Let's Talk About AI Art

[@CurtisHolt]

I have to keep writing the same response to comments about a Spotify payment model not being viable when there are so many possible solutions, please read the following before commenting: Two other possibilities: x.com/curtisjamesholt/status/1792534626180731286 x.com/curtisjamesholt/status/1792629213419249990 Think about it as compensation for access, not the end result. Alternatively for the possibility of a sub fee, splitting it by time usage of artist-specified models, rather than generations. There are lots of different ways to split compensation before even thinking about inspiration or copyright. For Blender, we sell access to things that belong to everyone all the time: GPL addons. In that case, people pay for access to the download service rather than the addon itself. None of this requires a decision to be made about inspiration vs ownership, or the licensing of a generation. Quite frankly, the results generated are irrelevant.

[@vizdotlife]

Thanks for sharing your thoughts and ideas on this. I found the video quite enjoyable to listen to and an interesting topic to discuss. It was also refreshing to hear this being brought up to talk about on KZbin in the Blender community.

[@CurtisHolt]

Leaving a couple of important timestamps here for people that comment after only a couple of minutes: 8:30 - AI tools may become a new norm as the idea of 'mental effort' compresses over time. 13:26 - (Job Market) Corporations jump on FOMO tech waves before homogenizing. For some additional thoughts: I had previously created a video teaching people how to combine AI art content with technical elements in Blender with a specific interest in procedural material generation (by the name of 'The Insane Potential of AI Art'), which I have since privated. I don't think I understood the complexities of the ethics at the time, and there wasn't much in the way of discussion, so I will hold off on doing educational content relating to this category of tools until I can feel more confident about the answers to some of the questions raised in this video, particularly relating to the access and compensation of artists. I do believe there's a way to introduce AI art tools into the art community in a way that's truly beneficial for artists that may offer them new types of independent careers, but it's not there yet. When the space feels more fair for artists, then I will be more open to doing educational content incorporating elements of the tools.

[@TheVolgun]

Fascinating video on this topic. It's great to see someone trying to think of solutions to the more problematic sides of generative AI!

[@MrEshen25]

It's a short sighted mistake to give AI a pass contingent on royalty payouts. Consider the stock photography and illustration market. There was boom of easy money early on in the 90's, but the industry is gutted now. YOU might get compensated for your work now as an early contributor. However, imagine the industry in 20 years. Eventually the models will be capable enough to service 99% of their customers and there won't be an incentive to increase the data set. The result being, new artist will be competing with this new AI, AND they won't get compensated like you did.

[@VertexRage]

🎨"The tools are surprisingly anti artists" - that's kind of key point in this new wave of generative ai. I doubt many people would complain if there would be a tool for ai-powered retopology or uv unwrapping for example...

[@YakOfArt]

These apps were not directly designed as art tools. Rather, a fun and playful piece of tech for your average person. What's happening though, is artists are trying to take control of them and use for themselves rightfully so. Kind of like how PS was originally designed to edit colors not paint lol.

[@DonChups]

Video denoisers and upscalers are kinda great, but can't shake the feeling, the good implementations are just side accidents. Just like nuclear power is a side producto of the intended atomic power use: bombs.

[@roni.rsnstn]

FInally someone who thinks the same way. I definitely think corporations will test the limits with AI, until they go to far, and AI art will be stigmatised. I'm so sick of AI fearmongering, this is a refreshing take

[@magnus6003]

In the US there's no copyright on AI generated content. I expect we'll see the same in Europe. There"s also a few pending lawsuits concerning the original content being used to train.. An excellent tool for creative people who wants someone else to do the creative work.

[@CurtisHolt]

I'm very interested in seeing how the EU approach legislation for generative AI tools in the future.

[@neuromancer_84]

EU AI Act is done already. It's in translation. Should be in work in a year. As far as I know it says image AI generator should credit the authors of images they use. Personally I'm not a fan of AI generators. They scanned whole accessable internet, were in it's dark places. Downloaded and learned on images that should never been taken. On images that you don't want to see, images after you have nightmares and fear for your closest, images of the worst thing a man can do to another man, child or animal. Even if the images were later deleted from AI generators database they were used to learn by AI. That darkness is a part of AI. A man would never use it but report it. AI has no consious, no moral values. Imagine what is happening in dark web with all these image generated possibilities. It's possible to evade every prompt block AI generator has to generate the sickest images someone wants. Dark web is growing cause of AI. And how do you go after it, how you make it illegal, what law to use when these are generated content? AI can do very very bad things. It has safeguard, created by men, in theory. But if it will be set loose it can strike with it's whole darkness. It was always a man greatest fear, the darkness which you can't understand cause you can't know it with your senses. It's there and you'll never know what will come from it.

[@WaterShowsProd]

Interestingly, though, Chinese courts ruled that the act of prompting, refining, and editing does constitute copyright for A.I. generated art. I've yet to hear what that will actually mean in practical terms, but Chinese law recognises generated art as belonging to the person who prompted it, as opposed to U.S. law which essentially states that all generated artwork is public domain. By the way, it's this waiving of copyright over generated artwork that causes me to take the stance that no image generator can prevent you from using generated images commercially, as no rights can be claimed to them by anyone, therefore the images are public domain. Obviously the legal landscape is a minefield that's only being freshly trodden, and things are bound to evolve over the next few years.

[@magnus6003]

@@WaterShowsProd Translated to the "real world" this would be the same as giving the copyright to whoever tells the artist what he wants him to make. Not viable IMO Let's not get started with China and copyright. :) They're not exactly the go to source for anything copyright related.

[@WaterShowsProd]

@@magnus6003 When an artist does a work for hire the copyright does indeed belong to the client who tells the artist what they want made. This is true in all fields of art. Just because China's copyright laws aren't the same, it doesn't mean they don't have them.

[@yportne6410]

I often have conversations with people who don't engage with art the same way we all do (they see an impressive image, "oh, that's cool", then they move on and forget about it) That's the vast majority of people. Unfortunately most people aren't art nerds and they can't tell if an image is AI or not. For us it's obvious - "AI gunk" is a good way of describing it, I think. The patterns that go nowhere, the airbrushed look. We art nerds can see that but the average person doesn't pay enough attention to notice. My concern is that the internet will get rapidly flooded with "AI gunk": Images, articles, everything. The scary thing is most people won't notice and we'll be the crazy ones. I already feel crazy when talking to people about it, and hell, maybe I am.

[@TheFrenchCizzle]

but the main problem is that people are no more impress by our work done because beautiful image are so common nowadays

[@KDawg5000]

In the lucid dream, instead of writer's block, you had dreamer's block. :) 🎨

[@YakOfArt]

This is a great video! I love the list of ideas towards the end. Just the simple one of listing a percentage breakdown of artists sourced, would not only be an openness from AI programs that would be welcomed, but also, I think it would actually be helpful and do what you're suggesting, like helping us discover new artists and talents. There's more you mentioned and they all got me thinking. Feeling very inspired 🙏

[@DonChups]

I would be super interested in a tool that generates abstract patterns following some patterns. But the problems are, as always, a labour and extraction problem. The sewing machine's pitch was to "free the women from the burden of manually sewing". And in the end we have sweatshops producing fast fashion and kids sewing cheap clothes for people in other countries.

