The training on particular characters from your own art has made for a dramatic improvement! However, it's noticeable how primitive the mouth animation is in this iteration - there's very little shaping of the lips, only opening and shutting of the mouth. I wonder if you had got Josh to draw a lot more examples of character mouths saying various phonemes whether they would have come out better.
@@JonahNelson7yeah but the audio syncing with the lip flaps could be better. I know, baby steps but this goes to show that even human animators are still needed for details AI does not grasp in finer details.
I really appreciate Corridor for addressing the rising controversy around AI with their own exploration from the position of artists themselves and finding a way to use it effectively, ethically, and while preserving their own expression and style and skill in their pieces.
Came to the comments to see how people would still complain after all the work they did to please them… happy to see a lot of the comments are positive!!!
The fact that this is trained on custom artwork and not artist who haven't consented puts every fear to rest. This is an incredibly powerful tool that still takes amazing skills and artistry to use, and if emotionally impactful art can be created with it then that's excellent in my book.
I’d love to see this built on with additional training for lip syncing. Like adding a library of animated expressions and mouth shapes (similar to the old school stuff they made for the back end production on The Nightmare Before Christmas) and see what kind of boost we can get from that. Plus, the challenge of figuring out how to implement that in the best ways would be interesting.
There's already software to adjust mouth movement for dubbed audio or fixed lines. Not sure how well it works on animation as I've only seen it for live action, and no idea the licensing cost, but the software solution is already out there.
I'm basically repeating what other people have said, but... They recorded the audio separate from recording the video. In the video recording, the actors did not try to speak the lines, they just made extreme mouth movements synced to the lines. The open mouths, snarls, etc. are exaggerated in anime so when they did their video pass, they tried to do the exaggerated mouth movements rather than actually talking. If the mouth movements don't sync with the audio, it's because of the ACTORS not syncing - it's not a software problem.
@@boring_ugly_dude3924 Yeah, I’m fully aware of all of that. I wasn’t referring to perfect lip syncing like some Disney animation, but to achieve a somewhat better repertoire of shapes beyond ventriloquist dummy status. All it needs is a couple variations of mouth shapes, even if it doesn’t perfectly sync to every word, but simply some variation of mouth shapes to shuffle in there at least.
My biggest issues with AI generated art always comes back to consent of the artist whose works are being used in the training model. Training a model off of you own art (or commissioned art in this case) is IMO the ethical way to go about it. I really appreciate that you guys put in the work to use these tools responsibly!
@@MrYTThor although it seems the machine is doing the same thing humans are doing from your perspective, it's pretty interpretable if it is or not. Our human minds aren't as straight forward as the digital neural networks that consist the learning models. I have a Computer Science degree and one of the classes specifically focused on this last year, they made sure we knew the neurons in a neural network are an extremely dumbed down version of our "analog" neurons. And not only that, they're deterministic. Whereas the jury is still out on if we're deterministic. Either way, a lot of factors, and a lot of reasons to believe either way of thinking
@@purplemossclump5505there's a difference between inspiration and straight up copying a style, it's not a non-issue and I think people are valid for having concerns about it
I think this view is going to be a double-edged sword. In the end, the only entities who are going to have the resources to be able to generate enough data to feed into these image models are the big corporations' gate keeping the jobs artists want to protect. Then, the technology is going to be prohibitive in cost and use rights for the public. Further empowering the largest corporations in their dominance over media.
What's interesting to me is that the human artist immediately looked at other art for inspiration, so he could make his version of another artists work... which is exactly what AI does. So what do we actually gain from the "ethical" route? A human doing the mimicry rather than a computer?
@@LayoutMaster In this specific case, which the Disney artist pointed out somewhat, the "ethical" made a better end product because the animation is a consistent output to a desired looked. As in most cases of production, the easy route everyone can do likely produces the most lackluster result.
@@NukeMarine Agreed. In this case, the "ethical" route produced a better result. I suppose my point is more to do with the question of why we consider this the "ethical" route in the first place, when both routes involve mimicking another person's art style. Why is it ethical for a human to do it, but unethical for a computer?
@@LayoutMasterhe wasn’t mimicking the style. He researched and used it to inform and mix with his own style. Creating something new. Whereas the ai will just make exactly what it’s given
@@LayoutMaster That's not what he said, he described that he widely studied the anime genre to see what defined it, then *brought that genre into his own style* (this is the important bit). That's very different from 1:1 copying and ruining someone else's style without understanding the artistical thought and intent behind it.
Much of my issue with AI art is the uncredited/unpaid use of source artists work (after which many suggest it is their work). In this case, an artist was paid to generate the training data for your model. That feels like the morally correct way to use AI.
That's not quite true. What they undoubtedly did is use the extremely general base AI that was trained on billions of images (most of which are photos, not drawings btw), and then finetune it with the style provided by this artist to make it consistent. The use of the artist wasn't because it makes it ethical, it's because the artist is needed to bring consistency. While it's possible to train a new AI on art deliberately supplied for training, we don't have that yet. We'd need at least tens of thousands of consenting artists, photographers, people just running around streets with cameras, to get an AI model I think most people would consider ethical. Overall, I thing they're taking steps in the right direction and making the most of what they have. I hope they have access to more ethical tools in the coming years.
You could circumvent that by training an AI model on textual descriptions of what an artist's style looks like, only pretraining the model using images that you have the rights to use.
I take issue with the judgement that training the AI is not ethically done unless you train it on your original work. As long as it adhered to the concept and content legally seen as FAIR USE it does not matter what and whose content was used and how much. Copyrighted material is often used for inspiration, for teaching and learning on your own. If somebody writes a book about a subject I like in a style I like and then write my own book about the same thing in the same style as long I don't use the same exact words in the same sequence It would perfectly within my legal right to do so. As long as people are honest about their influence and are not trying to closely copy something it's all good. Subtle or big differences will naturally appear due to different skills, tastes, needs and environment. Also if I read 3 books that inspire me to write one book containing different aspects of those books. nobody expects me pay royalties to the other authors unless I use exact elements in the exact way, why are people arguing that AI should be different, not to mention that even if we did this with AI most authors used for training would deserve to get a fraction of a fraction of a dollar at most. The way copyright is now used is rather perverse and unethical. It often just protects people and companies that have the money and time for lawyers and litigation and does little to protect common creators from non authorized use. Long copyright "protection" even makes very hard saving some older works for which their copyright holders can't be found.
"Your" issue. No, that's a talking point you found online. A talking point which doesn't really make any sense when you consider that human beings are also trained by looking at other people's art. Should artists be banned from visiting museums? After all, they might see some art that inspires them subconsciously. Which would lead them to create a piece of art without paying or crediting the piece that they derived it from.
The AI obviously still struggles a lot with the mouth animation, but it's a lot easier on the eye without the flickering lines. Also I'm really glad that you used the drawings of an artist who fully knew about the project and consented to it, because that's what usually makes me mad about how people use AI art. I'm still sceptic about AI art, especially for commercial use, but I know we can't put the cat back into the bag, it's a thing now and it's impossible to stop - so I think what you are doing now is a move into the right direction. I really hope we'll soon find a way to make it fair for artists. Edit: To clarify, I don't mean "slap copyright on styles so you can sue anyone who creates something close to it". I mean "let artists keep their jobs and make consent a defining factor in training AI." Would be nice if we could find a way to make that work.
Do you pay a license to the guy who designed and made the first snare drum in order to reproduce a specific sound? What this AI is mostly doing is learning how to recreate brush strokes, its not photoshopping someone elses work onto something else. For sure if the ai produces something that looks almost exactly like someones art you may have an argument but that basically doesn't happen because the dataset is too large.
@@WARnTEA Don't forget about how the amen break is used in nearly every single genre in existence and yet the man who made it got nothing people are really hypocritical to their own circumstances.
@@WARnTEA For digital instruments / samples, yes, you do pay for them to have them reproduce that sound for you. (as far as I know) Also, with a lot of modern AIs, you can tell them to generate things in the style of a particular artist and they will do it. (but the effectiveness varies based on the artist) But my personal perspective is that most of the current AIs are a product of taking massive amounts of material without having the rights to use it (let alone in a commercial setting) and using them to create a for-profit product which, to me, sounds at least questionable.
@@cameron7374 Yea but you are paying the person that hit the drum with a stick one time and recorded it. You aren't actually paying the person that designed the sound and built the drum. The only reason you are paying them is because they did work to assemble the sounds in one place, and they convinced you to buy it. No one is getting sued for recording their own sample of the exact same drumhit. It'd be like an artist drawing a bird and then suing anyone that draws a bird. These things exist in the real world, they are building blocks of life, they are not art on their own they are just things. It only becomes art when humans arrange those building blocks in a specific order and add a deeper meaning to it.
I do admit I love your passion and hype for this project. The ONLY thing that worries me so so much about this, is whenever the big studios get their hands on this, jobs for animators will be cut in half for sure. And jobs for animators are already barely a thing anymore, let alone jobs that pays well. BUT, maybe the introduction of this will allow traditional animation to become a rarity, thus being more sought out? I dunno. It's a scary world. However, if we look at this project secluded from traditional animation, I think it's a pretty cool project where you evolve techniques, push limits, and try something new.
Unfortunately usually tech can't be held back by the possibility of lost jobs-- it just doesn't happen. The tool exists now, and the industry and those within it will almost certainly need to adapt to it. At least historically, a tool hasn't been stopped from being used at the risk of jobs being lost.
As much as I agree with you, there is no denying that AI will eventually become the norm. Right now there's a lot of very strong opposition towards it, and I understand it very much as an art student. However, history has shown many times that technology ALWAYS triumphs. When digital art appeared, the traditional artists who used pen and papers also opposed it. Now Digital art is the norm. AI Art is a similar debate, IF we disregard the ethical aspect regarding the AI sampling on other artists without their consent (which is not a neglectable point). On a practical standpoint, we will have no choice but adapt. Marc Brunet said he would welcome it as a tool, as long as it is not used as the finality, and I agree on with him. Use the AI to help yourself, but you still gotta use your artistic skills to polish the final result. On a moral standpoint, I think our only hope is to have legal regulations on crediting sources / sampling consenting artists.
Look at it like this, if a company purely relies on these kind of programs without real talent to make it work just for the movie's story being bad in the end, then they weren't a good company to begin with. Meanwhile you have people like those in Corridor who are actually managing to pioneer this technology as the tool it was meant to be. The fact that they are willing to learn from the previous in order to improve goes to show that with enough drive and creativity, anyone can finally be able to see their ideas become reality without the need for big corporations who only ever cared about the bottom line. Now they just need to figure out the weird inconsistencies like with the eyes and etc. and there will be no telling where this will go.
@@Kenddamus The analogy that AI art is anything close to traditional artist opposing digital artist is stupid. sure, was there some pushback, but ultimately the basic need for skilled labor was still there. a digital Artist can still pull of great artwork on Paper, maybe the will struggle a little bit more, because they have to draw without all the quality of life tools digital art gave them, but AI art is not a quality of life tool right now. it literally replaces artists. that's it. And if the future of humanity is going to be that art and culture is AI controlled, then we as a species have no future.
I don't understand how anime rock paper scissors could be controversial. - You didn't violate any copyrights (as much as I disagree with copyright, different issue, another day) as there is no semblance of any of the original source. - As far as I can tell, all the frames are original, so how could the complaint of this "looking like" something else be made? - No, style can't be used as an excuse for controversy, no one person owns claymation nor anime nor cal art style, etc. If this were the case, there wouldn't be as much media as there is today. - If this is about the amount of human labor put into the project, that shouldn't be it either, buggy whip makers haven't been making buggy whips in the same number they have been in over a century and look at where we're at today, more jobs today than people back then. As a tool in art, I don't see why people are able to complain about this with a straight face. As a software dev who sees github co-pilot's output on occasion, I find it laughable that we consider ourselves replaced.
I think a big part of what this is missing is actual animation. Animation isn’t just trying to exactly emulate real world movement, it is being stylistic and specific with the animation choices it make, like what parts to include, what to tone down, what to exaggerate, and what to do that just looks better or more appealing even if it’s less realistic, so just having anime visuals on top of mostly real world motion will always look sort of strange, this is not really an anime substitute in my mind, just complex and very automated rotoscoping. It’s like wirh 60fps animation interpolation (which noodle made a great video about). It’s often times not really an improvement on the base animation as it does what is logical and direct but not really what is best artistically or purpose driven. I think though that that doesn’t mean that this side of things can’t be done or worked on with ai ever but it’s just not really now and it is id say a pretty large missing piece.
Its almost like we would need an Ai specifically trained on animation, ie the actual movement of animation, not just the stills. I wonder if that could be on the horizon? Every principle of animation can be interpreted in different ways and varying levels of exaggeration, the ai needs to learn about different ways of doing so.
We are still pretty early on in the realm of visual generative ai. Don’t count on that remaining permanent. My work is in large language models but this is nearly as interesting, and I’m not the only one who wants to push it forward more and more.
@@gmcubedexcept art is intent. Animation in particular is driven so much by the artists vision and intent, and you won’t get that with AI tools. Auto-generative AI tools will never know when to correctly use the 12 principles of animation, because when you use them is both heavily context dependent and intent driven.
Thank you for using commissioned artist work that they gave their consent for to be used in this project. This is an ethical use of AI and should be the standard.
Except someone will just hire an army of third world artists at 50c/week to make art in the style of well-known artists and then we're back in the same place.
@@chaos.corner look it's a start, i understand that there is a whole system of using studios with underpaid workers, and it's definitely something that needs to be addressed. For now I will give these guys a bit of credit for doing something right. We'll then work on unionizing animators around the world.
@@CaliMeatWagon I think they kinda covered that in their explainer video. I may have to double check it now again to see how they were able to work around that since they filmed themselves before rendering those shots into animated frames as well as using stock photos of castle interiors as opposed to using a prompt to magically yield out a frame.
@@CaliMeatWagon You are aware that every artist ever looks at other pieces of art for inspiration right? No one is just sitting in a bubble thinking of every little thing from scratch. That's literally why trends exist.
It looks great, one of the things I notice is the eyes still distort and break. We need a way to generate the eyes an have it stick to the same shape. One of the idea I had was to make clean versions of all the eyes from each angle, and have a seperate AI pass that just does eyes. Hopefully it can track them better with most angles and pupil directions. The other thought was to have the eyes rendered with a 3D AI, that actually makes the physical eyes and lighting based off the drawings, then applies them to the characters appropriately. Gen2 is currently doing some nutty stuff with video shots. Especially with stable models from frame to frame.
I do enjoy the previous art style more because I think it looks more gritty and visceral and I love that rotoscope look, but I appreciate the reasons why you changed it
I think if they had a larger data set to train their model on, the video could have come out better. The old Anime Rock, Paper, Scissors had potentially thousands of frames to learn from. I can't imagine the artist they worked with drew more than a few dozen images.
kinda affirms a thought I had about why artist protests won't stop the advance digital AI art: They can just hire a human to imitate/synthesize art styles and then train on that.
@@Deliveredmean42 : True, but this way you can get a famous artist's style without having to hire the *expensive* famous artist. I'm not saying it'll eliminate the need for artists, but that it's a workaround to recreate particular artists' styles.
@@Deliveredmean42 Until the library is built. Until the models get better. Then the ai models take over providing the library. And then corporations buy up competing ai models. And then big media company's get a practical monopoly and 99% of all payed artist gets kicked out of the industry because they cant compeat in their old product and they dont have access to the ai tools to compeat in the new ones. People are that are screaming about copyright with ai are only thinking of next week and not the next 50 years.
