Support NeoGAF

[nkarafo]

Whoever knows how traditional handmade animation works also knows how labor intensive it is.

You always get shortcuts because of this. Like cutting down the details, reducing the number of frames per second, having more static elements in the frame, etc.

Richard Williams was one of the handful animators who tried to not compromise. And he was the living example (RIP) that this is impossible. His dream project was never finished because it took too long and was too expensive to produce. All his best animations are only shorts or short segments (like in Roger Rabbit). He proved a human doesn't have a long enough lifetime to make a career of drawing 24 frames per second, detailed cartoons. Nor the resources.

It simply cannot be done.

And here's where AI comes for the rescue. From what i have seen it should be possible for it to take some extremely detailed cell, like those Spongebob static paintings:

And animate it while filling all the blanks. At any frame rate you want. While keeping all the details and varied shading, no flat colors.

Why stop there, it could even be something like a very detailed painting. Imagine a fully animated, 24 frames/sec scene like this:

That painting alone took how long for the artist to make? 6 months? A year? Now imagine a whole 1:30 hour movie at 24 frames. That's 130.000 such paintings. Yeah, i don't think anyone or any team or any collaboration would have the time and resources to make such a creation. And sure you could use custom made CGI that would try to emulate this. But so far, even random AI videos look far more convincing than your average custom made CGI in movies. I don't think a human can make CGI movie that looks like a detailed traditional cartoon and be convincing enough. Because humans can't help themselves, they always get something wrong.

As a huge fan of animation, this was always something i wanted to see but knew i would never do. But thanks to AI now i think i will be able to.

Thoughts?

 

[poppabk]

Forget this one, it is well known that AI is deathly afraid of drawing tits.

 

[yet_another_alt]

Forget this one, it is well known that AI is deathly afraid of drawing tits.

Someone finally found a way to stop Skynet but some coomer with a locally trained AI will upload it online and we're back to T-1000s hunting us down soon enough.

 

[Bashtee]

Forget this one, it is well known that AI is deathly afraid of drawing tits.

The AI erotic pics on Twitter with 7 fingers and 2 assholes would disagree.

 

[bitbydeath]

 

[Lambogenie]

It's still CG in a way in my opinion.

 

[jufonuk]

It's still CG in a way in my opinion.

If animation houses and shareholders can speed up the process and pay less for it. Hire less people you bet their ass they are gonna go with it.
Which is sad a people will be out of work.

 

[Ironbunny]

Yes its one way AI will be used. They had a great demo of using AI for game development when it comes to sprite graphics. In it AI animated characters from one sprite to full set of different actions. The result were amazing.

 

[Dacvak]

This is one of those examples where the end product is not the actual art. While I do think AI can and will replace a lot of art creation over the next few years, part of the reason why these hyper-realistic cels are so interesting is because they came from the brain and hands of a real person.

Generating this sort of semi-grotesque imagery inherently loses some of what makes this style interesting. At least for me.

 

[DeVeAn]

Tom & Jerry from 40's 50's is still some of the BEST animation ever. The orchestra music too. Can AI do that yet?

 

[hyperbertha]

Ai is still making glorified after effects clips. I'd wait to see some evidence it's even remotely capable of anything like this.

 

[efyu_lemonardo]

 

[Haint]

Ai is still making glorified after effects clips. I'd wait to see some evidence it's even remotely capable of anything like this.

Just 5 years ago it was only capable of making unrecognizable distortions, of course in several more years its going to be exceeding humanity's best works. That is why Nvidia is the most valuable company in the world.

 

[Grildon Tundy]

I think you're right, OP. It's probably not there yet because the available high-detailed animation data isn't as plentiful as other types of videos that AI has been trained on, but no reason to think that isn't a problem that can't be solved if there is money to be made.

 

[Dacvak]

Tom & Jerry from 40's 50's is still some of the BEST animation ever. The orchestra music too. Can AI do that yet?

The Mickey-mousing of older cartoons is truly what makes them timeless compared to post-2000s cartoons, imo. (Btw, that’s the actual term for “full synchronized scoring” of cartoons, most notably in cartoons like Tom & Jerry and Looney Tunes. It’s a lost art, and also insanely expensive in the modern age.)

 

[Hookshot]

Could an AI make a picture so detailed that AI couldn't copy it?

 

[Polygonal_Sprite]

Are there no early examples?

 

[nkarafo]

This is one of those examples where the end product is not the actual art. While I do think AI can and will replace a lot of art creation over the next few years, part of the reason why these hyper-realistic cels are so interesting is because they came from the brain and hands of a real person.

Generating this sort of semi-grotesque imagery inherently loses some of what makes this style interesting. At least for me.

