Reply With QuoteOr you could input song titles.
I think it says Lufoil Acia? Maybe we should stop for directions.
Wouldn't you know it.
You could be whimsical.
And if you have a zombie girlfriend by Valentines...
Or you could input song titles.
I think it says Lufoil Acia? Maybe we should stop for directions.
Wouldn't you know it.
That is a great idea! You can generate your own album covers. Why didn't I think of that?Originally Posted by Tocky
"Ninety percent of everything is crap." - Theodore Sturgeon
"Sturgeon was an optimist." - me, just now
System Shock 2 Walkthrough
Should have used the festive one. I've typed in all sorts of horror and it comes out cotton candy! Hours of fun. On the other hand I can't get "raining bloody axes" to show any recognizable body parts in even dark fantasy. And "pile of burning coffins" was also a disappointment. I was going to call it "Strange Eons" all literary like and had it in my head but the damn thing can't produce a coffin if it's life depended on it. No circle of people around bonfire of coffins in moonlit meadow or any other iteration would produce what I envisioned. I'll have to paint that one myself I guess.
Meanwhile here is festive armageddon-
I imagine their limiting the input images in the learning phase so they can keep it from churning out explicit pornography, gore, swastikas and the like, just even on pure liability grounds. But even within those limits you can get some explicit things wedged in if you play with it long enough.
I was thinking, if they wanted to take this the next step to have some more representational elements, like actual humans and buildings with structural integrity and not gaping holes or glitches, is you'd add extra layers of abstraction on top going into the design... We know that these neural nets can already procedurally make life-like humans and buildings. What they could do is generate those wholesale, and then drop the people and buildings in in this dreamy procedural way, so you could get a spectrum from really surreal like these over to representational art like a Hopper or Wyeth painting, and gradations in between.
I think that would lead to some really top shelf stuff.
While I'm posting, and on that note...
But coffins? Where's the fun?
One more because I found it amusing.
Kitties!
I recently learned about This Person Does Not Exist, and as I scanned through the images I felt an inexplicable growing sense of dread, as though it was a gallery of deceased.
Oh, and while we're at it, this cat does not exist, either.
I bet they have or do or will. There are only so many possibilities. On the other hand I've yet to see another Elvis.
Anyway it does understand there are some differences between the words you use.
It's insane how this technology has evolved in just a year or so of being available.
Using Dalle 2, Deviantart and Simplified AI:
Oh it is much less abstract now. And those bronze sculptures are very realistic.
Here is a try using previous words. So much better.
I'm sorry I can't seem to downsize this at work. I'll try once I get home.
Yeah Dream really improved. I found the Realistic, HDR, and Diorama styles give the best results.
Google is set to release the ultimate AI generator, Imagen, at some point, but it's not yet available to the public
Last edited by Azaran; 18th Nov 2022 at 16:46.
For anyone that doesn't like messing with Github, someone's finally come out with a version of Stable Diffusion called ArtRoom that you can just directly install locally like a normal app. https://artroom.ai/download-app
The model is something like 7GB, so you need some space for it.
I guess there are online versions that are better and more open than they were before. But I like having a local version anyway.
Another tip that's making the rounds these days are negative prompts. That's how you get rid of the weird eyes and extra arms and fingers and other artifacts. But you need to make sure the system you're using has a way to enter negative prompts.
The default ones for ArtRoom are: lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry.
----
The thing is, I think I mentioned this already before ITT, but I can already predict a kind of over saturation really soon that's going to change the scene. My AI Art folder already has well over 1000 works. They're almost all stunning. But there comes a point where ... well there's a theory behind this. As the generation of a thing becomes cheaper, at a certain point you don't have to collect the works. You just generate them on demand and trust they'll always be there. It changes the way you think about what art even is and how you interact with it. I think soon enough this is going to become a thing for most art forms. People will just take it for granted that the art or music or movie or whatever it is is just one click away when they want it. It's going to diminish the value or even point of saving art as a unique artifact in itself, since 4000 just like it are just a click away. I don't even know if that's as bad as it sounds, but it's inevitable I think. I also think at some point there's going to be a backlash where's going to be a fetish for "authentic creation", and a whole movement coming out of that.
Well, who knows what will happen. Who could have predicted the last year? We'll see.
I imagine this will spell doom for graphic designers and illustrators
Or, after looking at enough output from an AI, we'll begin to spot its tell-tales, and grow tired of it and move on. The earlier output from Dream already looks very samey now. And the set of generated images that you posted above all have perspective issues. I'm impressed by the sculpture though, unless the AI just copied something from its database.
But setting aside the recent NFT bubble as a one time thing, digital art has never been particularly valuable. In digital form, we use it mostly for background wallpapers. There are endless numbers of those available for free on image sharing sites. Some are used to make cheap prints sold at gift shops and novelty stores, but those have no lasting value. If I had an AI image generator and wanted to make money, I'd train it to generate logos.
In order for a piece of art to have value it has to be tangible, unique, and have a story behind it.
Last edited by heywood; 19th Nov 2022 at 12:12.
Well, there's nothing stopping someone from eventually using GPT-3 or a similar model to also generate a unique story for a piece of art. If you want tangible, you can just feed it to the next version of Ai-Da.
