Our vision of the future used to be one where every unpleasant job is automated, and humans are free to pursue higher callings such as art.
It's increasingly looking we'll all be working in construction, scrubbing floors, or performing maintenance on the machines that took our pleasant jobs.
The saving grace is it'll probably be just 10 hour work-weeks.
1) for their own creative fulfillment. Generative AI makes no difference.
2) for enjoyment / appreciation from other people. Generative AI makes little difference here since the average person's art is likely to be mediocre with or without generative AI
3) as a way to earn their living. Here generative AI will make a difference in some art professions, but that doesn't seem to be the implied scenario of the comment I was replying to.
Do you need to have the validation from others that your art exists for you to want to create it? If I read every book by only jumping to the final chapter, am I interpretating the work incorrectly or am I not interpretating it how you specifically look for it to be validated?
I keep photographs of moments as memories, eventually I'll collage them together, maybe I'll do this a few times to develop a style that I enjoy. For the argument, say that it looks horrendous but in a consistent way that I like. This collection and its output have significance to me and to others I'd share it with because of the attachment to the reflection of my identity, not because it was something I created "correctly". If AI art was generated of every moment of my life and done in the same style, it wouldn't have value not because it was made incorrectly, it wouldn't have value because it has no personal meaning to anyone.
If you view art without meaning to others being less financially valuable as an attack on all creative work, I don't think you view art much better.
If you didn't read the whole book you didn't experience the work in full.
You're allowed to read it however you like. No one is saying you're not. But, given those circumstances, is your interpretation of the art valid? At that point why even read the last chapter? Just make up your opinion based on the cover, you'll get through books much faster since that is your only goal.
But understanding an art piece frequently (usually?) doesn't end with that individual piece of art anyways. If Duchamp presented Fountain in 2024 would it have the same appeal? Could Fountain have been produced in 1200? Does the fact that Fountain started as a mass produced item fundamentally effect it's statement?
Your final sentence is just a completely unfounded ad hominem? No reason to attack me bud. I haven't told you how I view art (except for "in full") and I certainly didn't say anything like "art without meaning to others being less financially valuable is an attack on all creative work"
> Do you need to have the validation from others that your art exists for you to want to create it?
I wouldn’t be surprised if a good chunk of art is meant to be consumed rather than created. but the only way to know that is to gather all artist and ask them.
There are some art that insincerity meant to be seen of course. Such as murals , street, Large wall Paintings and it would be silly to act as if they’re not meant to be
No, it really doesn't. The value in Fountain has nothing to do with the tangible piece of art itself. That's what I'm trying to communicate to you.
Attempting to boil "art" down to any one quantifiable aspect is impossible. That is the point of the anti-art like Fountain.
In other words, I disagree with your assumption that highly polished corporate art is superior to home made "mediocre" art. I grew up with TV shows like 12 OZ Mouse, Home Movies, Shin Chan, Tom goes to the Mayor, etc. If you think running those shows through generative AI would improve the shows... I would say again you don't understand what made that art good.
At best you could try to make a case for "death of the author" but I'm not sure if that principal applies to art that never had an author.
> 3) as a way to earn their living. Here generative AI will make a difference in some art professions, but that doesn't seem to be the implied scenario of the comment I was replying to.
Generative AI can copy the style, if they's all you have as an artist, then it's barely art TBH. AI can't copy the thinking / motivation / juxtapositions behind the art.
Ethical concerns aside (which is a big topic by itself) having been quite invested in the generative scene for years now, no matter what some people say there is a core of more power users that push to experiment with the technology, the styles, the limitations and the weirdness of it. Not that it has crazy complexity, but some still.
But at the same time the barrier to entry is really low. I now see AI art everywhere: in articles, on LinkedIn, on posters etc. But most of the time in the very most basic style. It feels like instead of searching for an image of X people just type "image of X" in MJ / DallE3 and co, and go with whatever generic stuff they get
This tech became available to a mass audience a year ago, tops. I would be a little patient: it will take time for people to learn how to use it and not just use the default settings.
