Greater distinction needs to be made between "AI art" and "AI-assisted art". Images that are just generated by a prompt are low-effort and uncopyrightable. AI-assisted art, on the other hand, can have a lot of effort and human creativity involved. I see many images on e6ai that would be much better if an artist were driving and refining the piece, fixing unaligned elements and impossible poses. When effort is put in, the result is a lot more palatable, even knowing that AI was involved.

Most of the discussion about these new AI tools is about their ethics, but it's not a very interesting debate. Generative art as a field has the potential to reduce the labor, skill and time cost of artwork to almost nothing, which makes the true crux of the argument about the ethics of automation, which is an old debate with an answer we know all too well today.

The real important thing to look at is to predict the future implications, then decide what you're going to do about those implications.

I believe these things are near certain for generative AI in the next few years:
• Newer AI tools will be capable of generating at higher resolutions and achieve greater fidelity.
• The amount of data required for fine tuning concepts and styles will be reduced to only a few images, with superior results to fine-tuning today.
• Funding for generative AI tools will continue to explode as applicable uses for the tools continue to grow.
• Hardware will become more specific towards accelerating these tools, and become more available at a lower price.

In the furry community, I predict that as generative AI improves there are some effects in the short term (2-4 years):
• The proportion of newer artists starting out selling commissions as compared to the size of the community will likely diminish due to competition with AI tools.
• The vitriol towards usage of the AI tools will wear out and ultimately fail to impact their growth.
• Issues with specific, complicated concepts such as hands, feet will get ironed out.
• Existing artists will have the option of fine tuning an AI on their own artwork, but will be disincentivized to reveal that they're using AI in their workflow due to social pressures devaluing content creating with its use. However, they will be heavily incentivized by the potential labor savings, allowing them in the short term to work very low hours for high pay.
• Furries with lower incomes will increasingly turn to AI art, leading to an economic schism with commissioned non-AI art acting as a wealth status symbol. Furries already use commissioned artwork in this way (for instance Miles-DF commissioners).

In the long term (5-10 years):
• Generative AI tools capable of directly matching an artist's style with rapid fine tuning on small datasets will be capable of also drawing characters with fewer mistakes than human artists after tuning to that character.
• AI generations becoming extremely difficult to distinguish from a human artist, to the extent that "no-AI" rules become increasingly unenforceable. No way of verifiably telling if an artist has switched to using AI generations.
• Configuration options for AI tools becomes much more advanced, allowing input of poses, references for individual elements, allowing complete control of the final generation.
• Widespread use of generative AI due to increased fidelity and a new younger generation of furries who would have previously used generative AI when they were unable to pay for artwork, but are now used to using the AI tools.
• Status symbol artists still carry on, but for most artists competing against AI generations becomes increasingly taxing.
• Fields in the furry community that are very underutilized (e.g. game development, long-form animations) will grow rapidly as they can use AI to accelerate their output to levels that would previously require entire professional teams of people to complete.

That's roughly where we're headed, I believe, so what are we going to do about it? The current approach of stomping on people using AI tools will not stop people using them; in fact it might have the opposite effect.

I hate it, as how can i trust that say a photo of a girl on a dating site is actually her and not a computer generated one, and how do i know the pictures of a car on a classifieds site are real and not computer generated ?

Although i think the term AI is misused quite a bit, as AI (Artificial Intelligence) to me is a sentient machine such as the robots in the game Stray, Skynet in the Terminator movie and such, wheres what we call AI here is really just a script/algorithm like any other app on your pc/phone that generates an image based on a random reference photo/drawn image that the script/algorithm then goes online to find as part of the "generate image" process, which is why alot of ppl tend to call "AI" generated art theft.

I've always found it to be completely uninspired, generic, rather dull and most importantly, quite lacking in terms of personality. Moreover, I'm against the idea of using a machine to run algorithms to take various copyrighted artworks from the other artists, and then promptly refusing to give the said artists any credit whatsoever. Some people that I've known claim that it's "the same as studying art; the machine just does it faster" but I've to disagree over that notion. Studying other's work shouldn't be, and NEVER be compared to simply stealing work and mashing them together to generate something.