[@WaterShowsProd]

This was fascinating on so many levels. Regarding compensation: in kind of a similar situation, we, a group of voice actors, were approached by a recording studio that was contacted by a recording studio in Shanghai, and asked if we'd be willing to record text-to-speech A.I. voices to be used for NPC game characters. I understand what the studio in Shanghai is doing. We often record voices for games created by small studios in China, and rather than having us record 5 lines for each and every minor character, it's cost effective to save using actors for hero characters, and just generate the little minions/monsters/enemies/etc. using text-to-speech; everyone is trying to cut budgetary corners. The studio asked how much we would want in compensation. The sticking point of the whole situation is that the proposal was not presented to us properly. The people whose job it is to contact the talents were told it was a games studio and the work would be for one game. Then, when questioned, they said it would be for multiple games. It was actually I who looked up the client and saw that they are a recording studio that dubs games, anime, films, etc. I'm not suggesting that we were deliberately misled, I think it's a case that the project managers didn't provide the correct information, not the first time, and the taletn coordinators were merely repeating what they were told. We are going to meet with our studio and request proper contracts specifying what kinds of projects the voices would be used for, and prohibiting them from selling the voice models, or making them available, to other parties. One of the actors in our circle then raised the question: how could you police that? These, unfortunately, are the realworld issues that we're facing with this technology. We're not entirely opposed to it, but we are also careful about protecting our careers. The issue in this case wasn't the technology, but the business people not adequately communicating how that technology was going to be used, and that's where it gets scary. 🎨🖌🤖

[@Wendy3Dimensional]

I appreciate this video. It was interesting to hear your unique perspective on this topic. 🧐

[@jameshughes3014]

I love the way you used the AI as sort of a mirror, that lets you 'see' your art in a new way. I feel like that could be a great way for artists to improve. It's a bit like the trick of flipping an image horizontally, letting your brain see it in a new way, except with your style.🖼

[@wailfulcrab]

28:00 it's absolutely possible to do for local models, just training a style embedding takes a lot of time and sufficient machine, with correctly labeled and prepared images. The process takes at least few hours, revving the GPU to the max. As for me personally, I treat it as a glorified reference/iteration generator without the need of browsing through countless stock images. Most of AI imagery uses polished, "artificial" look, very distinct. It would take more time to mask the gloss from the image than making the image from scratch. But as a quick mockup to get the general idea what I want to do, its perfect.

[@ArielTavori]

IP_Adapter... Prompting is outdated for style and subject accuracy. IP adapter (which I use inside ComfyUI/Stable Diffusion), allows you to use an image as your prompt, with or without additional text prompting. There are optimized algorithms for style, composition, face/identity, and even regional masking so that you can define an image for the background, an image for the identity of the person standing on the left, and an image for the dog beside them on the right. You can also use multiple images for a single input, allowing you to blend identities or styles... This enables fairly effortless character and environment consistency, completely changing the scope of what these tools are useful for.

[@onesheisty]

🎨Great vid. You bring up good points about where this tech is going. Eye-opening for a novice, like myself. Interesting perspective you being the artist and providing specific examples of how your craft is being used. The idea of access + compensation for the artist is amazing. I just hope its not already past the horizon like the music industry point you brought up where the artist only sees pennies when it is their work that is the main source of value.

[@marcelzero3005]

I want artificial intelligence to solve my daily chores so i have more free time to do art, not artificial intelligence doing art while i do the dishes.

[@olivetree9920]

This is a sentiment I truly don't get. Just make the art. Is the AI slapping your hands when you try to make stuff?

[@go2viraj]

@@olivetree9920 they clearly stated that they would have more time to do art if AI did dishes. They are not saying AI is stopping them from making art.

[@johnfromberkeley]

This quote is stolen and not attributed to the original author.

[@bouncycow3010]

Very interesting discussion! 🖌

[@ericpatrick2779]

🎨 It's refreshing to see people talk about AI without urging the torch and pitchforks for either direction on for or against it. I personally don't like the way that current art models of AI sidestep the consent of artists, so I'm all for your proposed method of instilling a more collaborative community among artists with an incentive to opt-in for it. I know that would change my stance on it, for sure. As for the use of AI in medical matters, well, I can see how some doctors could improve their practice with its application in the way you have stated. However, I don't expect that to be that popular among "problem" doctors that are far too stubborn to accept that they may be wrong; even with a patient's health being at stake. So, I guess I could distill my comments down to agreeing with you on all points. 😅

[@olivetree9920]

I would be interested to see the full conversation you had with ChatGPT. Could you share the link?

[@Sajeas]

Stable Diffussion gives a lot of control to whoever trying to generate images.

[@PumpkinHook]

This was a very interesting take on AI art, love thinking of ways of improving the system to properly compensate artist

[@CGToonStudio]

Can you please drop a 15 minute long tutorial on texture projection mapping and texture painting? In blender 4.0

[@tregdemedia]

I really liked your point about references artists getting paid per generation. It would be like stock music or video creators who supply assets to Artlist or Motion Array. You get paid based on how popular your content is, and it legitimizes the platform as well.

[@bifrostbeberast3246]

Motto of Silicon Valley is: First do, then ask forgiveness. As long as artists are not backed by the law in precedence law suits, this theft will go on. Imagine the billions of dollars being poured into this technology while zero to none compensation goes to the very content creators used to create this technology. At the same time these companies charge users for their money while siphoning off every thiny bit of personal information a Scarlett Johannson voice LLM can flirt out of their users. Honestly, it is very sickening.

[@bububirb]

Quite an insightful talk. My favorite part was about the idea of overcoming limitations and how it's an important part of what makes something valuable. 🎨

[@barbi111]

Right and fair questions towards Open AI.

[@newyearstudios]

About a year and a half ago or so I trained a model on my art. I used stable diffusion as the base model but let my computer churn away for 2 days ingesting my work. When it was done I was totally weirded out that I could ask it for anything, even scribble a little doodle and then prompt it and it would turn that doodle into or create something that looked exactly like my work in a few seconds. I got very uncomfortable and after a period of time I deleted it. I turned my back on this tech and doubled down on traditional media, but since then the conversation around AI has changed. It is no longer simply about AI art. It is about the automation of all labor. These systems are amazing. I was angry and scared for a time, but I starting to feel positive about the long term effects of these systems. As you mention protein folding, material science and healthcare will be advanced beyond what we can even imagine I feel. I also, think in the short term things are going to get very uncomfortable as these systems replace humans across many, if not all fields with a monetary motive... I think the advantage will more than anything be a disruption of our economic system strangely enough. Corporations driven my profit will give into the race for efficiency and capitalism will consume itself out of existence. I think it will force us to turn away from a globalist mindset and back to local arts and efforts. People may no longer chase fame in the same way, and work that people don't want to do... they will simply just not do it, but, AI will. Our values and motivations for the things we will do will change over the next 5-10 years and ultimately these systems will be "better" than us at most things. That said art and expression of an individual subjective perspective will never be lacking in meaning and motivation from a human perspective. Even if art as a commodity loses its value it will never lose its ability to help humans process and explore our interests and emotions, test new visions for the future and bring delight to others by showing an individuals internal state. Humans are fascinated with other humans. When work as meaning is lost I think the world will be a better place and the best way to do that is to separate the human from the need to labor. I know a lot of people are afraid and angry about all this tech but in the long run at the rate things are progressing I think the vision of artists openly collaborating through AI as well as in person will be more accessible than it ever has been. The barrier of the need to monetize our efforts will free us to make our own limitations and create things that are truly unique. There is of course a chance this will not happen, and we will simply have our lively hoods stripped away from us and be forced to live in a false state of scarcity to prop up the ultra rich.... lol, but let's hope people are smarter than that and we can prevent that from happening. :D

[@chacecampbell2697]

Really appreciate the nuanced take! I think at the largest scales of foundation models it would be hell to enforce, however, I could absolutely see people going for custom Checkpoint and LoRA models trained by individuals and/or artist collectives. It's absolutely possible to do *right now* with the necessary coding experience to run and host Sable Diffusion in the browser. Like the Gen AI equivalent of farm to table. Not to mention that you'd *always* have the best and most up to date dataset to train from. Be it a way for fans who'd happily pay $5 for some generation credits to support you, or studios interested in licensing output with your stamp of approval, the potential for monetization seems quite vast.