What dema's alluding to is that any creative art form has the potential to be colonised by procedurally generated products from a sophisticated enough neural network or collection of AIs. It's inevitable that, in this version of our reality, that there's going to be a remixer that'll just take our parameters for entertainment and generate something we'll enjoy on the fly. Given enough training data, we could probably see something as good as a Dan Brown novel come out in the next five to ten years from one of these machines, which is very low-hanging fruit but paves the way for something possibly more profound eventually. Infinite content, but not exactly the way Arcade Fire imagined it, I'd wager.
The fetishisation of 'genuine' art already exists, dema, it's only going to be that in the divide of machine-generated remixes and human-produced art, we'll end up wanting to pledge our allegiances to one or the other, and quickly realise that in a world where it's genuinely difficult to distinguish between the two, and who's conning whom, it won't matter to the average person. While that's depressing, I think that once we grapple with the idea of what 'meaningful' really is in that future of extreme over-saturation and overloaded filters (which we've already begun on our own), we'll hopefully come out the other side a bit more evolved.
It's not "depressing" if the models can evolve by themselves.
It's depressing if they target only our entertainment, being something like a human mind eternal nursing (=Huxley's Soma)
It is a neat technical accomplishment to be able to generate an interesting image with as little as two words. But it also makes for a very shallow art form when the artist is limited to just a handful of words and parameters for input. The creative process is basically thinking up a general idea, and then trial-and-error testing it, and repeating until it spits out something you like. When it does, you can digitally share it with others, and they can have exact copies. That's only a step removed from mass produced art.
We already spend our lives surrounded by cheap art, so what's the significance of one more method for producing it?
I don't see it devaluing existing forms of art. The value of an original piece is determined by a lot of things: the stature of the artist, the story behind the work, the significance of the work to other artists, and the number of people who have seen it. It's not so much the physical item as the human story that goes with it.
One of my favorite artists is René Magritte. His paintings are technically and conceptually simple enough that any halfway decent painter could emulate him convincingly. Perhaps someday soon an AI will generate something in a similar style. But even if it could generate a new Magritte on demand, it wouldn't be of any interest to me. I like Magritte because of the questions he was asking through his paintings, in the context of when he painted them, as a member of the surrealist movement in Paris during the modernist period. Removed from that context, his best works are still clever. But in that context they are subversive.
I can see AI art becoming useful for people who wish to give form to their fictional works. E.g., writers of illustrated fantasy novels. The only issue is getting the image style to match. So for instance it's easy to generate the image of a castle. But then if you need a second image of that castle aflame, it may be tricky to get the AI to generate a matching castle, and could require dozens of tries. Unless you use a base image.
Every major movement is historically contextual. If Picasso/Magritte/Monet, &c emerged today, they'd be mere drops in the 21st century ocean of excess. You'd have to scream EXTREMELY loud to be heard in today's cacophony, and even then you'll be forgotten within a week thanks to the endless stream of info packed feeds we're all exposed to.
There are a lot of artists who failed to stand out in their time and were only recognised later.
This tech is leveling up in leaps and bounds. Midjourney v4 is coming on line soon, and lots beta images are coming out. I think one commenter summed it up by saying it's actually what people imagined Dalle 2 was going to be before they actually used Dalle 2. From what I've seen it's a big head above its predecessors. What's striking is just how hard and fast it's coming.
I remember like in the 1990s imagining what the 21st Century was going to be like, for me and the world, and my imagination carried me about to the 2020s, after which things got blurry. More or less things went as I imagined they would. But sure to form, I feel like we're in uncharted territory now. I don't know what to expect from anything.
Holy crap
I misread this as your talking about the normal creative process; the reason I did that is because this describes literally any creative process. Any creative person does exactly this.
The value of art is entirely notional, as you've outlined, and what you're talking about is the rarified medium of exhibited pieces in a gallery somewhere. They're important, but also not consumer entertainment, where the bulk of our speculations lie. Having said that, let's pivot to capital A Art and play the devil's advocate: why couldn't art created by a machine not have context and story attached? Would it be less worthy of the moniker of art because it wasn't created by human hands? What if the story of the neural network that birthed singular pieces of art was along the lines of something that it showed a preference for a subject, like a naiad, and spent the rest of its existence portraying them in various lights (and prolific excess) like a person smitten? Is the thing that provides meaning to art simply the fact that humans created it, or is it what it provokes in the viewer along with its unique context, human or not?We already spend our lives surrounded by cheap art, so what's the significance of one more method for producing it?
I don't see it devaluing existing forms of art. The value of an original piece is determined by a lot of things: the stature of the artist, the story behind the work, the significance of the work to other artists, and the number of people who have seen it. It's not so much the physical item as the human story that goes with it.
Stable Diffusion 2.0 meanwhile is having a rocky go of it. They're saying it's taking one step back so it can take 2 steps forward in the future, but it doesn't look good in the meantime, however long that lasts. Then again, it's free & open source.
Edit: To explain a bit more, they're replacing their entire model with one completely empty of potentially copyright protected images, e.g., anything by any artist after the 1930s, and most photos of people, which guts an awful lot out of it. And not just marginal content, but the really core stuff people would need to make anything useful with it.
Last edited by demagogue; 25th Nov 2022 at 07:51.
It's official. We're living in the Matrix
Posting Permissions
