Midjourney AI create art for events

[StupidDupes]

Just wondering if this can be integrated into the game. Create art from event descriptions.

[Morkonan]

Originally posted by VoiD:

I'd agree if you couldn't create, exactly, what you wanted through AI art just as well, I see it as a tool just like photoshop, it's just better, if AI art isn't art, photoshop isn't either, and neither is 3D modelling, or anything else.

"A.I. "Art"" is not "Art" without the human component and the "intent" behind the creation of it.

It's like a musical instrument. It may be capable of achieving notes a human can not, but it has no intent or ability to derive that, itself, to express anything, itself.

As your description shows, once the human component is removed, (and the intent, IMO) "everything" becomes "art" and that's not a definition of "anything."

Photoshop is a brush, paint is paint, a canvas is... whatever, and an A.I. Picture Maker are all just "tools."

.. and have literally been accepting anything as art,

This is a real issue.

We know that people will pay money for just about anything. I know for a fact that there has got to be one dude out there that would pay good money for a lump of cold crap... and he'd call it "art." (Probably has an extensive hospitalization history...)

If one goes to Instagram, there's fifty-eleven people there that claim they are celebrities and "influencers" and they don't have five ounces of skill or talent or craft between them... 'cause a bunch of people apparently just clicked "Follow."

Art can be crap, but all crap isn't art. A boatload of "artists" create crap, and we can say it's "art," but that doesn't mean its worth the effort they put into it. Right now, half the kids in the US have macaroni-art on refrigerators that is of better quality and presented with much more meaning than most of the people with Deviantart accounts.

This is bad because some "artists" build themselves up due to a few comments and never work on actually improving their skills. Some of them end up slitting their own wrists... because what they thought was true about their claims of talent and significance just wasn't true... because they didn't bother trying to learn to accept valid criticism or seek out appropriate instruction and knowledge.

Then, there's... NFTs. Which, is like... just no. Not gonna. :)

...while using AI generation does take a lot of technical knowledge instead of motor skills, very similar to 3D modeling training.

Dunno, there. Writing good prompts does take knowledge and acquired skill, that much is true. Now, about the back-end, I don't know much about that other than human-trained AI. (For 3D, I know a bit, but am not quite sure how you're applying it there.)

...Honestly, invokeAI is much easier to use, but it's rather limited, so you'll sort of end up at the mercy of the AI's references, you can use several functions, like drawing a stickman and using it to set up the exact position of your characters, or a rough sketch, and I'm not sure if it can be used to train your own patterns to teach it how to use a face/your face, or your own characters with it, but if you're not willing to be super hardcore it's a great program, once it evolves and can incorporate some of the features from automatic1111 then it will be much better. There are some videos showing how to quickly install and set it up but you can simply download it through the ItchIO platform too.

Interesting! So, you can use some kind of "guide" to give it some basic visual framework in that will be used for the final compositions?

The other one is... A pain in the ass to install, I don't quite understand every step in the process, so I'll just leave it here and hope it works for you as well: https://www.youtube.com/watch?v=3cvP7yJotUM

I Think Automatic1111 is also a bit heavier on the requirements, but I'm not sure, in any case look through the Checkpoints in Civitai to select a style you'd want to use, the base SD1.5 seems more focused on photorealistic images, which isn't very helpful if you want to do D&D style aventurers, or 3D fantasy, or anime, etc... Just keep in mind InvokeAI can't use most files you'll find there other than Checkpoints and text inversions, at least, since I last checked.

Cool, thanks!

All I'm really looking for is something I can throw words at and get a bunch of "ideas" from that I can then choose, fine tune the words, run again, trying to narrow down what I want for some visual styles to use as concept art for 3D projects. Pushing verts (3D) is my way of relaxing, even if it sometimes gets my blood-pressure up. (Last week I spent entirely too much time trying to organize some data points to then apply the sets to use with a programming language I didn't know to then give me a software tool I wasn't quite sure how I was gonna use... that certain Universities already have, but I wanted to do myself... 'cause I'm a stubborn idiot.)


On the OT, to be OT:

There's a bunch going on in our culture with AI, ChatGPT et al, that gamers are very quickly picking up on as "Holy Grail" solutions, envisioning them like "Star Trek: The Next Generation" "holo-suite" fantasies to yield the "Best Vidyo Gaem Evar."

There are a number of issues that need to have some caution put on them. The most Important that i can think of for people around here is:

Don't trust anything.

This is a high-profile netculture thing and that will attract predators who would love enthusiasts to download and interact with their creation to get the "perfect thing they always wanted." It happens every single time. All the time, every time, these trends are exploited and the marketing around them is rampant. Do not download and install junk that promises "everything" unless you have very good evidence it is trustworthy. The same goes for using online tools and phone apps - Be careful. Everyone with bad intent out there will most certainly be trying to entice gamers to "play a game you create with your own imagination using our new AI, just click here, download, install, ignore the "Warning" from your Anti-virus because it's stoopid and not as smhart as our AI, and get the most awesomest gaming experience evers." (Not saying any game here does that, but there will definitely be plenty of exploits riding on this meme.)