[@CurtisHolt]

I certainly appreciate those latter ideas. 👍

[@zana3d]

Thank you Curtis for your sober, balanced take on the whole divisive debate. 🎨

[@robinlovejoy5176]

I understand your opinion about the novelty of more organic, naturally made artwork. I find value in knowing that everything about the piece was intentional, even the shortcuts were done because the artist thought it would have diminishing returns to make it more complex. There's a lot that goes into an artwork, because of this, the final image is not the only thing that matters. There are some things I would like to address here about these models: 1. They do not reason or know anything at all; it is incapable of logic. What they do is correlate and interpolate between data points. When it tries to generate something "in the style of..." it pulls not only from that specific artist, but a wide body of artists around that data point that during training it has found to be closely related. There are a growing number of AI experts that conclude the same. These models do not possess the kind of architecture where logic and reasoning could ever arise, no matter how much data you put in them. 2. It does not discard the training data after the training process. All of these models, GPT's image model, Stable Diffusion, DALL-E, and Midjourney learn key pieces of data about an image during training so that they can recreate it. This data is then stored in a matrix, which is uses to recreate the original image closely. In effect, this is a very impressive and extremely heavy form of compression, being able to take the 5 billion images and store them in a few gigabytes. If you want to learn more about the process for training a diffusion model, I recommend Computerphile, there are a couple good videos about it there. Recently, there was a judgement in a court case, suing Midjourney I believe, where the judge thought the plaintiffs had good reason to believe the models do indeed contain their training data. During the process of generating an image, based on your prompt, it will splice together many images in a very fine way, in fact it's so fine it's difficult to detect the original images in the final mix, so it looks like an entirely new image. Since these models are essentially a natural language processor, strapped to a very large highly compressed dataset that is able to mix images together, I believe that puts them firmly in the realm of copyright breach. If it is concluded that these models have indeed breached copyright, they and the companies that made them likely will not be around for long. Twitter thread on the court case tentative ruling: x.com/LuizaJarovsky/status/1788569553145909599 3. Regarding payment and consent for these models, if these companies had the money to pay all these people (at least 10,000 in Midjourney's case) for their work, by now, with all the bad press and lawsuits, they probably would. The problem is fourfold: these models are very costly to run, the amount that would be fair is completely untenable for these companies (possibly in the 10s or 100s of billions), the potential money to be made and the greed of these companies is at an extreme level, and they also just don't care; from all of these developers and their user base, we consistently see aggressive anti-artist sentiment, "can't wait to see you lose your job to AI", "have fun on the street when you lose your home", "artists are gatekeeping elitists", and much worse. Some artists have been the victims of targeted campaigns to usurp their artistic identity (SamDoesArt). This video mentioned Spotify's payment system, which as I understand is widely criticised for being very stingy, a few pennies per thousand streams simply would not be enough. Most artists are barely scraping a living working ridiculous artists and still being treated like crap. Also, even a few pennies per data point call per generation from a model user would probably still bankrupt these companies, so they wouldn't even do that. 4. These companies and their spokespeople are doing everything they can to convince everyone that this kind of thing is inevitable, and they can't stop it. This is all part of their hype and terror culture. Legislation for these kinds of things looks to be coming in very soon, the EU recently passed its AI act which should come into effect within the next 2-3 years, one provision of which is the full disclosure of the data sets, which is found to contain copyrighted data (pending the outcome of whether it is found that these models can infringe copyright (which is looking even more likely)), they will not be able to operate models like this in the EU, and losing a huge economic sector like that is a complete no-can-do for these companies. People will also try to convince you that these models are only going to get exponentially better each update and that something like AGI is just around the corner. This is just a flat out lie, if anything they have very nearly plateaued, there simply isn't enough natural data for them to get much better. One of the potential solutions to this is to train on synthetic data, i.e. AI generated text and images, but because these models function by detecting patterns that humans can't see, they will find patterns in the generations, which at the lowest level is just noise, and eventually if training on enough synthetic data they will just stop returning good results, possibly just absolute nonsense. Being very generous, let's say that GPT-5 is 2x better than GPT-4, maybe GPT-6 will be a bit better than GPT-5. Moore's law is no longer in effect, for processors and for everything else. With that out of the way, here are some things about how these models function in the work environment right now: 1. every time someone wants a small change to the generation, the whole thing changes, which is absolutely no good at all in the production environment 2. you can't ask for very specific or niche things, the less prevalence something has in the dataset, the worse these models will be at making it. Plenty of cats and dogs, and women in the dataset, but not much of say a castle, underground, from 3/4 top view made of earthworms (I had no idea what example to give here lol). 3. you can't ask for technical drawings at high accuracy, such as orthographic breakdowns with material references and dimensions. More examples can be seen from FZDSCHOOL in his video: kzbin.info/www/bejne/h4XNYoxqf7xggas&t (I also left a similar comment there) I would like to end by saying, I and many other artists, Curtis Holt included as he said, are not against AI technology. What we are against is the misuse of our work to exploit us, which is exactly what these models do. Furthermore, when we talk about these kinds of models, I wouldn't refer to them as simply AI; Gen-Ai, generative models, or simply models would be better. It helps the companies that we have to keep talking about these things in the same vein as all AI, when they are just a small problematic subset. I hope anyone reading this got something useful out of it, thank you for reading :)

[@bifrostbeberast3246]

You've jeard of inpainting? Why would you need to generate a completely new picture for a small change? You can sinply mask the part you want changed via inpainting.

[@bifrostbeberast3246]

Also check civitai models and loras. You will be surprised how much more there is than just cats, dogs and women. Maybe you are just not as proficient in GenAI as you claim to be or maybe you didn't check in the possibilities for a while, but things look vastly dofferent now, from temporal consistency in videos to finely adjust via inpainting, deepfaking with just one image, etc. It's all about knowing the tool, just as photoshop or blender.

[@caryonplays9024]

About discarding images on the training. The images are discarded before the train starts. AI only uses them to identify patterns, after that, all the processes are done without any reference of the original material. The only thing stored on an AI neural network is the information of the processes it uses to "redraw" those patterns.

[@robinlovejoy5176]

@@bifrostbeberast3246 Hello, thank you for the reply. There are some things I would like to address: 1. I never claimed to be proficient in the use of these models, in fact I don't use them at all because I believe that to be unethical and possibly illegal. 2. I'm aware there are a lot more things with a large number of data points besides cat's, dogs, and women; they were just the only examples I cared to list off at the time. Regarding CivitAI, most of what I've heard about them is that they are known to host various fine-tuning models (some of which were highly unsavoury, if you catch my meaning), which are usually targetted to emulate a specific artist(s) style, I mentioned this happened to SamDoesArt. 3. It has been a little while since I checked in on the specifics of the workflow, most of what I keep track of is the actions of the companies and their practices, rather than the workflows themselves, though I do still try to stay somewhat informed on that as well to a lesser degree. 4. Temporal consistency has gone up, when OpenAI released SORA recently, I saw that most of the shots were consistent through time, but still subject to quite significant changes over their length. 5. Yes, I know about in painting, it is able to change smaller parts of the image, but still not with the finest of control. I was just looking through some videos of people swapping out the clothing on subjects and saw that frequently, things like the hands and body proportions would change along with the cloths, although the general pose was still mostly intact. I've seen some workflows where people are using image generation over a 3D viewport to achieve some sort of finalised result, but every time they make even a small movement, lots of details change (see: x.com/MartinNebelong/status/1724919110830633328 ) 6. I contend that see things are tools, rather I think they are replacements and intended as such. The user doesn't have complete control over the product, and many details will just appear without the user thinking about them because they happen to relate to a data point relating to a token in the prompt. Hope that clarifies some things, thank you for reading :)

[@robinlovejoy5176]

​@@caryonplays9024 Hello, a couple of things I think could use some clarification: 1. The data sets these models are trained on contain links to all the data, such as images, videos, or text, which are then processed by the model during training to draw connections between text and images to enable the prompting to work. The images are not wholly discarded, rather all unnecessary information for the model to reconstruct the image is discarded, leaving some sort of matrix of values that represents the original information only. In effect, the data has undergone a very heavy type of lossy compression, but this is still enough for the model to reconstruct a close representation of the original. So, I believe it can be said that these models do store the data they are trained on, just in a way that isn't recognisable or usable by anything else. You can also see this from the tentative court ruling I posted above. 2. I take issue with the term "redraw", as this seems as though it is tracing the features of the image. I prefer to think of this process as "seeing" the image. When training, noise is added in steps to a base image, the model learns what the image looks like at each step of noise, going all the way until the noise completely overwrites the image. It is trained with text associations as well, so you can tell it the word "apple" relates to this image of an apple. Training is complete, you can then tell it that somewhere in a field of noise, there is an apple by prompting "apple", and it cycles back through the noise processes, effectively denoising to find the apple you've told it is there. I think the more steps of noise it is trained on and steps back through during denoising reconstruction, the better the result is, but this takes longer to train and more resources. 3. The patterns it sees in the images are not the same patterns we see when we look at an image. This is why tools like Glaze and Nightshade are effective on these models during training without heavily distorting the image to our eyes. The patterns these models detect are so tiny we can't pick up on them. In case you didn't know, Glaze introduces small scale perturbations into the image, which to any model that then trains on that image makes it look like another image in the training set. For example, an image of a dog may look like a Van Gogh. This is why I believe that when a model starts training on synthetic data, it will pick up on the small scale pattern it itself introduces in its generations and increasingly obviously reproduce those patterns, an assertion seen in this paper: arxiv.org/pdf/2307.01850 Thank you for reading, hope that clarifies :)