Here's my 2 cents. I never really had much of an issue with the artistic integrity or value of using AI to create art. I think AI art can be a very interesting and distinct artform when it's not just trying to imitate existing ones. What I *did* have an issue with is the lack of consent involved with the training of latent diffusion models. It'd be one thing to train a model on copyright free images, but it's another thing altogether to just rip stuff from across the internet. The latter of which, is unfortunately precisely what Stable Diffusion has done. This isn't really an uncommon practice in AI, but when it comes to art, something with so much skill invested and so much of yourself put into it, I think it's work treading with more care, especially when the tech is so blatantly capable of devaluing the work of the artists that make that training data. StabilityAI got the memo with Dance Diffusion but apparently not with stable diffusion. I very much appreciate the effort of working with an artist who has agreed to have their art used, that's certainly a step in the right direction (and it's great that it actually helped produce a more stable result), but this doesn't change the fact that the underlying models still are built upon what I would consider to be an unethical form of data collection/use. And frustratingly, I don't think they need to be. (there are other issues I have with AI art, but I don't think they apply to Corridor's use case, so i digress) I can't help but find it upsetting (and if it were me personally doing it - more than a bit gross) to be promoting and using such technologies before the underlying issues at their foundation are resolved. The more we build upon the foundation the more difficult it becomes to change it, to make it better, and we *can* make it better. I've been following Corridor Digital consistently since I was a kid, always looked up to them, but this has unfortunately changed since the first Anime RPS. How it was handled was especially upsetting to see, especially after having admired how informed Jake's video on the technology was a while earlier. Idk. Hope Corridor really does figure out a way to do this which doesn't make me feel so icky abt it some day. Perhaps take a bit of time to look at any upcoming fringe versions of the tech more committed to ethical development, or consider alternatives like just working with artists on tech like EbSynth, dunno. I myself am an artist, but also working on research in deep learning, so if I perhaps seem too invested in the topic, it's because I am lmao. The general public is familiar with being not all that invested in ethical issues when they don't impact their lives directly, and I won't pretend to be a particular exception. But I think we ought to care about some of them at least. I care about art, I care about artists, and I care about developing AI safely and ethically. This ultimately doesn't effect me directly all that much, I don't post my art publicly online and my career path is more directed towards tech than art, but I reckon it'd be going against my values to not care about this, and would make me a bystander in building a world I don't want to see. idk, just food for thought ig. (as a tangent/aside, I regardless do still hope this AI version isn't a flat-out replacement for the anime-style videos. Corridor's previous more vfx-based anime-style videos had a lot of charm which tbh I think is actually held back by the AI image stuff a lot of the time.)
ITS NOT ABOUT THE TRANSFORMATION OR EFFORT ITS THE COMMUNITY OF THE LIVE ENVIRONMENTAL THAT CREATES A DIFFERENT EXPERIENCE. SAME AS SPORTS BARS. DONT ROB THE VIEWER OF THIS NEW FORMAT FOR THEIR FAV STREMAERS PUSH TO GET A STREAMLINED SYSTEM IN PLACE ON THE PLATFORM SO STREAMERS AND REACT CHANNELS CAN SHARE REVENUE OR PURCHASE REACT LICENSES
@@515JUKE You're 100% right. React content definitely exists in a nebulous place in terms of fair use and IP law but the fact of the matter is that there is a large audience for it, and the platform owners and rights holders are failing to capitalize on it. They need to develop methods for embedding and syncing other videos with view count and ad revenue sharing. And there should be communal viewing options incorporated into streaming platforms (IE "Netflix Couch" or "Twitch Theater") where the IP holders also get a cut of the revenue generated through subs, virtual ticket sales, cheering, or whatever other monetization methods they can figure out. And naturally the creators should get a cut of that too via royalties.
Flickering/consistency has been a huge problem for me. Fine tuning models that map one person to one character is a really interesting solution and clearly it's made a huge difference. This entire workflow is just on another level. Well done demonstrating the best this technology has to offer.
@@technus147 bro taking inspiration isnt stealing. same thing with references. ai takes ppls art and puts it in their dataset 90% of the time without the artists knowledge or consent
@@technus147not really since humans are not comparable to ai and ai is not comparable to humans the way you people compare it like “ oh stalking others and taking their photos without their knowledge is fine you look at those same people through your screen without permission or consent why should we “
I would be fascinated to see a breakdown of how long it takes to do this, versus a studio doing a more traditional animation technique. Obviously the Crew is still feeling out the work flow and everything, but the time comparison would be interesting i think.
I think it would be largely based around two large parts: 1) time to train stable diffusion with source material, and 2) time to export video shots into individual frames, run them through stable diffusion and recompile them
@@Spiker985Studiostraining speed can be solved by giving it more power. The slowest part is creating a dataset, describing every single image and everything contained in it. And then doing it again when training goes wrong. I trained some loras on datasets of two dozen images and even that was pretty tedious and slow process
But., everytime there are a new tools to make the workflow faster, all the big studios will use it as a license to make a half assed product, new workflow mean that they could McDonalds-ify the product, make a mid story, mid vfx, and assign a cheap, clueless and mid people to create the product . fast production mean fast money, therefore it is not about the Product anymore, it's about making a generic product that addressing the current trends because they saw that technology as a way to replace expensive people in their company we already saw this phenomenon time and time again, in the movie industry, in the game industry. there are a massive lost in the knowledge base, because the guy who create the masterpiece was too expensive to hire, and the knowledge wasn't getting passed down to the next generation. therefore we saw a massive degradation of quality in the past 10 years despite the promise of giving you the power to make a jawdropping scene and great storytelling easier, giving people the liberty of creation. we saw less and less Masterpiece each and every year
@@jensenraylight8011 the big studios were doing exactly this for years, well, effects were good but story usually worse then mediocre to compensate for this:)
From what I gathered the AI in the 1st animation grabbed info from VHD and that style often uses squint eyes, closed eyes then animating them wide open in times of shock surprise etc, this is a key style while the AI couldn't allocate where these changes should accour, so even tho you could take care of the flicker fixing the face is a nightmare. So if you are going to train something I guess a code needs to exist that covers opening and closing the eyes, because it pulls it all over the place at random and if you spend time fixing it frame by frame might as well make everything from scratch since you will get more clean results lightyears better in comparison and original style too. The time spent training, editing rendering versus the time making everything from scratch by an capable artist burns less money and less resources. I honestly don't want to know what these people pay their electricity bills on all these badly optimized amateur programs and AI scripts ... A year ago I was playing around with learning AI things and not only it nearly fry my SSD, RAM, GPU and CPU basically whole PC was burning, my electricity bill was 100 times more then during winter time heating ... I had to stop burning time and money also had to repair my PC's cooling asap. With other words I burned years worth of winter heating in just few weeks time. So yeah no thanks ... I rather do nothing or crowdfund and pay actual artists if I want something done properly ... F--k AI and these amateur programs too ...
If a team of humans watched D and made a new movie faithful to that art style, the community would love it and praise it for making more of what they love. If you do the same with a machine, they lose it regardless of the insane amount of work you put in. This is the final equalizer for 2D art. The barriers of skill, time, manpower and budget are stripped away, so anyone with creativity can now get their vision onto the screen despite the real life limitations they face. If you're an artist who gets that unique satisfaction out of the process of your work, you are unaffected. If you draw for profit, you now know how the auto worker felt when Japan introduced the robotic arm.
When you create your own art stylebook specific to the project and train the AI to do that specifically for each character, I'd day there's no longer any problem at all. It's now performance capture with AI-assisted animation. 100% your own product.
People will still find problems with it, I guarantee it. Because "real art" is more than just the style being your own, its the execution too. The moment you give up control to an AI on that level, it opens things up to a lot of backlash.
It's not 100% their own. It's still using a base model like SD 1.4 or 2.1, with the anime style Josh made layered over it. If they weren't using the base models, there would be no way to make this look coherent without an impossible amount of art. For reference, LAION-5B, the dataset which Stable Diffusion's base models were trained on, has over 5 billion reference images. That being said, it's still a big improvement from their first project in terms of ethics, and Stable Diffusion is probably one of the more ethical AI models of any kind purely by virtue of being open source. It's a cool project and more interesting than just using AI to replace people's jobs.
Training the AI on art specifically made for this project is genius. I hope this will become the new standard for using AI. Well done guys, really great work!
I wouldn't call it genious, its a great Idea but they arent the ones who came up with it, it was actually the main concern with people who are anti AI useage and we are actually advocating that specific artists get paid and have their art used with their premission in ai models, instead of the models that art scraping millions of artworks from google images without artists being compensated or even without their permission.
@@VJArt_ They still used stable diffusion and just trained a lora on their own stuff. Their process still uses SD because they know they couldnt make this work without all the stolen content from SD.
@@VJArt_ True - I mean it was a reaction to the backlash they faced for their first video. I still think it's great that they share their workflow for a more ethical way of using AI, considering they have a huge audience.
@@2darkiBut that's a whole different conversation. They are not replicating any copyrighted style on this creation, which was the point of controversy from the last video.
You guys nailed the only Ethical way of using this technology. Hire an artist and train the model based on the artworks he created for that specific project, with his consent. BRAVO
It amuses me that you would say that, because the artist explicitly states that he trained himself on the artwork of several other people, none of whom were asked permission.
@@duckshallrule6937thats different though, as when a human processes that artwork into their own its transformative, as a human can not have a perfect memory, cannot replicate everything it sees perfectly, but an ai artwork is directly derivative of the information it is fed, it processes everything it sees perfectly, it only applies transformations using the information it has been given, and it itself can not be creative in any way.
@@jc_art_ "it itself can not be creative in any way." You can truly make this exact same philosophical argument for humans though. If the only difference is that humans are imperfect then you can just tweak the AI to make "mistakes" too.
@@dustyfoxboy442 Machine learning is on every single level different then human learning is, 1st an foremost in the number of neurons, humans have enough to be fully sentient and make decisions, and decisions in art is what creativity is, just as studying referance and other artwork, its in the decisions of what you study and what you get inspired by. Second of all, the way those simulated neurons are made is also different from biological neuron connections, just as the name implies, its just a simulation, and it can never replicate a biological brain, its can be its own thing, but a machine learning something is nothing like a human learning something. Do some research before you start spewing stuff you dont know about.
Corridor showing what can be done when you put effort into the process of making stuff with AI instead of “this is the future of animation, fire your VFX artists”. Truly good stuff
I don’t think it’s necessarily pertinent to call it “AI art” or “AI animation” since the only parts AI is involved in is applying a glorified filter onto live-action or pre-modeled images. Still, a certain level of humility and consciousness should be held vis-à-vis the media this is meant to replicate, as studios will use any and all tools available to cut costs and workers. You may not personally plan to replace regular 2D-animation or make the lives of 2D animators harder, but presenting stylized filters as a potential substitute for the art form opens a Pandora’s Box that you must be aware of. An ungodly amount of artistry and effort goes into your work, so it may be very misguided to emphasize the importance of the only part in the process that isn’t artistic and requires little creativity: applying the filter. Respect yourselves, your own craft and talent!
@@pendaco Well they are still just playing around with the tech. The aim is to improve the process while making a cool interesting video They are not trying to make a perfect anime product that actually compeat in the normal way. And doing things like hand animating eyes would be a large sink of time, energy and money when that kind of perfection is not the goal. The selling point of the videos for now is the ai novelty not the art alone.
The eyes, the shadows, the teeth are still really wonky.... the general face.......... Even the cracks around the hands feels like it's lacking impact and........... Yeah, the more I think about it, this just doesn't look great at all. I'm not sure we'll ever get this up to the standard of actual animation, but it's better than nothing, I guess. I just hope big corporations don't see this and just use something that looks this rough to replace actual artists.
Now that all your processes are in place, I'd love to see you take a 30 second random scene and then in real time (or as reasonable as possible for UA-cam) show us how quickly it can be converted into a finished render with full audio and effects
Well you can just think about it. Its not gonna be fast. Days of work perhaps. The final editing and pre planing more so then the filming. But still far faster then animation. Just look at the release schedule of any bad animation studio and compare. And like with everything ai this is the worst the quality and tools will ever be. THey will only improve making the workflow faster.
@@RasakBlood Disagree. This level of animation is not that hard, assuming you have the initial artwork, which they do. This episode took 6 artists 4 months? That is a good time, but not unthinkable. If you approached it by actually animating the artworks by hand, which is a normal thing in cheap animation projects, you could probably do it in 6 months with 6 artists, but the quality would be slightly better, mouth animations especially, and the perspective consistencies, which are all over the place in their animation. We've done 30 second animations like this in a week for commercial clients - I'm not saying they were good. In fact they were complete crap IMHO. But that's what a client gets if he wants a 30 second animation in a week. And their project absolutely reminded me of these animations.
Well, 4 months is the time between release of both shorts and they have been doing a lot more work beside this short. I think that they coild achieve this within a month if they were fully dedicated just to this video.
The main thing I don’t like about generative AI is that they are trained on stolen artwork from actual artists. If you create everything that goes into it for the purpose of training an AI model, then that sounds fine.
The people who dogpiled on the last video were just virtue signalling. The complaints about AI using art without compensation are silly. If you make art to put art into the world then you should be glad you're still accomplishing your goal while also contributing to the future. If you're making art to make money, you got paid when you sold it. Maybe you believe computers can't create true art, then your personal contributions will always be valuable, and you've got nothing to worry about. Or you're worried that computers can truly create art better than any human and you will be replaced. Well welcome to humanity, there's a lot of that going around right now.
all you people thinking this would "replace" people's jobs , this is just a form of technology. stuff like this have happened many many times in history. machines , computers , etc. when cars came along , people who breed horses thought "I'm gonna lose my job!" aaaand that wasn;t the case. It's the same. "History repeats Itself"
I appreciate that they’re not copying another art style through ai, but at the same time I do think the art style outputted from the ai looked much better, more detailed and serious, than their own new style, even if the flickering itself was improved
I agree with you, but Vampire Hunter D´s model is so much more complex. They could hire a veteran anime artist to create a dataset for them, but the cost would be much higher, and they´re taking the first baby steps in a new workflow. I assure you we´re going to see such datasets from the big companies like Bandai very soon.
To me this is the crux of why compensating the creators of the source data is important. With full respect to the artist who created their new style the work in vampire hunter D is stellar and the result of professionals at the top of their field. The ai output is only as good as the source data you give it so it does have value. If people want to train models based off the work of an artist that artist should be compensated for it, which then allows them to create more art which can then contribute to other models. There's a real possibility for everyone to benefit here.
I find it especially intriguing that with a solid pipeline, projects certainly seem so achievable that were never previously. It really let's your creativity go wild with the possibilities. It could even be as draft or a strong pitch for a much deeper, more expensive, and involved full feature project.
If you remove the word AI from the equation, you guys just created your own stylized filter. If people are butthurt over that, then we have a different problem. This is incredible work you guys have done. This is exactly how AI should be used to create art.
It is a neat filter, and if they make the reference material for it then awesome have your fun.... The problem is they are not acting like it's a filter or treating it like one
@@kingdomkey63 They're playing it safe. Majority of people are still not sure what to do with this "AI art" which is understandable. Its new. Corridor can't exactly call it a "filter" because thats not exactly it either. Its a lot more complicated than that. But to your average joe, it is basically an advanced filter. If they called it something along the lines of a "filter" directly then people would assume Corridor is not taking responsibility of the fact that they are indeed using AI to avoid controversy. It would've been worse actually lol What they did here is smart and how it should be done. AI art exists if done right. For example, generating an image from Dall-E and manipulating it a little and calling it art is not art. Training your own AI with your own reference material, generating an output and then manipulating it is art because now you've put your own spin to it. Idk, thats just my opinion as a creator and a viewer.
