AI generated art

[Meaning Less]

Something that might work is taking AI art, cutting it up and fixing it, then painting over it. but I think it'll still be bruteforcing through hundreds of tries before you get anything close to what you want.

And honestly, at this point it probably would be faster to just google for existing artworks yourself and make edits on top of them instead of relying on pure AI randomness.

 

[Geralt_R]

yeah, that's a cool picture of a mystical ass-faced wizard standing on a desert made of flower-skulls, but what I needed was a girl bending over at a well on a meadow.

Dall-E however actually does that, it will give you a picture of a girl bending over a well on a meadow, if that's what you ask for. It won't give you a random picture of a wizard instead. But you never know what the girl, the well or the meadow will look like.

Also, Dall-E only takes 10 seconds to generate pictures (it will always generate a selection of pictures to choose from), so you can further refine the prompt if you find it's not quite yet what you want and say "I want a girl wearing a red hoodie bending over a well made of stones on a meadow at sunset".

And honestly, at this point it probably would be faster to just google for existing artworks yourself

Check out this video

You must be registered to see the links

The graphic designer wins 2 out of 3 times against Dall-E, but Dall-E made its results in 10 seconds, the designer needed 3+ hours for each image he did. And the goat in Mona Lisa style taking a selfie, while looking much better in the version created by the artist, would still work really well as concept art if you just use the Dall-E version.
And it's still very early stages for Dall-E. The result with the robot woman guarding a wall of computers tells you that in certain instances Dall-E is already extremely helpful and even better. And it's much faster, again, 10 seconds vs 3+ hours.
It is certainly a great tool for creating quick concept art, for storyboarding maybe, for sketching a rough idea... the human artist can then eventually refine the idea and make the final image. But you just need a rough sketch of your idea? Dall-E can absolutely do that, no more humans required.
And results will only get better. If I were a graphic artist beginning my career now, I would not be sure I can still compete in 10 or 20 years. Same goes for photographers.

 

[anne O'nymous]

That entirely depends, in my opinion, on one thing... will AI eventually gain true consciousness, i.e. will it be an "artificial" human, however with the added ability for self-improvement and much higher computational speeds, that is, it can change and improve its own algorithms, learn new things without external input? If yes, then no such restrictions will exist anymore.

Being able to make a comic/game is not a question of algorithm, nor a question of computational capacities.
The main problem encountered when making a comic/game is that you need to use the same location and characters from start to stop, but represent them from different angles and, for the characters, in different clothes and poses. And whatever the algorithms and computational capacity, an AI can't do this because it's build to generate each images independently to the others.
What permit a human to make a comic is his memory of the visual representation of the subject, something that AIs don't have ; it's why I talked about meshes, because it's what correspond the most in AI terms. As long as AIs don't have this important feature, they'll never be able to draw the same character in each scenes of the comic/game.

 

[Geralt_R]

The main problem encountered when making a comic/game is that you need to use the same location and characters from start to stop

Absolutely true, as of now. AI can't do that. And this is not even touching on the writing aspect. So for now artists can feel safe. However, I would argue even with the limited abilities we have now, you could certainly tweak the algorithm so that the AI would generate consistent looking images, using the same characters, same locations. You don't even need strong generalized AI for that.

That poor graphic artist in the video I linked was really stressed out during the experiment. And he competes with a very early version of image generating AI, it's just version 2. Imagine what version 5.0 or 10.0 will be able to do.

I wouldn't feel too safe about my future career as a graphic artist. Just as truck drivers will eventually be replaced by self driving trucks (not ready for prime time yet either, self driving is much harder than people generally assume, but we'll get there eventually).

 

[FleshSac]

Think about all the things the AI would need to cover for a basic visual novel. First, it would need to be able to consistently produce the same cast of characters. Second, it would need to be able to consistently produce the same environments. If it could do both of those things, how would you produce a scene?

Okay, AI... I need X character and Y character walking down hallway 6. X should be wearing outfit q and have her head tilted slightly towards Y, with her gaze fixed on Y. Mouth is slightly open and her expression should be pensive. Y is wearing outfit 9a and has her head tilted slightly towards X with her gaze fixed on X and has a slightly curious expression. It is midday. Go.

...

No, that doesn't look quite how I want. Let's start over.

Fuck. That.

