Make your own AI-generated art

[Aja]

I wonder if there's a parallel, dema, between using AI to create art and using a synthesizer or other software to create music. When I'm using my eurorack, I almost never have a sense in advance what I want to do. I experiment, mess around, and when I hear something I like, I focus on it and embellish it until I'm satisfied. I've heard it said that the key to being a good electronic musician is not necessarily mastery of an instrument but rather the ability to curate your own work, to know when what you have is good and what to do with it. In that sense using AI to create the kinds of videos that Azaran posted above doesn't feel especially different; it's all about experimenting with your tools and knowing when to keep what's good.

[demagogue]

As you know I synth a lot too, especially now with my Deluge. The thing is made for experimentation. Anyway that's why I stopped myself and said I recognize that using AI as a tool can be legitimate and inspired too. But I think there's still a difference between playing the tool and the tool playing you, so to speak. Really I think it's not the AI itself, it's a culture that can build around it where the user isn't really putting any thought into it anymore. They could, but in practice they're not as much.

I don't think it's even new. Even with a normal instrument, people might coast thoughtlessly playing, e.g., the pentatonic and few blues licks they know over and over. So it's the equivalent of that, but the difference is the AI can take that and add a lot of polish. I suppose there's technique in curating the sound over iterations, so they still could put thought into that. I guess the difference is there's less incentive to put thought into one's craft when you're starting out and you don't know what's what. With an instrument and especially synths, the learning curve forces you into uncomfortable places where you have to learn your way around. AI might trick many just getting into it that they don't have to go through that learning curve to get some result, even though a lot may still not be satisfied and want to push through learning the system better.

[Azaran]

They just updated our Dall E 3, and now it only generates about half the time it used to. Otherwise, it just hangs in generating mode forever, which I imagine is its deceptive way of censoring prompts.

I wonder what's so objectionable about "group of people dressed in black, inside a cave with frescos"?

The censorship is really getting absurd.

And they're not even transparent about what exactly they censor. I guess they don't want the public to know their inner algorithm?

Anyway, here's a redo of an old prompt I had done with Wombo years ago, dogs playing with sapphires on beach

[mxleader]

They just updated our Dall E 3, and now it only generates about half the time it used to. Otherwise, it just hangs in generating mode forever, which I imagine is its deceptive way of censoring prompts.

I wonder what's so objectionable about "group of people dressed in black, inside a cave with frescos"?

The censorship is really getting absurd.

And they're not even transparent about what exactly they censor. I guess they don't want the public to know their inner algorithm?

Anyway, here's a redo of an old prompt I had done with Wombo years ago, dogs playing with sapphires on beach

This is completely unrelated to this thread but I was in the middle of trying to defeat you last night during mission 10 of the Black Parade. Need to get back to it.

[demagogue]

Anyway this is cool. It took me a bit of time to even realize it wasn't real.
(It's 3D. You can scroll around. I don't think the forum code will allow me to actually embed it in this post.)

I'm wondering how long it will be before someone figures out how to generate 3D architecture and texture it based off of a Midjourney, et al, render. It could be amazing to walk around inside some of these worlds.

[PigLick]

Github has stuff like stable diffusion source code available. You can run it locally (if you have enough power) with absolutely no censorship, you can even train it with your own set of images. I have dabbled around with it a bit and it is rather frightening. The Taylor Swift porn thing is just the tip of the iceberg.

[Azaran]

This is completely unrelated to this thread but I was in the middle of trying to defeat you last night during mission 10 of the Black Parade. Need to get back to it.

So that's why I felt so tired yesterday!

Github has stuff like stable diffusion source code available. You can run it locally (if you have enough power) with absolutely no censorship, you can even train it with your own set of images. I have dabbled around with it a bit and it is rather frightening. The Taylor Swift porn thing is just the tip of the iceberg.