Imagine you are an artist and have a specific style and one day you wake up and the computer program that ripped off your style gets credit for that particular style, as in "4000 midjourney artistic styles". Horrible.
As an artist, your style is not the only thing that defines you. It's about how you use that style, similar to how one utilizes a patent. It involves adding your unique human touch, from your voice to your hand on the canvas. it's not the end of the world.
See it’s weird since I would consider this a form of immortalization. Some artists are seeing this way, but very few. It seems crazy to me more artists aren’t being amazed that their entire style is now “canon”
This is all so wild to me. All of human art is derivative of other art and all human artists wildly steal from one another.
Further, "style" isn't copyrightable or sellable.
Artists don't sell a style, they sell artwork.
I just think it's wild that the entire human art scene is allowed to blatantly rip each other off (all pop sounds the same, all anime looks the same, all comic books look the same, etc etc) and it's OK, but the second that a computer does it, it's the most horrifying crime against art in history.
I think the fascinating part to me will be when AI is good enough to pick apart human art and point out all the thefts. "You stole this chord, these lyrics are basically ripped from here" "Your style is 98% this artist and you've only really change two elements" etc.
Artists hide behind their influences and it's going to be a new world when a computer can ascertain every single reference that was stolen, going to be a lot of interesting law suits.
A model that could reliably match up your work with similar artists and similar works (and point out the details) would be cool the have, tbh. Especially in music, where it's so easy to subconsciously copy some tune you vaguely heard somewhere before. Could set aside that worry of accidental copying, as well as give an objective measure in trying to steer towards a more unique niche.
Sure, I’m just saying I’d have expected more artists to say “I know this will harm me financially but as an artist it is the most transcendent kind of success I could have imagined.”
Eg Grimes sees it this way, anyone else? It’s very surprising to me this isn’t a dialectic within the global community of artists. It points to me as a sign that perhaps art has become so intertwined with commercial and financial interests the latent mental models artists have of art have shifted over time, broadly speaking, away from art for art’s sake or art to speak.
See it’s weird when somebody is upset when corporation steals and uses their code I would consider this a form of immortalization. Some programmers are seeing this way, but very few. It seems crazy to me more programmers aren’t being amazed that their entire codebase is now “canon”
If I had a style of coding (whatever that means) and it got embedded into the models in a way that would cause it to echo for eternity long past my death, and was recognized as mine, I would consider it a successful work of art.
I guess at the end of the day, they're just fakes like any other fake reproduction. They'll be sold on marketplaces, occasionally a few buyers will be duped. But most are buying the style, not the artist, and wouldn't care.
The issue for this site is that they're using the artist names against the reproductions -- that could be argued as intentionally misrepresenting the origin of the works. I imagine, in no short order, they'll remove the artist's names or otherwise change the language to reflect that these are not the artists original works.
I'm not sure what people would use these for, other than their own personal use? Maybe inspiration, mood-boarding, or concept art... Some printing at home? I don't know.
Somewhat of an aside - it occurs to me there's an analog in population ethics here, e.g. Parfit's "Repugnant Conclusion".
Would you rather have a few people thriving and the rest not doing so well? Or is it better to have more people doing marginally better, at the cost of those few thriving exceptionally?
And the people most likely to use AI to copy artists probably wouldn't have hired the artists in the first place. Some people are losing on opportunities, but I think. in most cases, those opportunities didn't exist to begin with.
I think artists who create art out of love and passion will be valued even more. Much like how handicrafts in an industrial and mechanized day and age still have their place. Also the countless “amish furniture” references I’ve encountered are a testament to that.
The artistic record is going to be stunning. “Now you can see an explosion in craftsmanship and diversity of perspectives from the Renaissance, when art became professionally viable outside of the patronage of religious/political institutions, to about 2020, when a new technology made it totally unviable for anyone to dedicate their life to developing artistic works. But hobbyist artists kept making it in their free time, so this is what we’re left with!”