Maybe it will all be simplified and improved someday, but right now it seems like an extra programming language that I don't want to learn. I'm old. I don't have enough patience to teach a computer how to properly pose a German suplex...

Now, can an AI do the job a graphic designer does for big businesses who only need one image for their logo or magazine ad? Quite possibly, sure. I don't think SkyNet is ready to infiltrate the world of adult games, though. Not yet, anyway.

 

[Meaning Less]

The graphic designer wins 2 out of 3 times against Dall-E, but Dall-E made its results in 10 seconds, the designer needed 3+ hours for each image he did.

Remember that you are seeing a very biased video trying very hard to make the AI look good, DALL-E sponsors those videos for a reason.

Can AI replace lazy graphic designers that create random logos with zero creativity but just copy other existing artwork? maybe.
But actual artists that rely not only on creativity to create new things but also consistency to reproduce more of that if needed? not really.

 

[Geralt_R]

But actual artists that rely not only on creativity to create new things but also consistency to reproduce more of that if needed? not really.

Not yet.

What about in 10 years from now? 20 years?
I wouldn't be so sure about humans having the monopoly on creativity in 2040.

We already have the tools to fool people, i.e. speech synthesis has become near perfect, we can also create photorealistic human avatars, if you watch videos of conversations with OpenAI's GPT-3 model with the text output used as input for speech synthesis software and using human digital avatars for the visual presentation you will think you watch a conversation with a real person. Some people watching these videos who don't read the description even think an actor is speaking the lines, but everything you see is digital and everything you hear was written by a "chatbot", more like a sophisticated language model of course.

Once things like OpenAI or Google's LaMDA evolve even more nothing and no one will be unaffected by it. And those language focused projects and image generation AI projects will certainly merge at one point. And it is kind of scary that the likes of Google or Facebook or maybe some state run lab in China are working on strong generalized AI.

 

[Meaning Less]

What about in 10 years from now? 20 years?

Well, you know we are far from it when the best google AI available today still can't distinguish a car from a bus in a 2d image without relying on a massive database and frequent human input.

Meanwhile this same task can be accomplished by a human baby with more precision by just showing him a single image of a car and a single image of a bus.

Computer vision is still the achilles heel of computers.

 

[Geralt_R]

Computer vision is still the achilles heel of computers.

Self driving is another proof of that. Things that humans find extremely easy are incredibly difficult to recreate for AI. You can show a human one chair and the person will immediately understand the concept of "chair" and then easily recognize other chair like objects. An AI needs a bazillion images to kind of learn what a chair is.

But just because it's like that now doesn't mean it will be like that forever. I think it's inevitable that AI will eventually master all these things. And I think this will happen in our lifetime. Not tomorrow, not next year... but what about 2030? 2040?

Of course strong AI has been teased for virtually forever, it was always just "a few years away" and then everything fell apart like a house of cards, resulting in the "AI winter" where the field was almost dead. But we have made progress with neuronal networks. I remember one AI project that tried to literally feed every single piece of info by hand to the AI (Cyc), that never went nowhere really like all the other projects at the time and IBM's Watson that won at Jeopardy was mostly smoke and mirrors and the few hospitals that bought it as an expert system were apparently less than satisfied. But again, that was years ago.

LaMDA or GPT-3 are a on different level already. And image creating tools are just in their infancy, and they already can create some impressive results (with caveats). I would guess it's still some time off before any AI would ever put out smutty VNs But AI tools may certainly significantly speed up development times long before that.

 

[woody554]

Not yet.

What about in 10 years from now? 20 years?
I wouldn't be so sure about humans having the monopoly on creativity in 2040.

oh, absolutely. in time AI will do everything better, even all creative work. and while I think it's gonna be a little furter down than 2040 I wouldn't be shocked if it happened that fast.

but the current ones are not really even trying to do that. they use a very simple decomposition of images that's been around already 20 years ago. they are superficially mimicking the input, not doing any kind of analysing or deduction on it. that requires some form of 'understanding' of what they see, simple classifying isn't enough. I use 'understanding' here very loosely and I don't want anyone to think it means becoming sentient or self-aware, but it does require building blocks they simply don't have yet and we don't know where to even begin with it.