I tried it but I never got it to work, kept giving me python errors or something, eventually I gave up

[DuatDweller]

I keep trying to imagine Sylvester Stallone as the terminator, or at least appearing in a good movie as of lately he's been quite a let down.



"Like a good Neighbaaa!" -Arnie.

[Sulphur]

Well, I guess we've arrived.

The future's only going to get more chaotic from here.

[Azaran]

Well, I guess we've arrived.

The future's only going to get more chaotic from here.

Holy shit, the videos on that page

It's all over.

I'm telling you, it's gonna wreck the film and tv industry in the next few years

[demagogue]

I guess I posted some comments about this on FB and not here.

One, I'm thinking the last substantial step in this line will be rendering a dynamic 3D world populated with AI that you can walk around in. That will be something I'd be really interested in seeing, and that'll be my "we've arrived" moment.

Two, like with Midjourney et al, I think this tech will still be finicky for a while. Well, I was trying to make a comic out of Midjourney, and it was a serious challenge to make panels that were consistent and coheasive and blocked and framed to tell the overall story. I imagine that this will have similar issues, like it'll be good for these 10 second clips, but it looks like it'd also be a challenge to make an actual movie short out of them for the same reason it's hard to make a comic out of Midjourney or Dall-E2. But then I think that's the next bit of tech they may work on.

Three, I thought there's going to be a particular kind of backlash as this stuff gets churned out. I'm already seeing AI art all over the place and a backlash brewing in comments to it. I think it's not only going to be a privileging of non-AI art, but a movement towards schools or master-discipline-style circles that put relationships in the center of art production, so you know it's human; and art becomes more about the production side than the consumption side (which will be flooded with this stuff).

I think even expecting that to happen, though, it may still be a fringe kind of response. But it'd be one that'd interest me. Adorno wrote a book called The Culture Industry that developed some of my thinking on this kind of stuff, so I was trying to think about how the socio-economics of cultural production might evolve in light of this. If you divorce art from profit, is it even an industry anymore, and if it's not, what is it becoming, and is that good for culture or not? Questions like that come up.

[Sulphur]

What I mean by 'we've arrived' is that we've landed on the exact spot that we were predicting, which is the point where we can make complete fiction look convincing to the untrained eye. There's issues with those videos that you can spot if you pay attention, but the border between uncanny and realistic is pretty small now compared to where we were a very short while ago.

So yeah, you won't be able to make an entire film project with it just yet, because like you mentioned, shot for shot it just isn't going to be working with the same parameters to compose something consistent. But that's a problem for the future. Right now, with just ten second clips, I'm more intrigued by the fact that this will enable people to churn out an exponentially higher amount of awful and dangerous dis/misinformation at will to make the internet an even bigger wasteland than it has ever been. It's still missing voice generation, but that's probably just waiting to be welded on.

[Azaran]

Three, I thought there's going to be a particular kind of backlash as this stuff gets churned out. I'm already seeing AI art all over the place and a backlash brewing in comments to it. I think it's not only going to be a privileging of non-AI art, but a movement towards schools or master-discipline-style circles that put relationships in the center of art production, so you know it's human; and art becomes more about the production side than the consumption side (which will be flooded with this stuff).

I think even expecting that to happen, though, it may still be a fringe kind of response. But it'd be one that'd interest me. Adorno wrote a book called The Culture Industry that developed some of my thinking on this kind of stuff, so I was trying to think about how the socio-economics of cultural production might evolve in light of this. If you divorce art from profit, is it even an industry anymore, and if it's not, what is it becoming, and is that good for culture or not? Questions like that come up.

So I will say AI may be good in the long run for real artists trying to make it - 'real' art may be more valued. I have two people close to me who are painters and well tapped into the art scene, and they've told me how toxic the art world is. Your talent has almost nothing to do with how successful you are; most artists who make it do so because they endear themselves to X prominent gallery, which then propels them to fame. "Don't show me what you paint; tell me what bigshot gallery owners you're cozy with" in a nutshell.