My second gripe over this is that the person behind it calling themselves an "artist". In no given way the person generating AI images is ever an artist. With this tech, every guy who generated something with CrAIyon will start calling themselves "artists" and it's rather insulting to us, the actual artists in my opinion.

lazyoldmutt said:
I hate it, as how can i trust that say a photo of a girl on a dating site is actually her and not a computer generated one, and how do i know the pictures of a car on a classifieds site are real and not computer generated ?

Although i think the term AI is misused quite a bit, as AI (Artificial Intelligence) to me is a sentient machine such as the robots in the game Stray, Skynet in the Terminator movie and such, wheres what we call AI here is really just a script/algorithm like any other app on your pc/phone that generates an image based on a random reference photo/drawn image that the script/algorithm then goes online to find as part of the "generate image" process, which is why alot of ppl tend to call "AI" generated art theft.

It's called Artificial Intelligence because it tackles tasks that are associated with human intelligence that are hard for machines to perform, as opposed to tasks that computers can always do better than humans, like mathematical operations. Artificial neural networks are not like traditional programs or algorithms, in that they are not told how to solve a problem, but are put through a training process with inputs, adaptation and scoring of their output to guide them towards producing better solutions.

There is a misconception that a trained model is simply mashing together things that it has previously seen, and cannot exhibit deeper understanding of concepts. In actuality, when trained models are given larger datasets and more parameters they tend to spontaneously "develop" unpredictable new capabilities. Google DeepMind studied this effect extensively with both image generators and LLMs, finding that capabilities such as rendering accurate text in images or being able to translate text suddenly appear once you give it enough power. Early neural networks were incapable of handling data that wasn't included in their datasets, but modern, bigger ones can. Something that's especially striking is that if you increase the amount of data for one type of task to a neural network, it gets better at the other tasks in its training data as well.

At present the capability limits are nowhere close to being met, Stable Diffusion cost $600,000 to train and can have inference run on consumer laptops. What people were able to do with it also came a long way in the past year as fine tuned models, LoRAs and hypernetworks came along, and then SDXL released which handles many difficult concepts better than the original model, at a higher resolution.

These models can and will be built without needing your™ artwork
Adobe Firefly uses only public domain content and stock photos it has license to. The outcome will be the same, because it's the technology itself that's disruptive. 70% of Chinese game dev artists lost their jobs because of this in a year, because they found that artists using AI in their workflow could produce the same amount of work as 5 people each. If they were using Adobe Firefly instead of Stable Diffusion, that doesn't give all those people their jobs back, it doesn't matter how it works or how it was trained, only what it does and how good at doing the task it will be in 3-5 years.

v01d said:
Generative art as a field has the potential to reduce the labor, skill and time cost of artwork to almost nothing

• Status symbol artists still carry on, but for most artists competing against AI generations becomes increasingly taxing.

Most artists in this fandom don't make art just for the sake of profit and efficiency, it's a hobby and personal expression. AI generation just isn't a replacement for making art themselves or commissioning work no matter what the quality is, the competition just isn't there when it comes to hobby artists, and artists working within fandoms on commission. There's a reason why most of the people posting and making AI generations regularly were not artists, and it's not just because of the negativity or because of the visual issues with AI generated images.

• Furries with lower incomes will increasingly turn to AI art, leading to an economic schism with commissioned non-AI art acting as a wealth status symbol. Furries already use commissioned artwork in this way (for instance Miles-DF commissioners).

Not all commissions are that expensive, there isn't as much stratification between "low income" and "people who commission" as you seem to think, artists like Miles-DF are "memes" because their art prices are so outrageous that they stand out from the average. Most commissions I think are under $200

For most, the choice of getting something unique and interesting from someone else, tailor made for them with feedback to the artist, potentially developing a relationship with that artist, and getting attention for their commission by being created by that artist for less than $200 (probably even less than $100), or having to spend hours of their own time troubleshooting prompts to get a substandard product, only to post it and have it ignored in a flood of other generated images... what are most going to choose? If they will increasingly turn to AI art, why haven't most of them done it already? Because they don't care for the results, because it's time consuming to set up, time consuming to get the prompts right, it requires technical know-how, and having the hardware that can do the generations.

