Love the video and the concept, much more in tone with the album than some guitars and planets, but only thing I find weird is the intro being too long and perhaps looking a bit emptier, an explosion of colours would be so nice.
This may be silly conspiracy, but it really does make me wonder if the judges or were paid or sponsored by an ai generator company to select an ai winner just to try and convince people that they can get money if they use ai. Deviantart does the same thing. They promote their ai users earnings in efforts to promote their ai generator.
It's this. There's a big conspiracy about big ai companies trying to take as much credit away from artists. On the other side, it could be that the judges have no idea about art, and se AI as a new cool label to use.
@@jeanpierrepolnareff9919 this is most likely the case. i mean, literally every tech company right now is trying to tie AI into their products, even if it is just repackaged ChatGPT lol
The competition organizers are the real wrong doers here, it was incredibly unethical even without the added layer of selecting an ai generated video as a winner. I mean I wouldn't want to be that guy who made the ai video but he's just the tip of the iceberg
@@GrumpDog I will never do that again, I swear. I only did it for The Dark Side of the Moon. Now I feel exploited as an artist and as a fan alike. I was lucky enough that my video got attention by some awesome people, giving support and cheering me up. Other artists were not so lucky.
looking at the honourable mentions playlist they made, nearly half of the one's on the playlist are Ai, just like the Any colour you like video. In fact all the any colour you like videos in the honourable mentions playlist were Ai which shows they had absolutely no care on picking the winners
@@GrumpDog then good. The issue here isn't the number of applications he made. Rather, it's a clearly displayed lack of effort. As mentioned in the video, the AI animation lacks any reference to the beat of the song - failing to accomplish what is baseline required for pretty much every music video. You could make a better animation if you prompted images from ai and then cut transitions on the beats or on the verses even. It's just an ai ken burns slideshow with similar themes to the song.
he deserves that money the terms and conditions do not state any regulations nor bans on AI either partially or as a whole to get mad over this is simply pathetic
It's absolutely insane that this won and one of the terrible things about it is that it's completely over shadowed the winning submissions as everyone is just talking about the AI one
As a IT bachelor, majored in new generation of AI models, i can without a doubt, ASSURE you, that this man did not actually train his own model. This would be the general pipeline if he did: 1. Find at least few milion images that correspond to what you generally need. If the model is made for one specific pourpose and one specific prompt in mind, it doesn't require as many as stable diffusion due to the fact that we only want one style of images, in one theme. Still, even with the best internet connection, it would take a really long time to even find those images, let alone keep them for future use. If we want different styles as the man made, we would require more and more milions of images. It's not possible for him to create a model on his work alone, and buying enough copyrights would bankrupt even the top1%. 2. Pre-process the images. Images come in different formats, sizes and trying to build an AI that can take in every possible size and format exponentially increases workload for those poor programmers that will create the model. Also, if the model is supposed to generate 10 different results by using a prompt (Or if we want to group the images for 10 different models) those images will require tagging, which is basically attaching key words to every single one of those images. Doing it by hand is not possible due to the amount of them, and hiring another model for detecting things on the images may prove costly for a single person. I can assure you, it would cost more than 10k $. One image is cheap, a lot less than 1 cent. anything more than a milion may prove costly. 3. Program the model. I don't think i have to explain that for a model to create an animation, dude would need to either be a extradimentional being, or hire an entire team of experienced engineers in those types of models. Spoilter alert. AI engineers don't come cheap. Especially experienced ones. 4. Train the model. This is very much automatic process, but for it to get any results within this lifespan, let alone before the deadline of the competition, dude would need a supercomputer. A beast of a personal computer requires little less than a 1 day to teach simple and very efficient AI to draw black and white eyes in 256x256 resolution. Colorful, 2k images with few different objects? More than few weeks. Turn those images into actual animations, and expanding the possibilities of the model further, with different styles and even more objects? I think you can extrapolate that much. This only accounts for a situation where everything went smoothly, no issues no bugs, god and anime is on his side. HOWEVER. With all of that said. Fine tuning stable diffusion is a lot easier than you think. By using a very smart system called LoRa, adding new keywords to the model and being very skilled at that, can result in amazing things in less than a week worth of work. I myself did that for some extra credit. Not hard at all, there are a lot of tutorials on that available online. tl:dr -> Creating a new model is impossible. Fine tuning is easier than most folks would like to admit. Thank you and good night! :>
Training your own model is as simple as using a program to just trial and error acceptable and non-acceptable actions... it's almost exactly like hackers who buy something for $50 just to use it to grief everyone else in seconds just to get kicked off the network and banned. They don't put any work into anything all they do is pick the right choice for each path and that's it. Even then, they aren't the ones imagining or making any of the art so bullshit it doesn't matter they didn't draw they didn't do anything to make the art itself.
@@TheAnxiousOwl while i agree with you on the second part, i can't agree with you on the first part. Its not simple. For useful models, its a very complex process, that takes months of work, done by people in many different fields.
If you want to protect your work even more effectively the team that made Glaze now also has Nightshade, a tool that can actually damage AI models' understanding of what certain descriptive terms mean, thereby incentivizing their owners to actually request consent before scraping.
@@stabbycorvidif someone wants to train a Lora for your artstyle, they would handpick the images and do a lot of testing afterwards. Even the big datasets are curated by the real people. Old base models also won't just disappear.
The one thing in the video(s) that was *definitely* done by a human is the Windows Movie Maker-ass title captions. Awful colour and font choices, and the background is straight out of the 2000s.
Honestly if a contest does not explicitly, and visibly, rule out any kind of AI submission, I do not get involved. They aren't worth the effort that would go into it. The worst part about the winning submission is that it isn't even any good, AI or not. It's the graphical equivalent of lexical noise.