[@dragon3dnet]

As a professional 3D artist in architecture, I have been very interested in AI art for a while, and I have played around with Midjourney and Stable Diffusion. However, I mostly generate images for my own entertainment because the "creativity" of the AI is often very far from the vision of what I wanted or expected. For this reason, I often give Midjourney open-ended concepts as prompts, just to see what it comes up with but I have found very little practical application to use anything created. And I can't bring myself to claim any of it as mine. I suppose there may be an edge case in the future where an AI generator might help to fix an image, but in my job the output has to reflect the boss's vision exactly: I can't see using whole-cloth AI images will get the job done.

[@spenzakwsx4430]

did you try to write a prompt where you put "in the style of curtis hold" at the beginning or in the 1st half of the prompt?

[@dieutownsmith1668]

Do you use Blender AI render?

[@KatereenArt]

These were some interesting thoughts🎨 I wonder what it would generate for someone who has no style yet or someone who does too many different artworks that can't be put in one style category (maybe it would show possible style elements that a human didn't notice before 😄, "patterns" as you said). Also it would be interesting to give the same prompt to AI and to a human artist about copying a particular style. Although that would demand a strong sense of self-worth from artist side and their readiness to possibly "loose" to AI (although if one owns a sense of self-worth, no results from AI would shatter their believe in themselves). P.S. I love the dichotomy 😄. I wonder if it was on purpose, but I agree with both statements in the context. 0:54 -01:05 "don't expect "it's good" or "it's bad!"..." 29:29 "Healthy conversaion - good, unhealthy argument - no good"

[@zachhoy]

Really appreciate hearing your input on this minefield subject, and I too hold a nuanced view. Weird to see your own style replicated isn't it.... 🎨

[@spejarn]

I wouldn't mind seeing a free/local version for amateurs just messing about, and a paid tool for professionals looking to use it in their workflow. Getting compensated for someone just messing about doesn't feel realistic, whereas with people actually using it professionally feels a different story. As far as i'm concerned, if someone is making money off your work you need to be compensated, this is also true for the service itself, especially the service. The risk becomes in implementing such a systems, many companies in the past has bankrupted themselves trying to do the right thing, whereas their rivals flourished.

[@jackky_john.]

This is one of the best take on AI art I seen so far. ❤❤❤

[@DaveTimperley]

If one doesn't make use of the latest and greatest tools, there is a good chance that one will make oneself obsolete. A hundred years ago, I was apprenticed to a 'layout designer' in an advertising agency. The job was all boards on drafting tables/Rotring pens/enlarger and ordering type from a Linotype agency. There was only one Mac in the building and it was used by the secretary/receptionist. I guess nothing says 'we're a modern forward thinking ad co' quite like the clients seeing a Mac when they walk in the door? I was the only other person in the agency to use this thing, and I did, after hours. After a while, I decided to leave and start a 'desktop pub biz' using the original Laser-writer hooked up to a Mac+. Five years later, I still had a job, and the 'layout artist' that had trained me had taken early retirement. I have this 'thing' I call 'deliberate ignorance' which I tend to reserve for 'people of religion' and 'boomers'. It's where ones emotions and fears overrides their perception of reality. This AI denial smells a bit like 'deliberate ignorance' to me :-)

[@Stanisslayer]

How to check there u in dalee database or not? I have buch of viral artworks.

[@ForcePF]

🎨 I don't have a strong opinion either way about AI art, but I have been using AI to help me with very tedious / repetitive stuff such as rotoscoping in my visual effects projects. It's been extremely helpful.

[@Lluc3D]

I have been thinking about this idea of "AI spotify" althou in my mind it was more like a "KZbin" of AI art. The big problem come from the human nature, many will try to take advantage and may upload art from artists that are not theirs to take profit, especially those that are not that interested in digital stuff, for example sculptors or graffiti artists. One thing is for sure those companies should be way more open in how they obtain datasets and how they train their models.

[@o0alois0o]

loved your thoughts about this, great video! ✏

[@jasonfarmer5193]

AI vs Artists - The Biggest Art Heist in History - I suggest watching this for a better idea of the impact generative AI is already having. kzbin.info/www/bejne/kHuYappqhbtkd6c

[@greyhooves3959]

I'm looking forward to the point when AI music generators get close to being as easy to work with as the art generators, because I'm really interested in how an AI would interpret my music, but the ones I've tried so far just aren't on the same level as art generators have gotten, as far as user interface, accuracy, or quality of output goes. Also, when it comes to art generators, I do wonder about the compensation part. Say, for example, someone puts in their prompt for a generation, but they decide to use a bunch of artists names as "modifiers," but they use, for an exaggerated example, 100 artist names for the generation. Would that complicate things, or are there perhaps limits on what the AI uses for the final result? I know that the word order in the prompt typically matters, but maybe there's also a cut-off point to how many artists can be used as an influence per generation. ....I dunno. I just know I would want AI to fill in some gaps between the work people do and the pay they receive by simply automating the process 🖌

[@vizdotlife]

It would be interesting if we could find out for traditional artist artworks created that aren't AI generated how closely they look to other artworks by other artists. The vast majority of artists use references from other artists for inspiration and many are following a similar style. One example that comes to mind is the amazing Blender 3D artist Max Hay. I've seen many 3D artists now create scenes that look very close to his. In that case who would get the small royalty commission if they were used in AI models? I don't know the answer, but interesting to think about.

[@olivetree9920]

Traditional artists don't really even know how much they're influenced by other artists. I had a close colleague who made really cool and wild illustrations, but one day going back through some old google docs he found a pdf he made in high school like 15 years prior of an artist he was into at the time, and his current work all these years later was nearly a 1:1 stylistic copy. He genuinely almost quit art because he felt so bad about inadvertently copying, but also the direct confrontation of his own originality really threw him for a loop. Tbh I even think a few of the similarities to his own work Curtis pointed out were a bit of a stretch. Yes many of them looked a it like his images, but he himself also borrows heavily from very common tropes and stylistic themes in popular sci-fi. It would be interesting if he shared the whole conversation he had with it

[@vizdotlife]

​@@olivetree9920Yes. I personally see nothing wrong with using references to help inspire creativity with your own twist. There are still plenty of traditional artists that do indeed have a pretty good idea of how much they're work is influenced by some of their favorite artists styles. Many of them spend countless hours studying and learning from their favorite artists works and techniques so they can improve themselves, which ultimately can heavily influence their art creations. And then overtime they develop their own art styles as they improve and grow. It becomes vague on how AI models would pay royalties to artists work that are used in models because so many of those artworks from traditional artists were borrowing styles from other artists that they didn't originally create themselves.

[@TomWDW1]

🎨🖼 Loved the video, Curtis. A lot to think about ❤

[@CurtisHolt]

Thanks Thomas! ❤️

[@ameknite]

For me these AI generative tools will always be unethical until 4 things happen: 1. The creators of the generative AI tool get a license from every author they are using their work. (Needs to be explicit and opt-in) 2. Make public all the dataset they are using, with authors and license agreements. 3. The user of the AI doesn't get the rights about the work is generated, only a license to use the work. 4. Every AI work generated needs to be marked that is AI generated and which AI tool was used to created it. with "work", I refer to every medium: images, music, books, games, etc. But I know this will never happen, if they really care about the authors, they will do this already. But this will not make their business win money, and they only care about that.

[@CGMatter]

🖼️

[@besknighter]

You know when you talk with a utterly eloquent friend that can express complex ideas and concepts with simple terms and clarity? And then there someone else (usually us) who stumble upon the words, painting a really muddy but understandable picture of what they wanna say? I feel like when asking an AI Img Gen tool to mimic someone's style is the same as asking for this second friend to cite with their own words an opinion of the first friend's each top 3 topics of interest.