• The vitriol towards usage of the AI tools will wear out

The vitriol towards NFTs, crypto, the Metaverse, and other Web3 "innovations" hasn't gone anywhere in several years, generative AI, whether accurate or not, has been hitched to that wagon, there's a solid negative bias amongst artists and art appreciators that doesn't seem to be changing - in fact, the negativity has been growing louder among the mainstream, experts and professionals as well. This bias is probably even heavier in the furry fandom, which is focused on creators - especially artists - and creativity. That isn't going anywhere anytime soon.

It's nihilism.

accipere_facile said:
It's nihilism.

It's kafkaesque.

maplebytes said:
Most artists in this fandom don't make art just for the sake of profit and efficiency, it's a hobby and personal expression. AI generation just isn't a replacement for making art themselves or commissioning work no matter what the quality is, the competition just isn't there when it comes to hobby artists, and artists working within fandoms on commission. There's a reason why most of the people posting and making AI generations regularly were not artists, and it's not just because of the negativity or because of the visual issues with AI generated images.

By proportion most of the furry fandom cannot draw, it's representative that most of the AI generations would not be from artists. Beyond just regular negativity, admitting to using AI in a workflow now is admitting to using less labor, which devalues what you've created: even if it's something as simple as backgrounds which generative AI excels at. Chunie experimented with AI to generate artworks in his style, but nobody is going to pay Chunie price for something he worked on for 1/10th the amount of time, so my point was that there's a growing incentive to save on time but not be honest about how you created things.

This fandom is home to commission scams on a large scale as much as it is full of people who make personal expression their main goal. People who take massive amounts of commissions and then bail, or ponzi-likes that use recent deliverables to issue refunds to 1/20th of their 200-long backlog while keeping the rest of the money. That's not just unethical, these are flat out criminal schemes that operate non-stop, every day.

Comparatively, not admitting you used AI fine tuned on your own work as a baseline for pieces you're selling is much less black and white, especially when you consider that artists might have to choose between honesty and being able to make a living with reasonable work hours (10 hours a week versus 60 or 80). Many of them are barely getting by, to the extent that "emergency commissions" have become a common concept for a couple of decades now. I spoke to a graphic designer that started secretly using generative AI to save on time at his company and he now works only a few hours a week.

maplebytes said:
Not all commissions are that expensive, there isn't as much stratification between "low income" and "people who commission" as you seem to think, artists like Miles-DF are "memes" because their art prices are so outrageous that they stand out from the average. Most commissions I think are under $200

Almost all are not, that's true. Miles isn't so much a meme as he is the definition of a specific concept; you don't buy his artwork for the quality or even the style, there are countless copycats in the past who drew exactly like him, but because he's the name in expensive commissions. What I was trying to say is that people like Miles don't have to compete with AI, and that the future of artists that don't use AI in their workflows is to try and enter that market, but that it's not big enough for everyone to join.

maplebytes said:
For most, the choice of getting something unique and interesting from someone else, tailor made for them with feedback to the artist, potentially developing a relationship with that artist, and getting attention for their commission by being created by that artist for less than $200 (probably even less than $100), or having to spend hours of their own time troubleshooting prompts to get a substandard product, only to post it and have it ignored in a flood of other generated images... what are most going to choose? If they will increasingly turn to AI art, why haven't most of them done it already? Because they don't care for the results, because it's time consuming to set up, time consuming to get the prompts right, it requires technical know-how, and having the hardware that can do the generations.

In the past, if you wanted to do AI image generation you had to read papers and tinker with research models to get results that almost never worked properly, with little to no control. This is how it was with StyleGAN in 2018, and later the precusors to "This Fursona does Not Exist". Now AI is a lot more consistent, we went from making 20-30% "okay" results 2% of the time to 70-80% "okay" results 30-40% of the time. It also became much easier to use; Dall-E 2 could produce excellent results with natural language prompts, while Midjourney is a discord bot you interact with.

Artists are also rather slow at adopting new technologies, for instance it took a long time for people to start moving on from the original Paint tool SAI 1 which hasn't been updated in 8 years, but there is a consistent shift towards advanced tools like Clip Studio Paint with time-saving features. It's this sort of pattern I'd expect to repeat itself, first it will be using it in a limited fashion for things it's good at like backgrounds and concept references (we are here), and then later as it gets more consistent and easier to control it can do more of the heavy lifting.