Literally just looks like generic procedural video. When it was new in the mid 2010s it was it's own aesthetic but now it just looks like your editing software had a stroke
@@stitchfinger7678we liked the old ai because it generated stuff that was bizarre and very difficult to recreate by hand, but then they leaned more into art and also believed that most people liked ai because it's gonna generate lifelike images and not because, y'know, we found it to be strange and funny seeing a computer trying to process and understand things and just making impossible photos
Great stuff!!!! Pink floyd should be ashamed. The judges in particular. Terry Gilliam ( who was one judge ) likely voted for this slop.. That is also extremely disappointing.
Thank you for making this video explaining the AI assault on animation art. I made a fake AI entry in 5 hours to disprove the claims of some of the biggest con 'artists' from the competition. Cheers. You are a hero!
Thanks for so many warm and positive comments! 😯 I'm reading them all. I got some feedback that it was too hard to find Álvaro Lomba's submission, so I've added it as a "i" card in the upper right corner. You can also find it in the description.
I tried to submit a video right before the deadline, but the submission was broken... It was such a time sink but honestly after seeing the results and reviewing their policies (which imply they OWN your animation for free) I'm glad I didn't.
messing around with stable diffusion and training loras is pretty fun and imo pretty cool but it should never ever be passed for human art... or even be considered something you can own and sell. It's just a generation. Everyone and no one made it, so everyone and no one gets to own it.
I was genuinely convinced he actually did sh*t with Blender. I thought he created the shapes in Blender and used AI to create colours and shape on top of it and use the randomness as inspiration. Very sad that it turns out he didn't
Really good video! Very informative and well spoken, and really well edited. A geniunely good video in a genre that's super oversatured with slop. Also I love your avatar it reminds me of Nyx from Hades
I think it would be an effective way to send a message, that AI art isn't art, nor is it welcome, if large scale contest like these would flat-out and immediately stamp these with *[Disqualified for AI Generated]* but the sad truth is, usually the contest holders aren't visual artists themselves, so they simply don't _see_ the problem. On another note, I'm sick of how AI-bro's are so full of themselves. _"Oh yeah, _*_I was very good_*_ at clicking the button on the machine that did the entire thing for me. Good job, me!"_ 🙄
Amazing video, very well explained. Shame on the members of the jury, it's a punch in the face of all the artists who spent numerous hours/months creating real art. I hope that the winner will get sued into oblivion, some day when it comes out how AI developers train their machines with copyright material.
10 out of 10 times someone says they trained their own model, it's just a fine tuned LORA over Stable Diffusion. Training a model from scratch requires computational power compared to entire render farms or bit mining floors. It's not something you can do from your laptop while playing Minecraft in another window.
So... the part with the truck sticking out of a wall ... that's half life 2 episode 1. there's so much theft going on of both ideas AND art here. This guy should be banned from ever doing anything remotely labeled for sale or 'art'. lol and that other one in the list of all the entries: that's Gman from Half Life: Alyx. how are people thinking that this is okay?
This seems like a contest MADE to set precedent for AI winning in a "valid " way while getting FREE work without realizing maybe how PISSED people would be...that or dont care.
This whole situation was an absolute travesty if you ask me. The fact that PINK FLOYD of all people are trying to mooch free labour out of artists, that so many people submitted AI prompts, that AI submissions weren't outright banned, that the judges are somehow stupid enough to choose an AI submission as a winner... I have no respect for anyone involved, and I think it's so fucking embarrassing that musicians cry about AI audio and then turn around and support it in visual mediums. Brilliant display of absolutely no solidarity between groups that should have each others backs.
Gonna see if I can embed them in the video frame. Here's the "Any Colour you like" entry, th-cam.com/video/oWZSg5dr7r8/w-d-xo.htmlsi=jl3jePkWRAoPWshV and here's the "Eclipse" entry: th-cam.com/video/1F8MogHSw-k/w-d-xo.html
I'm sorry for not making it easier. 🙇🏻♂️ Since I wasn't verified, youtube didn't let me add links to the description. Gonna see if I can embed them in the video frame. For anyone else, here is links to the videos I talked about. th-cam.com/video/oWZSg5dr7r8/w-d-xo.htmlsi=jl3jePkWRAoPWshV th-cam.com/video/1F8MogHSw-k/w-d-xo.html
awesome video! super disappointed in Pink Floyd for letting ai slop win considering their rich history with great art and animators in the past, I'm happy more people are talking about it and calling them out for this
It's honestly disgusting to see this AI crap get awarded. On the other hand the Brain Damage music video is amazing and beautiful- actually amazing art
Gonna see if I can embed the videos in the video frame. Sorry for not making it easier, had trouble with TH-cam not allowing me to add links to the description. Here's Álvaro Lomba's "Any Colour you like" entry, th-cam.com/video/oWZSg5dr7r8/w-d-xo.htmlsi=jl3jePkWRAoPWshV and here's the "Eclipse" entry: th-cam.com/video/1F8MogHSw-k/w-d-xo.html
On my channel I have a Pink Floyd music video made from clips of an anime. I make it crystal clear where I am sourcing the work so it falls under fair use. Projects that use AI should also have some means to credit the work of the artists that went into training the model. Trouble is, that is damn near every artist on earth, so AI corpos should have their work cut out for them.
An AI generated video _may_ be copyrightable in Great Britain, but it also _may_ violate copyright laws over there to use it to make money. One of the clips you showed looked more like a fractal video than modern AI art; still low-effort, but not unethical, since fractals don't usually require someone to take other's art.
This is true! And it looks like a fractal video, but if you look closely it bears the unmistakeable frame-to-frame muddyness of a diffusion model. They COULD have used any animation software to make the fractals, but for some reason they did it an AI image [re]generator model. My suspicion is that they might have downloaded a fractal image/video somewhere, then ran that through Image-to-image in SD, because I don't know if SD is good at generating accurate-ish fractals on it's own.