[@vizdotlife]

Thanks for sharing your perspective, I enjoyed listening to it

[@remii_online]

I appreciate you contributing to the conversation, very thoughtful and cogent words.

[@AllExistence]

Does gpt just use a prompt "in a style of" or does it actually add all the elements that belong to your style in it? I think it's an important question.

[@reinnardbartholiusdinata9174]

Really good insight on recent AI stuff. Truthfully i start to look into experimenting on ai with stable diffusion just a week ago and its kinda fun to play around and to get inspiration for your artworks. outside of all the legal rights problems. 🎨

[@somrender]

🎨

[@Yipper64]

AI art might be a black box but while I dont know that much about it, SURELY you could get some kind of output for the liminal space, and properly credit artists. I mean if you can ask for it in the style of a certain artist then certainly that information is somewhere. I think with GPT 4o being multimodal that type of crediting might be more possible.

[@makkon06]

the hopeful mythology with NFTs compensating artists for their contributions ended up being completely bunk because, in reality with tech made and hyped by corporations, the real thing is always going to be the shittiest possible version. it's already the same with genAI. I know you talk about the hopeful possibility of artists being compensated for their work in these genAI models, but it will never happen. it's not within the interest of the corporations that built these models, it never will be, it's too expensive. it might have been possible at the very beginning, but that ship has long sailed past the horizon. unless all current models utterly collapse, the ai bubble catastrophically bursts, and the corporations that made these models cease to exist and the tech effectively has to be made again from scratch, we cannot expect a future where a consensual and fair system of compensation will exist with this technology. it will always be the shittiest version. getting paid for all the art they stole from me would be amazing! I'm not holding my breath. 🖌 [edit] I recommend watching the video "The AI Revolution is Rotten to the Core" by Jimmy McGee

[@CurtisHolt]

A new and fairer system may not necessarily need to come from the current largest corporate players, it might be possible to start a new, more grassroots approach to art generation in the future. x.com/curtisjamesholt/status/1792534626180731286

[@jimhewitt4843]

I total agree with a ai Spotify idea

[@AdamEarleArtist]

Nice idea ♥

[@smokytopia6354]

I us Ai images all the time. It has never occurred to me to ask for anything in anyone else style. I do sometimes ask for images in my style...

[@AndyThirtover]

Good arguments, 🏔, my personal test is concepts. Can AI determine which prompted concept applies to what part of the output. "Refactor this code to use a class and make it multi-threaded for MicroPython" .. there's two concepts that AI mostly gets wrong -- but give it another 3 months and I expect AI to get this right more often than not.

[@lastchildz8567]

AI is a tool that can boost the work of those who have ideas. I use it everyday.

[@Dmitry2184]

i like how people keep comforting themselves that it's only a tool - ai tools, they say. Well, guys, in this case I believe you're a tool as well, in the eyes of the employer, and ai is just a better one.. if we dismiss all the "ai art has no soul" crap, which really means nothing and no one really cares about if the end result is decent, which it is. that is the problem.

[@MunishKumarCGI]

🎨🖌🖼 👍

[@gordonbrinkmann]

I sometimes feel like my opinion on these things is not seen as legitimate since I'm no artist and not having to make a living from art, but here are my thoughts anyway. So although I understand the hate AI gets from the artists community, I like to play around with it and see what happens. I sometimes find it fascinating with what it comes up with, because I usually do things like not describing specifically what I want to get like many users do, I give song titles, book titles or lyrics and quotes, often very vague and ambivalent and see what happens, only directions I give are sometimes the mood or the artistic style like photography, oil painting, watercolor etc. I also never use artist names, of course I know the AI has to draw from somewhere its "inspirations", but then again, when in school I was also taught what oil paintings were by being shown paintings from other artists, not straight been given a brush and paint and then doing my own picture. Of course I understand it is more valuable if I could create art with my own hands. But still, I'm doing this in my spare time for fun, I'm not selling this (especially not claiming it to be my own) and if I had no AI to use, I would surely not pay artists just for hours and hours of dabbling around with ideas, so they are losing a client on me. So I do not see why I (and millions of other non-commercial users out there) should be prohibited to use AI, even more so that I do not pay some service but do it all for free on my own PC, so I'm neither earning nor paying anything. 🎨 (Interesting how many people commenting have either not watched to the end or simply ignore your request...)

[@CurtisHolt]

Thank you for sharing, and for the last point: yes, it's frustrating. I feel like a lot of people just click on the video with an assumption for my stance is and start commenting before finding out whether they were right or not.

[@gordonbrinkmann]

@@CurtisHolt Yep, and some comments definitely read like you're right about that

[@LoriGriffiths]

Another video of high intellectual content. Thanks for that. I have no doubt that there is an equitable solution to the AI art generation problem, however, I don't think that's of any concern to a corporation. Why give you money if they can keep it all? Now for some constructive criticism. I had issues watching this video because of the excessive jump cuts. There were so many that I started to get dizzy and had to look away and just listen. You're not the only creator doing this and I don't understand the purpose. It doesn't add value to the video in any way and visually breaks up the conversation, in my opinion. Still love the content, just can't watch it.🖌

[@mind_of_a_darkhorse]

I doubt companies will be willing to compensate the artists. They may claim that it is not your work and make excuses, and then use the legal system as a weapon against you!

[@leandrorp6642]

I don't care at all for AI art. For me is just big tech company taking and using our work without compensation and consent, and never will. It is just one more big hindrance you have to deal with in addition to the competitive entertainment and art market. You as an individual could never use someone else's IP to profit without getting sued but big corpa bent the reality in her favor, that's peak capitalism. But thanks for the philosophical perspective you put in, I hope people appreciate human art more as a result of more processed AI one.

[@ArielTavori]

Compensated by whom exactly? The whole conversation around compensation, axiomatically assumes that someone, somewhere, is making money at some point from which that compensation would be deducted. Who is making any profit right now or in the foreseeable future from any of this? The whole industry is burning venture capital like it's going out of style, banking on the exponential curve to carry them out of the red.

[@CurtisHolt]

Users. It's really not as complicated as everyone is trying to make it. How do both the Blender Foundation and Addon developers make money from Blender? There's no money to deduct from anyone, no royalties to pay. There's nothing a user *needs* to pay for. Everything is GPL, and no one in the chain of creation owns or has any claim over what users generate with the tools. The users own what they create. And yet... Somehow people get compensated for their contribution to the tools. Without any contract or agreement with the users, without even a unified platform to work with. Artists could participate in a method of selling customized CC0 art generation models (or access, akin to paying for the download service of a GPL addon) adjusted for their own style, or any style according to their explorations. CC0 diffusion models already exist (Mitsua Diffusion One). Separately, others are already experimenting with either paid modifications (Loras) or parallel funding methods for creators (Civitai / Buzz system, who messaged willing to provide information after I released this video). Making the creation and adjustment of these tools more accessible are the two main requirements for allowing a new support philosophy to grow around the tools, focused this time on artists, as it has already for open source software. Literally all it takes is a few people to agree on a philosophy and act on it, on their own. A bubble where artists and users agree to support each other. The exact reason this channel still exists is because it has been funded by a bubble built on a philosophy of support, in an industry traditionally grown on capital and proprietary ownership, full of licenses subscription fees. When open access reaches a threshold, a new philosophy will be born, regardless of any contrarian naysayers, perhaps even in defiance of people who complain without any ability to visualize or propose alternative solutions. Whatever other companies (or other users) do on their own machines is completely irrelevant, and will never suppress a communal desire for fairness. It will not suppress the ability for people to come together under their own philosophies - helping artists to create art generation 'product' from their work, and support them in the endeavour. Saying something is impossible is not, and has never been, impressive or useful. It contributes nothing of value. It helps no one. Finding solutions, however, is impressive.