Artists could still make the original concept art, the still images (like the backgrounds) and some key frames. The AI would fill the rest.

Repetition is the biggest hurdle in traditional animation. To achieve smooth animation you need to draw the same thing over and over again, with only slight variations. That's why cartoons never have the detail shading of a good looking still.

The AI seems good at this. Take a few key frames and produce all the repeating images, thousands of them. With all the shading, details, etc. Artists could still review them and fix errors. It would still need manual labor to polish such product but reviewing thousands of cells is faster than creating them from scratch.

 

[newtypepilot]

no, machine learning requires huge amounts of data or else it can only produce garbage. Furthermore, the traditional animation you are thinking of is done in layers. Each single frame is split up into multiple layers to ease workload.

Even if you were to do the keyframes and have AI do the in-betweens, the results will be unsatisfactory because it won't understand timing or expression. Good animation is able to convey the character's personality through its motion. The AI will have trouble dealing with basic animation concepts such as squash and stretch, anticipation, follow through, etc.

 

[IDKFA]

This is one of those examples where the end product is not the actual art. While I do think AI can and will replace a lot of art creation over the next few years, part of the reason why these hyper-realistic cels are so interesting is because they came from the brain and hands of a real person.

Generating this sort of semi-grotesque imagery inherently loses some of what makes this style interesting. At least for me.

That's where I stand. If a human hasn't made it then it isn't art.

 

[nkarafo]

Even if you were to do the keyframes and have AI do the in-betweens, the results will be unsatisfactory because it won't understand timing or expression. Good animation is able to convey the character's personality through its motion. The AI will have trouble dealing with basic animation concepts such as squash and stretch, anticipation, follow through, etc.

All true.

But aren't key frames enough to give the AI this information? All those things you mention, the most expressive frames, are usually the key frames and those could still be done by the artists. Sure, even the in-betweens need good timing though but who knows how smart AI can be. I mean, it already does things i thought impossible 10 years ago.

Or how about this... Artists only make the rough animation. Just the initial pencil drawings that have the basic shapes, etc. Without colors or shading or most of the fine details. The AI does the rest.

I think there's plenty of room for an AI to cover a lot of the work it needs to be done for traditional animation because animation, by it's nature, uses a lot of repetition that requires more time and patience than talent. It's not that Richard Williams was more talented than many other animators. Drawing 24 frames rather than 12 or 6 isn't a matter of more talent. And if you can draw, you can add as many details as you want in a drawing. But re-adding them in every frame is, again, a matter of time and patience. Most animators wouldn't bother.

And this is why CGI animation took over. Because there is no concept of repeating frames. You just create the models once and move them around. That's how a lot of 2D animation is done too nowadays, which is why it looks like clinical shit.

But the AI doesn't care. It doesn't burn out or get bored like humans do. And its much, much faster. It can draw all these nice, detailed frames.

 

[Reizo Ryuu]

If you are going for that much detail in motion you'll just wind up using CGI anyway, we already have many examples of cgi being very stylised and they are only getting better.

 

[nkarafo]

If you are going for that much detail in motion you'll just wind up using CGI anyway, we already have many examples of cgi being very stylised and they are only getting better.

These are nice and all but i'm talking about traditional 2D animation.

 

[Reizo Ryuu]

These are nice and all but i'm talking about traditional 2D animation.

I know you are, hence my first sentence.

 

[nkarafo]

I know you are, hence my first sentence.

3D cgi looks nothing like 2D traditional animation. It's not that i only want detailed visuals, i love detailed 2D animation in particular.

 

[Reizo Ryuu]

3D cgi looks nothing like 2D traditional animation. It's not that i only want detailed visuals, i love detailed 2D animation in particular.

Yet.
And they are animated with the same principles, the difference mostly comes down to lighting/shadows/FX, since those are basically free in CGI.
Disney's paperman has pretty much already done what you suggested, hand drawn keyframes on top of cgi animation and automatically generated in betweens

 

[nkarafo]

Yet.
And they are animated with the same principles, the difference mostly comes down to lighting/shadows/FX, since those are basically free in CGI.
Disney's paperman has pretty much already done what you suggested, hand drawn keyframes on top of cgi animation and automatically generated in betweens

I can still see the pretty big difference between this and the stuff i'm talking about.

Also, that's not even that detailed. It's still missing the fine detail and shading of the still paintings i'm talking about in the OP.

But thanks for the suggestion either way.

 

[Reizo Ryuu]

Also, that's not even that detailed. It's still missing the fine detail and shading of the still paintings i'm talking about in the OP.

That's just an art style difference, they could've made it as detailed as they wanted, but then it wouldn't look like disney anymore.