There is a difference in material. Mass industrialised processes clearly can’t always replicate what small artisans produce. But when the output is digital most of material differences are gone. There is no reason AI cant generate 100% same digital image or text as human would do. But if you wanted to recreate “amish furniture” you probably can automate some parts but you still end up having the person/artisan. Industrialization didn’t go all all the way as AI did yet if you read people like William Morris it’s clear we’ve lost so much.
idk, social networks are kinda brutal to artists nowadays. In soo many ways.
And too much ai art would make discoverability harder. Even now sometimes if you google some style/artist/a character, top sites on google are AI generators...
A lot of these are very pleasant to look at.. yes, it's easy to spam them so it can get old very fast. But I'm in Midjourney socials and not gonna lie, everyday I see something that stands out and amazes me. Should I feel bad because I'm looking at a flashy AI fake?
A lot of other professions produce essentially solely economic value. Art and journalism are sort of inverted: people happen to be willing to pay for it which allows it to be produced, but the real value to society is external to that transaction. Thus: museums and archives.
It doesn’t make sense to preserve analogous records for most other professions.
I think it's a consequence of abundance, which can lead to excess and undue waste if we begin to lose control of self-regulation.
Resources will have to fairly be being distributed; not through authoritarian collectivism mechanism, control and suppression isn't the answer - but through "a rising tide lifts all ships."
Tyranny is extremely expensive, and abundance generates exponential riches for all - so long as the greedy and authoritarian don't capture systems and implement policy that suppresses various populations; why I believe Elon Musk said he thinks UBI is inevitable, however how that manifests is important. A corruptable-capturable centralized state government is too much of a single point of failure to risk such a mechanism, and if learning from the efficiencies of free market systems then in reality it will be the controller(s) of systems who then employ and fund their workers enough [who voluntarily decide they are the best employer to support to accelerate the values-processes of the organization's hierarchy] via their inherent built-in profit mechanisms to the organizational structure, to allow for those families to thrive, and fuel evolution towards such a culture winning out; otherwise "leave the dead to bury the dead [who are misguided off a cliff by tyrants]" - as Jesus said.
I agree too, the "flashy AI fakes" are in my mind equivalent to porn, crafted to maximize excitement but void of soul-spirit-authenticity, etc; at least so far - I could see, as AI avatars begin to become a thing that an individual can refine to respond how they deem acceptable, could also craft or refine AI art generator styles to weight various aspects in much more nuanced way - that then reflects their own tastes and sensibilities.
I think gradually things will settle down. People are beginning to understand that many of these image generators are built on theft, and are pushing back accordingly. We need one or two Kim Dotcom adventures and the rampage will calm down.
Doubt any companies who needed commercial art will care at all. Most likely they will force artists to use AI to hyper speed up the process and fix some of the imperfections at the end. But there is no way they will be keeping artists exponentionaly so they can produce the same result the “inefficient” way.
I am not against using machine learning, nor am i against companies doing so. What I dont want is stolen property, which a lot of ai currently relies on.
Generative AI are trained on "stolen property" the same way as human artists are trained on "stolen prolerty". Both study previous masters, learn to copy their styles, "inspire" their own art from some combination of existing styles.
Perhaps you are not familiar with how ai and machine learning work, and that’s okay, but ai doesn’t “study” and most certainly doesn’t “inspire” itself. It builds token databases that it then regurgitates. There no resemblance of human behaviour, just a simulation. Software always did that.
Exactly and LLMs dont have emotions, audiovisual experience, human experience or any experience at all.
For humans its impossible to create same art as someone elses. For LLMs its impossible to create anything else than copy of art thats in its database.
Even if you model yourself on someones work and try to replicate it the results will be different. Its not that artists want to always create work different from each other its actually impossible for them.
Saying that art is “just” recombination of prior art is extremely dehumanising. Its like saying are ideas are just recombination of the ideas before… ok so general theory of relativity is just recombination of caveman thinking about fire and finding out how to create it?
> Exactly and LLMs dont have emotions, audiovisual experience, human experience or any experience at all.
> For LLMs its impossible to create anything else than copy of art thats in its database.