I think we'll need a more 'complete' approach into designing neural nets, in that instead of building singular systems we need to design a whole hierarchy or ecology of neural nets which all interact and change each other in semi-mysterious ways. and it'll probably be impossible to design it top down by our logical thinking, but rather we'll have to 'grow' it bottom up and let it design itself. which is a very obvious thought for anyone studying them, and I'm sure it's been tried thousands of times by thousands of people, without results. but we'll get there, and it won't be nearly as long as people think. and when we crack that, things will happen very quickly and world will change in vast ways.

but we're not there, and the current AIs are far far dumber than bacteria. we almost haven't even begun taking the first steps on that cognitive ladder.

 

[♍VoidTraveler]

Basically this AI is similar to the one that can spit stories - good for mass-producing something simple, but no more than that.
At least, for now.

 

[Meaning Less]

But just because it's like that now doesn't mean it will be like that forever. I think it's inevitable that AI will eventually master all these things. And I think this will happen in our lifetime. Not tomorrow, not next year... but what about 2030? 2040

But even things like self-driving are only possible because of lots of extra 3D scanners and sensors that make computers a bit less dumb.

But now when it comes to a pure 2D picture without any extra information? Trying to interpret images and create original art? Well that's so far away that we really can't fathom it replacing the human brain anytime soon.

 

[woody554]

An AI needs a bazillion images to kind of learn what a chair is.

they're not really learning what a chair is, they're classifying the input. they're building a n-dimensional function where n is the number of inputs, or pixels in an image's case. the training is approximating that function iteratively. after training when you feed it new data it'll drop it into that function and see where it lands, and if it lands where all the 'chair' data landed it'll classify it as a 'chair'.

at no point will it even attempt to 'understand' what the idea of 'chair' is, it doesn't do any analysis or deduction. it just classifies, and how well it hits the mark is entirely dependent on the training data.

and that's the part we need to figure out, how do we make it conceptualize. how do we make it form 'ideas'. and we have no clue.

 

[Count Morado]

The advances we are seeing in AI are incremental when considering the whole of man-made computational machines.
Starting with Babbage in 1883, we're but 140 years since the first modern "thinking" device. If we were to consider what is necessary For to have AI advanced enough to do what is being hypothesized in this thread, it would need an exponential boom that is still decades away. One could easily take a timeline of the complexity of evolution in living creatures on earth and scale it appropriately for better visualization:

 

[woody554]

The advances we are seeing in AI are incremental when considering the whole of man-made computational machines.
Starting with Babbage in 1883, we're but 140 years since the first modern thinking device. If we were to consider what is necessary For to have AI advanced enough to do what is being hypothesized in this thread, it would need an exponential boom that is still decades away. One could easily take a timeline of the complexity of evolution in living creatures on earth and scale it appropriately for better visualization:

while I don't really disagree, I don't think framing it as some continuous 'growth' with some pacing describes the wall we need to surpass. what we need is a paradigm shift. and when we figure that out the change will be abrupt and happen almost overnight. and what's on the other side won't be in any way connected to this side. we'll go from apples to oranges.

and to clarify I'm not talking about singularity, which is nonsense. I'm talking about going from classifying to conceptualizing. teaching the AI to 'understand' instead of blindly mimicking.

 

[Geralt_R]

and that's the part we need to figure out, how do we make it conceptualize. how do we make it form 'ideas'. and we have no clue.

Really good point... at no point does the AI "understand" what a chair, dog or cat is. I don't work in the field but I believe the latest neural networks have become so complex that understanding what's really going on inside the network in detail is either very difficult or even impossible.

But things like DALL-E etc certainly have no idea what it is they are creating.

And that brings us back to the matter of sentience and consciousness... Without AI will never understand what it does. We don't even know how that works in humans. How are we to recreate it in AI? Unless it's a self-emerging property of complex neural networks, independent of the substrate, i.e. it does not matter if your neurons are wetware or hardware. If it's not a self-emerging property it may take a long, long time until we know how to recreate consciousness in AI. We'd have to understand it in humans first.
No one really knows yet. Most would say it's substrate independent, so silicon neurons can in theory form networks that are sentient and understand the world.

If we were to consider what is necessary For to have AI advanced enough to do what is being hypothesized in this thread, it would need an exponential boom that is still decades away.

