights of others.“

If I take two storebought cakes and put one on top of the other, I am not a baker or cake decorator.

If you type in a few words and click a button and get an AI to create an image for you that is nothing but a pixel-mashed reimagining of other images you didn't take, you are not an artist.

Yes, AI art is fun. Yes, AI art has very real-world possibilities in terms of referencing, inspiration, ect. But you are not an artist for literally nothing nothing more than prompting an AI to create something. That's a ridiculous and frankly deeply insulting statement to make.

No, you are not an artist if you literally do nothing but sit there manually prompting an AI to create something. You're just not.

Speaking as if coding and configurating an AI to make hard is easy I see

"Literally nothing more than prompting an AI to create something"

I... look I just woke up and my brain is scrambled but. Lol kinda feels elitist and lowkey ableism towards people who might make AI art because they cannot work with a pen for some health reasons.

I'm a digital artist myself and I don't think it's insulting. My best friend is an artist and he doesn't think it's insulting.

Fun fact: photography was not seen as 'real art' in the very beginning of it because people vould just say "you're just pushing a button". Sounds familiar?

"Making reimaginated images of an image you didn't take/create isn't art." Wow. I guess fanart is not art then. Or remixes / mashups in music. Or synth music because yknow these are not real instruments after all.

Oh also, you know what "just pushing a button isn't art" reminds me of? All the people saying contemporary art isn't art. Pollock? "Oh he's just making paind lines on a canvas even a toddler could do that".

Prompting an AI isn't just giving it words, it's finding the specific words to be able to get the right result and choosing appropriate image banks yadda yadda

If you use AI, you’re not an artist,

No matter how much you cry and squeal,

Over how you didn’t steal,

If you use AI, you’re not an artist.

Typing a prompt will never be art,

Typing a prompt will never be art,

You have failed the definition,

All you made was a commission,

Typing a prompt will never be art.

It doesn’t fucking matter how it works,

It doesn’t fucking matter how it works,

No one asked to be explained,

How the data set was trained,

It doesn’t fucking matter how it works.

Because no artist gave consent,

Because no artist gave consent,

No one asked for their permission,

Just plowed ahead with greedy ambition,

Not a single fucking artist gave consent.

Claiming you’re disabled is not an excuse,

Actually being disabled; still not an excuse,  

If writing a prompt is what it took,

*** Then you can write a fucking book,***

I’m sorry if you’re disabled, it’s not an excuse.

Using progressive language doesn’t work,

Using progressive language doesn’t work,

You can quote from Lenin and from Marx,

Your pics still have artist watermarks,

Using progressive language doesn’t work.

Besides we’ve already seen your NFTs,

Oh yes we’ve already seen your NFTs,

All over Twitter and Instagram,

You were neck-deep in that scam,

Also, did you wait for me to reblog art from this AI artist to actually reply? Sounds awfully petty from you

And just like photography and film, we shouldn't film or take pictures of things that the owners of those things don't want in films or pictures. The same is true of putting something in your AI training data. I mean if you fundamentally disagree that film and photography can be art, then nothing is going to convince you one way or the other. But this EXACT same rhetoric was being shouted to the rooftops when cameras were invented.

You’ve been here five fucking minutes because the usual place you huckster assholes use to push your bullshit, Twitter, is failing. Guaranteed you’re fresh off the NFT scam, and this is just the next grift you hopped to.

you lying pieces of shit have been stringing one line of bullshit after another over this, and yet you still can’t get around the fact that typing a prompt isn’t work in the very terms and standards of every single goddamn art medium you try and compare it to.

I SUPPORT AI ARTISTS.

I SUPPORT AI ART TOOLS BEING TRAINED WITH FREE, NOT STOLEN IMAGES.

I DO NOT SUPPORT AI ART BEING MADE FROM AI TRAINED ON EXISTING ARTWORKS FROM "REAL HUMAN BEINGS" WHO REFUSED TO HAVE THEIR ART USED LIKE THIS.

I BELIEVE AI ART IS A VALID FORM OF ART. MANY ARTISTS ACTUALLY USE AI FOR CERTAIN WORKS.

YOU DO NOT NEED TO BE SOME KIND OF TORTURED SOUL WITH WICKED CREATIVITY AND IMAGINATION TO BE A REAL ARTIST.

YOU DO NOT NEED TO GIVE PURPOSE OR SIGNIFICATION TO YOUR ART TO BE A REAL ARTIST.

THINKING ART IS SOME KIND OF MEDIA RESTRICTED TO PEOPLE WHO ARE ABLE TO DRAW/PAINT EXTREMELY WELL, HAVE MESSAGES TO SHARE OR ALL THAT KIND OF STUFF IS LAME.

ART IS SUPPOSED TO BE FUN.