@@stealcasePerhaps. Google's AI isn't great at generating fractal images. (Just tried.) Don't know about an other AI; I don't want to give more of my info to _another_ online service. There were some programs designed to create in-betweens from key frames. Do those look different than the current crop of AI generators? I really need to find some fractal program; they look cool! And they can make mountain ranges & planets. Terragen looks good, but it's not exactly free, & I'm cheap. :)
@stillbuyvhs the in-between program you're talking about is probably EbSynth, which is awesome and has its own graphical "glitch" look, but it's subtly different from the Stable Diffusion artefacts. As for fractals: there's probably dedicated software, but for these types of problems if your requirement is "free", then Blender is my go-to. If you're familiar with blender already, then this tutorial basically just tells you to copy 12 lines of code and you have a free fractal generator. If you DON'T already know blender, then this is too complicated for you. th-cam.com/video/H8nHxRO7eX0/w-d-xo.htmlsi=F3NRt1sL_xf5nF4z Blender is great for any kind of procedural graphics. But if you want to make fractals, you can likely do it in any software that accepts code input to generate graphics, cause fractals is just math
@@stealcase I really should try Blender; I've looked at it before, but it always seemed daunting. Still, thanks for the link! Maybe, with time & patience, I can get something to work. :)
I actually believed for a moment he did train the AI on his own work. It's still lazy and shouldn't be called animation, but shame on me. Also YAY Nightshade!
I have had a life long career in photography, film, and animation. I embraced and learned AI Animation in early 2023, primarily for the Dark Side of the Moon Competition. Why? Because it looks amazing! Is it art? It is to me and anyone looking at it honestly, and that's all that counts. Many nay sayers (most of whom have no experience with it) claim it is theft. It is no more theft than photography and film making which, by the way, frightened illustrative artists when it was invented in the 19th Century (it did not kill painting for example, it liberated it). That AI requires no skill? It does and like all skill requires knowledge and discipline. That AI is going to destroy art? It won't. If you want to "future proof" your art and your career as an artist, the same rules still apply: Make sure you focus on high-end excellent work. Make sure your creation involves a lot of “human” tasks, like picking story angles, persuading, researching, architecting, strategizing, selling, which AI can’t do well and it doesn’t seem like it will anytime soon.
The video is great, thanks, but it is of my understanding that both Glaze and Nightshade are useless snake oil. Aren’t they? Edit: Sometimes it feels like the folks at U Chicago are playing the same game as OpenAI and Stability, stealing code for early versions of GLAZE and ignoring detractors who genuinely want to know their shit works.
Hey Juantoledo. They aren't snake oil, I personally know Ben and many in the team working on this. Glaze basically analyzes your image USING stable diffusion and CLIP, and modifies the pixels in subtle ways so that Stable Diffusion THINKS the image is actually a picture of something else. Because AI doesn't look at images the same ways as humans, it identifies the image completely differently than you and me. They NEVER send any data anywhere, the software is completely offline. Glaze you can see the effects of: if a consumer or individual tries to finetune on your work, they'll have a lot of trouble. But it also has more visual "noise" thats more visible to the human eye. Nightshade is designed to harm bigger models and companies like OpenAI. As a result, it relies on people adopting and using it, but we don't even need a majority of artists to use it to be a deterrent. But the more who use it, the more damaging it will be. If you're worried about the motivations of the creators, this interview with Ben Zhao (one of the main professors working on this) is excellent. themuse.substack.com/p/the-poison-well As far as the "stealing code" bit: One of the programmers on Glaze used an OpenSource library, but didn't consider the license properly that they would have to share the entire source-code of the Glaze software. They THOUGHT they were following the license, but didn't realize they weren't. Once they were alerted of the issue, they worked almost around the clock to fix it to respect the license. They took the Glaze download offline until they fixed it. The UI of GLAZE is literally more ugly to-this-day because of it. Personally I think this shows their character: everyone makes mistakes, but it's how you handle those mistakes that matter, and this is a textbook version of how to handle this correctly IMO. You can read more about it here. x.com/ravenben/status/1636439335569375238 As far as the detractors: A ton of them are AI bros or prompters, but not all. Remember, there's a lot of fearful people who are highly incentivized to make people believe this won't actually work. If they genuinely wanted to know, understand and reproduce, there are literally 2 research papers explaining everything about the process. Here's the Glaze Paper people.cs.uchicago.edu/~ravenben/publications/abstracts/glaze-usenix23.html And the nightshade paper: people.cs.uchicago.edu/~ravenben/publications/pdf/nightshade-oakland24.pdf There's a lot of reproducible evidence that this works, and I've seen AI researchers reproduce broken models with nightshade, confirming it works. I've seen some people are incorrectly finetuning their own models with Glazed images and claim that it doesn't work: Often, their finetune doesn't actually work as they wanted, yet they claim success. I've seen others do finetuning, and completely destroy their Stable Diffusions ability to generate the target images. So to me, I think all of this points to credibility.
Another sad attempt at a human being trying to take credit for an Artificial Intelligence's art. Blasphemy. Ngl any contest including artists and animators should exclude AI altogether...
I more or less agree with the statements in this video. However, slot machines and AI sampling are apples and oranges. A reasonable assumption using a slot machine is that we are sampling over a uniform distribution, whereas with a diffusion generator we are manipulating the distribution through a variety of means. So it's like the difference between someone gambling and someone rigging the dice. If you reduce CFG to zero, but keep denoising strength at 1, you get a slot machine. Btw, if our AI-guy had spent a large amount of time on the piece, would that make it respectable?