@Grabster no I would say that filter is actually the perfect word for it. If you can't accept that this is an augmented reality filter then I really don't know what to tell you, other then to hand you a dictionary.
The main controversial thing isn't the tool itself makin it easy, but the ethics of the training data which was taken from. Which i give you props to commission an artist!
The problem the haters have is that by that logic no human artist would ever be allowed to learn by studying and learning to mimic the style of Bloodlust, and yet that’s exactly how artists learn. Copyright gives artists ownership of their expression of a style, but not ownership of the style itself.
@@truejimThere is a major difference between a human learning from others and AI simply copying what it has been fed. Current AI does not actually comprehend what it is "learning" and simply regurgitates, while a human actually does comprehend what they learn and is able to create something new that isn't just copying. AI in its current form is NOT actual intelligence, and can not be directly compared to human learning as they are fundamentally extremely different. The only people suggesting otherwise simply do not understand the technology or are purposely trying to misrepresent it. Therefore in its current form yes the main controversy with AI art is that it is trained on source material without permission, because the end product is simply reproduction and only imitates being transformative by mixing a ton of different sources together but it is still just copying what it is trained on. So if permission is granted to train the AI on source material that is fine, but if permission was not granted it literally copying someone's work without permission and without credit. An important note to end on is that art style is not protected by copyright but the final product is. Therefore training an AI to copy that final product by feeding it that final product IS copyright infringement. It's just too new of a technology for court cases about it to make it through judicial systems yet, but that doesn't mean it is legal or properly transformative.
@@hazelberry You’re making my case for me. You can’t sue a computer, you can only sue people(s). There’s zero legal precedent that says people can’t use tools (even computers) to study prior art. If literally nothing is being copied, then copyright doesn’t even apply.
@@hazelberry It literally is learning and it is intelligence, regardless of what you want to believe. It’s not copying and it’s legal until determined otherwise, which is very much up in the air.
@@truejim You clearly are neither an artist nor a developer because you are making a wild stretch to connect what actual artists do when learning vs what an AI does. It is fundamentally different.
I personally LOVE the new artstyle, it's such a great blend between anime and cartoon, which seems perfect when using tools that struggle with clarity and consistency :D
10:51 That's what I think is the problem right there. The issue with Ai isn't artists like you making new and exciting content for the sake of creativity, I do fully agree you're on to something entirely different, and in fact is something quite intriguing and awesome. However, sadly, artists like you aren't the ones who lead the world of entertainment. The concern I personally have (and the one I've heard other animators like me have) is that companies and studios will only see the last part of your statement "Here we are, with a full episode of an anime, made with 6 people over 4 months". That's what's scary, that because of a new form of Vfx, the side of animation is in danger of being seen as something worthless by corporations, all while animators are already being neglected and exploited as I'm sure you know. So I think that's the fear- that this fantastic tool is going to be exploited by corporations by kicking out animators and artists in exchange of something more efficient. Leaving artists with almost nothing to fight back- no matter what we do AI is always going to be faster than traditionally drawing and more efficient- and that's all that corporations see, and that's all that gets out to the world, making it all that people will see devaluing artists even more so than they already are- so If I were to assume why the controversy exists, it's just that... we are scared that something wonderful you are making is going to be used to harm other artists, is nothing against you or Ai really, it's fear of loosing our livelihood. It isn't just that AI is stealing art and what a machine is doing is being referred to as "art" (which is demeaning and terrible because it carries no emotion or creativity, it simply blindly replicates what others do and blends it together in a big slush of an image) Right now they are writers on strike, and what Disney is doing is trying to replace them with Ai. You, with a small team (relative to Disney) was able to improve Ai this much, imagine what a corporation with ridiculous amounts of money can do with Ai- So it isn't even an irrational fear anymore... I'm happy that you've used an artist to create your style from a new, and I'm happy you're excited about the progression of Vfx with the use of Ai- but this doesn't really make it any better, if anything you're kind of proving the point that if anything the art world is becoming even more competitive at best and becoming truly horrific at worst, so while this isn't anything to do with corridor, it is a situation which may bring consequences might be a lot bigger and dangerous than one might imagine. So are you responsible for any of the things that might happen? Not really, but by normalizing the use of Ai it's discrediting what artists like me or other animators can do. So if anything, I'd just ask that anyone using Ai can really understand that the commercial use and normalization of it hurts artists, not because of you, but because of how corporations and the public might react to something like this. That's what I'd like to say, personal opinion but based on what I've seen with the art world I'm in... I hope it helps shed some light in what I think is the other problem you didn't get to address here. But again, thanks for getting an artist to make an art style, if the entertainment world wasn't so focused on making money and cared about art, that would solve the issue, but we don't live in that wonderful imaginary world were artists are valued... But that's what I think, regardless thank you for the video addressing the artists, I know it isn't for me but as an artist I appreciate the effort :) it was a good progress to witness and it was great to hear your point of view
I think the problem isn't ai, it's more about capital owners firing workers to get more profit. Human greed always ends up messing up new technology (we had the same issue decades and even centuries ago with new technology.)
@@no_less03 You do understand that everything you interact with daily is new technology? And that almost no jobs that existed 150 years ago exist today? Redundant jobs getting eliminated benefits us all.
@@no_less03 I'm not sure if used this way it'll make artists lose their jobs. Rather, it'll change artists' jobs. Instead of animating everything, they can now draw a stylebook for the AI to use and still get paid for it. Plus they'll likely need to touch up/redraw some bits and pieces even after the AI. And of course, you can't do performance capture without hiring actors either. At the end of the day, I don't think it'll spell the end of traditional animation either. Computer animation didn't kill drawings when it came out in the 90's, it simply became another tool for artists to use.
@@Ruzzky_Bly4tonly when the people working those jobs are either reallocated in the workforce and/or taken care of somehow until they can learn a new trade, but guess what, companies don't want more expenses, they want more profits, so making jobs obsolete without an attentive eye will just lead to more struggling people
@@luigiff3431 You just described the reason why capitalism will have to change or be replaced. Stopping all progress because it can replace humans is not a good solution. But if the current systems won't change, people will be unemployed, and the 1% contorlling the machines will have all the power.
you know what? god bless yall and i love seeing people work on something they're passionate about. you're hardworking and you'll be the one to break through this new gen of production and animation
Thanks for making this! it's a great demonstration of how these tools can be used creatively, and the best part is that you still needed experts for it to work.
Im SO glad you saw the point about your OWN art style. Effectively training on your own art with the new updated to the method is crazy for true transformative works
They just got an artist to redo the Vampire Hunter D style, how is that different from a diffusion model doing it? It's just an extra step but instead of via silicon its via carbon.
@@tylerpixel"I took inspiration from vampire hunter d, AND ALSO GUNDAM, COWBOY BEBOP AND BLENDED IT WITH MY OWN CARTOON STYLE" Skip that part did you? But for sake of argument, say they DID pay an artist to just "redo" the style. Its still better then taking art through ai, as, 1. They are paying the artist. 2. there is no full seperation of style from artist, even having a full team making a show in one style its possible to pick what animators drew what scene if you're familiar enough with their hallmarks (sakuga fans can attest to such) pure ai art simply rips it from them, generalizes and spits it back out. Which is why using an artist helped it look better (shock horror artists are good at art) Ai and art can coexist, done ethically. This video is case and point which is why its way less controversial than the predatory methods of the rising corperations
@@gojicandle8188 Yeah so they added more "prompts" and mixed it, when I use Midjourney I merge different art styles together, Robert McCall, Simon Stalenhag, Alex Ross and others. I then add to it in Photoshop then reupload it into Midjourney to generate more variations and ideas. What I am trying to get across is the other sides argument that both human creativity and digital creativity are the same thing, latent diffusion is the same process that is happening in a human mind.
@@gojicandle8188 The AI model did all of that but also considered 1000 OTHER styles as well, it's less of a copy if anything. Anyway a lot of artists are Narcissists apparently, who don't admit that their work is in one way shape or form a derivative of the works of art that inspired them. PS: The only reason it looked better with this workflow is because they made their own AI model with LESS data/inspiration to draw from, producing a more reliable generation between frames. - It's literally MORE of a copy cat this way and some artists still had their style ripped off.
With how the render turned out, I feel like this is underachieving what mocap with cell-shaded 3D models would achieve. It's nice that you're only using art from your on board artist for training, but the "style change" for backgrounds is still an iffy portion. Again, a 3D environment would've solved that to begin with.
I loved the jittery art style because it literally looked like just that, a style. It reminded me of 'a scanner darkly'. And since the last ep I've been curious what a scanner darkly could have looked like had it used this Ai technique instead of the hand drawn stuff. Could be an experiment for a future project maybe? Scanner darkly with todays technology
I think what makes this one more ethical then the last is that y’all decided to train on an art style from an artist who gave permission for y’all to do so and was involved with the project as well. That is one of the main reasons artists like myself don’t like ai “art” and the way the tools are being used at the moment is because people’s art get put into these tools without consent or proper credit, then people try to pass off these collages of stolen art as their own. There is still a long way to go in terms of regulations of these tools to protect people from being exploited by them. And with y’all deciding to take in the perspectives of those who are being exploited by these tools and look to improve upon them is something that is needed.
You could not have said it better!! The choice to use a consenting artist is a hugely good thing because now it’s not really stealing from another group. Plus with the extra work that corridor did besides the art the AI would be trained on, it’s really truly it’s own thing now in my opinion. Edit: I love the deep discussion happening in this comment’s replies
As an outsider who’s not an artist or anything, here’s my question. What’s the difference between what this hired artist did, using reference from multiple different styles of anime without the creators permission to make his own style, versus an AI, who trains from different images of anime without the creators permission to output its own interpretation? That’s been my biggest question about this AI debate. Is there something I’m missing that can help me see your point of view?
When the internet exist, there were tons of criticsm circulating everywhere, but we are forced to cope with it and now it’s a norm. I dont see any difference with this Ai things. At first there is going to be many criticsm, eventually people just accept the way it is and move on. Instead of listing disadvantage of the tools. Maybe we should look at the brighter side and think how we can use this tool on the positive side.
This is it. The general AI tools and their coders take art and then ask later versus asking first and letting the artist be involved. One stregnth of the general tools is someone like me writing a book, but I'm very visual. So I can use them to get an idea for the look of my characters, but only for internal, initial use. If I decide to publish, I will get an artist to make the covers and redo the characters in a different way for future projects. I'm very happy Corridor decided to revisit RPS using an artist and their own style.
this is the thing with AI. it's an incredible tool and you prove it here. You used it AS A TOOL! WITH AN AIM!.... the problem is that some people feed images most people would found beautiful to the algorithm, insert a prompt and call themselves artists ... which is nonsense!
How did they use it as a tool tho? It wasn't the actual artist who used it to help him. it wasn't an animator who used it. They paid an artist for some concept art basically. took that, ran it through the AI and managed to now kick out any specialized people. They literally only had to hire one artist to make this all. How many animators and artist would have worked on a project this scale and would have made a living off of it? And they are a small studio that don't even specialize in actual 2D animation. Is even one of them a solid artist? Probably not. All I see is a dozens lost job opportunities for real artist who actually learned to draw and animate the hard way. 2D is probably going to be dead. This technology will only improve and become better and better. And at some point the entire entertainment industry is going to be majority AI controlled. We already have AIs for generating texts. We have AI for generating art. Only a few more steps and soon humanity will be relegated to labor jobs. Human creativity will basically be reduced. And those labor jobs at some point will also replaced with machinary and AI as well. "What's the point of trying to be creative, when an AI is going to make sure I will never be able to use it to make a living?"
Those kind of people won't be around when the fad fades. And those that know what they're doing do actually respect the artists that made their work possible, despite the differences regarding their workflow. The jerks who use AI art tools to feed their own ego would usually be quietly shunned by the community, since it did give them a bad name disrespects both sides.
Its art. It's literally a r t. Or else you can't say some paint splashed on a canvas is art either. Because that's just some people throwing beautiful paint on a canvas and calling themselves artists.
@@DatAsianGuylook dude, I know you think the worlds gonna fucking end when AI creates “Imbiamba Jones 10: The Lost Whip”, but be fucking realistic, AI is only gonna affect Hollywood and few animation studios around the world. EVERYTHING ELSE is gonna use traditional medium, like every single indie studio trying to make movies with passion aren’t going to use AI as a “cheap” option, which is what this is shaping up to be, despite obv being waaay more efficient than hiring 20 inbetweeners, you get me?
I think this is the most genius way to respond to the controversy and says exactly what was on my mind at the time - all the actual art/craft you put in every step of the way separate from the anime style transfer. Thanks guys.
Proves that there is no controversy. AI is a tool and can't be used to actually replace anyone outside using dangerous machinery and menial min wage task jobs that both should not exist in the modern era.
I can see Corridor are desperately trying to justify themselves to the losers who cry about AI being theft and immoral. You don't have to justify yourself to those morons. You have done absolutely nothing wrong. People complaining about the AI stealing from other people's work are just bitter. All artists steal from other artists. Your "unique" style was trained by a lifetime of exposure and learning from every other artstyle. No one is "original", in the true sense. People are just jealous because instead of a lifetime of having to train their model, the AI can train a model much more rapidly. Why hire an artist to make your own style? Where did that artist get his style from? A combination of other artstyles he's seen before? Why not just train your AI to do the same thing? You guys are pushing the envelope forward, which is exactly what true artists should always be doing. Keep on exploring AI and how it can make art better and easier. Don't ever feel bad for it. Forget people on Twitter and Reddit crying about nothing.
Hiring a REAL artist not only makes this morally right, but also makes it look BETTER than stealing from an anime movie. Yet still, kinda shitty. It still needs an in-between animator.
First, I agree that making money another's art without due attribution/compensation is unethical, but I'm about to play devil's advocate for a moment. I was just rewatching this and at 6:50 the artist they hired said he took his style and, since he is not learned in the style of anime, used anime like, Vampire Hunter D, Gundam, Cowboy Bebop, and more to influence his art. Shouldn't they have hired an anime artist to use just their style to make the art? Didn't they just essentially do the exact same thing, but pay a person instead of an AI to do it? I am not trying to troll or get under anyone's skin. I genuinely want to know others opinions.
"Are we witnessing the birth of a new art form?" I sure as fuck hope not. AI is/has already become a HUGE threat to people working in the creative space, and the last thing we need is even more of that. Real artists are being drowned out by a flood of AI shlock, writers are at high risk of losing their jobs to something like ChatGPT, and now animation is being threatened more and more as the days pass us by. This AI has the potential to lose people countless jobs because the higher ups see no need for them anymore. Why hire a team of a few dozen artists who have honed their craft for decades to make you something beautiful when you could have one artist and a guy who manages an AI? Sure, the studios might love it at first since they're able to pump out episode after episode on a fraction of the regular cost, but what about the people who ACTUALLY matter? Those animators will now have trained at their craft for decades and are now out of a job. The fans are receiving soulless AI slop and they'll eventually fall off, losing the studios more than they've been saving. I mean, animators are already paid horribly and overworked as is, now you're telling me they can easily be replace with an AI by the people who overworked and underpaid them in the first place? Whether you see it or not, this is showing a grim future for artists as a whole. Nowadays it seems like there's no reason to pursue your creativity, as you'll just be replaced and drowned out by a huge flood of AI. You can't even say this is just a hypothetical or that it will never happen, because it IS happening, and it has been for months and months on end. This isn't a potential future, this IS THE PRESENT DAY! If you have any sense of worth as an artist, or if you value the work of not only the greats in the industry, but also the unsung heroes of animation, you owe it to not only yourself but EVERY artist to not support something like this. It may all be fun and games now, but this can and will result in the loss of thousands and thousands of jobs, and there will almost certainly be no coming back from it any time soon after.