[@deepblender]

The way image generators work today, I don't think there is a technical way to figure out who's work contributed to the final image. Figuring out what influenced the outcome is essentially impossible as far as I can see. You might use the text prompt to figure out which artist's work (that was used for the training dataset) fits the description in some way. But you would still not know whether or how much an image was influenced by an artist's work. I have trained generative neural networks and if you told me you wanted to hire me to do something like this, I would have to tell you that I don't know how to achieve it. I can think of many ways to have indications, but at the end those would be worthless as soon as you started to dissect them. The core issue is, as you are training a neural network, every image can influence everything within the neural network.

[@CurtisHolt]

Maybe it's time for a new type of architecture. I like to think that nothing is impossible. x.com/curtisjamesholt/status/1792534626180731286

[@deepblender]

I believe this sort of boils down into the problem of interpretability of neural networks. There is quite some research about it, but I haven't seen results that were practically relevant on a large scale or related to what you are suggesting. I have been following the research for many years. It became obvious that there are massive improvements for language models, just like for generative techniques, which could lead to something amazing. Those things did not just appear out of nowhere, there was a clearly visible trajectory which lead to it and took many years. But there were many signs years ago that there was something. When it comes to interpretability or to understanding how an image was generated based on the training data, I am not aware of ideas of how it could be achieved. It might not be impossible, but it might also be that it is completely unknown how to achieve it.

[@CurtisHolt]

I think there might be a simpler way about it. I'll copy my response to another comment in case it applies here: The way I see it working in practice would be compensation more for the product of access to a process, rather than the inspiration of an image. I don't think the end result is necessarily something an artist can claim ownership over, but the process can be made more fair through a platform of access. Maybe there's a parallel to draw with how Blender GPL addons are free to distribute and re-use - they technically belong to everyone, and no one is suing each other for the code - yet, somehow, people make a living from them. Without seeing that system in practice, people would likely think it was ridiculous. The storefronts, like Blender Market, provide a community focused way for developers/artists to be compensated for the contribution. Technically (and specifically for GPL addons), the only thing people are really paying for is access to a download service. Why could the same not apply here? Models based on CC0 content, modified by a specific artist's work, becoming a product to provide compensation, irrelevant of the ownership or rights provided by the end result. x.com/curtisjamesholt/status/1792534626180731286

[@moonthurst]

🖌 interesting view of AI thanks

[@naturalia71]

I would think the reason for the pennies not coming is the same as the reason artists in general are not paying pennies to every artist their own art is inspired by. Can someone plaese explain the difference? I literally would like to know as I want this argument to be a good one if I ever need it myself

[@The_Trident_Master]

I have a question I’ve been arguing with my friend for the past hour now. Is ai art actually art? I saw yes, he says no. I argue that it’s low art at best, but still art in the end, if it’s a more advanced ai that can blend samples well. What are your opinions?

[@KiaAzad]

It's not as easy as: Just pay the artists. Let's say Ai companies start paying artist a share, and the whole art industry relies on those payments. Wouldn't new artists be left out? How about the artists that used Ai to create some art, just to make money by including it in the databases? Should they be paid as well? How about non artists with AI tools and lots of time? Should they be paid too? What if a company just hires a bunch of minimum wage workers to create a database of only AI art to use in training? Can they keep all of the money? Because that will be the most likely case.

[@CurtisHolt]

I see most of those concerns as irrelevant to artists receiving compensation for contributions higher in the chain (not the generation end). In theory, no corporation is required to pay artists for them to receive compensation in the case of a community-focused open source alternative. See pinned comment for details.

[@paulstanley7292]

You're right that there will always be a desire for art created by a person because it has a story to it. I liken it to thinks like sports stars signatures, if you have a picture of your favourite star with a printed signature on it it will look the same when in a frame on the wall than a hand signed version. We all know the hand signed version is worth more. I still think that images that are 'inspired' by other's work is ok, it's how we all learn. If I painted a picture in the style of David Hockney would he own it???? As long as I don't try to pass it off as his work then it's mine. Same should apply with AI. What would be your thoughts if you were inspired by one of these AI generated images and created an artwork based off of it? Who owns that???? :) you squared :)

[@CurtisHolt]

I agree. I don’t think copyright can necessarily be applied to AI generated images. All I know from my subjective feeling during the video is it ‘feels’ like I own a small part of it because I invoked my style. Other than that, I don’t think it can be measured. But artists could still be compensated either at the generation level, or (if not technically feasible) at a higher level of the service chain, such as producing and selling tools based on CC0 (public) content that is then modified by their own work.

[@paulstanley7292]

@@CurtisHolt I would love to see you be paid for works commissioned by AI, but I think that the model is outputting works based on EVERYONES input (i.e. it's training) so unless the prompt specifically has your name in it I don't see how you can decide how much of that artwork is based on your work. Time will tell though!

[@hippapasworkshop146]

🖌️ love hearing your view, as always! I think I have to disagree with you about being compensated. That seems to be a slippery slope into a lawsuit land of stagnation. How would the descendants of DaVinci, Rembrandt or anyone else in history be compensated for using their works as inspiration? Or more recently, Bob Ross, Warhol, and others. Just a thought. Humans tend to want to drag others down rather than lift themselves up. In this imagined future, everyone would be suing everyone else for inspiration compensation.

[@CurtisHolt]

The way I see it working in practice would be compensation more for the product of access to a process, rather than the inspiration of an image. I don't think the end result is necessarily something an artist can claim ownership over, but the *process* can be made more fair through a platform of access. Maybe there's a parallel to draw with how Blender GPL addons are free to distribute and re-use - they technically belong to everyone, and no one is suing each other for the code - yet, somehow, people make a living from them. Without seeing that system in practice, people would likely think it was ridiculous. The storefronts, like Blender Market, provide a community focused way for developers/artists to be compensated for the contribution. Technically (and specifically for GPL addons), the only thing people are really paying for is access to a download service. Why could the same not apply here? Models based on CC0 content, modified by a specific artist's work, becoming a product to provide compensation, irrelevant of the ownership or rights provided by the end result. x.com/curtisjamesholt/status/1792534626180731286

[@Devilin-Pixy]

What defines one's style? Reminds me of the 'chicken or the egg' causality. I guess we all 'copy' in our own unique way, over and over again, such is life. AI is becoming good enough to copy things we do. Can be a great tool in many ways, but it is intention that worries me most. With ability speeding up, we are bound to lose sight of the consequences.

[@Richard_Baylon]

12:06 i believe it might have a possibility of replacing many, or any jobs that requires files at it's main thing, wich is probably everything in the internet since all you need is a large quantities of file, and combine.. well not combine, more like mutate. as far as what i know with ai, is in wich it works like a root of a tree finding the nearest source of water. this root paths is controlled, improved and updated. first chaotic, and then near perfect. and the scary thing is not the path of the roots but the rate in Wich it's growing. compare it like this, humanity is a tree 🌲 that grows everywhere and ai is a tree 🎋that grows in a straight line but it's growing at exponential eventually speed of light the problem is it stabs the human tree 🎠that is growing thus the **demotivation of artists** and stuns human development because the mass gets less motives and it grows slower there's only a few ai trees that grow that can stab it majorly imagination,audio,speech, the only way to make it grow again fast is for the ai to wrap around the root wich is basically turn into brain with metal parts that help it. ¯⁠\⁠_⁠(⁠ツ⁠)⁠_⁠/⁠¯ in wich there is a gamble either most of humanity is useless or humanity is useless and but there's ai, because humanity cannot go faster than AI technology and human will need to become ai to be at society, there will be ai chip in your brain if you want to improve in the future, ai chip is just basically google in your brain but faster and just like google it's really fucking dangerous and controlled by massive corporations, good luck with that you might improve but you will be controlled just like the ai, and of course there's no way out because the one who has the chip can make better machines in the real world, go be optimistic and believe humanity does not get violent and control people this argument is so godamn crazy because brain chip is just way too sci-fi and any file related jobs argument are just crazy but it's true any file related jobs will eventually be replicated, and brain chips exists and for now it's just on standby and i don't know what to do with it. and yes it's obviously gonna improve human health but that's just health not ideas. for context ideas are violent and Inspirational so don't expect a great world idea from a great torture idea , and holy shit I can write a book in this comment section I will stop arguing, just remember this, ai is fast ai in your brain and end. there are many possibilities in this end part dystopia or utopia.