1 I oppose IP legislation. 2 yes he put low effort, I acknowledge that, you are 100% right, others put time and sweat into theirs, this guy pooped his out (literally). It is still good. Sad reality is that AI is scary good these days and will soon take over. Instead or targetting this guy, why don't you call for AI to be open source or regulate it?
There's a reason for that. They're using the exact same AI software. To the Iron Lung's credit, at the very least the visuals are somewhat synced to the song.
Available for free does NOT mean Copyright free! This is a misconception. It's only free for you to see, not to use/sell. If you took someone's art that they published online for free and started selling prints of it without asking them, it would obviously be theft, and grounds for getting sued. Current AI is doing the same thing with an extra step of data laundering in between (the vast majority of pictures they used in their training datasets weren't licensed, they should never even have been there in the first place).
That is correct: "if I *[sold]* prints of them". The people mentioned here supposedly "scrape the internet and steal art from artists". I'm pretty sure "scraping" doesn't involve "selling". By the way, this video and it's comments weren't specifically licensed for quotes. I still did repeat some of its words. That's not considered theft or illegal either. When someone publishes information it automatically stops being private and it's perfectly legal to observe it.
@@quantumangel "Observing" and processing data are two different things. AI models are not human and lack the eyeballs required to observe anything. This is "my camera recorder observes things the same way I do so its fine to take it to the movies" levels of playing dumb. Who do you believe is falling for this? You're losing all credibility when you imply that this scraping doesn't go hand in hand with the goal of processing it, to out-compete and push the people it scraped from out of their own market. Think very carefully about whose interests you're serving when you propagate this kind of debunked propaganda. You're being used.
@@IRLlosersQ Yes. Observing and processing are 2 different actions. They are, however equally legal in this context, which is why I'm comparing them. Also I didn't imply scraping goes alone. We're clearly talking about ai training. Also, movies aren't published for free on the internet. Movies studios are very strict as to who can watch and distribute their content and their commissions. That's why it's illegal to record movies in the cinema (as it is to watch them without paying). As for the interests, were talking about what covitai users do, so I suppose it's them? Not sure what you mean by propaganda though. Are you talking about my argument? Or is there a pro-ai propaganda campaign in unaware of? Cause so far I've mostly seen complaints against ai...
@@IRLlosersQ Yes. Observing and processing are 2 different actions. They are, however equally legal in this context, which is why I'm comparing them. Also I didn't imply scraping goes alone. We're clearly talking about ai training. Also, movies aren't published for free on the internet. Movies studios are very strict as to who can watch and distribute their content and their commissions. That's why it's illegal to record movies in the cinema (as it is to watch them without paying). As for the interests, were talking about what covitai users do, so I suppose it's them? Not sure what you mean by propaganda though. Are you talking about my argument? Or is there a pro-ai propaganda campaign in unaware of? Cause so far I've mostly seen complaints against ai...
I’m what you would call slot gambler. My contest video was created 100% with AI. I have no artistic skill what so ever, but there is a reason why I started to work with these AI videos. I do music licensing somewhat normally. All little commercial music jobs are disappearing(AI or my music is not good enough anymore?) and I alway wanted to make videos for my music this was my chance to do something remotely close to that. I’m really slow and detailed as old people sometimes are and I been out of work last year. I took a loan so I can pay all my software licences for 6 months and I put 800 hours on this video. I figured track is called “Time” so what would be more valuable to give to this project that my time. Looking back at the project, I’m very happy with it and I still think it looks amazing. And I learn new skill last year? Or can you call it a skill? Or did I just waisted my year on nonsense? th-cam.com/video/PTMnrxigMfo/w-d-xo.htmlsi=nzRjdn-R89ZWaU2-
I'm the "Alvaro Lomba" behind the volumetric caustics simulations entry. Thank you very much for the recognition!
How did you simulate it? Using an existing 3D renderer or building your own in some language?
@@gandjalfthegreen2143It was rendered on a well known physically-correct engine. The shaders and environment were custom-made.
Love the video and the concept, much more in tone with the album than some guitars and planets, but only thing I find weird is the intro being too long and perhaps looking a bit emptier, an explosion of colours would be so nice.
If the 'judges' didn't know it was AI-generated... they have no business judging art competitions period!
unfortunately, one of the judges was Gerald Scarfe, also known as the man behind the animations and drawings of The Wall. They knew
This may be silly conspiracy, but it really does make me wonder if the judges or were paid or sponsored by an ai generator company to select an ai winner just to try and convince people that they can get money if they use ai.
Deviantart does the same thing. They promote their ai users earnings in efforts to promote their ai generator.
You know, I'm wondering the same thing. Look what happened with Global Game Jam over the past year, it screams the same way.
It's this. There's a big conspiracy about big ai companies trying to take as much credit away from artists.
On the other side, it could be that the judges have no idea about art, and se AI as a new cool label to use.
@@jeanpierrepolnareff9919 this is most likely the case.
i mean, literally every tech company right now is trying to tie AI into their products, even if it is just repackaged ChatGPT lol
Na its more likey the judges are not tech savvy to know that AI art is not that creative or labour intensive.
The competition organizers are the real wrong doers here, it was incredibly unethical even without the added layer of selecting an ai generated video as a winner. I mean I wouldn't want to be that guy who made the ai video but he's just the tip of the iceberg
It's up to us to call them out on it.
You're free to not enter competitions that allow AI.
@@GrumpDog I will never do that again, I swear. I only did it for The Dark Side of the Moon. Now I feel exploited as an artist and as a fan alike. I was lucky enough that my video got attention by some awesome people, giving support and cheering me up. Other artists were not so lucky.
looking at the honourable mentions playlist they made, nearly half of the one's on the playlist are Ai, just like the Any colour you like video. In fact all the any colour you like videos in the honourable mentions playlist were Ai which shows they had absolutely no care on picking the winners
the fact that he uploaded a video for every song is so damning
What if someone who was just incredible fast, did the same thing, without AI?