Preach it. A.I. comes at a price measured in humanity. Once the corporations realize they can save money by firing real workers, they can and they will. Only have to look at the automation in manufacturing industries and the "self-serve" kiosks everywhere.
I feel you guys have been a lot more creatively free and playful with your compositions and scenes thanks to this new work method, It will be interesting to see where you go with it and if you could blend it with your "traditional" way of making videos.
A bit of topic, but I wanted to acknowledge what an eloquent speaker Niko is. He doesn't stutter, or "um" and it's clear a lot of thought went into every word of the script. I know how hard that is, so I wanted to pay some respect to it.
It's a good move because it really made the complainers look ridiculous. The End result is not only the same, it actually delivered a more faithful "rip-off" of the chosen styles they were looking to replicate because the AI model had LESS interfering styles to use as reference.
This looks really good! I think the 2 pieces that are the most jarring afterwards are the mouth movements and the motions. Animated character movement is generally not bound by human anatomy, which gives this movie a distinct feel in how the characters move within their environments.
So... For the first one, did they use a model built on copyrighted material? Because they kinda glossed over that. If that's the case, is the video monetized? Finally... If those two are true... How did it get approved when there should be members of Corridor that are well aware that the use of copyrighted material for the creation of tools or research without permission from the owners of said copyrighted material, if used for monetary gain or market research is legally dubious and has basis in previous cases, aka "case law"... Which means, again, IF they did use copyrighted materials without permission for the training of tools used in a project that resulted in monetary gain... They could be in deep shit. And honestly... They should? But... Idk. They never really answered those questions and I'm not sure there's any monetization on that first video.
one of the things i respect corridor is doing is their continuous pursuit in new forms of digital art. most people would've folded and apologized with that backlash but they just went ahead and continued forward and instead improved and provided proper insight. I'm an artist too and have been fascinated about ai art and have never hated it just cause some people use it with malice. that technology has a lot of potential while still being a good tool for artists and corridor is one of the very few group of artists that are showing that amazing middle ground.
@@Mente_Fugaznah, artists are just scared that anyone can get better images in an instant ad for almost free compared to what they have to offer. Which is understandable, but still a stupid reason to hate something.
@@amunak_ no man.. you guess that because you think that, magically, the artists are not capable to use AI, Everyone can learn to use AI in just 1 week, the problem is not that artists don't know how to use it, But the fact that it's wrong and it doesn't feel like a tool because it kills your authenthicity, being a product of hyperconsumism and destroying the original purpose of the art of ilustrate the reality and non reality with your own vision and personality, using without consent the hard work of those artists who really want to make a live with their own vision. If the problem were the ad and stuff, artists would just useAI because of the money, and artists would make a better work, But is not okay until the models works in ethical manners. There's a lot of layers about why is wrong the actual AI ... Everyone wanna make more money making art, so why you think artists didn't switched to that? Being more easy and profitable without efforts right now? It's because the problem is more complex than that
Imo you guys are pushing the realistic use of AI and machine learning for art further than anyone else I've seen trying. It's great to see you guys iterating on the process to find something that is fair and works for everyone.
@tylerpixel anyone still mad wants to be mad about anything. They went out of their way to fix the main gripe which is artists being unpaid for their work or not involved in the process, but having an artist develope a unique style willingly and passionately completely changes the AI narrative.
Looking at different ways, processes and pipelines is the real gem here. Did everyone congratulating them on using a commissioned artist think they were immoral for the first one? It was nice to see some ones love of the D world make it back into the limelight. The internet proves again how quick people are to judgement.
You guys are really setting the standards for the rest of the industry. Stable diffusion exists now and it will never “un-exist” so we need to figure out how to live with it and use it ethically and I think this is the sweet spot. Keep it up!
By using ai to create art you're still making something original. It's really not stealing anything unless specified to and you can easily find out through reverse image searching. Ai art isn't inherently stealing art
I'm glad that you have now used AI and fed it your own art. This is the most ethical way of how to use machine learning software that uses images. The controversy stems from using copyrighted art and was used without the permission of the artist. I for one am now all in for the support of this because you've reached out to an artist and the artist created art with the knowledge and intention to have their art be used on the software.
@@OGPatriot03 It is just called "inspiration". The artist even explained it in the video that he looked at a bunch of anime and added his own spin to the style. Were you not watching the video? AI doesn't have inspiration because it just straight up copies several artworks pixel by pixel.
@@d0decadice That's literally NOT how the AI works, it doesn't have a database of pixels anywhere that it uses during the generation process. Just like the artist the only reference is the inspiration gleaned from learned principles during the training process.
I'm neutral with AI and I love how you're handling this! Making your own art was definitely the right call but I would still love to see more articulated mouths because I think being able to do those would elevate this technology even more.
Regarding the mouth animation, they actually did that on purpose! They've said in a previous video (Did We Just Change Animation Forever?) that they purposefully didn't act out the lips to make it look more cartoony
Some people seem to be confused; this version of stable diffusion is built on their OWN artworks this time around. I can't say much for photoshop or software's like that when it comes to stealing artwork, but they essentially have their own version which has NONE OF THE STOLEN ARTWORK just the code that makes it work with their OWN artwork for it. so essentially this is the best possible way for them to use it. Now in terms of artists job security on the future for animators I can't speak on that topic cause it's pretty up in the air with how that'll play out in the future, but I think overall with how corridor wants to do things this is the BEST possible thing for them to do. AI is going to still be hated but, in my opinion, they at least went down the most ethical route they could.
@@WeeklyComedianthat’s not how it works, each AI software has its own data set, depending on the software it could be using millions of stolen artwork but that’s so it can generate anything, corridors data set is just their own stuff meaning they can only generate what they own and variations of what they own nothing about any of their images has stolen property anymore
May I offer an idea: Long ago, in a remote corner of the universe... painters were realistic. Then PHOTOGRAPHY came along and put those master painters out of business. They had to reinvent themselves...impressionism.
I don't feel like this replaces hand-drawn or traditional cartoons. This reminds me of A Scanner Darkly (or even Rango), where all frames of the movie shot with live actors were replaced by hand-drawn cartoons. This just feels like an automation of that same process.
While I greatly appreciate the use of your own custom art for the fine-tuning it also needs to be acknowledged that the general stable diffusion model was trained on other peoples art without their permission in the first place.
I am really happy to see you using an artist who consented for the reference. That makes a big difference. Keep trying to be great and ethical at the same time.
So Josh got inspired by other art to create new art. So a non-artificial intelligence was trained on existing art. Why is it a problem, when an artificial intelligence is trained? Me as a musician and composer also get my inspiration from other people's work. So?
In the fairest way possible, I think a big problem people have is that an Ai can only replicate; it cannot innovate. When we replicate someone’s style, we do so imperfectly, and those imperfections become our own style. While an Ai may not be “copying” in the traditional sense, it is not adding its own style into the mix.
This for me is how you address the plagiarism aspect of AI and create a project that uses AI ethically to speed up the process. To do this process you'd still need costume designers and creators, you'd still need actors, you'd still need writers and artists to create the art style, you'd still need a studio crew to set up cameras and green screens and film the work. It just removes the stumbling block of needing an artist (or studio of artists) to spend hundreds of hours drawing thousands of individual frames per episode. I think the problems with AI ethically is that it creates a world in which you can take someone else's work and use it without their permission for something they did not intend for it to be used for. The artist of Vampire Hunter D created his style for use on Vampire Hunter D and did not intend or licence it for your use in the original Rock Paper Scissors, by hiring someone to create an art style for your show you move past that dilemma and show a way in which studios can perhaps begin to use this pipeline to create shows while still keeping the creative industry in business and hiring people. It's still not perfect and I can see a lot of ways in which big studios could abuse this new system against the actors and writers and artists, but it's not necessarily worse or better than the current abusive system and at least shows a way forward for ethical use of AI in creative projects.
I think you are onto something amazing, first congratulations to the team. As feedback I think that less flicker might not be the perfect solution, because then some flickers might become more visible, it would be similar to reducing the frame rate of a fast paced animation. Keep up the good work!
It's really cool seeing how this tool is used. I loved the first RPS and I look forward to seeing this one. Honestly though, I'm not too big a fan on the art style of this one. The Vampire Hunter D style looked so good. I understand why y'all went with your own art style. I just wish it was more on the anime side instead of the cartoon side
It’s because of new court cases they could get sued for the vampire hunter d art used. (Atleast if they tried using it for commercial reasons) What art used to teach it effects if it’s legal or not.
So is your dataset ethically sourced and only contains art from Josh? Edit: Gave the video a proper watch, let it never be said I write off things I disagree with, had to stop at 8:22. They are still using Stable Diffusion which is trained on art without the artist's consent. That is where I draw the line and will always draw the line.
This is really eye opener for many people who watched the Part 1. Because when i showed my friend he said u guys acted it out and put it in an AI tool and entered a bunch of prompt which for felt too simple for u guys to do So there are hundreds and thousands of people who think creating an art form with AI is just some prompt typed in without any hardwork. Good job bringing it up and making this video
@@GoharioFTW true true Its a new thing Not everyone is super upto date And everyone being into more short format news, getting info from there its hard to know the real truth indepth So i hope everyone will understand not all AI generated art and works are bad But also saying that there is a form of unpassionated, very lazy works lurking on youtube and other social media apps and its ez to find because u find no soul to it pretty easily
I must say, the original RPS video reinvigorated my waning love of film and encouraged me to step out of my comfort zone and embrace technologies that I had previously shyed away from, like Unreal Engine. The work you are doing is inspiring.
@@miniwhiffy3465 Have you not watched the behind the scenes work on this? The AI is a tool they're using, but this still takes a tremendous amount of work to accomplish. I am not a fan of anime, but this was one of the most impressive artistic feats I've seen done by filmmakers in a long time. The embracing and utilizing of the technology is brilliant and I can't wait to see how they keep improving it.
I find it interesting how people complain that an AI is trained on existing art styles to them mix them into something new and then we see a human artist doing exactly the same thing. Including the trial and error until the style is acceptable.
Unpopular AI opinion - Photography is at the same level as AI generated images. Think of a photo of a tree on a mountain. The photographer did not build the mountain, tree, sun for lighting or clouds for shading. In fact, the machine did most of the work in the generation of the image base on inputs the "artist" had zero in created aside from curating the the angle of view and focus/zoom/exposure settings. Even then, many photographers take many photos of the subject then curate the best looking shot later (along with photoshop). If you believe that photography gets full copyright protection then a curated AI generated content should enjoy those same protections. Personally, I that unless an artist be it photographer or videographer demonstrates the images on their screen were directed and guided by them, then their copyright to the video or images are weak and should be considered public domain. Basically paparazzi can get bent given they mainly deal with "right place/right time" which is at the same level of creativity as telling MidJourney "Make cool image".
The training on particular characters from your own art has made for a dramatic improvement! However, it's noticeable how primitive the mouth animation is in this iteration - there's very little shaping of the lips, only opening and shutting of the mouth. I wonder if you had got Josh to draw a lot more examples of character mouths saying various phonemes whether they would have come out better.
In their first video, they said they did basic puppet Mouth opening as an anime style choice.
Yeah the mouths look really bad. Can't beat traditional animation.
@@Spyro_2076 that's just the style they're going for
@@JonahNelson7yeah but the audio syncing with the lip flaps could be better. I know, baby steps but this goes to show that even human animators are still needed for details AI does not grasp in finer details.
@@reaganmonkey8 but as they mention in to the first one it has more of a rotoscoped feel, so it ended up with more lip shapes getting through
I really appreciate Corridor for addressing the rising controversy around AI with their own exploration from the position of artists themselves and finding a way to use it effectively, ethically, and while preserving their own expression and style and skill in their pieces.
people on twitter are probably still going to find reasons as to why this isn’t art and it’s actually stealing 🙄🙄🙄
@@Johnny_JDmany don't know how it works and think it just copy and paste. They aren't interested in learning only seething
@@Johnny_JDlmao
Addressing, yeah, I'd say so. But clearly, they say their stance on the whole thing is, "If you like it and an AI made it, then we think it's fine."
@@Johnny_JD@johnny_jd5337 Hey, Mr negative Nancy, it's not called Twitter anymore either.
Came to the comments to see how people would still complain after all the work they did to please them… happy to see a lot of the comments are positive!!!
The fact that this is trained on custom artwork and not artist who haven't consented puts every fear to rest. This is an incredibly powerful tool that still takes amazing skills and artistry to use, and if emotionally impactful art can be created with it then that's excellent in my book.
Exactly my feelings. Train off your own character sheets
Wholeheartedly agree! Now, Could you imagine the results if Corridor Crew had used character sheets created in 3D? 😲
@@DannyVdesignThis. Anything involving illegal scraping is still unregulated theft.
Art is free. There is no consent.
@@bedinor No.
I’d love to see this built on with additional training for lip syncing. Like adding a library of animated expressions and mouth shapes (similar to the old school stuff they made for the back end production on The Nightmare Before Christmas) and see what kind of boost we can get from that. Plus, the challenge of figuring out how to implement that in the best ways would be interesting.
There's already software to adjust mouth movement for dubbed audio or fixed lines. Not sure how well it works on animation as I've only seen it for live action, and no idea the licensing cost, but the software solution is already out there.
It was a stylistic choice. They purposefully just opened and closed mouths, not articulating the words.
I'm basically repeating what other people have said, but... They recorded the audio separate from recording the video. In the video recording, the actors did not try to speak the lines, they just made extreme mouth movements synced to the lines. The open mouths, snarls, etc. are exaggerated in anime so when they did their video pass, they tried to do the exaggerated mouth movements rather than actually talking. If the mouth movements don't sync with the audio, it's because of the ACTORS not syncing - it's not a software problem.
@@boring_ugly_dude3924 Yeah, I’m fully aware of all of that. I wasn’t referring to perfect lip syncing like some Disney animation, but to achieve a somewhat better repertoire of shapes beyond ventriloquist dummy status. All it needs is a couple variations of mouth shapes, even if it doesn’t perfectly sync to every word, but simply some variation of mouth shapes to shuffle in there at least.
@@MOYAHORROR There were a lot of shots in this episode where the characters mouth barely move if at all when they're speaking
My biggest issues with AI generated art always comes back to consent of the artist whose works are being used in the training model. Training a model off of you own art (or commissioned art in this case) is IMO the ethical way to go about it. I really appreciate that you guys put in the work to use these tools responsibly!
Artist do the same. Copying things they have seen. So don't cry just because a machine is making artists useless. Go with the time.
It's a non-issue. Artists look at other works for inspiration or reference all the time.