When it comes to tailor-made content, AI workflows can gain a significant advantage. For instance making example references for what to have drawn: AI today can currently create a hundred variations of different poses, outfits, scenes. Comparatively a traditional workflow has a few suggested sketches, or no options at all.

The flood of generated images is going to flood out almost all images, not just other AI generations. It's something that will only continue to get worse as AI gets better at the task to where if someone doesn't admit it's AI, it would take a significant time investment and knowledge to prove it's AI generated. Worse still, people who don't use AI being mistaken for using it.

maplebytes said:
The vitriol towards NFTs, crypto, the Metaverse, and other Web3 "innovations" hasn't gone anywhere in several years, generative AI, whether accurate or not, has been hitched to that wagon, there's a solid negative bias amongst artists and art appreciators that doesn't seem to be changing - in fact, the negativity has been growing louder among the mainstream, experts and professionals as well. This bias is probably even heavier in the furry fandom, which is focused on creators - especially artists - and creativity. That isn't going anywhere anytime soon.

The Web3 craze was different in that most of it was a scam; they claimed metaverse platforms and applications for crypto and NFTs that didn't have proof of concepts or real products. It wasn't the hatred of NFTs on social media that caused their collapse, but that it was revealed to be a paper tiger with no real world applications.

The "Metaverse" as a general concept has gone a massive distance in the past 5 years with increasing consumer spread of VR, with VRChat tripling their player averages in the past 3 years, after 20 years of pretty much only Second Life. Meta embarrassingly overestimated the short term growth of their VR business, missed their market with the quest pro and released the technologically inferior "Horizon Worlds", but that's one company's blunders in the short term and I wouldn't sleep on the idea of a "metaverse" displacing social media.

But not to get too fixated on the other fields here, generative AI is a beast on a totally different level. It's proven to work unlike Crypto and NFTs, and it does things that people already want done. "Creativity" is always brought into question regarding AI fields because it's a philosophical debate with unsettling consequences; there's nothing in known science that prevents a silicon machine from performing any human task at any level, but the idea of it isn't pleasant. In fields where AI has beaten humans, it has shown uncannily ingenious ways of achieving those results and AI approaches problems in different ways than humans.

Final note, I think vitriol is not effective when it comes to stopping anything, there are many people in this community who are subject to (deserved) vitriol far greater than is directed towards generative AI, but continue to have success. I'm increasingly convinced that social media vitriol might be one of the most ineffective ways of getting changes made. It drowns out the actual conversation and encourages a sea of escalating demands that cannot be met, ultimately resulting in no changes being made. When it runs out of stamina, you're still left in the same place you started. For generative AI, demanding everyone stop using the technology under threat of harassment does not stop people using it. In fact it generates sympathy for the people using it and a slow growing fatigue of wondering if it's really doing the right thing to hop onto social media every day to hurl abuse at everyone who made an AI generation of their fursona.

lazyoldmutt said:
what we call AI here is really just a script/algorithm like any other app on your pc/phone that generates an image based on a random reference photo/drawn image that the script/algorithm then goes online to find as part of the "generate image" process

Common technical misconception. The weights in the model doing the image generation were fixed at training time, as a description of the statistical properties of its training set. It doesn't go out and find references at inference time, it just produces a prediction of another image that statistically might be found in its training set, if you kept running the same process that originally produced it for long enough. See, well, any number of sources about diffusion image generation works, it's not hard to find plenty of them. Or actually, don't - GANs were all the rage a few years ago, maybe MaskGIT will turn out really popular next year, diffusion is just one of many ways to set things up so that you can generate images out of those statistics.

(This line of questioning tends to lead down a path of asking what's actually *in* the weights and whether there's exact copies of other people's work in there.

I'm a bit less certain than a lot of people seem to be about the "it's 4 gigabytes, you couldn't possibly fit all of its training data in there" argument: Stable Diffusion's architecture is an incredibly more dense way to describe image dataset statistics than a list of JPEG files is, that's kind of the point. Training data extraction attacks already worked in 2020 on the barely larger GPT-2, too.