 

[nkarafo]

That's just an art style difference, they could've made it as detailed as they wanted, but then it wouldn't look like disney anymore.

Well, i'm looking for traditional looking animation with details and shading like the images in the OP. Basically animated detailed paintings that move like traditional 2D animation at 24+ fps. Not skeletal/3D animation like that Disney movie.

 

[nkarafo]

R Reizo Ryuu

Take a look at this short. This is close to what i'm talking about. See how different it looks/feels than your CGI examples.




Also, watch this video if you care:

 

[Reizo Ryuu]

The animation principles are all the same, it doesn't matter if it's 2D or 3D, you can do all the squashing and stretching and animating on 1s/2s/4s etc if you want to, the biggest difference is as I said the things you can have the computer calculate for you instead of having to draw it all manually, like shading.

 

[nkarafo]

The animation principles are all the same, it doesn't matter if it's 2D or 3D, you can do all the squashing and stretching and animating on 1s/2s/4s etc if you want to, the biggest difference is as I said the things you can have the computer calculate for you instead of having to draw it all manually, like shading.

You are still missing the point.

Look at the video i posted. Does that look anything like computer generated 3D at all?

It's that particular look i'm searching for. I don't care about 3D animation because it always looks like 3D animation, that's my whole point. You can post any 3D animation with a "2D art style" that exists, it still looks like 3D animation. I can't put the difference into words though, i'm not that good expressing it, especially in English. But that video should be enough.

So, if humans can't do it with 3D animation, let alone actually drawing all these images traditionally, i believe the AI can do a convincing job with it. That's the point of this topic.

 

[tamago84]

when can ai remove itself from that god awful saturated scheme with soft edges

 

[Reizo Ryuu]

You are still missing the point.

I'm not, you're looking at a difference that only exists because artists are taking advantage of the benefits of 3D animation, I'm saying it looks exactly the same if they'd strip away all those benefits; keyframes for a normal > surprised face, are going to look exactly the same whether your draw it by hand or manipulate a control rig by hand, it's going to have the same anticipation, the same bounce, the same relax etc.

 

[nkarafo]

I'm not, you're looking at a difference that only exists because artists are taking advantage of the benefits of 3D animation, I'm saying it looks exactly the same if they'd strip away all those benefits; keyframes for a normal > surprised face, are going to look exactly the same whether your draw it by hand or manipulate a control rig by hand, it's going to have the same anticipation, the same bounce, the same relax etc.

And yet, if it's hand made it looks completely different.

At least it looks different to me.

 

[ResurrectedContrarian]

Pixar (and similar) movies will be rendered with AI eventually; the notion of manually constructed rendering pipelines of polygons/textures/etc that we have today is temporary for such things, and will vanish. I probably won't be long, either.

 

[hyperbertha]

Just 5 years ago it was only capable of making unrecognizable distortions, of course in several more years its going to be exceeding humanity's best works. That is why Nvidia is the most valuable company in the world.

Ai improvement rate is slowing down. Video generation hasn't really improved in over a year. You need to understand there's a good chance it's not going to get better because we've hit the limits of the current architecture. Well likely need a breakthrough, like the transformer a few years ago

 

[IDKFA]

Ai improvement rate is slowing down. Video generation hasn't really improved in over a year. You need to understand there's a good chance it's not going to get better because we've hit the limits of the current architecture. Well likely need a breakthrough, like the transformer a few years ago

I was going to reply, but then I thought why not be meta and let ChatGPT respond. Here was its reply to your post:

I think there's some truth to the idea that the rate of improvement in AI, particularly in certain domains like video generation, might seem slower compared to the rapid leaps we saw a few years ago. However, I wouldn’t say we've hit the absolute limits of current architecture. The pace of development is still impressive, but the nature of improvements is changing.

With video generation, the challenges are more complex than with static images. Generating high-quality, realistic video involves not only improving individual frames but also maintaining temporal coherence, handling long-term dependencies, and dealing with the massive computational demands. These are tougher nuts to crack, and while we've seen impressive strides, it's harder to achieve the same visible breakthroughs that we did with things like GPT or image-based models.

You're also right in suggesting that significant future progress could require a new breakthrough, much like how the transformer architecture revolutionized NLP and other fields. There are still areas where AI is progressing, such as in techniques like diffusion models for generation, and advancements in model efficiency or multimodal capabilities, but it’s also possible that a breakthrough innovation is required to take things to the next level.

So, while it may feel like we're plateauing, I think there's still a lot of room for creative progress—whether that comes from refining existing architectures or from something entirely new.

 

[FeralEcho]

The AI erotic pics on Twitter with 7 fingers and 2 assholes would disagree.