@@MrYTThor although it seems the machine is doing the same thing humans are doing from your perspective, it's pretty interpretable if it is or not. Our human minds aren't as straight forward as the digital neural networks that consist the learning models. I have a Computer Science degree and one of the classes specifically focused on this last year, they made sure we knew the neurons in a neural network are an extremely dumbed down version of our "analog" neurons. And not only that, they're deterministic. Whereas the jury is still out on if we're deterministic. Either way, a lot of factors, and a lot of reasons to believe either way of thinking
@@purplemossclump5505there's a difference between inspiration and straight up copying a style, it's not a non-issue and I think people are valid for having concerns about it
I think this view is going to be a double-edged sword. In the end, the only entities who are going to have the resources to be able to generate enough data to feed into these image models are the big corporations' gate keeping the jobs artists want to protect.
Then, the technology is going to be prohibitive in cost and use rights for the public.
Further empowering the largest corporations in their dominance over media.
This is the way, you really rose to the ethical challenge by supplying the model with your own work. It looks incredible gang, bravo!
What's interesting to me is that the human artist immediately looked at other art for inspiration, so he could make his version of another artists work... which is exactly what AI does. So what do we actually gain from the "ethical" route? A human doing the mimicry rather than a computer?
@@LayoutMaster In this specific case, which the Disney artist pointed out somewhat, the "ethical" made a better end product because the animation is a consistent output to a desired looked. As in most cases of production, the easy route everyone can do likely produces the most lackluster result.
@@NukeMarine Agreed. In this case, the "ethical" route produced a better result. I suppose my point is more to do with the question of why we consider this the "ethical" route in the first place, when both routes involve mimicking another person's art style. Why is it ethical for a human to do it, but unethical for a computer?
@@LayoutMasterhe wasn’t mimicking the style. He researched and used it to inform and mix with his own style. Creating something new. Whereas the ai will just make exactly what it’s given
@@LayoutMaster That's not what he said, he described that he widely studied the anime genre to see what defined it, then *brought that genre into his own style* (this is the important bit). That's very different from 1:1 copying and ruining someone else's style without understanding the artistical thought and intent behind it.
Much of my issue with AI art is the uncredited/unpaid use of source artists work (after which many suggest it is their work). In this case, an artist was paid to generate the training data for your model. That feels like the morally correct way to use AI.
That's not quite true. What they undoubtedly did is use the extremely general base AI that was trained on billions of images (most of which are photos, not drawings btw), and then finetune it with the style provided by this artist to make it consistent. The use of the artist wasn't because it makes it ethical, it's because the artist is needed to bring consistency. While it's possible to train a new AI on art deliberately supplied for training, we don't have that yet. We'd need at least tens of thousands of consenting artists, photographers, people just running around streets with cameras, to get an AI model I think most people would consider ethical. Overall, I thing they're taking steps in the right direction and making the most of what they have. I hope they have access to more ethical tools in the coming years.
@@empresslithia Ah, yeah that would make sense. A step in the right direction .. indeed I agree.
You could circumvent that by training an AI model on textual descriptions of what an artist's style looks like, only pretraining the model using images that you have the rights to use.
I take issue with the judgement that training the AI is not ethically done unless you train it on your original work. As long as it adhered to the concept and content legally seen as FAIR USE it does not matter what and whose content was used and how much. Copyrighted material is often used for inspiration, for teaching and learning on your own. If somebody writes a book about a subject I like in a style I like and then write my own book about the same thing in the same style as long I don't use the same exact words in the same sequence It would perfectly within my legal right to do so. As long as people are honest about their influence and are not trying to closely copy something it's all good. Subtle or big differences will naturally appear due to different skills, tastes, needs and environment. Also if I read 3 books that inspire me to write one book containing different aspects of those books. nobody expects me pay royalties to the other authors unless I use exact elements in the exact way, why are people arguing that AI should be different, not to mention that even if we did this with AI most authors used for training would deserve to get a fraction of a fraction of a dollar at most.
The way copyright is now used is rather perverse and unethical. It often just protects people and companies that have the money and time for lawyers and litigation and does little to protect common creators from non authorized use. Long copyright "protection" even makes very hard saving some older works for which their copyright holders can't be found.
"Your" issue. No, that's a talking point you found online. A talking point which doesn't really make any sense when you consider that human beings are also trained by looking at other people's art. Should artists be banned from visiting museums? After all, they might see some art that inspires them subconsciously. Which would lead them to create a piece of art without paying or crediting the piece that they derived it from.
The AI obviously still struggles a lot with the mouth animation, but it's a lot easier on the eye without the flickering lines. Also I'm really glad that you used the drawings of an artist who fully knew about the project and consented to it, because that's what usually makes me mad about how people use AI art. I'm still sceptic about AI art, especially for commercial use, but I know we can't put the cat back into the bag, it's a thing now and it's impossible to stop - so I think what you are doing now is a move into the right direction. I really hope we'll soon find a way to make it fair for artists.
Edit: To clarify, I don't mean "slap copyright on styles so you can sue anyone who creates something close to it". I mean "let artists keep their jobs and make consent a defining factor in training AI." Would be nice if we could find a way to make that work.
Do you pay a license to the guy who designed and made the first snare drum in order to reproduce a specific sound? What this AI is mostly doing is learning how to recreate brush strokes, its not photoshopping someone elses work onto something else. For sure if the ai produces something that looks almost exactly like someones art you may have an argument but that basically doesn't happen because the dataset is too large.
@@WARnTEA Don't forget about how the amen break is used in nearly every single genre in existence and yet the man who made it got nothing
people are really hypocritical to their own circumstances.
@@WARnTEA For digital instruments / samples, yes, you do pay for them to have them reproduce that sound for you. (as far as I know)
Also, with a lot of modern AIs, you can tell them to generate things in the style of a particular artist and they will do it. (but the effectiveness varies based on the artist)
But my personal perspective is that most of the current AIs are a product of taking massive amounts of material without having the rights to use it (let alone in a commercial setting) and using them to create a for-profit product which, to me, sounds at least questionable.
@@Sammysapphira Because you can't copyright a drum fill. Someone else came up with it before it was done by greg coleman.
@@cameron7374 Yea but you are paying the person that hit the drum with a stick one time and recorded it. You aren't actually paying the person that designed the sound and built the drum. The only reason you are paying them is because they did work to assemble the sounds in one place, and they convinced you to buy it. No one is getting sued for recording their own sample of the exact same drumhit. It'd be like an artist drawing a bird and then suing anyone that draws a bird. These things exist in the real world, they are building blocks of life, they are not art on their own they are just things. It only becomes art when humans arrange those building blocks in a specific order and add a deeper meaning to it.
I do admit I love your passion and hype for this project.
The ONLY thing that worries me so so much about this, is whenever the big studios get their hands on this, jobs for animators will be cut in half for sure.
And jobs for animators are already barely a thing anymore, let alone jobs that pays well.
BUT, maybe the introduction of this will allow traditional animation to become a rarity, thus being more sought out? I dunno.
It's a scary world.
However, if we look at this project secluded from traditional animation, I think it's a pretty cool project where you evolve techniques, push limits, and try something new.
Unfortunately usually tech can't be held back by the possibility of lost jobs-- it just doesn't happen. The tool exists now, and the industry and those within it will almost certainly need to adapt to it. At least historically, a tool hasn't been stopped from being used at the risk of jobs being lost.
@@MobiusStudiosJack Crazy to think that 2d animation in the future will not be made by actual animators apparently.
As much as I agree with you, there is no denying that AI will eventually become the norm.
Right now there's a lot of very strong opposition towards it, and I understand it very much as an art student. However, history has shown many times that technology ALWAYS triumphs.
When digital art appeared, the traditional artists who used pen and papers also opposed it. Now Digital art is the norm. AI Art is a similar debate, IF we disregard the ethical aspect regarding the AI sampling on other artists without their consent (which is not a neglectable point).
On a practical standpoint, we will have no choice but adapt. Marc Brunet said he would welcome it as a tool, as long as it is not used as the finality, and I agree on with him. Use the AI to help yourself, but you still gotta use your artistic skills to polish the final result.
On a moral standpoint, I think our only hope is to have legal regulations on crediting sources / sampling consenting artists.
Look at it like this, if a company purely relies on these kind of programs without real talent to make it work just for the movie's story being bad in the end, then they weren't a good company to begin with. Meanwhile you have people like those in Corridor who are actually managing to pioneer this technology as the tool it was meant to be. The fact that they are willing to learn from the previous in order to improve goes to show that with enough drive and creativity, anyone can finally be able to see their ideas become reality without the need for big corporations who only ever cared about the bottom line. Now they just need to figure out the weird inconsistencies like with the eyes and etc. and there will be no telling where this will go.
@@Kenddamus The analogy that AI art is anything close to traditional artist opposing digital artist is stupid.
sure, was there some pushback, but ultimately the basic need for skilled labor was still there. a digital Artist can still pull of great artwork on Paper, maybe the will struggle a little bit more, because they have to draw without all the quality of life tools digital art gave them, but AI art is not a quality of life tool right now.
it literally replaces artists. that's it.
And if the future of humanity is going to be that art and culture is AI controlled, then we as a species have no future.
I don't understand how anime rock paper scissors could be controversial.
- You didn't violate any copyrights (as much as I disagree with copyright, different issue, another day) as there is no semblance of any of the original source.
- As far as I can tell, all the frames are original, so how could the complaint of this "looking like" something else be made?
- No, style can't be used as an excuse for controversy, no one person owns claymation nor anime nor cal art style, etc. If this were the case, there wouldn't be as much media as there is today.
- If this is about the amount of human labor put into the project, that shouldn't be it either, buggy whip makers haven't been making buggy whips in the same number they have been in over a century and look at where we're at today, more jobs today than people back then.
As a tool in art, I don't see why people are able to complain about this with a straight face.
As a software dev who sees github co-pilot's output on occasion, I find it laughable that we consider ourselves replaced.
I think a big part of what this is missing is actual animation. Animation isn’t just trying to exactly emulate real world movement, it is being stylistic and specific with the animation choices it make, like what parts to include, what to tone down, what to exaggerate, and what to do that just looks better or more appealing even if it’s less realistic, so just having anime visuals on top of mostly real world motion will always look sort of strange, this is not really an anime substitute in my mind, just complex and very automated rotoscoping. It’s like wirh 60fps animation interpolation (which noodle made a great video about). It’s often times not really an improvement on the base animation as it does what is logical and direct but not really what is best artistically or purpose driven. I think though that that doesn’t mean that this side of things can’t be done or worked on with ai ever but it’s just not really now and it is id say a pretty large missing piece.
They don't call it "The 12 Principles of Animation" for nothing. Having no knowledge of how animation actually works will end with bad results.
Its almost like we would need an Ai specifically trained on animation, ie the actual movement of animation, not just the stills. I wonder if that could be on the horizon? Every principle of animation can be interpreted in different ways and varying levels of exaggeration, the ai needs to learn about different ways of doing so.
We are still pretty early on in the realm of visual generative ai. Don’t count on that remaining permanent. My work is in large language models but this is nearly as interesting, and I’m not the only one who wants to push it forward more and more.
@@gmcubedexcept art is intent. Animation in particular is driven so much by the artists vision and intent, and you won’t get that with AI tools. Auto-generative AI tools will never know when to correctly use the 12 principles of animation, because when you use them is both heavily context dependent and intent driven.
@@gadget2622Less than 10 years later you’ll laugh at your own comment
I will never get tired of watching this channel.
Yep
I never get tired of your mom
@@wulfrachebe careful brave warrior for the night is dark and full of terrors.
wait till you hear about their website
Facts.
Thank you for using commissioned artist work that they gave their consent for to be used in this project. This is an ethical use of AI and should be the standard.
Except someone will just hire an army of third world artists at 50c/week to make art in the style of well-known artists and then we're back in the same place.
@@chaos.corner look it's a start, i understand that there is a whole system of using studios with underpaid workers, and it's definitely something that needs to be addressed. For now I will give these guys a bit of credit for doing something right. We'll then work on unionizing animators around the world.
DId their artists get the consent of the Vampire D animators to study their works?
@@CaliMeatWagon I think they kinda covered that in their explainer video. I may have to double check it now again to see how they were able to work around that since they filmed themselves before rendering those shots into animated frames as well as using stock photos of castle interiors as opposed to using a prompt to magically yield out a frame.
@@CaliMeatWagon You are aware that every artist ever looks at other pieces of art for inspiration right? No one is just sitting in a bubble thinking of every little thing from scratch. That's literally why trends exist.
It looks great, one of the things I notice is the eyes still distort and break. We need a way to generate the eyes an have it stick to the same shape. One of the idea I had was to make clean versions of all the eyes from each angle, and have a seperate AI pass that just does eyes. Hopefully it can track them better with most angles and pupil directions. The other thought was to have the eyes rendered with a 3D AI, that actually makes the physical eyes and lighting based off the drawings, then applies them to the characters appropriately. Gen2 is currently doing some nutty stuff with video shots. Especially with stable models from frame to frame.
00:40 "Why? Because FUCK EM that's why!"
I do enjoy the previous art style more because I think it looks more gritty and visceral and I love that rotoscope look, but I appreciate the reasons why you changed it
The flickering on the first one sold it more for me. More of a human touch. New looks like a cartoon style on a 3d animation.
@@TheMonthlyJack or an anime style on a cartoon
Agreed, it felt more "real" vs a cartoon trying to be an anime.
All previous "flickering" remembers me experimental European animation of years ago...
I think if they had a larger data set to train their model on, the video could have come out better. The old Anime Rock, Paper, Scissors had potentially thousands of frames to learn from. I can't imagine the artist they worked with drew more than a few dozen images.
kinda affirms a thought I had about why artist protests won't stop the advance digital AI art: They can just hire a human to imitate/synthesize art styles and then train on that.
But that suggest that they still need them. As they still are a lot of imperfections that needs to be corrected by hand.
@@Deliveredmean42 : True, but this way you can get a famous artist's style without having to hire the *expensive* famous artist. I'm not saying it'll eliminate the need for artists, but that it's a workaround to recreate particular artists' styles.
@@Deliveredmean42 Until the library is built. Until the models get better. Then the ai models take over providing the library. And then corporations buy up competing ai models. And then big media company's get a practical monopoly and 99% of all payed artist gets kicked out of the industry because they cant compeat in their old product and they dont have access to the ai tools to compeat in the new ones. People are that are screaming about copyright with ai are only thinking of next week and not the next 50 years.
Here's my 2 cents.
I never really had much of an issue with the artistic integrity or value of using AI to create art. I think AI art can be a very interesting and distinct artform when it's not just trying to imitate existing ones.
What I *did* have an issue with is the lack of consent involved with the training of latent diffusion models. It'd be one thing to train a model on copyright free images, but it's another thing altogether to just rip stuff from across the internet. The latter of which, is unfortunately precisely what Stable Diffusion has done. This isn't really an uncommon practice in AI, but when it comes to art, something with so much skill invested and so much of yourself put into it, I think it's work treading with more care, especially when the tech is so blatantly capable of devaluing the work of the artists that make that training data. StabilityAI got the memo with Dance Diffusion but apparently not with stable diffusion.
I very much appreciate the effort of working with an artist who has agreed to have their art used, that's certainly a step in the right direction (and it's great that it actually helped produce a more stable result), but this doesn't change the fact that the underlying models still are built upon what I would consider to be an unethical form of data collection/use. And frustratingly, I don't think they need to be. (there are other issues I have with AI art, but I don't think they apply to Corridor's use case, so i digress)
I can't help but find it upsetting (and if it were me personally doing it - more than a bit gross) to be promoting and using such technologies before the underlying issues at their foundation are resolved. The more we build upon the foundation the more difficult it becomes to change it, to make it better, and we *can* make it better.