But in practice, what we're actually trying to do is pull new images out of the model, not existing ones, and it clearly works for that and generalizes quite well. So the relevance of the question seems unclear to me, and I don't know why arguments about it seem so popular.)

I'm just starting to learn digital art but I've kind of accepted the stereotypical AI art as a reasonable minimum standard for the most part. Stuff like this though just makes me excited for it.

It just tickles me that AI porn started being done approximately five minutes after AI art became a reality.

'What kind of art would you like to generate?'

'I want some porn.'

'...I can create fantastic fantasy worlds, breathtaking dreamscapes, gritty visions of an alternate reality...'

'That's nice, I want some porn.'

iwasnthere said:
I'm just starting to learn digital art but I've kind of accepted the stereotypical AI art as a reasonable minimum standard for the most part. Stuff like this though just makes me excited for it.

If you're trying to learn digital art, AI generated images are a very bad benchmark for the "minimum standard" to hold yourself to.

v01d said:
Artists are also rather slow at adopting new technologies, for instance it took a long time for people to start moving on from the original Paint tool SAI 1 which hasn't been updated in 8 years, but there is a consistent shift towards advanced tools like Clip Studio Paint with time-saving features.

That's just comparing shifting popularity of art programs, that's not really comparable to jumping from the actual process of drawing and painting, to entering prompts and using the results to further generate towards what you want, that's essentially a different artistic medium. Clip Studio isn't as new as you think either, it was on the market as far back as 2001, known as Manga Studio. What changed was that it became better known in the digital art community, and so it grew in popularity.

Additionally, many people continue to use Sai 1, Sai 2, Photoshop, Krita, Medibang, FireAlpaca, GIMP... there's not a "constant shift" to time saving programs in digital art merely for the sake of time saving features. On top of that, not everyone uses the time saving features in Clip Studio at all, some people never use the stock effects, comic panels, or 3d model abilities, but use CSP simply because they like the interface and brushes. I guarantee that there are many artists using Sai and Photoshop, who, despite seeing time saving features like comic templates, speech bubble templates, and stock images, shrugged and continued with their "less efficient" workflow, without a care, and continue to be successful.

Again, I really doubt most artists are going to jump to using AI any time soon, if ever. More will use AI in some aspects... but regardless of whether it can generate ideas, or make backgrounds, some people simply just prefer to sketch 15 thumbnails, or collect photos, or other people's art, and use those for reference and idea generation. AI can make backgrounds, but they usually have to be altered, and they have to work with your art style.

The mistaken argument here seems to be that near universal adoption of AI in the art process is inevitable on the grounds that it's inefficient not to, and that there's a lot of it, some people simply don't have a want or need to be as efficient "as possible" with their art — look at the millions of artists that work mostly or entirely traditionally (or in addition to digital art), and in spite of digital art probably being most artwork online today.

I wouldn't sleep on the idea of a "metaverse" displacing social media.

I'm fast asleep on a "metaverse", the overwhelming majority of people are simply not interested in using virtual reality just to browse the Internet, it doesn't offer any advantage over sitting at a computer, or opening an app on your phone or tablet, and introduces a lot of tediousness. Why would the average person want to strap on a VR headset to walk around a virtual Walmart to do their shopping? That's putting aside the sterility and aggressively corporate nature of the "Metaverse".

The flood of generated images is going to flood out almost all images, not just other AI generations

I really and truly doubt this, people are constantly creating artwork and photos of their own, to claim there's going to be such a massive shift to only uploading AI generated images, and it'll spam every corner of the internet until there's no room sounds like a absurd argument someone would make against AI image generation.

Although there is a grain of truth here, in that sites like rule34, Pixiv and DeviantArt are currently flooded with AI content, which seems to irritate the majority of users, to the point where there are settings specifically to filter them out, or requirements to have tags on AI generated images to filter them out.

Final note, I think vitriol is not effective when it comes to stopping anything, there are many people in this community who are subject to (deserved) vitriol far greater than is directed towards generative AI, but continue to have success.

I'm not sure who you're referring to, but there are a number of people in the Furry Fandom who have essentially become persona non grata and they disappear, or lose a significant portion of their audience and only have an audience among people who tolerate whatever reason they faced vitriol or are somehow unaware of the vitriol.