PLEASE SUPPORT AI ARTISTS. DO NOT SUPPORT AI ART MADE WITH STOLEN IMAGES. DO SUPPORT AI ART AS ANOTHER FORM OF ARTISTIC EXPRESSION.

BE ANGRY AT BIG CORPORATIONS MAKING QUICK AI ART APPS FOR EASY MONEY, NOT AT INNOCENT AI ARTISTS.

AI image generation is a process where computers are able to create images based on certain information and data. One way this is done is through text-to-media AI, which involves feeding machine-learning algorithms a large number of image-text pairs and allowing them to learn how to connect them. This creates a complex knowledge base that is not directly understandable to humans, but can be navigated through the use of prompts.

While text-to-media AI can mimic an artist's style and create images that look similar to their work, it does not copy their work exactly. The way text-to-media AI works is by learning from a large number of image-text pairs and creating a database of knowledge about different images and what they look like. When given a prompt, the computer uses this knowledge to generate an image that it thinks best matches the prompt. However, because the computer is not directly copying an existing image, but rather using its own knowledge and understanding to generate a new image, it does not replicate the work exactly.

Some people may worry that using text-to-media AI could be considered stealing from artists, as it is able to generate images that look similar to their work. However, existing copyright law does not consider the use of text-to-media AI to be theft. This can leave artists feeling vulnerable and raises questions about the ethical implications of using AI in the arts.

To address these concerns, it is important for artists and developers to consider the ethical implications of text-to-media AI and work towards solutions that respect the intellectual property and creativity of artists. This may involve creating more robust legal protections for artists, developing ethical guidelines for the use of AI in the arts, and finding ways to fairly compensate artists for the use of their work.

To summarize:

AI image generation is a process where computers create images based on data and information

Text-to-media AI is one way this is done, by learning to match up pictures and descriptions

While text-to-media AI can create images that look similar to an artist's style, it does not copy their work exactly as it is using its own knowledge and understanding to generate a new image

Some people may worry that using text-to-media AI could be considered stealing from artists, but existing copyright law does not consider it to be theft

It is important to consider the ethical implications of using AI in art and work towards solutions that respect the intellectual property and creativity of artists

Examples:

AI image generation: An example of AI image generation could be a computer program that is trained to generate realistic pictures of cats based on a database of cat photos and descriptions.

Text-to-media AI: An example of text-to-media AI might be a program that is given a large number of images of flowers along with descriptions of the type of flower, the color, and other details. The program could then learn how to generate new images of flowers based on different prompts, such as "generate a blue rose."

Mimicry of an artist's style: Let's say an artist creates a series of paintings of landscapes with a distinctive style, using bold brushstrokes and bright colors. If a text-to-media AI program is trained on these paintings and given a prompt to generate a new landscape, it might create an image that has similar brushstrokes and color choices, but it would not be an exact copy of the original paintings.

Not copying work exactly: Using the same example as above, let's say the text-to-media AI program is given a prompt to generate a new landscape based on a specific photograph of a mountain range. While the program might use its knowledge of the artist's style to generate an image that looks similar to the artist's work, it would not be copying the photograph exactly. Instead, it would be using its own understanding of the photograph and the artist's style to create a new image.

HOWEVER - In light of the protest against AI we are NOT for theft or cheating with AI art.

Consider this when you're generating AI Art: Is it a TOOL or is it your whole workflow? Are you claiming these as your own, or working it into something different?

AI art is technology. Ai art is not the next best thing to try and fool someone into getting a job that you don't have the skills for. AI art should be ethically trained, but the arguments around this are difficult and complex. Corporations will not ask for your consent, however there is a website that in this large complex protest that we even believe in: https://haveibeentrained.com/ Check to see if your data has been trained in the LAION-5B training data.

You'll notice a theme happening: Deviant Art, Pinterest, Smugmug and many other public data sets such as google images, Wix and other websites.

That isn't to say that we shouldn't ethically find a way into the future of AI art and work with artists to PROPERLY train these models and make it so the future of AI isn't what people force it to be.

AI ART is not dehumanizing, AI art is not abelist.

However, AI Art is not a full replacement for art. It's a tool, just like Photoshop, just like a pattern for cross stitch. Many people can just craft a prompt, and generate an image - but it's what you do with it after is the thing.

We are indeed FOR the the ethical future of AI art, but we will stand in the middle of this forked debate. We are FOR the ETHICAL future of AI art, we are with artists and yet aside from that we are with the future of the technology.

Not for the crypto maniacs who PURPOSELY generate signatures and try and profit off stylized AI art. If you're going to commercialise generated pieces, at least learn to use photoshop or learn to paint something new out of it. And remember: Corporations do not care about you. Your data is already stolen the minute you upload it to the internet. That does not mean people should use or access it incorrectly, but the ethics debate shouldn't be "DONT DO IT" - it should be, how do we create a better future. As for why it's not abelist? It's far less abelist then the time we were on reddit, and someone said it's better to just pick up a pen and paper than to even do digital art let alone AI art.