I've been following Corridor Digital consistently since I was a kid, always looked up to them, but this has unfortunately changed since the first Anime RPS. How it was handled was especially upsetting to see, especially after having admired how informed Jake's video on the technology was a while earlier.
Idk. Hope Corridor really does figure out a way to do this which doesn't make me feel so icky abt it some day.
Perhaps take a bit of time to look at any upcoming fringe versions of the tech more committed to ethical development, or consider alternatives like just working with artists on tech like EbSynth, dunno.
I myself am an artist, but also working on research in deep learning, so if I perhaps seem too invested in the topic, it's because I am lmao. The general public is familiar with being not all that invested in ethical issues when they don't impact their lives directly, and I won't pretend to be a particular exception. But I think we ought to care about some of them at least. I care about art, I care about artists, and I care about developing AI safely and ethically. This ultimately doesn't effect me directly all that much, I don't post my art publicly online and my career path is more directed towards tech than art, but I reckon it'd be going against my values to not care about this, and would make me a bystander in building a world I don't want to see. idk, just food for thought ig.
(as a tangent/aside, I regardless do still hope this AI version isn't a flat-out replacement for the anime-style videos. Corridor's previous more vfx-based anime-style videos had a lot of charm which tbh I think is actually held back by the AI image stuff a lot of the time.)
In the next one there better be a "got your nose" or "you've got something on your shirt." Great work guys!
There's still more transformative work going into this than 90% of react channels.
ITS NOT ABOUT THE TRANSFORMATION OR EFFORT ITS THE COMMUNITY OF THE LIVE ENVIRONMENTAL THAT CREATES A DIFFERENT EXPERIENCE. SAME AS SPORTS BARS. DONT ROB THE VIEWER OF THIS NEW FORMAT FOR THEIR FAV STREMAERS PUSH TO GET A STREAMLINED SYSTEM IN PLACE ON THE PLATFORM SO STREAMERS AND REACT CHANNELS CAN SHARE REVENUE OR PURCHASE REACT LICENSES
@@515JUKECalm down, we’re just talking
@@515JUKE You're 100% right. React content definitely exists in a nebulous place in terms of fair use and IP law but the fact of the matter is that there is a large audience for it, and the platform owners and rights holders are failing to capitalize on it.
They need to develop methods for embedding and syncing other videos with view count and ad revenue sharing. And there should be communal viewing options incorporated into streaming platforms (IE "Netflix Couch" or "Twitch Theater") where the IP holders also get a cut of the revenue generated through subs, virtual ticket sales, cheering, or whatever other monetization methods they can figure out. And naturally the creators should get a cut of that too via royalties.
@@SuzakuX I love you
@@lonelylama5222 I’m sorry I was so loud I haven’t had much sleep I do apologize and hope you have a good days your joke was funny
Flickering/consistency has been a huge problem for me. Fine tuning models that map one person to one character is a really interesting solution and clearly it's made a huge difference. This entire workflow is just on another level. Well done demonstrating the best this technology has to offer.
creating your own dataset also fixes the whole stealing other people's art issue.
What's weird is because the animation is so much better now, you can tell the wigs and flappy mouth movements which were better hidden in the first.
@@userunfriendly9304no because if you make your own art you took inspiration from other art so youre still stealing
@@technus147 bro taking inspiration isnt stealing. same thing with references. ai takes ppls art and puts it in their dataset 90% of the time without the artists knowledge or consent
@@technus147not really since humans are not comparable to ai and ai is not comparable to humans the way you people compare it like “ oh stalking others and taking their photos without their knowledge is fine you look at those same people through your screen without permission or consent why should we “
I would be fascinated to see a breakdown of how long it takes to do this, versus a studio doing a more traditional animation technique. Obviously the Crew is still feeling out the work flow and everything, but the time comparison would be interesting i think.
I think it would be largely based around two large parts: 1) time to train stable diffusion with source material, and 2) time to export video shots into individual frames, run them through stable diffusion and recompile them
@@Spiker985Studiostraining speed can be solved by giving it more power. The slowest part is creating a dataset, describing every single image and everything contained in it. And then doing it again when training goes wrong.
I trained some loras on datasets of two dozen images and even that was pretty tedious and slow process
But., everytime there are a new tools to make the workflow faster,
all the big studios will use it as a license to make a half assed product,
new workflow mean that they could McDonalds-ify the product,
make a mid story, mid vfx, and assign a cheap, clueless and mid people to create the product
.
fast production mean fast money, therefore it is not about the Product anymore,
it's about making a generic product that addressing the current trends
because they saw that technology as a way to replace expensive people in their company
we already saw this phenomenon time and time again, in the movie industry, in the game industry.
there are a massive lost in the knowledge base, because the guy who create the masterpiece was too expensive to hire,
and the knowledge wasn't getting passed down to the next generation.
therefore we saw a massive degradation of quality in the past 10 years
despite the promise of giving you the power to make a jawdropping scene and great storytelling easier,
giving people the liberty of creation.
we saw less and less Masterpiece each and every year
@@jensenraylight8011 the big studios were doing exactly this for years, well, effects were good but story usually worse then mediocre to compensate for this:)
From what I gathered the AI in the 1st animation grabbed info from VHD and that style often uses squint eyes, closed eyes then animating them wide open in times of shock surprise etc, this is a key style while the AI couldn't allocate where these changes should accour, so even tho you could take care of the flicker fixing the face is a nightmare.
So if you are going to train something I guess a code needs to exist that covers opening and closing the eyes, because it pulls it all over the place at random and if you spend time fixing it frame by frame might as well make everything from scratch since you will get more clean results lightyears better in comparison and original style too.
The time spent training, editing rendering versus the time making everything from scratch by an capable artist burns less money and less resources.
I honestly don't want to know what these people pay their electricity bills on all these badly optimized amateur programs and AI scripts ...
A year ago I was playing around with learning AI things and not only it nearly fry my SSD, RAM, GPU and CPU basically whole PC was burning, my electricity bill was 100 times more then during winter time heating ... I had to stop burning time and money also had to repair my PC's cooling asap.
With other words I burned years worth of winter heating in just few weeks time.
So yeah no thanks ... I rather do nothing or crowdfund and pay actual artists if I want something done properly ... F--k AI and these amateur programs too ...
If a team of humans watched D and made a new movie faithful to that art style, the community would love it and praise it for making more of what they love. If you do the same with a machine, they lose it regardless of the insane amount of work you put in. This is the final equalizer for 2D art. The barriers of skill, time, manpower and budget are stripped away, so anyone with creativity can now get their vision onto the screen despite the real life limitations they face. If you're an artist who gets that unique satisfaction out of the process of your work, you are unaffected. If you draw for profit, you now know how the auto worker felt when Japan introduced the robotic arm.
Controversial drop.
When you create your own art stylebook specific to the project and train the AI to do that specifically for each character, I'd day there's no longer any problem at all. It's now performance capture with AI-assisted animation. 100% your own product.
You will be surprised how people will still find something negative to say
People will still find problems with it, I guarantee it. Because "real art" is more than just the style being your own, its the execution too. The moment you give up control to an AI on that level, it opens things up to a lot of backlash.
The stylebook inside the artists mind is still Vampire Hunter D, so in effect its the same thing but with a human middleman.
@@gmcubed Artists do not own their style, only their works.
It's not 100% their own. It's still using a base model like SD 1.4 or 2.1, with the anime style Josh made layered over it. If they weren't using the base models, there would be no way to make this look coherent without an impossible amount of art. For reference, LAION-5B, the dataset which Stable Diffusion's base models were trained on, has over 5 billion reference images. That being said, it's still a big improvement from their first project in terms of ethics, and Stable Diffusion is probably one of the more ethical AI models of any kind purely by virtue of being open source. It's a cool project and more interesting than just using AI to replace people's jobs.
Training the AI on art specifically made for this project is genius. I hope this will become the new standard for using AI. Well done guys, really great work!
I wouldn't call it genious, its a great Idea but they arent the ones who came up with it, it was actually the main concern with people who are anti AI useage and we are actually advocating that specific artists get paid and have their art used with their premission in ai models, instead of the models that art scraping millions of artworks from google images without artists being compensated or even without their permission.
@@VJArt_ They still used stable diffusion and just trained a lora on their own stuff. Their process still uses SD because they know they couldnt make this work without all the stolen content from SD.
@@VJArt_ True - I mean it was a reaction to the backlash they faced for their first video. I still think it's great that they share their workflow for a more ethical way of using AI, considering they have a huge audience.
@@2darkiBut that's a whole different conversation. They are not replicating any copyrighted style on this creation, which was the point of controversy from the last video.
the "AI" has already got thousands of pieces not made by them
You guys nailed the only Ethical way of using this technology. Hire an artist and train the model based on the artworks he created for that specific project, with his consent. BRAVO
It amuses me that you would say that, because the artist explicitly states that he trained himself on the artwork of several other people, none of whom were asked permission.
@@duckshallrule6937thats different though, as when a human processes that artwork into their own its transformative, as a human can not have a perfect memory, cannot replicate everything it sees perfectly, but an ai artwork is directly derivative of the information it is fed, it processes everything it sees perfectly, it only applies transformations using the information it has been given, and it itself can not be creative in any way.
@@jc_art_ That's just speciesist, it learns the same way you do.
@@jc_art_ "it itself can not be creative in any way."
You can truly make this exact same philosophical argument for humans though. If the only difference is that humans are imperfect then you can just tweak the AI to make "mistakes" too.
@@dustyfoxboy442 Machine learning is on every single level different then human learning is, 1st an foremost in the number of neurons, humans have enough to be fully sentient and make decisions, and decisions in art is what creativity is, just as studying referance and other artwork, its in the decisions of what you study and what you get inspired by. Second of all, the way those simulated neurons are made is also different from biological neuron connections, just as the name implies, its just a simulation, and it can never replicate a biological brain, its can be its own thing, but a machine learning something is nothing like a human learning something.
Do some research before you start spewing stuff you dont know about.
Corridor showing what can be done when you put effort into the process of making stuff with AI instead of “this is the future of animation, fire your VFX artists”. Truly good stuff
I don’t think it’s necessarily pertinent to call it “AI art” or “AI animation” since the only parts AI is involved in is applying a glorified filter onto live-action or pre-modeled images.
Still, a certain level of humility and consciousness should be held vis-à-vis the media this is meant to replicate, as studios will use any and all tools available to cut costs and workers. You may not personally plan to replace regular 2D-animation or make the lives of 2D animators harder, but presenting stylized filters as a potential substitute for the art form opens a Pandora’s Box that you must be aware of.
An ungodly amount of artistry and effort goes into your work, so it may be very misguided to emphasize the importance of the only part in the process that isn’t artistic and requires little creativity: applying the filter. Respect yourselves, your own craft and talent!
Next thing to "fix" are your King's eyes. They keep crossing every time he moves. Great job on everything else.
4:35 because controversy= money to you
the eyes still breaks the magic for me but it's quite cool to see how far did you come from rps 1, great job everybody
Yeah indeed. I would opt to redo them manually, at least for the pupils.
@@pendaco Well they are still just playing around with the tech. The aim is to improve the process while making a cool interesting video They are not trying to make a perfect anime product that actually compeat in the normal way. And doing things like hand animating eyes would be a large sink of time, energy and money when that kind of perfection is not the goal. The selling point of the videos for now is the ai novelty not the art alone.
The eyes, the shadows, the teeth are still really wonky.... the general face.......... Even the cracks around the hands feels like it's lacking impact and...........
Yeah, the more I think about it, this just doesn't look great at all. I'm not sure we'll ever get this up to the standard of actual animation, but it's better than nothing, I guess. I just hope big corporations don't see this and just use something that looks this rough to replace actual artists.
Now that all your processes are in place, I'd love to see you take a 30 second random scene and then in real time (or as reasonable as possible for UA-cam) show us how quickly it can be converted into a finished render with full audio and effects
Well you can just think about it. Its not gonna be fast. Days of work perhaps. The final editing and pre planing more so then the filming. But still far faster then animation. Just look at the release schedule of any bad animation studio and compare. And like with everything ai this is the worst the quality and tools will ever be. THey will only improve making the workflow faster.
@@RasakBlood Disagree. This level of animation is not that hard, assuming you have the initial artwork, which they do. This episode took 6 artists 4 months? That is a good time, but not unthinkable. If you approached it by actually animating the artworks by hand, which is a normal thing in cheap animation projects, you could probably do it in 6 months with 6 artists, but the quality would be slightly better, mouth animations especially, and the perspective consistencies, which are all over the place in their animation. We've done 30 second animations like this in a week for commercial clients - I'm not saying they were good. In fact they were complete crap IMHO. But that's what a client gets if he wants a 30 second animation in a week. And their project absolutely reminded me of these animations.
Well, 4 months is the time between release of both shorts and they have been doing a lot more work beside this short. I think that they coild achieve this within a month if they were fully dedicated just to this video.
Theres an hour long making of on their website. Very cool technically but unless you want to learn SD its like watching paint dry.
The main thing I don’t like about generative AI is that they are trained on stolen artwork from actual artists. If you create everything that goes into it for the purpose of training an AI model, then that sounds fine.
The people who dogpiled on the last video were just virtue signalling.
The complaints about AI using art without compensation are silly. If you make art to put art into the world then you should be glad you're still accomplishing your goal while also contributing to the future. If you're making art to make money, you got paid when you sold it.
Maybe you believe computers can't create true art, then your personal contributions will always be valuable, and you've got nothing to worry about.
Or you're worried that computers can truly create art better than any human and you will be replaced. Well welcome to humanity, there's a lot of that going around right now.
all you people thinking this would "replace" people's jobs , this is just a form of technology. stuff like this have happened many many times in history. machines , computers , etc. when cars came along , people who breed horses thought "I'm gonna lose my job!" aaaand that wasn;t the case. It's the same. "History repeats Itself"
My thoughts exactly
I appreciate that they’re not copying another art style through ai, but at the same time I do think the art style outputted from the ai looked much better, more detailed and serious, than their own new style, even if the flickering itself was improved
I agree with you, but Vampire Hunter D´s model is so much more complex. They could hire a veteran anime artist to create a dataset for them, but the cost would be much higher, and they´re taking the first baby steps in a new workflow. I assure you we´re going to see such datasets from the big companies like Bandai very soon.
To me this is the crux of why compensating the creators of the source data is important. With full respect to the artist who created their new style the work in vampire hunter D is stellar and the result of professionals at the top of their field. The ai output is only as good as the source data you give it so it does have value. If people want to train models based off the work of an artist that artist should be compensated for it, which then allows them to create more art which can then contribute to other models. There's a real possibility for everyone to benefit here.
I find it especially intriguing that with a solid pipeline, projects certainly seem so achievable that were never previously. It really let's your creativity go wild with the possibilities.
It could even be as draft or a strong pitch for a much deeper, more expensive, and involved full feature project.
These behind the scenes videos are just as exciting to me as the show. The passion behind the project is palpable.
If you remove the word AI from the equation, you guys just created your own stylized filter. If people are butthurt over that, then we have a different problem. This is incredible work you guys have done. This is exactly how AI should be used to create art.
It is a neat filter, and if they make the reference material for it then awesome have your fun....
The problem is they are not acting like it's a filter or treating it like one
@@kingdomkey63Filter? I thought this was an "anime" 😂
@@kingdomkey63 They're playing it safe. Majority of people are still not sure what to do with this "AI art" which is understandable. Its new. Corridor can't exactly call it a "filter" because thats not exactly it either. Its a lot more complicated than that. But to your average joe, it is basically an advanced filter.