You forget that LLMs don't generate images out of its own volition. You need to prompt them. That's the injection of emotions, human experience, into the LLM. LLM provides technical means to create an image, the prompt author provides the idea behind the art, the composition, juxtaposition etc.
> Saying that art is “just” recombination of prior art is extremely dehumanising.
Based on your previous lines, you did notice that I mentioned "prior art, audiovisual experience, emotions, human experience", so I won't reply to that strawman.
Promp is injection of experience how? You pick from artist you like https://midlibrary.io/styles and try to describe what you want on the picture. Are you aware that most artists already work this way - you hire them with an assingment. No client would claim they are authors or that they injected emotion into the work.
Execution is the creative and hard part not thinking out the content of the artwork. Most artworks content is super basic like face of a person. Midjourney will make even the basic prompts look awesome.
I am sorry but you wont experience what artists do by prompting LLMs. We just stole, copied their work and possibly destroyed their lives. Shame.
We don't know that much about how exactly the humans learn and how AIs learn. Both are still to a degree black boxes.
But you're asserting that in case of AI training it's "stealing property", but imply it's not stealing for human learning. I think you need to substantiate this claim/difference. "Humans learn differently from AIs" is not enough.
I am surprised by the cognitive dissonance these people exhibit. On one hand they confidently claim that humans are stochastic parrots, on the other hand they claim they dont know how humans learn and think. They claim ai is intelligent but dont know how to define intelligence.
Right not with a high degree of confidence we can state that they are taking what’s not theirs, apply software to it, and resell it.
Everything else is their defence of said process, gaslighting people and spreading schizoid fear to obfuscate it.
> On one hand they confidently claim that humans are stochastic parrots, on the other hand they claim they dont know how humans learn and think. They claim ai is intelligent but dont know how to define intelligence.
I am surprised at the prevalence of building armies of strawmen here.
Here is a list of various artist/style cheat sheets available for SD 1.5 and SD XL. There may be other sheets for MidJourney but I have been into SD for a while and that's why I have collected these. https://gist.github.com/SMUsamaShah/218e602d508e891a123929ce...
I do wonder how many distinct artistic styles Midjourney and other similar image generators actually have. Say that you you were to cluster all the styles it knows about based on similarity and then generate images of the same things with all of them - how far apart would the styles have to be before the difference between them is larger than the normal variability in output? It's my general understanding that getting a reasonably decent imitation of a particular artist's style is usually a lot more difficult than just putting their name in a prompt, that what that copies is generally fairly superficial and high-level stuff.
Can anyone share artistic projects that have used generative AI? Where it is used as a tool, and not as a goal.
ie something that’s not:
- Poster/illustration where stock photo would have been used beforehand
- NSFW
- Experiments trying to display model quality (like on /r SD)
The closest I’ve found so far was Rock paper scissors (found from HN) [0], but it’s unclear if it was really less work than usual CGI/animation techniques, and for worse quality.
I’d be particularly interested to know it has been used by (amateurs?) webtoons/comics, and how they have solved the issues related to consistency of style and character.
we produced a connection card game in a couple of months end-to-end using AI. text is a true mixture of our ideas and GPT's 'ideas'. we gave GPT the vibe we wanted, let it generate tons of new stuff, then we heavily curated and remixed it.
art is created and curated by various AI tools. it does not rely on any particular artist's style. we bootstrapped the game using our own money, and this was only possible because we were able to produce our own visual assets.
AI is a tool in our kit just like figma and photoshop. people consistently have fun with the game in our playtest, and we're bringing our first print run to the world soon.
We’re going to get a first run printed at the start of March. Ship would probably be April-May. Doing some final content polishing and playtesting before then.
Last year I used Stable Diffusion + ControlNet to create a ~8x3 in sticker for my pal's motorcycle. I won't share the result here for privacy (and possible copyright/trademark) reasons.
I created the basic design in InkScape, then used SD and several ControlNet adapters to render photorealistic images from the design until one was deemed perfect. My friend loved the result, and since I can only do some sort of glorified programmer art, I wouldn't have been able to deliver a high quality image like that without using SD.