@@GrumpDog then good. The issue here isn't the number of applications he made. Rather, it's a clearly displayed lack of effort.
As mentioned in the video, the AI animation lacks any reference to the beat of the song - failing to accomplish what is baseline required for pretty much every music video. You could make a better animation if you prompted images from ai and then cut transitions on the beats or on the verses even.
It's just an ai ken burns slideshow with similar themes to the song.
he deserves that money
the terms and conditions do not state any regulations nor bans on AI either partially or as a whole
to get mad over this is simply pathetic
@@lucid4entrelisesall the guy did was write some words into a video generator and that's it
@@trashfur1656 so? It's still within the rules
It's absolutely insane that this won and one of the terrible things about it is that it's completely over shadowed the winning submissions as everyone is just talking about the AI one
As a IT bachelor, majored in new generation of AI models, i can without a doubt, ASSURE you, that this man did not actually train his own model. This would be the general pipeline if he did:
1. Find at least few milion images that correspond to what you generally need. If the model is made for one specific pourpose and one specific prompt in mind, it doesn't require as many as stable diffusion due to the fact that we only want one style of images, in one theme. Still, even with the best internet connection, it would take a really long time to even find those images, let alone keep them for future use. If we want different styles as the man made, we would require more and more milions of images. It's not possible for him to create a model on his work alone, and buying enough copyrights would bankrupt even the top1%.
2. Pre-process the images. Images come in different formats, sizes and trying to build an AI that can take in every possible size and format exponentially increases workload for those poor programmers that will create the model. Also, if the model is supposed to generate 10 different results by using a prompt (Or if we want to group the images for 10 different models) those images will require tagging, which is basically attaching key words to every single one of those images. Doing it by hand is not possible due to the amount of them, and hiring another model for detecting things on the images may prove costly for a single person. I can assure you, it would cost more than 10k $. One image is cheap, a lot less than 1 cent. anything more than a milion may prove costly.
3. Program the model. I don't think i have to explain that for a model to create an animation, dude would need to either be a extradimentional being, or hire an entire team of experienced engineers in those types of models. Spoilter alert. AI engineers don't come cheap. Especially experienced ones.
4. Train the model. This is very much automatic process, but for it to get any results within this lifespan, let alone before the deadline of the competition, dude would need a supercomputer. A beast of a personal computer requires little less than a 1 day to teach simple and very efficient AI to draw black and white eyes in 256x256 resolution. Colorful, 2k images with few different objects? More than few weeks. Turn those images into actual animations, and expanding the possibilities of the model further, with different styles and even more objects? I think you can extrapolate that much.
This only accounts for a situation where everything went smoothly, no issues no bugs, god and anime is on his side.
HOWEVER. With all of that said. Fine tuning stable diffusion is a lot easier than you think. By using a very smart system called LoRa, adding new keywords to the model and being very skilled at that, can result in amazing things in less than a week worth of work. I myself did that for some extra credit. Not hard at all, there are a lot of tutorials on that available online.
tl:dr -> Creating a new model is impossible. Fine tuning is easier than most folks would like to admit. Thank you and good night! :>
Training your own model is as simple as using a program to just trial and error acceptable and non-acceptable actions... it's almost exactly like hackers who buy something for $50 just to use it to grief everyone else in seconds just to get kicked off the network and banned. They don't put any work into anything all they do is pick the right choice for each path and that's it. Even then, they aren't the ones imagining or making any of the art so bullshit it doesn't matter they didn't draw they didn't do anything to make the art itself.
@@TheAnxiousOwl while i agree with you on the second part, i can't agree with you on the first part. Its not simple. For useful models, its a very complex process, that takes months of work, done by people in many different fields.
So? Even if you just alter an existing model so that it only outputs stuff like the images you've given it. I'd say that's good enough.
If you want to protect your work even more effectively the team that made Glaze now also has Nightshade, a tool that can actually damage AI models' understanding of what certain descriptive terms mean, thereby incentivizing their owners to actually request consent before scraping.
Nightshade is briefly mentioned at 6:22
Not only that but it poisons the samples so it looks cooler and more original. I like the poisoned art more than the familiar copyrighted art.
While theoretically interesting, those tools are practically irrelevant.
@@scottpreston3892 why would they be irrelevant when they protect artists' works from AI scraping?
@@stabbycorvidif someone wants to train a Lora for your artstyle, they would handpick the images and do a lot of testing afterwards.
Even the big datasets are curated by the real people. Old base models also won't just disappear.
AI bros trying not to be scumbags challenge
The one thing in the video(s) that was *definitely* done by a human is the Windows Movie Maker-ass title captions. Awful colour and font choices, and the background is straight out of the 2000s.
Honestly if a contest does not explicitly, and visibly, rule out any kind of AI submission, I do not get involved. They aren't worth the effort that would go into it.
The worst part about the winning submission is that it isn't even any good, AI or not. It's the graphical equivalent of lexical noise.
Literally just looks like generic procedural video.
When it was new in the mid 2010s it was it's own aesthetic but now it just looks like your editing software had a stroke
@@stitchfinger7678we liked the old ai because it generated stuff that was bizarre and very difficult to recreate by hand, but then they leaned more into art and also believed that most people liked ai because it's gonna generate lifelike images and not because, y'know, we found it to be strange and funny seeing a computer trying to process and understand things and just making impossible photos
Clear. Comprehensive. Informative. Amazing video. … What’s happening right now to creative industries is disgusting.
Great stuff!!!! Pink floyd should be ashamed. The judges in particular. Terry Gilliam ( who was one judge ) likely voted for this slop.. That is also extremely disappointing.