If they called it something along the lines of a "filter" directly then people would assume Corridor is not taking responsibility of the fact that they are indeed using AI to avoid controversy. It would've been worse actually lol
What they did here is smart and how it should be done. AI art exists if done right. For example, generating an image from Dall-E and manipulating it a little and calling it art is not art. Training your own AI with your own reference material, generating an output and then manipulating it is art because now you've put your own spin to it.
Idk, thats just my opinion as a creator and a viewer.
@Grabster no I would say that filter is actually the perfect word for it. If you can't accept that this is an augmented reality filter then I really don't know what to tell you, other then to hand you a dictionary.
Semantics. Actual semantics. And like normal most dont care.
The main controversial thing isn't the tool itself makin it easy, but the ethics of the training data which was taken from. Which i give you props to commission an artist!
The problem the haters have is that by that logic no human artist would ever be allowed to learn by studying and learning to mimic the style of Bloodlust, and yet that’s exactly how artists learn. Copyright gives artists ownership of their expression of a style, but not ownership of the style itself.
@@truejimThere is a major difference between a human learning from others and AI simply copying what it has been fed. Current AI does not actually comprehend what it is "learning" and simply regurgitates, while a human actually does comprehend what they learn and is able to create something new that isn't just copying.
AI in its current form is NOT actual intelligence, and can not be directly compared to human learning as they are fundamentally extremely different. The only people suggesting otherwise simply do not understand the technology or are purposely trying to misrepresent it.
Therefore in its current form yes the main controversy with AI art is that it is trained on source material without permission, because the end product is simply reproduction and only imitates being transformative by mixing a ton of different sources together but it is still just copying what it is trained on. So if permission is granted to train the AI on source material that is fine, but if permission was not granted it literally copying someone's work without permission and without credit.
An important note to end on is that art style is not protected by copyright but the final product is. Therefore training an AI to copy that final product by feeding it that final product IS copyright infringement. It's just too new of a technology for court cases about it to make it through judicial systems yet, but that doesn't mean it is legal or properly transformative.
@@hazelberry You’re making my case for me. You can’t sue a computer, you can only sue people(s). There’s zero legal precedent that says people can’t use tools (even computers) to study prior art. If literally nothing is being copied, then copyright doesn’t even apply.
@@hazelberry It literally is learning and it is intelligence, regardless of what you want to believe. It’s not copying and it’s legal until determined otherwise, which is very much up in the air.
@@truejim You clearly are neither an artist nor a developer because you are making a wild stretch to connect what actual artists do when learning vs what an AI does. It is fundamentally different.
I personally LOVE the new artstyle, it's such a great blend between anime and cartoon, which seems perfect when using tools that struggle with clarity and consistency :D
10:51 That's what I think is the problem right there.
The issue with Ai isn't artists like you making new and exciting content for the sake of creativity, I do fully agree you're on to something entirely different, and in fact is something quite intriguing and awesome. However, sadly, artists like you aren't the ones who lead the world of entertainment. The concern I personally have (and the one I've heard other animators like me have) is that companies and studios will only see the last part of your statement "Here we are, with a full episode of an anime, made with 6 people over 4 months". That's what's scary, that because of a new form of Vfx, the side of animation is in danger of being seen as something worthless by corporations, all while animators are already being neglected and exploited as I'm sure you know.
So I think that's the fear- that this fantastic tool is going to be exploited by corporations by kicking out animators and artists in exchange of something more efficient. Leaving artists with almost nothing to fight back- no matter what we do AI is always going to be faster than traditionally drawing and more efficient- and that's all that corporations see, and that's all that gets out to the world, making it all that people will see devaluing artists even more so than they already are- so If I were to assume why the controversy exists, it's just that... we are scared that something wonderful you are making is going to be used to harm other artists, is nothing against you or Ai really, it's fear of loosing our livelihood. It isn't just that AI is stealing art and what a machine is doing is being referred to as "art" (which is demeaning and terrible because it carries no emotion or creativity, it simply blindly replicates what others do and blends it together in a big slush of an image)
Right now they are writers on strike, and what Disney is doing is trying to replace them with Ai. You, with a small team (relative to Disney) was able to improve Ai this much, imagine what a corporation with ridiculous amounts of money can do with Ai- So it isn't even an irrational fear anymore... I'm happy that you've used an artist to create your style from a new, and I'm happy you're excited about the progression of Vfx with the use of Ai- but this doesn't really make it any better, if anything you're kind of proving the point that if anything the art world is becoming even more competitive at best and becoming truly horrific at worst, so while this isn't anything to do with corridor, it is a situation which may bring consequences might be a lot bigger and dangerous than one might imagine. So are you responsible for any of the things that might happen? Not really, but by normalizing the use of Ai it's discrediting what artists like me or other animators can do.
So if anything, I'd just ask that anyone using Ai can really understand that the commercial use and normalization of it hurts artists, not because of you, but because of how corporations and the public might react to something like this.
That's what I'd like to say, personal opinion but based on what I've seen with the art world I'm in... I hope it helps shed some light in what I think is the other problem you didn't get to address here. But again, thanks for getting an artist to make an art style, if the entertainment world wasn't so focused on making money and cared about art, that would solve the issue, but we don't live in that wonderful imaginary world were artists are valued...
But that's what I think, regardless thank you for the video addressing the artists, I know it isn't for me but as an artist I appreciate the effort :) it was a good progress to witness and it was great to hear your point of view
The improvement on the sequel is absolutely wild.
Keep pushing the boundaries crew 💪
I have a degree in animation and think it's great what you created! It's a new tool in the artists playbook not a replacement of art.
I think the problem isn't ai, it's more about capital owners firing workers to get more profit. Human greed always ends up messing up new technology (we had the same issue decades and even centuries ago with new technology.)
@@no_less03 You do understand that everything you interact with daily is new technology? And that almost no jobs that existed 150 years ago exist today? Redundant jobs getting eliminated benefits us all.
@@no_less03 I'm not sure if used this way it'll make artists lose their jobs. Rather, it'll change artists' jobs. Instead of animating everything, they can now draw a stylebook for the AI to use and still get paid for it. Plus they'll likely need to touch up/redraw some bits and pieces even after the AI. And of course, you can't do performance capture without hiring actors either. At the end of the day, I don't think it'll spell the end of traditional animation either. Computer animation didn't kill drawings when it came out in the 90's, it simply became another tool for artists to use.
@@Ruzzky_Bly4tonly when the people working those jobs are either reallocated in the workforce and/or taken care of somehow until they can learn a new trade, but guess what, companies don't want more expenses, they want more profits, so making jobs obsolete without an attentive eye will just lead to more struggling people
@@luigiff3431 You just described the reason why capitalism will have to change or be replaced. Stopping all progress because it can replace humans is not a good solution. But if the current systems won't change, people will be unemployed, and the 1% contorlling the machines will have all the power.
you know what? god bless yall and i love seeing people work on something they're passionate about. you're hardworking and you'll be the one to break through this new gen of production and animation
I really like the artstyle Josh came up with. very contrast, vibrant, and clear.
Still they use stable difussion that its
Illegal in everyway
Still not half as good as Bloodlust tho... But I prefer it this way
@@oliverfalco7060 the guy said he'd never made anime before, so considering that I think he did an excellent job.
Thanks for making this! it's a great demonstration of how these tools can be used creatively, and the best part is that you still needed experts for it to work.
Im SO glad you saw the point about your OWN art style. Effectively training on your own art with the new updated to the method is crazy for true transformative works
They just got an artist to redo the Vampire Hunter D style, how is that different from a diffusion model doing it? It's just an extra step but instead of via silicon its via carbon.
@@tylerpixel difference being you cant take it to court as new intellectual property
@@tylerpixel"I took inspiration from vampire hunter d, AND ALSO GUNDAM, COWBOY BEBOP AND BLENDED IT WITH MY OWN CARTOON STYLE"
Skip that part did you? But for sake of argument, say they DID pay an artist to just "redo" the style. Its still better then taking art through ai, as, 1. They are paying the artist. 2. there is no full seperation of style from artist, even having a full team making a show in one style its possible to pick what animators drew what scene if you're familiar enough with their hallmarks (sakuga fans can attest to such) pure ai art simply rips it from them, generalizes and spits it back out. Which is why using an artist helped it look better (shock horror artists are good at art) Ai and art can coexist, done ethically. This video is case and point which is why its way less controversial than the predatory methods of the rising corperations
@@gojicandle8188 Yeah so they added more "prompts" and mixed it, when I use Midjourney I merge different art styles together, Robert McCall, Simon Stalenhag, Alex Ross and others. I then add to it in Photoshop then reupload it into Midjourney to generate more variations and ideas. What I am trying to get across is the other sides argument that both human creativity and digital creativity are the same thing, latent diffusion is the same process that is happening in a human mind.
@@gojicandle8188 The AI model did all of that but also considered 1000 OTHER styles as well, it's less of a copy if anything. Anyway a lot of artists are Narcissists apparently, who don't admit that their work is in one way shape or form a derivative of the works of art that inspired them.
PS: The only reason it looked better with this workflow is because they made their own AI model with LESS data/inspiration to draw from, producing a more reliable generation between frames. - It's literally MORE of a copy cat this way and some artists still had their style ripped off.
With how the render turned out, I feel like this is underachieving what mocap with cell-shaded 3D models would achieve.
It's nice that you're only using art from your on board artist for training, but the "style change" for backgrounds is still an iffy portion. Again, a 3D environment would've solved that to begin with.
I loved the jittery art style because it literally looked like just that, a style.
It reminded me of 'a scanner darkly'.
And since the last ep I've been curious what a scanner darkly could have looked like had it used this Ai technique instead of the hand drawn stuff.
Could be an experiment for a future project maybe? Scanner darkly with todays technology
I think what makes this one more ethical then the last is that y’all decided to train on an art style from an artist who gave permission for y’all to do so and was involved with the project as well. That is one of the main reasons artists like myself don’t like ai “art” and the way the tools are being used at the moment is because people’s art get put into these tools without consent or proper credit, then people try to pass off these collages of stolen art as their own. There is still a long way to go in terms of regulations of these tools to protect people from being exploited by them. And with y’all deciding to take in the perspectives of those who are being exploited by these tools and look to improve upon them is something that is needed.
Great. So using AI is not a problem. It's misuse is...just like any tool.
You could not have said it better!! The choice to use a consenting artist is a hugely good thing because now it’s not really stealing from another group. Plus with the extra work that corridor did besides the art the AI would be trained on, it’s really truly it’s own thing now in my opinion.
Edit: I love the deep discussion happening in this comment’s replies
As an outsider who’s not an artist or anything, here’s my question. What’s the difference between what this hired artist did, using reference from multiple different styles of anime without the creators permission to make his own style, versus an AI, who trains from different images of anime without the creators permission to output its own interpretation? That’s been my biggest question about this AI debate. Is there something I’m missing that can help me see your point of view?
When the internet exist, there were tons of criticsm circulating everywhere, but we are forced to cope with it and now it’s a norm.
I dont see any difference with this Ai things. At first there is going to be many criticsm, eventually people just accept the way it is and move on.
Instead of listing disadvantage of the tools. Maybe we should look at the brighter side and think how we can use this tool on the positive side.
This is it. The general AI tools and their coders take art and then ask later versus asking first and letting the artist be involved. One stregnth of the general tools is someone like me writing a book, but I'm very visual. So I can use them to get an idea for the look of my characters, but only for internal, initial use. If I decide to publish, I will get an artist to make the covers and redo the characters in a different way for future projects. I'm very happy Corridor decided to revisit RPS using an artist and their own style.
this is the thing with AI. it's an incredible tool and you prove it here. You used it AS A TOOL! WITH AN AIM!.... the problem is that some people feed images most people would found beautiful to the algorithm, insert a prompt and call themselves artists ... which is nonsense!
The difference being they are experienced professionals trying to actuly make something, not cheat something.
How did they use it as a tool tho? It wasn't the actual artist who used it to help him. it wasn't an animator who used it. They paid an artist for some concept art basically. took that, ran it through the AI and managed to now kick out any specialized people.
They literally only had to hire one artist to make this all. How many animators and artist would have worked on a project this scale and would have made a living off of it?
And they are a small studio that don't even specialize in actual 2D animation. Is even one of them a solid artist? Probably not.
All I see is a dozens lost job opportunities for real artist who actually learned to draw and animate the hard way.
2D is probably going to be dead. This technology will only improve and become better and better.
And at some point the entire entertainment industry is going to be majority AI controlled. We already have AIs for generating texts. We have AI for generating art. Only a few more steps and soon humanity will be relegated to labor jobs. Human creativity will basically be reduced. And those labor jobs at some point will also replaced with machinary and AI as well.
"What's the point of trying to be creative, when an AI is going to make sure I will never be able to use it to make a living?"
Those kind of people won't be around when the fad fades. And those that know what they're doing do actually respect the artists that made their work possible, despite the differences regarding their workflow.
The jerks who use AI art tools to feed their own ego would usually be quietly shunned by the community, since it did give them a bad name disrespects both sides.
Its art. It's literally a r t. Or else you can't say some paint splashed on a canvas is art either. Because that's just some people throwing beautiful paint on a canvas and calling themselves artists.
@@DatAsianGuylook dude, I know you think the worlds gonna fucking end when AI creates “Imbiamba Jones 10: The Lost Whip”, but be fucking realistic, AI is only gonna affect Hollywood and few animation studios around the world. EVERYTHING ELSE is gonna use traditional medium, like every single indie studio trying to make movies with passion aren’t going to use AI as a “cheap” option, which is what this is shaping up to be, despite obv being waaay more efficient than hiring 20 inbetweeners, you get me?
I think this is the most genius way to respond to the controversy and says exactly what was on my mind at the time - all the actual art/craft you put in every step of the way separate from the anime style transfer. Thanks guys.
Proves that there is no controversy. AI is a tool and can't be used to actually replace anyone outside using dangerous machinery and menial min wage task jobs that both should not exist in the modern era.
the next phase is making the faces expressive. especially since this is meant to be anime, the faces are usually extremely expressive
This is really great and a good example of how more traditional forms of art creation and the use of AI don't need to be adversarial.
I can see Corridor are desperately trying to justify themselves to the losers who cry about AI being theft and immoral.
You don't have to justify yourself to those morons. You have done absolutely nothing wrong.
People complaining about the AI stealing from other people's work are just bitter.
All artists steal from other artists. Your "unique" style was trained by a lifetime of exposure and learning from every other artstyle. No one is "original", in the true sense. People are just jealous because instead of a lifetime of having to train their model, the AI can train a model much more rapidly.
Why hire an artist to make your own style? Where did that artist get his style from? A combination of other artstyles he's seen before? Why not just train your AI to do the same thing?
You guys are pushing the envelope forward, which is exactly what true artists should always be doing. Keep on exploring AI and how it can make art better and easier. Don't ever feel bad for it. Forget people on Twitter and Reddit crying about nothing.
Hiring a REAL artist not only makes this morally right, but also makes it look BETTER than stealing from an anime movie. Yet still, kinda shitty. It still needs an in-between animator.
First, I agree that making money another's art without due attribution/compensation is unethical, but I'm about to play devil's advocate for a moment. I was just rewatching this and at 6:50 the artist they hired said he took his style and, since he is not learned in the style of anime, used anime like, Vampire Hunter D, Gundam, Cowboy Bebop, and more to influence his art. Shouldn't they have hired an anime artist to use just their style to make the art? Didn't they just essentially do the exact same thing, but pay a person instead of an AI to do it? I am not trying to troll or get under anyone's skin. I genuinely want to know others opinions.