[@GreaterStuff]

Thats alot of words for, " I want to be payed for my work."

[@Shackrah]

🎨Great ideas and opinions about the whole generative AI concepts, as well as the concept of value. I'm not sure if what you talk about is really achievable in the current context, but I really like the ideas. What do you do about totally free generative AI tools? Nobody is paying anything as such , so nothing can be given to the artist. are those tools and models condemned to disappear? That would be bad since those are the open source tools and models that can be freely modified and improved by the users.

[@CurtisHolt]

I'll copy a response I put to another comment, but one of the solutions proposes a new open source tool which I think would fit with that: Two other possibilities: x.com/curtisjamesholt/status/1792534626180731286 x.com/curtisjamesholt/status/1792629213419249990 Think about it as compensation for access, not the end result. Alternatively for the possibility of a sub fee, splitting it by time usage of artist-specified models, rather than generations. There are lots of different ways to split compensation before even thinking about inspiration or copyright. For Blender, we sell access to things that belong to everyone all the time: GPL addons. In that case, people pay for access to the download service rather than the addon itself.

[@buda3d2007]

I’m hoping for animation it could enable a small team to create what a much larger studio could do, stable diffusion 3 and their rumoured superior control nets sound close or over the target to get this to happen, of course trained on one’s own art style and empowering smaller creators with open source models and not tech poison that’s controlled by the world of shareholding tech bros 🎨

[@rickh6963]

From the perspective of someone with no artistic talent, what is the difference in someone typing "In the style of Curtis Holt" into a AI art prompt and someone with some talent looking at your works and imitating your style? Customer walks in with a bunch of Curtis Holt pictures and tells the artist that he wants a picture of "X" in this style, what then? I recently used Dall-E to generate a thumbnail for me to use. I typed in my prompt and it output something well beyond my expectations. I did not use any "in the style of" prompts. Do you feel that this is an acceptable use of AI generated imagery? Or do you still feel that every artist that who's work was used to train the AI needs to be compensated? Will any kind of artist compensation lead to an AI that is only trained on art from artists that are dead? The AI companies will just wait for you to die and then use your art. With the power of home computers, and the fact that the newest processors have built in NPUs, how will we prevent the average creator (or clever businessman) from spinning up an AI and training it on "everything" on the internet? Is the only way to stop unapproved training with your work to put it behind a pay wall and turn off all the search bot access so no one knows you exist? Maybe just not post any of your work on the internet to the point of only sending encrypted images of your work to those that YOU want to see it? Is being a paid artist no longer a viable business model, like Fast Food in our current economy, just not a profitable venture? Tough questions that I wish I had answers to.

[@KTMICD2]

The artist you hired started his work from scratch after years of training and is a human and as a self respecting artist will try to give you something in the style but not outright copy, the other is an unfeeling machine that literally steals the content others made, gives them no credit, takes seconds to make and robs humanity of the desire to be creative, making the art world incredibly repetitive and unimaginative and ultimately dead.

[@wsippel]

I don't think compensation can work. Everything generated by AI is public domain, which means the generations have little commercial value. Things change when generative AI is used in pastiche works, but you wouldn't get compensated for those, either. And with local models like Stable Diffusion, Kandinsky, PixArt or Stable Cascade, you can't track what users add to the models or generate to begin with, and those models are pretty much the future, anyway - especially in commercial settings.

[@CurtisHolt]

Compensation does not have to encapsulate every tool or local usage, just one platform that gives artists something to rally around. This would also be regardless of the licensing of outputs. x.com/curtisjamesholt/status/1792534626180731286/photo/1

[@GaryParris]

Visual arts and other arts are not just about pretty or whether value is inherent in it it's purity or bespoke nature, that is a commodification not an actual reason or limit on what art is and is done for and by whom, AI=LLM=processing uint's is also not what people prescribe to this current AI models, Ai has been around longer than most people think way before you got interested in it, i was doing AI back at Uni in the mid 90's, what has changed is the processing power, and some small developments in algorithmic inference, which all eventually breakdown to entropy when over trained Still very much based in what you ask is what you get. much of the time that is generic and repetitive and boring, use as a tool for sure, and some artists can use them effectively as a new generative experience, but mostly it will be used for consistently bland copycat imagery! it has more validity within science but should never be without humans checking it's validity for algorithmic bias! Opensource art libraries for AI art generation, by people who are willing to donate to it, but AI should be not be allowed to be monetised and IP, until AI libraries pay artists. been talking about this since the beginning of AI art and NFT🎨

[@ДаниилЖуков-э3э]

I don't think it's necessary to mention or pay authors whose work was used to train AI models. Because when real people learn from the work of other authors and are inspired by their styles, they don't have to pay for it. And if compensation for the use of other people's works is made mandatory, this will hit small authors who want to use AI as a tool in their works. After all, then it will become impossible to use stable diffusion models for free, and large companies will be able to easily deduct part of the money they earn from providing paid access to their AI.

[@pookienumnums]

i cannot stress enough that youre not asking Dalle-3. youre asking an llm that has access to the web and can search the web. and that can manipulate your prompts after searching the web for information related to your prompt. dalle has no idea who you are.

[@pookienumnums]

oh, and you cant copyright a style. if you want to be paid for a style generated by an ai, pay the people who you got your styles from first.

[@caryonplays9024]

I think is hard to explain but I'll try to explain why this model of monetisation won't work. When we pay for music on Spotify, we pay to "reproduce" the music, so we pay the licence of the copyright material. By the law, when a copyrighted work is reproduced, distributed, performed, publicly displayed, or made into a derivative work, we need a licence of the author or will be a copyright infringement. AI doesn't do nothing of that. I will skip distribution, performance and publicly displaying because don't fit on the discussion so, reproducing: to create a copy we need the original, all the images are discarded before the train process start, AI just identify patterns and never uses the original material again. What AI storages in the neutral network is the processes it learned to "redraw" patterns of the original image. So, by design, AI is no capable of "reproducing". Derivative work is when you make a similar work that's have the same purpose, AI training create a model which could be use to create a derivative work (not necessarily), but the responsibility of using it to the finality is of the final user, so, to be a copyright infringement, we need to judge each image created by it, not the model or the training database. You could monetize Loras, but i don't think you would be payed for your style, just for the work you did to create the model and the distribution, because, technically, you don't hold copyright on your style, only on individual images you create with it. Imagine creating a model and a big company (Disney for example) claims your model is way to similar to their, that could create a bigger problem than a automatic tool could.

[@CurtisHolt]

I'll copy a response I put to another comment, but the end result of the generation could be completely irrelevant for a compensation model. It does not require any definition of ownership or licence. The two other possibilities: x.com/curtisjamesholt/status/1792534626180731286 x.com/curtisjamesholt/status/1792629213419249990 Think about it as compensation for access, not the end result. Alternatively for the possibility of a sub fee, splitting it by time usage of artist-specified models, rather than generations. There are lots of different ways to split compensation before even thinking about inspiration or copyright. For Blender, we sell access to things that belong to everyone all the time: GPL addons. In that case, people pay for access to the download service rather than the addon itself.

[@caryonplays9024]

@@CurtisHolt Sorry, I posted my answer before it was finished. So I edited it. I understand what you want, but, before inspiration and copyright, there is nothing to compensate for. What I was trying to explain is, AI uses images just to identify patterns and information, after that, AI will never use the original artwork ever again, so, unless people hold copyright on information and patterns, we, technically don't need to compensate for using the original material. The way we can use it to "compensate" someone for the work is not directly linked with the original images. You could create a Lora of your own art and monetize the access of it(what I don't think is possible by the terms of service of stable diffusion, not sure), but, legally and technically, I could create a Lora using art of lots of artists styles that look like yours and create and and "abstract art style" Lora. The compensation, in that case, would be about the work you did on the model, not about the copyright art. Here is where I think the problem is, to be compensated for your art style, you need to hold copyright on it, not your images, your style, and, if you know what is the problem with copyrighting styles and how harmful this is to art art, this could be way more dangerous than an automatic tool.