I don't think it's like that. From a pure hardware perspective we are capable (even if the human brain is still much more powerful than the best supercomputer currently available), what we lack is a proper understanding of what consciousness, actual understanding of the world really is. So far all approaches use some kind of machine learning, where we kind of brute force tons of data into the machine which then forms connections and "learns" what a chair is (it doesn't really know what a chair is though).
Once someone figures out how to emulate the way animals (humans) learn things would change overnight. So I think it's more of a question of finding the right algorithm, the right way to set up your neural network. And the first proper conscious AI may be something like a small mammal, maybe a mouse. But once we are there things will escalate quickly. Maybe for true AI to emerge it may need a body, so it can interact with the world. Who knows.

 

[Carpe Stultus]

Well it's finally happened. AI generated art is just vastly superior to whatever humans can draw, never mind the artists who make porn games lmao.

What do you all think will be the implications this has on video games here and better yet, how will patreon milkers manage to maintain their 5 month update output?

You sound happy if AI's would make your jerk off material now. I'd be very concerned about that instead of happy alone because it threatens the existence of every single artist. I wonder if you'd be happy too if something like that would threaten your livelihood, not that either will happen soon.

I'm not even going to call you out for your bullshit "5 month update output milker" because you have obviously no fucking clue what you are talking about.

 

[Geralt_R]

You sound happy if AI's would make your jerk off material now.

The truth is, whether we are happy about it or not.... many things point to the possibility that we will have to learn how to live with strong generalized AI in our lifetime. And it will change everything. If we don't know how to adapt it may disrupt societies to the point they are crumbling. How would you feed the millions and millions of people who are now out of a job, because strong generalized AI can do everything better, won't need vacation time, can work 24/7, doesn't get sick. Basic general income may be something societies have to look at, there are already experiments underway for that in several countries, because everything points to a future where there won't be enough jobs left for people. Even in the creative arts.

Some things go hand in hand with robotics, but we also make huge advances there, so once we have capable robot hardware and strong AI it's "goodbye" to the world as we know it. The thought it may happen in our lifetime can be very scary. We have to hope any strong AI will like humans. And that maybe it's not developed in China first. Or by Facebook. Maybe not even by Google. But chances that some corporation like Google or a country like China will develop strong AI first is pretty high.

And yes, some people don't know how much work it is to pose models, to tweak models, to make them look a bit more unique, to search for clothing, assets in general, add realistic lighting, maybe animate a few things, render everything and write the scenes to begin with.

 

[Count Morado]

what we need is a paradigm shift.

what we lack is a proper understanding of what consciousness... Once someone figures out how to emulate the way animals (humans) learn things would change overnight.

At no point in my argument did I say that hardware was the problem. I apologize that I wasn't clear.
The exponential growth is based upon exactly what you two have stated:

• A paradigm shift, and
• proper understanding of consciousness

Once that occurs, the growth I mean through the use of the chart of of the complexity of life should occur.
The problem isn't in the machine, it's the "person at the keyboard." And people are far behind that curve of growth.

The thought it may happen in our lifetime can be very scary.

This is highly improbable, incalculably small.

*Edit: many "doomsayers" have been saying what you are for many years (decades?)... this is going to be on the far horizon for many to come.

 

[Carpe Stultus]

The truth is, whether we are happy about it or not.... many things point to the possibility that we will have to learn how to live with strong generalized AI in our lifetime. And it will change everything. If we don't know how to adapt it may disrupt societies to the point they are crumbling. How would you feed the millions and millions of people who are now out of a job, because strong generalized AI can do everything better, won't need vacation time, can work 24/7, doesn't get sick. Basic general income may be something societies have to look at, there are already experiments underway for that in several countries, because everything points to a future where there won't be enough jobs left for people. Even in the creative arts.

Some things go hand in hand with robotics, but we also make huge advances there, so once we have capable robot hardware and strong AI it's "goodbye" to the world as we know it. The thought it may happen in our lifetime can be very scary. We have to hope any strong AI will like humans. And that maybe it's not developed in China first. Or by Facebook. Maybe not even by Google. But chances that some corporation like Google or a country like China will develop strong AI first is pretty high.

And yes, some people don't know how much work it is to pose models, to tweak models, to make them look a bit more unique, to search for clothing, assets in general, add realistic lighting, maybe animate a few things, render everything and write the scenes to begin with.

AI's could do a lot of good things, if they would be used to make work easier and safer for people, but yes it could also destroy the lives of many by costing them their jobs.

Besides that, the possibility that a AI will make porn games in the near future goes torwards 0 simply because it is very niché and the people with the power over an AI like that would probably not make it accessable for everyone and their dog.