Thank you for making this video explaining the AI assault on animation art. I made a fake AI entry in 5 hours to disprove the claims of some of the biggest con 'artists' from the competition. Cheers. You are a hero!
This youtube video is much better and artful than his submissions. Congrats!
My shit would be better than his submissions
Thanks for so many warm and positive comments! 😯
I'm reading them all.
I got some feedback that it was too hard to find Álvaro Lomba's submission, so I've added it as a "i" card in the upper right corner. You can also find it in the description.
Disappointed to hear this news. Thank you for posting.
I tried to submit a video right before the deadline, but the submission was broken... It was such a time sink but honestly after seeing the results and reviewing their policies (which imply they OWN your animation for free) I'm glad I didn't.
i'm so happy that someone is actually talking about this, i havent seen anyone talking about it so far and you did it perfectly. earned my subscribe.
messing around with stable diffusion and training loras is pretty fun and imo pretty cool but it should never ever be passed for human art... or even be considered something you can own and sell. It's just a generation. Everyone and no one made it, so everyone and no one gets to own it.
I was genuinely convinced he actually did sh*t with Blender. I thought he created the shapes in Blender and used AI to create colours and shape on top of it and use the randomness as inspiration. Very sad that it turns out he didn't
Really good video! Very informative and well spoken, and really well edited. A geniunely good video in a genre that's super oversatured with slop. Also I love your avatar it reminds me of Nyx from Hades
really, thank you.
ALso re: Nyx heh, now I see the resemblance. x)
I think it would be an effective way to send a message, that AI art isn't art, nor is it welcome, if large scale contest like these would flat-out and immediately stamp these with *[Disqualified for AI Generated]* but the sad truth is, usually the contest holders aren't visual artists themselves, so they simply don't _see_ the problem. On another note, I'm sick of how AI-bro's are so full of themselves. _"Oh yeah, _*_I was very good_*_ at clicking the button on the machine that did the entire thing for me. Good job, me!"_ 🙄
AI SHOULD HAVE BEEN
ELIMINATED FROM SUBMISSION!
It’s like winning the Nobel. Prize for literature using a ghost writer!
Amazing video, very well explained. Shame on the members of the jury, it's a punch in the face of all the artists who spent numerous hours/months creating real art. I hope that the winner will get sued into oblivion, some day when it comes out how AI developers train their machines with copyright material.
10 out of 10 times someone says they trained their own model, it's just a fine tuned LORA over Stable Diffusion.
Training a model from scratch requires computational power compared to entire render farms or bit mining floors. It's not something you can do from your laptop while playing Minecraft in another window.
oh my god. This story is so bad and the man who won is an obvious scammer.. I really hate this timeline we live in
What i hate the most is, that song title at the first second of the video, its f ing ugly
0:28 For this section specifically, I don't really understand the argument because all of the artists could've just not worked on the contest
I like your video. The subject matter, your style, and most of what you were saying. I'm surprised you don't have more subscribers.
Wow, thank you for saying that!
Really appreciate it. ^_^
So... the part with the truck sticking out of a wall ... that's half life 2 episode 1. there's so much theft going on of both ideas AND art here. This guy should be banned from ever doing anything remotely labeled for sale or 'art'. lol and that other one in the list of all the entries: that's Gman from Half Life: Alyx. how are people thinking that this is okay?
This seems like a contest MADE to set precedent for AI winning in a "valid " way while getting FREE work without realizing maybe how PISSED people would be...that or dont care.
AI *MUST* be distorted by any means necessary
You went viral on twitter, congrats
I nearly spent 3 months making an entry, glad I didn't bother now
This whole situation was an absolute travesty if you ask me. The fact that PINK FLOYD of all people are trying to mooch free labour out of artists, that so many people submitted AI prompts, that AI submissions weren't outright banned, that the judges are somehow stupid enough to choose an AI submission as a winner... I have no respect for anyone involved, and I think it's so fucking embarrassing that musicians cry about AI audio and then turn around and support it in visual mediums. Brilliant display of absolutely no solidarity between groups that should have each others backs.
Side note but I really love the way your avatar looks, very cool! Subscribed
Thank you! Was feeling nervous about using it, but this is very validating.
Im actually not against using ai to speed up work flow. But, imo it shouldnt be used in any competition.
Would loe to see the competition entry you preferred over the ai slop. are you able to link. Searched but couldn't find ☹
Gonna see if I can embed them in the video frame.
Here's the "Any Colour you like" entry,
th-cam.com/video/oWZSg5dr7r8/w-d-xo.htmlsi=jl3jePkWRAoPWshV
and here's the "Eclipse" entry:
th-cam.com/video/1F8MogHSw-k/w-d-xo.html
I had it as "Unlisted". It should be public now.
finding the video you praised was kind of annoying, but I agree that was a cool piece.
I'm sorry for not making it easier. 🙇🏻♂️
Since I wasn't verified, youtube didn't let me add links to the description. Gonna see if I can embed them in the video frame. For anyone else, here is links to the videos I talked about.
th-cam.com/video/oWZSg5dr7r8/w-d-xo.htmlsi=jl3jePkWRAoPWshV
th-cam.com/video/1F8MogHSw-k/w-d-xo.html
awesome video! super disappointed in Pink Floyd for letting ai slop win considering their rich history with great art and animators in the past, I'm happy more people are talking about it and calling them out for this
great video, thanks for the breakdown
It's honestly disgusting to see this AI crap get awarded. On the other hand the Brain Damage music video is amazing and beautiful- actually amazing art
Im faded and the lip syncing is killing me
can you link Alvaro Lomba's submission?
Same! I've been looking for it, but can't find it since it's been unlisted
Gonna see if I can embed the videos in the video frame. Sorry for not making it easier, had trouble with TH-cam not allowing me to add links to the description.