[@CurtisHolt]

@@caryonplays9024 You’re almost there but we’re close to agreeing on some things. On the point about there being no copyright so nothing to compensate for - that is exactly how the GPL paid addon model works. They are free to distribute and use by everyone, and yet developers still get compensated for creating them. Because we as a community respect the effort of creating the tool, and choose to use marketplaces established to compensate them. On paper, it sounds pointless, but in practice, it works very well. Yes, the same can apply with models and loras. No, they don’t have to be based on stable diffusion. They could be based on a model entirely trained on CC0 material. Is it technically unfeasible today: yes, impossible in the future: I don’t think so.

[@caryonplays9024]

@@CurtisHolt I get that, and I agree, but, in that case, I, a non artists fella, could create a model using your art and be compensated by it. I don't think this is exactly what you mean. I think you want the artist to be compensated.

[@CurtisHolt]

That’s fine, that’s just how digital products have always worked. People take my products and try to resell them all the time. That’s illegal. And proving it’s illegal doesn’t require any artistic copyright, nor does it rely on any measurement of copyright on any generated outputs. The tool is the product. Alternatively, if the license requires, the ability to download the tool is the product. Both the tool and the generated outputs can adopt different licenses and rights, just like Blender. As we agree that AI art can’t be copyrighted, that is fine. And yes, I do believe that AI art can’t be copyrighted. People seem to have intentionally missed that point in the video. I said I feel like I don’t completely own the output, and I’m okay with people replicating my style if they want to, but I would like some form of compensation somewhere in the chain.

[@andripopa13]

🎭🎨

[@H8KU]

If you feel threatened by AI art existing then just remember that your art still exists. AI art is the next art renaissance, already in full swing, you can use it to your benefit, or be a luddite. And remember that everything anti-AI art people say today was said about people who use computers to make art with 30 years ago. They said digital art was not real art, it was fake, it had no value, it was a threat to traditional art. You using a stabilizer is computer assisted art, not actually done by you. The only pure form of art is then pixel art where you intentionally click every single dot - even traditional brush strokes are tool assisted. Even photography is not by an artist, there was no intentionality in those megapixels only a broad prompt by the so-called photographer.

[@cyborgzloth]

Art emoji

[@matet7772]

6:50 It is one thing that God is lack - Limits.

[@igorsiwanowicz512]

📷🔬

[@KevinMerinoCreations]

Thanks for sharing your thoughts in this video! I find it interesting that as the image generation learning models ingest more of their own output, they become less "creative". For me I think of it in terms of it being a form of digital inbreeding. At least for now the models are dependent on the creativity of humans. I am still concerned how AGI will change that part of the equation. 👏👏👏

[@Meta-Gnome]

🤖👽✏🖌

[@AllExistence]

I always want people to pay some attention to the "stolen" part of the argument. Because it's accepted without hesitation. But it's not a fact. What is stolen? There may be resemblance in the artstyle, but artstyles cannot be owned. And nothing recognisable actually remains after training the model. Is it REALLY different from the way artists take inspiration from stuff they see on the internet? Does taking someone's painting, grinding it into dust and making a new one from those really considered stealing art? What about people who create art from arranging photo cutouts in Photoshop, for example?

[@CurtisHolt]

It's interesting, because I didn't know how I would feel about that until I started experimenting with 'my style' in DALL-E. I have no idea whether it's stolen or inspiration, but the only conclusion I could come up with from the subjective experience of using it is "I feel like I own this result more than someone else generating the same result, because at least part of it was made using my work as an input." In this way, I think the idea of ownership being a completely binary and absolute thing might not completely apply. I'm not smart enough to conceptualize an alternative way of measuring ownership, but I find the conversation interesting.

[@AllExistence]

​@@CurtisHoltI think you should take note whether Dall-e knows your style, or is it Chat-Gpt. Because style can be defined by combination of features, and gpt knows those exactly, perhaps models wasn't actually trained on any of your pictures, but prompt contained all of the features that encapsulate it? Of course maybe it was. Can anyone tell? There might be unique styles that cannot be defined with words, but what about yours?

[@ameknite]

The compensation will not work, those companies that built those tools don't care about anyone except money. If they can get away with stealing they will do it, that's why the law and regulations are important. It's similar to ntfs, just capitalism

[@theafricanrhino]

You’re effectively asking for sound alike songs to pay royalties… not derivative work, but sure let’s pay royalties for inspired by, thing is thought artists should do that too , any time you create something that “gives” another artist a feel of their work then you should pay…

[@theafricanrhino]

I can already see every artist that uses pink and purple wanting a chunk of your income. I mean it makes sense wanting to be paid for influence

[@EdLrandom]

Personally, I love the video it's interesting to see ppl speculating about AI and talking about it and the results are impressive sometimes, but when it comes down to anything actually useful I don't think AI will do much, I'm completely over the hype, it can't drive a car better than some sort of pre-scripted solution would it can't hold a conversation and it doesn't generate images it finds them in the database. I think the entire AI conversation is a part of tech start-up grift that will first convince the public and investors about the next big thing is this new tech knowing well it's not the case and then we have a bunch of failing solutions in search of a problem, it been this way with Crypto, with "3D" stereoscopic movies, with ray-tracing in video games, voice assistants, phone app dev craze and so much more. We'll end up using some machine learning in limited extremely boring cases that will not improve anyone's life and move on to another pseudo-futuristic "technology". Maybe it will be flying cars next, or something even dumber.

[@gordonbrinkmann]

"...when it comes down to anything actually useful I don't think AI will do much" I think you are not aware of how much AI is already been used for and of course the development is not finished yet. And the most recent hype is just now that more people can actively use it with things like Dall-E or ChatGPT instead of having it working behind the scenes where most people do not realize it or are not affected by it. And with the controversies about stolen art it has gotten even more attention of course. But that is just a small aspect of AI. Will it be able to replace artists soon? I do not think so. Are there a lot of advancements and progress made with AI in all kinds of technical, medical and general scientific fields? Sure enough and this will continue and get more and more sophisticated. But as said in another comment, if AI would help with UV unwrapping or curing terminal diseases, it would be welcome but probably get less attention. It's always the scandals and drama that leads to debates.

[@EdLrandom]

@@gordonbrinkmann Yes, it's being used to do scams and slop "content" to gain social media algorithms. It's that every time I try to use AI to do CG-related stuff like rotoscoping some VFX footage or motion capture some animation, it can do a simple example, anything else and I end up re-doing all of the work myself. Same with chat GPT, everyone at this point knows it just makes stuff up presenting as a fact, the most fun/use I had with it is testing RPG roleplaying, but it gets old quickly and you kind of have to do all of the funny/interesting stuff yourself, it's in the role but it's like you are talking to someone very stupid basically. Medical research or chemistry, people keep bringing it up but to be honest, there are a lot of issues in science right now regardless, so until someone actually brings up an actual meta-analysis study on AI use in scientific research I won't believe it, it's just that most ppl like to imagine AI as some kind of genius that can solve any issue with more computer power but it's simply not how it works. BTW did you notice how AI de-noising didn't get better, even though we had a crazy headstart with that compared to other AI, there is a limit to what machine learning can and can't do.

[@gordonbrinkmann]

@@EdLrandom Have you really dug deep into the subject? Since you are so skeptical about the example of AI in medical fields, maybe read the Wikipedia article about "Artificial intelligence in healthcare" or the general Wiki article about "Applications of artificial intelligence". But these are just examples, a tip of the iceberg. You talk about (and I said, sure I understand that this has been mostly in the public focus in recent years) scams, deepfake, ChatGPT or AI generated images. But that's not all there is, on the contrary there is a lot more other stuff. So saying AI will not do much when it is already becoming used more and more (allegedly, there is still a lot to develop and improve upon) seems to me very narrowed down to these negative examples. AI might not become a genius solving all issues, but that is not really expected from the scientist and researchers in the field. You say "it gets old quickly" and you have to do all the "funny stuff" yourself. But that again is very narrowed down. Especially when it comes down to boring stuff, AI can do tasks which might need a little more than simple algorithms to be fulfilled, but are still (to the human) very boring, annoying and time consuming things which are predestined by the simple fact that routine and fatigue can cause errors. But anyway, I'm not going to continue an endless discussion here.

[@silkemyk3178]

AI can make better versions of your art in seconds than you can do in hours. AI is already taking artist jobs and will continue to do so.