My example is kind of trivial, but I've used Midjourney with its --tile option to make backgrounds for web pages. One case is an internal tool where I just wanted to spice up the homepage and so I put in a rotating background, and another is a game where I put in one of ~50 backgrounds based on the environment.
I don't think many people use the tile option in Midjourney, but it's fun because it's not easy to make tiling images, and the result is generally aesthetic and not representational so the AI aspects are fine.
I have a friend who is using Midjourney to illustrate some instructional material she's writing. It's a little like stock photography, though without AI she wouldn't have used stock photography, and the art itself is more like graphical design. She keeps it fairly abstract but has artists and themes she uses consistently (lots of Ezra Jack Keats). She's probably creating dozens of images to get one usable one. But both of us enjoy the process, there's something a bit meditative to making Midjourney images sometime... a quirk of their Discord interface too, I suppose. Dall-E is much more instructable, but doesn't have any feeling of flow.
I've toyed recently with the idea of creating a massive number of pre-built avatar images for NPCs: https://hachyderm.io/@ianbicking/111740198056655468 – for my use case the most consistency I need is to change the age of characters, which works well with a fixed seed and no other pose changes. I'd probably use Stable Diffusion because Midjourney isn't scriptable and Dall-E is too expensive and lacks controls for aspect ratio and seeds. Obviously lots of people do live character image creation, but the cost and operational complexity kind of annoys me.
I’m creative but have never been adept at sharing visual ideas in my head with the rest of the world. I’ve found genAI image tools empowering because for the first time I can turn an idea into my head into something other people can see and enjoy. I’ve used it for everything from portaits and memes to recreating scenes from my memories that I don’t have photos of to help me remember.
I use a website with good night stories for children, the texts are written by humans, but the illustrations are AI generated. There's no consistency in style between stories and it doesn't matter. My daughter loves the illustrations.
It's ad-supported, low budget (cause small market). There's a new story every day, I don't think the site could afford to hire an illustrator (which would also cause friction in publishing).
There's just one picture in every story, but they usually look nice, and my daughter never forgets to ask to see it, seems to be a highlight for her. I think it's a great use case for generative AI.
This is key. The unethical part is not the use of AI (which is neutral), but corpos taking billions in profits from creators using creators' own copyrighted works. BigAI currently feels entitled to simply take these works for their own use, without permission or licensing.
This isn't hard stuff. Adobe Firefly is an example of an ethical, safe AI-powered art product that has been trained on owned stock images, licensed content, and public domain content.
Linkin Park used it in 4 videos recently and seemed to use a combination of img2img and leaning into the glitchy style. [0] (These are 9-11 mo old so using much older models and techniques)
Peter Gabriel used it extensively in some recent music videos. The artists behind it also leaned into the glitchy style but are probably using some pre trained stuff to keep style.[1]
A writer/creative named Austin McConnell used AI art to make a 50 minute anime short to help market a book he wrote using AI [2]. Not sure how he kept consistency but this video got some flak and he has another video addressing his techniques.
Corridor crew did a second video which is a lot better but still a lot of work. [3]
I think a lot of projects using it are still kind of in the spec stage and are usually using a combination of loras, clip, generating multiple angles, and aggressive use of img2img and controlnets. And a few companies (Scenario [4] etc) are working on consistency. I think you won't see a ton of big projects using it yet because the tech is still early days and the early versions were really a bear to work with.
A lot of people on YouTube are using it like theyd use stock art (YouTube thumbnails, backgrounds, ads, or story boards for writing focused/story YouTubes).
There is still a stigma so a lot of people using it aren't announcing their use broadly, so I've found I usually have to stumble on their projects. Also, a lot of major companies aren't using it for that reason. I know Wizards of the Coast has had some arguments about that recently. Also consistency is still a problem as you mention which limits it's use in bigger projects.
Video games I think will be the first place we'll see it widely accepted. As people have been using AI generation techniques for background tools for like 20 years, see speedtree.