If someone has pain, or inability to create via normal conventions - who are we to stop them from learning to create via generative AI formats? Remember: Disabilities come in all shapes, sizes, invisible and visible. It's not up to us to gatekeep technology.

Also a hilarious reminder you CAN feed Stable Diffusion and Midjourney your own artwork, and it can come up with wildly varying and disgustingly weird results. This was one we used on Stable Diffusion 1.4/1.5 and used the IMG2IMG function with our own art - prompt is unknown at this stage but we used NightCafe Studio.

If you're against AI ART, we urge you to reply or reblog with your feelings regardless of if you disagree or hate this post. Don't just blindly repost something without sharing your full feelings.

And also a reminder: NOT ALL AI ART IS THEFT. Referencing work in the normal art world, photo manipulation is a form of "THEFT" in the eyes of some artists. Inspiration can be copyrighted, and ergo even in the music industry having to reference, credit and pay royalties is a thing. Think twice before bandwagon jumping! We have a 3 year degree in graphic design, and understand the nuance of this issue, fear not we're not even blindly for AI art in the way people believe.

If you use AI, you’re not an artist,

No matter how much you cry and squeal,

Over how you didn’t steal,

If you use AI, you’re not an artist.

Typing a prompt will never be art,

Typing a prompt will never be art,

You have failed the definition,

All you made was a commission,

Typing a prompt will never be art.

It doesn’t fucking matter how it works,

It doesn’t fucking matter how it works,

No one asked to be explained,

How the data set was trained,

It doesn’t fucking matter how it works.

Because no artist gave consent,

Because no artist gave consent,

No one asked for their permission,

Just plowed ahead with greedy ambition,

Not a single fucking artist gave consent.

Claiming you’re disabled is not an excuse,

Actually being disabled; still not an excuse,  

If writing a prompt is what it took,

Then you can write a fucking book,

I’m sorry if you’re disabled, it’s not an excuse.

Using progressive language doesn’t work,

Using progressive language doesn’t work,

You can quote from Lenin and from Marx,

Your pics still have artist watermarks,

Using progressive language doesn’t work.

Besides we’ve already seen your NFTs,

Oh yes we’ve already seen your NFTs,

All over Twitter and Instagram,

You were neck-deep in that scam,

AI art is a complex and fascinating subject that raises many questions about the nature of originality and ownership in the digital age. On the one hand, the artist who prompts the AI to create an image is the one who initiated the creative process, which could be seen as a form of originality. On the other hand, the resulting image is generated using other images created by manual artists and can be recreated if the same prompts and settings are used, which raises questions about the level of originality of the work.

However, it's important to recognize that AI art is not simply a matter of reproducing existing art. The artist who prompts the AI to create an image has the opportunity to bring their own creative vision and personal expression to the process, and the resulting image is likely to be unique and distinct from the original images that were used to train the AI. Furthermore, the artist can take the AI-generated image and make further changes to it using editing software like Photoshop, which adds an additional layer of originality to the work.

Despite these complexities, there is no doubt that AI art has the potential to be an exciting and innovative form of art, and it's important for artists and AI developers to work together to find mutually beneficial solutions that respect the rights and needs of both parties. One promising solution is the use of NFTs (non-fungible tokens) to certify the ownership and authenticity of AI art, which can help to ensure that the original artist is properly credited and compensated for their work and that collectors can be confident in the authenticity of the art they are purchasing.

In conclusion, AI art is a complex and multifaceted subject that raises many questions about the nature of originality and ownership in the digital age. While there are valid concerns about the use of art to train AI models, it's also important to recognize the potential benefits of this technology and to find ways to work together to ensure that artists are respected and compensated for their work. NFTs offer a promising solution for addressing these issues, and it will be interesting to see how they are used to shape the future of AI art.

————

My thoughts fed to ChatGPT then rewritten as a blog post for tumblr by ChatGPT.

the statements "ai art can exist" and "it is perfectly fine to steal others' artwork and use it for training without their consent or any compensation" do not have to be both false or both true

So for several years now, I've been working on a project about a group of artist AI whose user left them to their own devices trying to create their own fictional universe... only now there are "actual" art ai that are stealing art without credit and potentially nullifying my entire industry in the near future... So why is it I want to incorporate them into my project? Don't get me wrong! Stealing art is terrible and humans should always be part of the equation when it comes to depicting the human experience. But... what if there was a way to make a generator that only fed from a specific set of assets instead of a huge stolen library? TLDR: I want to create an AI that only feeds off of my own art so I can use those prior images to more easily stylize my digital work to look more like my traditional art. Or does this actually exist? Is there an AI that can be downloaded to a local drive and only learn from a specific set of files instead of the larger databases? One can dream...