I'm increasingly convinced that social media vitriol might be one of the most ineffective ways of getting changes made. It drowns out the actual conversation and encourages a sea of escalating demands that cannot be met, ultimately resulting in no changes being made.

In fact it generates sympathy for the people using it

This sounds like an indirect complaint about "cancelling" on social media... which generally does actually seem to work? Sometimes too well, if there's a genuine misunderstanding.

There are very few instances where I've seen sympathy come about for people who face online vitriol, except among people who share whatever opinion that individual had in the first place that upset so many people. I don't want to get political, but I think everyone in this thread can think of some sort of situation like that.

Someone who plays with AI image generation, who makes it clear that they're making AI generated images, and doesn't act obnoxious about what they're doing, isn't going to face much, if any, hate directed at them specifically. I've yet to hear of any broad and concerted harassment against a private individual making image generations merely for the reason that they're making image generations. What I have seen is hate levied at people spitefully using a particular artist's work as a data set to "replace them" and mock them for not being positive about AI image generation, and vitriol towards companies for using AI generated images instead of hiring an artist.

For generative AI, demanding everyone stop using the technology under threat of harassment does not stop people using it.

I haven't seen too many people who are against AI generation images make any serious demands like this, nor threaten harassment, what I do see is vitriol toward people who claim they are better than artists, who say they are replacements for artists, for people who spam AI images where they're unwanted, and vitriol against using data scrapping (which is already seeing incoming legal backlash).

maplebytes said:
If you're trying to learn digital art, AI generated images are a very bad benchmark for the "minimum standard" to hold yourself to.

It occurs to me the janky AI I was thinking of are pretty much old and irrelevant, the newer AI imitates artists way more experienced than I am (Always has but more accurately than before). 🤡

Oy, besides the obvious "Keep improving! 😃👍" I need to sleep on that or something. 🤔

AI is a new tool. And just like any other new tool for artistic expression that gets a modicum of attention, a bunch of people who hadn't touched art before are using it, and a bunch of people resistant to change are opposing it. Photoshop et. al. didn't invalidate their predecessors. Digital art didn't invalidate photography. Photography didn't invalidate manual artistic media. And so on and so forth. A few years from now, the trend of people mass-producing AI art and flooding art repositories of it will die down - it's a novelty right now, after all. Imagine being one of the first people to own a camera! You'd likely be taking pictures of everything, and likely wouldn't have even close to a modern understanding of photographic composition. The big difference with AI is that, unlike all these other tools when they were new, it's not purely accessible to the rich. Anyone with an internet connection and a half-decent device can access some manner of AI generator. So rather than a small number of people firing blanks in the dark early on, it's on a much larger scale, and with the internet, it's much easier to see. Will people lose jobs? Yes. Technological progress generally invalidates some manner of profession. But I imagine that, before long, people will start finding ways of using AI generators to assist in art, as another tool in an ever-growing toolbox.

nuclear_furry said: ...
Now, I’m a believer in a future where artificial beings can make actual art too. I would LOVE to see an AI’s actual perspective on the world. But frankly, we are not there yet and won’t be for a while.

Your whole post is spot on. I especially think people need to realize that we don't have real AI (yet). My simplistic way to describe what we have now is "glorified search engine".

People like this are the reason this stuff's gotten so rampant, I think.

They don't seem to understand the magnitude of the situation and believe certain artists are angry about it for no reason. I may not be able to do anything about any of it, but at least I'm not like the user I linked to and I actually understand what's going on.

https://m.youtube.com/watch?v=-MUEXGaxFDA

milabue said:
Your whole post is spot on. I especially think people need to realize that we don't have real AI (yet). My simplistic way to describe what we have now is "glorified search engine".

I wouldn't even call it a search engine, as that implies it's capable of presenting its source material. These models only contain enough mathematical data to answer questions like "what colors would likely be appropriate here given the data from the previous iteration as well as the prompt?"

My opinion of this whole AI art thing is that, well, the box has already been opened. Whether it will mean good things or bad in the end, it can't be ignored, and fighting against it will only cause harm in the long run. For example, if professional artists are discouraged from learning and incorporating AI-assisted tools in their workflows, many are going to have a harder time remaining competitive.