Here's Álvaro Lomba's "Any Colour you like" entry,
th-cam.com/video/oWZSg5dr7r8/w-d-xo.htmlsi=jl3jePkWRAoPWshV
and here's the "Eclipse" entry:
th-cam.com/video/1F8MogHSw-k/w-d-xo.html
On my channel I have a Pink Floyd music video made from clips of an anime. I make it crystal clear where I am sourcing the work so it falls under fair use. Projects that use AI should also have some means to credit the work of the artists that went into training the model. Trouble is, that is damn near every artist on earth, so AI corpos should have their work cut out for them.
We don't even know how dangerous such applications are, we stumble across con-men who show us.
Hey, could link that playlist with all the animation entries. I'm interested in viewing the runner ups.
Great to see LGBT+ TH-camrs making great content on here. Nothing can stop you Stealcase!🏳🌈
An AI generated video _may_ be copyrightable in Great Britain, but it also _may_ violate copyright laws over there to use it to make money.
One of the clips you showed looked more like a fractal video than modern AI art; still low-effort, but not unethical, since fractals don't usually require someone to take other's art.
This is true! And it looks like a fractal video, but if you look closely it bears the unmistakeable frame-to-frame muddyness of a diffusion model.
They COULD have used any animation software to make the fractals, but for some reason they did it an AI image [re]generator model.
My suspicion is that they might have downloaded a fractal image/video somewhere, then ran that through Image-to-image in SD, because I don't know if SD is good at generating accurate-ish fractals on it's own.
@@stealcasePerhaps. Google's AI isn't great at generating fractal images. (Just tried.) Don't know about an other AI; I don't want to give more of my info to _another_ online service.
There were some programs designed to create in-betweens from key frames. Do those look different than the current crop of AI generators?
I really need to find some fractal program; they look cool! And they can make mountain ranges & planets. Terragen looks good, but it's not exactly free, & I'm cheap. :)
@stillbuyvhs the in-between program you're talking about is probably EbSynth, which is awesome and has its own graphical "glitch" look, but it's subtly different from the Stable Diffusion artefacts.
As for fractals: there's probably dedicated software, but for these types of problems if your requirement is "free", then Blender is my go-to. If you're familiar with blender already, then this tutorial basically just tells you to copy 12 lines of code and you have a free fractal generator. If you DON'T already know blender, then this is too complicated for you.
th-cam.com/video/H8nHxRO7eX0/w-d-xo.htmlsi=F3NRt1sL_xf5nF4z
Blender is great for any kind of procedural graphics. But if you want to make fractals, you can likely do it in any software that accepts code input to generate graphics, cause fractals is just math
@@stealcase I really should try Blender; I've looked at it before, but it always seemed daunting. Still, thanks for the link! Maybe, with time & patience, I can get something to work. :)
Very interested video, subscribed!
Awesome! Hope I can make stuff you enjoy watching. =)
I actually believed for a moment he did train the AI on his own work. It's still lazy and shouldn't be called animation, but shame on me. Also YAY Nightshade!
Stuff like this got me thinking I have to be a garbage person to pay the bills
I have had a life long career in photography, film, and animation. I embraced and learned AI Animation in early 2023, primarily for the Dark Side of the Moon Competition. Why? Because it looks amazing! Is it art? It is to me and anyone looking at it honestly, and that's all that counts. Many nay sayers (most of whom have no experience with it) claim it is theft. It is no more theft than photography and film making which, by the way, frightened illustrative artists when it was invented in the 19th Century (it did not kill painting for example, it liberated it). That AI requires no skill? It does and like all skill requires knowledge and discipline. That AI is going to destroy art? It won't. If you want to "future proof" your art and your career as an artist, the same rules still apply: Make sure you focus on high-end excellent work. Make sure your creation involves a lot of “human” tasks, like picking story angles, persuading, researching, architecting, strategizing, selling, which AI can’t do well and it doesn’t seem like it will anytime soon.
Since this is AI. You can steal and reupload it as much as you want, as you can't copyright something that wasnt made by a human :)
U a real one
The video is great, thanks, but it is of my understanding that both Glaze and Nightshade are useless snake oil. Aren’t they?
Edit: Sometimes it feels like the folks at U Chicago are playing the same game as OpenAI and Stability, stealing code for early versions of GLAZE and ignoring detractors who genuinely want to know their shit works.
Hey Juantoledo. They aren't snake oil, I personally know Ben and many in the team working on this.
Glaze basically analyzes your image USING stable diffusion and CLIP, and modifies the pixels in subtle ways so that Stable Diffusion THINKS the image is actually a picture of something else. Because AI doesn't look at images the same ways as humans, it identifies the image completely differently than you and me.
They NEVER send any data anywhere, the software is completely offline.
Glaze you can see the effects of: if a consumer or individual tries to finetune on your work, they'll have a lot of trouble. But it also has more visual "noise" thats more visible to the human eye.
Nightshade is designed to harm bigger models and companies like OpenAI. As a result, it relies on people adopting and using it, but we don't even need a majority of artists to use it to be a deterrent. But the more who use it, the more damaging it will be.
If you're worried about the motivations of the creators, this interview with Ben Zhao (one of the main professors working on this) is excellent.
themuse.substack.com/p/the-poison-well
As far as the "stealing code" bit:
One of the programmers on Glaze used an OpenSource library, but didn't consider the license properly that they would have to share the entire source-code of the Glaze software. They THOUGHT they were following the license, but didn't realize they weren't. Once they were alerted of the issue, they worked almost around the clock to fix it to respect the license. They took the Glaze download offline until they fixed it. The UI of GLAZE is literally more ugly to-this-day because of it.
Personally I think this shows their character: everyone makes mistakes, but it's how you handle those mistakes that matter, and this is a textbook version of how to handle this correctly IMO.