This will sound like a stupid question, but is there a downloadable list of all the art styles understood by these models as an offline PDF?
If there isn't, I'm very, very tempted to make one myself. Each style entry consists of a description and a list of models that understand it and any special tags or keywords the model needs to generate output with that style.
I know we have a lot of websites dedicated to this, but this information is more suited for a wiki
It appears like. Since ai doesnt create art, it relies on someone else’s work to generate novel looking content. But by looking carefully you can see the hundreds of intermixed originals and most important the unique styles they’ve stolen. Pretty sad how ai has been overtaken by sociopaths.
Cloning a style doesn't replace an artist.
the AI doesn't get things like proper lighting, emphasis, motion, perspective and, above all else, intention.
They're also not copyrightable. if you use an AI piece in your product, anyone can take it and make a competing product with the same piece or prompt chain.
They're good to replace crude doodles or making some references to have a real artist work on your idea, though. that way humans keep their jobs and machine improves the process.
How can I use this? For example: if I have a picture of my wife, how can I apply one of these styles to her picture? I've downloaded InvokeAI and it runs, but I can't get it to just apply a style to an existing picture.
I wanted to give her a birthday present and her picture using the style of her favorite artist in a frame would be amazing.
I asked DALL•E 3 if he takes commissions and he said no. I tried screaming the question at photoshop too but it didn’t answer. I’m still awaiting MSPaint’s reply.
I call BS -- every attribute of the "non-profit, volunteers" is calculated to deflect hostile rejection.. "community" has AI chat agents ready to help you?
how many actual art school people do you know who tried to make their showcase website.. Is it like this one? Notice the Andrei says "studio photography" .. a tech skill, not an illustrator.
as a representative of actual Art School people -- this is awful socially.. really
If you like anime artists, there is this: https://rentry.co/NAIDv3artisttagtest, a list of 470 artists using the last model of NovelAI (this time they trained their model without removing artist names).
Fun fact, this model is based on SDXL, finetuned probably on a cluster of H100s (from coreweave).
The consensus seems to be that the lack of open source SDXL anime fine tunes is simply due to laziness and lack of resources from the interested parties in the community. Datasets like from the boorus are already floating around online to train on. Always funny to see that furries are ahead of weeaboos in the war for the best open source nsfw generator
Well to be honest I don't really know what's the best furry SD XL model out there but recently this model has been released: https://huggingface.co/cagliostrolab/animagine-xl-3.0, it's probably far from the NovelAI checkpoint but I would be curious to know how it compares to the furry model you're talking about.
It’s an interesting example of a new technology that is inferior to the old one because the old one is superior in important ways (speed, ease of manipulation, and established toolsets).
Last I heard, everyone was struggling to fine tune SDXL on furry art too. The only community fine tunes that seem to have succeeded are ones aimed at taking things it was already designed to generate and trained on and making it actually good at those, or improving the aesthetics of its output.
That's mildly surprising, given how large and well tagged e621 in particular is. Though I assume most of the major furry art hosts are blocking scraping by this point, given how divisive GenAI is?
Is there a big difference in the result from fine tuning a model compared to using a LoRA? I thought the idea with LoRAs was that updating all model weights are unnecessary?
Lots of people (including potential trainers) struggle with SDXL I think. It just doesn't give you good results if you shove it into SD 1.5 frameworks and prompt it like SD 1.5.
The devs behind NovelAI have been talking about their training on the discord a little bit. According to their messages, it took about the same amount of compute to finetune the model, as it took for StabilityAI to train SDXL itself.
No wonder other people struggle to train it, if it takes >=128 A100s to get the model to perform better than SD1.5 on only slightly OOD tasks (anime pictures, in this case.)
If you're talking about anime, every popular SD v1.5 model out there has a backbone based on the leaked NovelAI model, so what SD XL lacks the most is people with compute power finetune it, that being said, this model: https://huggingface.co/cagliostrolab/animagine-xl-3.0 was recently released and it was finetuned for +500 A100-hours.