This is even more common with ab-ex\extreme minimalist art, which anyone can make, and is entirely dependent on a narrative or underlying symbolism, not the work itself. The other day I was at a gallery, and one work there was just a blank canvas....by a very successful abstract painter who makes a living from her art.

Then again, in a world where there's SO MANY artists, this type of thing is expected.

[demagogue]

What I mean by 'we've arrived' is that we've landed on the exact spot that we were predicting, which is the point where we can make complete fiction look convincing to the untrained eye.

Ah yeah, that reminds me, the first time I saw screenshots of Gothic II, that was the first moment that for a fraction of a second I thought a game scene looked real, but it didn't last long. And I thought, someday there's going to come a game where I can't tell it's not real except for a fraction of a second when I notice some glitch. And that came for me with the 2020 Microsoft Flight Simulator, and I thought, welp, we're there now.

I don't know if this tech will be consistently at that same kind of level, but I think in the hands of someone that knows what they're doing, it will be good enough for a lot of purposes. Although I think a lot of people's AI-sense will be honed, like it is for AI art. Actually I can already see a divide, with some people really attuned to AI art and other people that take everything at face value, even if they do suspect it. That latter group has me a little concerned, if it's paired with them being easily worked up, which it often is, I think.

[Sulphur]

I mean, if we're talking video games, you can all the way back to an old LGS game, Flight Unlimited 2, for pushing 'realism' in graphics. It was pretty good for its time. But that's what video games are good at, things that aren't people. The specific problem Sora is tackling is people, and while it still doesn't do hands and other things right at various points, its model interpolates human faces and details like skin really, really well. Take a closer look at some of those people-centric videos; if you don't peer at them too hard, the uncanny valley isn't in effect until your brain turns on and goes, 'wait, why is that reflection, no that walk...' and then it sort of breaks down. That it's gotten to this point at all is, on its own, both impressive and one of the largest issues we might ever face as it gets better.

[PigLick]

I shudder to think how much energy was used to create those videos. I honestly don't think sora will be released into the wild for quite some time yet, this announcement was probably just to jack up the share prices.
Is Openai going to be gatekeeping this tech?

[Azaran]

Well, I guess we've arrived.

The future's only going to get more chaotic from here.

Gotta love the 'safeguards':

We’ll be taking several important safety steps ahead of making Sora available in OpenAI’s products. We are working with red teamers — domain experts in areas like misinformation, hateful content, and bias — who will be adversarially testing the model.

We’re also building tools to help detect misleading content such as a detection classifier that can tell when a video was generated by Sora. We plan to include C2PA metadata in the future if we deploy the model in an OpenAI product.

"Ok guys, I know we gave out nuclear warheads and bioweapons to the public, but we also made protective suits!"

[mxleader]

I was chatting with Gemini this morning and I asked it to translate several insults from German to English and it started showing the translations and telling me how bad they are to say. Then on the last request it spit out the English translation along with a few sentences in Western Frisian but in Chinese characters. I think it was getting mad at me.

[Azaran]

Using Ideogram

[Azaran]

My thing now is generating new paintings by Old Masters.

Far better nowadays

[mxleader]

[SD]

So it turns out AI has a sense of humour

[DuatDweller]

Descarted I mean Discarded.

[Azaran]

[demagogue]

Dream Machine is pretty good.

For my first experiment, I was trying to reproduce the scene of the 1688 acquittal of the Seven Bishops in England (I was reading about it; never mind why), and you see the seven bishops and three judges. It doesn't distinguish them all too well. But I can see how I should have been more detailed in my prompt. But all the same, it's looking good.

What's funny to me is that it looks like a 1950s movie of the scene, rather than say contemporaneous paintings of the scene or scenes like it, as I gather there are more data points on the former.

https://lumalabs.ai/dream-machine/cr...c-b7df93ca4fa0