You can read more about it here.
x.com/ravenben/status/1636439335569375238
As far as the detractors:
A ton of them are AI bros or prompters, but not all. Remember, there's a lot of fearful people who are highly incentivized to make people believe this won't actually work.
If they genuinely wanted to know, understand and reproduce, there are literally 2 research papers explaining everything about the process.
Here's the Glaze Paper
people.cs.uchicago.edu/~ravenben/publications/abstracts/glaze-usenix23.html
And the nightshade paper:
people.cs.uchicago.edu/~ravenben/publications/pdf/nightshade-oakland24.pdf
There's a lot of reproducible evidence that this works, and I've seen AI researchers reproduce broken models with nightshade, confirming it works.
I've seen some people are incorrectly finetuning their own models with Glazed images and claim that it doesn't work: Often, their finetune doesn't actually work as they wanted, yet they claim success. I've seen others do finetuning, and completely destroy their Stable Diffusions ability to generate the target images.
So to me, I think all of this points to credibility.
Another sad attempt at a human being trying to take credit for an Artificial Intelligence's art. Blasphemy.
Ngl any contest including artists and animators should exclude AI altogether...
I more or less agree with the statements in this video.
However, slot machines and AI sampling are apples and oranges.
A reasonable assumption using a slot machine is that we are sampling over a uniform distribution, whereas with a diffusion generator we are manipulating the distribution through a variety of means. So it's like the difference between someone gambling and someone rigging the dice.
If you reduce CFG to zero, but keep denoising strength at 1, you get a slot machine.
Btw, if our AI-guy had spent a large amount of time on the piece, would that make it respectable?
1 I oppose IP legislation. 2 yes he put low effort, I acknowledge that, you are 100% right, others put time and sweat into theirs, this guy pooped his out (literally).
It is still good. Sad reality is that AI is scary good these days and will soon take over. Instead or targetting this guy, why don't you call for AI to be open source or regulate it?
It's not. It's looks like pixel diarrhea
Idk i found the video worked pretty well reminded me of kglw iron lung
There's a reason for that. They're using the exact same AI software. To the Iron Lung's credit, at the very least the visuals are somewhat synced to the song.
Not true everybody has to use a/I one way or another. So I'm calling cap to you
How do you steal art that is published for free on the internet?
Available for free does NOT mean Copyright free! This is a misconception. It's only free for you to see, not to use/sell.
If you took someone's art that they published online for free and started selling prints of it without asking them, it would obviously be theft, and grounds for getting sued.
Current AI is doing the same thing with an extra step of data laundering in between (the vast majority of pictures they used in their training datasets weren't licensed, they should never even have been there in the first place).
That is correct: "if I *[sold]* prints of them". The people mentioned here supposedly "scrape the internet and steal art from artists".
I'm pretty sure "scraping" doesn't involve "selling".
By the way, this video and it's comments weren't specifically licensed for quotes. I still did repeat some of its words. That's not considered theft or illegal either. When someone publishes information it automatically stops being private and it's perfectly legal to observe it.
@@quantumangel "Observing" and processing data are two different things. AI models are not human and lack the eyeballs required to observe anything. This is "my camera recorder observes things the same way I do so its fine to take it to the movies" levels of playing dumb. Who do you believe is falling for this?
You're losing all credibility when you imply that this scraping doesn't go hand in hand with the goal of processing it, to out-compete and push the people it scraped from out of their own market.
Think very carefully about whose interests you're serving when you propagate this kind of debunked propaganda. You're being used.
@@IRLlosersQ Yes. Observing and processing are 2 different actions. They are, however equally legal in this context, which is why I'm comparing them.
Also I didn't imply scraping goes alone. We're clearly talking about ai training.
Also, movies aren't published for free on the internet. Movies studios are very strict as to who can watch and distribute their content and their commissions. That's why it's illegal to record movies in the cinema (as it is to watch them without paying).
As for the interests, were talking about what covitai users do, so I suppose it's them?
Not sure what you mean by propaganda though. Are you talking about my argument? Or is there a pro-ai propaganda campaign in unaware of? Cause so far I've mostly seen complaints against ai...
@@IRLlosersQ Yes. Observing and processing are 2 different actions. They are, however equally legal in this context, which is why I'm comparing them.
Also I didn't imply scraping goes alone. We're clearly talking about ai training.
Also, movies aren't published for free on the internet. Movies studios are very strict as to who can watch and distribute their content and their commissions. That's why it's illegal to record movies in the cinema (as it is to watch them without paying).
As for the interests, were talking about what covitai users do, so I suppose it's them?
Not sure what you mean by propaganda though. Are you talking about my argument? Or is there a pro-ai propaganda campaign in unaware of? Cause so far I've mostly seen complaints against ai...
Womp womp
Naw. Get over it. Y'all need to stop complaining when AI art wins a competition it was allowed to enter in the first place.
Naw. Get over it. It's low effort trash. The visual diarrhea wasn't even synced up to the song.
100% Deserves the hate.
skill gap
I’m what you would call slot gambler. My contest video was created 100% with AI. I have no artistic skill what so ever, but there is a reason why I started to work with these AI videos. I do music licensing somewhat normally. All little commercial music jobs are disappearing(AI or my music is not good enough anymore?) and I alway wanted to make videos for my music this was my chance to do something remotely close to that. I’m really slow and detailed as old people sometimes are and I been out of work last year. I took a loan so I can pay all my software licences for 6 months and I put 800 hours on this video. I figured track is called “Time” so what would be more valuable to give to this project that my time. Looking back at the project, I’m very happy with it and I still think it looks amazing. And I learn new skill last year? Or can you call it a skill? Or did I just waisted my year on nonsense? th-cam.com/video/PTMnrxigMfo/w-d-xo.htmlsi=nzRjdn-R89ZWaU2-