"Are we witnessing the birth of a new art form?"
I sure as fuck hope not. AI is/has already become a HUGE threat to people working in the creative space, and the last thing we need is even more of that. Real artists are being drowned out by a flood of AI shlock, writers are at high risk of losing their jobs to something like ChatGPT, and now animation is being threatened more and more as the days pass us by.
This AI has the potential to lose people countless jobs because the higher ups see no need for them anymore. Why hire a team of a few dozen artists who have honed their craft for decades to make you something beautiful when you could have one artist and a guy who manages an AI? Sure, the studios might love it at first since they're able to pump out episode after episode on a fraction of the regular cost, but what about the people who ACTUALLY matter? Those animators will now have trained at their craft for decades and are now out of a job. The fans are receiving soulless AI slop and they'll eventually fall off, losing the studios more than they've been saving. I mean, animators are already paid horribly and overworked as is, now you're telling me they can easily be replace with an AI by the people who overworked and underpaid them in the first place?
Whether you see it or not, this is showing a grim future for artists as a whole. Nowadays it seems like there's no reason to pursue your creativity, as you'll just be replaced and drowned out by a huge flood of AI. You can't even say this is just a hypothetical or that it will never happen, because it IS happening, and it has been for months and months on end. This isn't a potential future, this IS THE PRESENT DAY! If you have any sense of worth as an artist, or if you value the work of not only the greats in the industry, but also the unsung heroes of animation, you owe it to not only yourself but EVERY artist to not support something like this. It may all be fun and games now, but this can and will result in the loss of thousands and thousands of jobs, and there will almost certainly be no coming back from it any time soon after.
Preach it. A.I. comes at a price measured in humanity. Once the corporations realize they can save money by firing real workers, they can and they will. Only have to look at the automation in manufacturing industries and the "self-serve" kiosks everywhere.
I feel you guys have been a lot more creatively free and playful with your compositions and scenes thanks to this new work method, It will be interesting to see where you go with it and if you could blend it with your "traditional" way of making videos.
A bit of topic, but I wanted to acknowledge what an eloquent speaker Niko is. He doesn't stutter, or "um" and it's clear a lot of thought went into every word of the script. I know how hard that is, so I wanted to pay some respect to it.
Exciting to see the progress!
Massive props to addressing the controversy and then even going so far as to create your own style. Seriously.
I appreciate this so much too
It's a good move because it really made the complainers look ridiculous. The End result is not only the same, it actually delivered a more faithful "rip-off" of the chosen styles they were looking to replicate because the AI model had LESS interfering styles to use as reference.
The directing and effects still are art, thats stuff still takes some talent.
This looks really good! I think the 2 pieces that are the most jarring afterwards are the mouth movements and the motions. Animated character movement is generally not bound by human anatomy, which gives this movie a distinct feel in how the characters move within their environments.
So... For the first one, did they use a model built on copyrighted material? Because they kinda glossed over that. If that's the case, is the video monetized? Finally... If those two are true... How did it get approved when there should be members of Corridor that are well aware that the use of copyrighted material for the creation of tools or research without permission from the owners of said copyrighted material, if used for monetary gain or market research is legally dubious and has basis in previous cases, aka "case law"... Which means, again, IF they did use copyrighted materials without permission for the training of tools used in a project that resulted in monetary gain... They could be in deep shit. And honestly... They should? But... Idk. They never really answered those questions and I'm not sure there's any monetization on that first video.
one of the things i respect corridor is doing is their continuous pursuit in new forms of digital art. most people would've folded and apologized with that backlash but they just went ahead and continued forward and instead improved and provided proper insight. I'm an artist too and have been fascinated about ai art and have never hated it just cause some people use it with malice. that technology has a lot of potential while still being a good tool for artists and corridor is one of the very few group of artists that are showing that amazing middle ground.
Artists don't hate AI itself,
Artists just hate the AI being trained without consent or compensation...
@@Mente_Fugaznah, artists are just scared that anyone can get better images in an instant ad for almost free compared to what they have to offer. Which is understandable, but still a stupid reason to hate something.
@@amunak_ Bullshit nonsense based on no evidence.
@@amunak_ no man.. you guess that because you think that, magically, the artists are not capable to use AI,
Everyone can learn to use AI in just 1 week, the problem is not that artists don't know how to use it,
But the fact that it's wrong and it doesn't feel like a tool because it kills your authenthicity, being a product of hyperconsumism and destroying the original purpose of the art of ilustrate the reality and non reality with your own vision and personality, using without consent the hard work of those artists who really want to make a live with their own vision.
If the problem were the ad and stuff, artists would just useAI because of the money, and artists would make a better work,
But is not okay until the models works in ethical manners.
There's a lot of layers about why is wrong the actual AI ...
Everyone wanna make more money making art, so why you think artists didn't switched to that?
Being more easy and profitable without efforts right now?
It's because the problem is more complex than that
@@Mente_Fugaz to be fair there are plenty of comments that appear to support amunak's statement
Imo you guys are pushing the realistic use of AI and machine learning for art further than anyone else I've seen trying. It's great to see you guys iterating on the process to find something that is fair and works for everyone.
AI haters will still be ignorant about it though so this doesn't matter.
@tylerpixel anyone still mad wants to be mad about anything. They went out of their way to fix the main gripe which is artists being unpaid for their work or not involved in the process, but having an artist develope a unique style willingly and passionately completely changes the AI narrative.
@@tylerpixel
There will always be haters, let them hate. We can just ignore them so this still matters.
@@orionlucas3272 I don't have to pay an artist to look at their artwork. Additionally, artists do not own their styles, they only own their work.
@@orionlucas3272Your scientists were so preoccupied with the fact that they could, they never stopped to think if they should.
Looking at different ways, processes and pipelines is the real gem here. Did everyone congratulating them on using a commissioned artist think they were immoral for the first one? It was nice to see some ones love of the D world make it back into the limelight. The internet proves again how quick people are to judgement.
There's still no AI you can type in:
Make kick awesome anime
Still a lot of hand work involved.
The end isn't near, kids.
You guys are really setting the standards for the rest of the industry. Stable diffusion exists now and it will never “un-exist” so we need to figure out how to live with it and use it ethically and I think this is the sweet spot. Keep it up!
Props for you taking the criticism in stride and not being discouraged.
By using ai to create art you're still making something original. It's really not stealing anything unless specified to and you can easily find out through reverse image searching. Ai art isn't inherently stealing art
I'm glad that you have now used AI and fed it your own art. This is the most ethical way of how to use machine learning software that uses images. The controversy stems from using copyrighted art and was used without the permission of the artist. I for one am now all in for the support of this because you've reached out to an artist and the artist created art with the knowledge and intention to have their art be used on the software.
Artists who believe they made up their own style devoid of all the media/art that inspired them are narcissistic liars.
@@OGPatriot03 It is just called "inspiration". The artist even explained it in the video that he looked at a bunch of anime and added his own spin to the style. Were you not watching the video? AI doesn't have inspiration because it just straight up copies several artworks pixel by pixel.
@@d0decadice this.👍
@@OGPatriot03 you dont know what you are talking about of course we know we sample each others work, workflows, and techniques.
@@d0decadice That's literally NOT how the AI works, it doesn't have a database of pixels anywhere that it uses during the generation process. Just like the artist the only reference is the inspiration gleaned from learned principles during the training process.
I'm neutral with AI and I love how you're handling this! Making your own art was definitely the right call but I would still love to see more articulated mouths because I think being able to do those would elevate this technology even more.
Regarding the mouth animation, they actually did that on purpose! They've said in a previous video (Did We Just Change Animation Forever?) that they purposefully didn't act out the lips to make it look more cartoony
Doesn't matter you are neutral or not
ai art still a threat to creative jobs
@@BloppTheIraeBlob How? Companies trying to remove large swaths of their workforce? IMO that sounds like ... a company / corporate problem to me.
Some people seem to be confused; this version of stable diffusion is built on their OWN artworks this time around. I can't say much for photoshop or software's like that when it comes to stealing artwork, but they essentially have their own version which has NONE OF THE STOLEN ARTWORK just the code that makes it work with their OWN artwork for it. so essentially this is the best possible way for them to use it. Now in terms of artists job security on the future for animators I can't speak on that topic cause it's pretty up in the air with how that'll play out in the future, but I think overall with how corridor wants to do things this is the BEST possible thing for them to do. AI is going to still be hated but, in my opinion, they at least went down the most ethical route they could.
But wasn't the ai initially trained on millions of stolen images?
@@WeeklyComedianthat’s not how it works, each AI software has its own data set, depending on the software it could be using millions of stolen artwork but that’s so it can generate anything, corridors data set is just their own stuff meaning they can only generate what they own and variations of what they own nothing about any of their images has stolen property anymore
May I offer an idea: Long ago, in a remote corner of the universe... painters were realistic. Then PHOTOGRAPHY came along and put those master painters out of business. They had to reinvent themselves...impressionism.
I don't feel like this replaces hand-drawn or traditional cartoons. This reminds me of A Scanner Darkly (or even Rango), where all frames of the movie shot with live actors were replaced by hand-drawn cartoons. This just feels like an automation of that same process.
While I greatly appreciate the use of your own custom art for the fine-tuning it also needs to be acknowledged that the general stable diffusion model was trained on other peoples art without their permission in the first place.
been around for ages. crazy to see how you guys manage to always push the cutting edge. can't wait to watch RPS 2, you guys are doing great work
You forgot some words before "been around for ages."
Who's been around for ages?
@@GraemeGunn you forgot that i didn't ask
I am really happy to see you using an artist who consented for the reference. That makes a big difference. Keep trying to be great and ethical at the same time.
POSTED 2 SECONDS AGO??? COUNT ME IN
love how you guys are always pushing the medium, inspiring!
4:56 "... for the axe was clever and convinced the trees that because his handle was made of wood, he was one of them."
AI has yet to discover what a tounge is. Artists remain safe for another day.
So Josh got inspired by other art to create new art.
So a non-artificial intelligence was trained on existing art. Why is it a problem, when an artificial intelligence is trained?
Me as a musician and composer also get my inspiration from other people's work. So?
In the fairest way possible, I think a big problem people have is that an Ai can only replicate; it cannot innovate.
When we replicate someone’s style, we do so imperfectly, and those imperfections become our own style.
While an Ai may not be “copying” in the traditional sense, it is not adding its own style into the mix.
Absolutely love this pipeline! Keeping the artists involved improves it so much 😍
This for me is how you address the plagiarism aspect of AI and create a project that uses AI ethically to speed up the process. To do this process you'd still need costume designers and creators, you'd still need actors, you'd still need writers and artists to create the art style, you'd still need a studio crew to set up cameras and green screens and film the work. It just removes the stumbling block of needing an artist (or studio of artists) to spend hundreds of hours drawing thousands of individual frames per episode.
I think the problems with AI ethically is that it creates a world in which you can take someone else's work and use it without their permission for something they did not intend for it to be used for. The artist of Vampire Hunter D created his style for use on Vampire Hunter D and did not intend or licence it for your use in the original Rock Paper Scissors, by hiring someone to create an art style for your show you move past that dilemma and show a way in which studios can perhaps begin to use this pipeline to create shows while still keeping the creative industry in business and hiring people.
It's still not perfect and I can see a lot of ways in which big studios could abuse this new system against the actors and writers and artists, but it's not necessarily worse or better than the current abusive system and at least shows a way forward for ethical use of AI in creative projects.
Thank you someone who isn't just saying ooh video bad because AI art
I think you are onto something amazing, first congratulations to the team.
As feedback I think that less flicker might not be the perfect solution, because then some flickers might become more visible, it would be similar to reducing the frame rate of a fast paced animation.
Keep up the good work!
It's really cool seeing how this tool is used. I loved the first RPS and I look forward to seeing this one. Honestly though, I'm not too big a fan on the art style of this one. The Vampire Hunter D style looked so good. I understand why y'all went with your own art style. I just wish it was more on the anime side instead of the cartoon side
It’s because of new court cases they could get sued for the vampire hunter d art used. (Atleast if they tried using it for commercial reasons)
What art used to teach it effects if it’s legal or not.
They didn't want people to think they were "stealing" from other works, that kinda stuff is a pretty hot topic rn
@@AL-lh2ht The courts have yet to determine anything.
Well sucks for you I guess
@@AL-lh2ht yeah, I get that. I just wish the art style was a bit more anime than cartoon. It still looks absolutely stunning
So is your dataset ethically sourced and only contains art from Josh?
Edit: Gave the video a proper watch, let it never be said I write off things I disagree with, had to stop at 8:22. They are still using Stable Diffusion which is trained on art without the artist's consent. That is where I draw the line and will always draw the line.
there's no such thing as "unethical datasets" that's something ya'll made up.
Didn't they literally say that in the video?
@@BioshadowXHe said something like that but I'm not sure if it's 100 percent just Josh's art.
@@ReaperTheRager it is for the characters .. backgrounds used the default anime datasets
This excites me so much because it could bring the "a scanner darkly" style to sub 30k budget movies
This is really eye opener for many people who watched the Part 1.
Because when i showed my friend he said u guys acted it out and put it in an AI tool and entered a bunch of prompt which for felt too simple for u guys to do
So there are hundreds and thousands of people who think creating an art form with AI is just some prompt typed in without any hardwork.
Good job bringing it up and making this video
THIS
But that's unfortunately to be expected since this is still a relatively new area and not everyone is informed of it
@@GoharioFTW true true
Its a new thing
Not everyone is super upto date
And everyone being into more short format news, getting info from there its hard to know the real truth indepth
So i hope everyone will understand not all AI generated art and works are bad
But also saying that there is a form of unpassionated, very lazy works lurking on youtube and other social media apps and its ez to find because u find no soul to it pretty easily
I must say, the original RPS video reinvigorated my waning love of film and encouraged me to step out of my comfort zone and embrace technologies that I had previously shyed away from, like Unreal Engine. The work you are doing is inspiring.
they ain't doing any work it's the ai doing it
@@miniwhiffy3465 Have you not watched the behind the scenes work on this? The AI is a tool they're using, but this still takes a tremendous amount of work to accomplish. I am not a fan of anime, but this was one of the most impressive artistic feats I've seen done by filmmakers in a long time. The embracing and utilizing of the technology is brilliant and I can't wait to see how they keep improving it.
I really appreciate that they address the controversy by just mentioning it and not actually addressing anything
I find it interesting how people complain that an AI is trained on existing art styles to them mix them into something new and then we see a human artist doing exactly the same thing.
Including the trial and error until the style is acceptable.
Unpopular AI opinion - Photography is at the same level as AI generated images. Think of a photo of a tree on a mountain. The photographer did not build the mountain, tree, sun for lighting or clouds for shading. In fact, the machine did most of the work in the generation of the image base on inputs the "artist" had zero in created aside from curating the the angle of view and focus/zoom/exposure settings. Even then, many photographers take many photos of the subject then curate the best looking shot later (along with photoshop).
If you believe that photography gets full copyright protection then a curated AI generated content should enjoy those same protections. Personally, I that unless an artist be it photographer or videographer demonstrates the images on their screen were directed and guided by them, then their copyright to the video or images are weak and should be considered public domain. Basically paparazzi can get bent given they mainly deal with "right place/right time" which is at the same level of creativity as telling MidJourney "Make cool image".
Well it’s finally happened. Major CC content locked behind a paywall. I can only hope it eventually comes to UA-cam.