[VoiD]

Originally posted by Morkonan:

Interesting! So, you can use some kind of "guide" to give it some basic visual framework in that will be used for the final compositions?

That's what the big, general checkpoints are, trained in very specific styles (or not, you can train it to do a bit of everything too).

Then there are the inputs, and then there's the order of inputs, and then there's the weight to each input which can be both positive or negative, so say you have an image, let's ignore CFG, seeds and negative prompts for now, picking a random picture from civitai using these prompts:
yorha no. 2 type b, 1girl, absurdres, black blindfold, black dress, black hairband, blindfold, blue sky, boots, building, city, cloud, covered eyes, debris, dress, feather-trimmed sleeves, feather trim, from below, gloves, grass, hairband, high heel boots, high heels, highres, juliet sleeves, katana, leather, leather boots, long sleeves, nier \(series\), nier automata, outdoors, overgrown, pod \(nier automata\), post-apocalypse, puddle, puffy sleeves, rubble, ruins, scenery, sky, thigh boots, thighhighs, thighhighs under boots, water <lora:yorha_noDOT_2_type_b:0.5>

So let's say the AI just isn't creating the gloves, even though you're using a lora for the character 2B that should wear gloves, you can increase the weight of gloves with () or even more with (()), ((())), if you're using automatic1111 you could even use numeric weights for 1:2, even crazy weights that tend to bug the picture going over 2.0, every input is weighted, and creating a picture from nothing exactly how you want it requires you to change the order and weights accordingly.

If you're trying to create different chracters the AI often can't really do it very well, as it's using the same seed for both, they tend to look very much alike, take this picture I've made as an example: https://pbs.twimg.com/media/Fr8ABgVWIAAlVI5?format=jpg&name=large

I tried inputing different characters, different hair colors, different sizes, etc... But the results aren't ideal.

So, to create these compositions you'd probably need to create one image, with 1 character, or one big background in the way you want it, then inpaint an area with a mask telling the Ai only to change that area, use a different input and a different seed to generate a character there, doing what you want it to do (or use dreambooth to use a stickman to get the exact position, or aides for hands), then inpaint another area and repeat the process with different inputs, then keep adjusting until it's the way you want it to be.

You can potentially inpaint to get everything exactly right from the minimal details, or just attempt to generate one big perfect picture, or just slightly alter an image using image to image several times, the less work you put into it, the less control you'll have over the final product.

Also, higher CFG means closer to your inputs, too high and it won't know what to do unless you're extremely descriptive to the most minimal details, low CFG takes your inputs more like suggestions and tries to create something closer to it's references.

Originally posted by Morkonan:

Cool, thanks!

All I'm really looking for is something I can throw words at and get a bunch of "ideas" from that I can then choose, fine tune the words, run again, trying to narrow down what I want for some visual styles to use as concept art for 3D projects. Pushing verts (3D) is my way of relaxing, even if it sometimes gets my blood-pressure up. (Last week I spent entirely too much time trying to organize some data points to then apply the sets to use with a programming language I didn't know to then give me a software tool I wasn't quite sure how I was gonna use... that certain Universities already have, but I wanted to do myself... 'cause I'm a stubborn idiot.)

yeah, it's a good hobby for sure, real pain in the ass to add any extensions or anything else to it, but other than that it's pretty cool.

[Henry of Skalitz]

Originally posted by Morkonan:

Originally posted by ᲼⁧⁧Smelly Fish:

...
This game for example has everything randomly generated by AI as you play:..

Not "everything." It is not writing game code and new mechanics determined by an AI. It has to operate from within a set of rules. Sure, a lot of the content may be generated by an AI as well as some of the variables that are already coded into the back-end mechanics, and basic operators for combat and the like, but read the description... It's a "text-based game."

Introducing AI Roguelite, the world's first text-based RPG where every location, NPC, enemy, item, crafting recipe, and game mechanic is 100% determined by artificial intelligence.

Some "AI Everything Game" writing its own executable code by itself on my computer is going to cause fire and brimstone to fall upon it from on high...

Hey, it's an interesting premise and a nice achievement, no doubt. But, it's not "everything."

ChatGPT could be incorporated into Rimworld right now as long as Micro$oft allows its free and open use... Some modder out there may have already done it. (So many reports of modders doing that in games now that I've forgotten who's done what.)

Okay I maybe should have clarified that I meant the playable parts of the game and not actual inner workings of it :P

[Morkonan]

Originally posted by VoiD:

...
That's what the big, general checkpoints are, trained in very specific styles (or not, you can train it to do a bit of everything too)....
yeah, it's a good hobby for sure, real pain in the ass to add any extensions or anything else to it, but other than that it's pretty cool.

Thanks for that demo pic! Nice!

@$^@^ I'm old... The first thing i looked at in detail were the hands. :)


I don't know the characters and/or may not have been able to correctly pick their prompts from your example. But, from what I interpret and have seen in various walkthroughs/demos, I understand that prompts can be weighted and a number of parameters set. What I've seen detailed demos of usually included a GUI, but command-prompt-shenanigans are also popular. :)


These characters used - The AI was trained on these specifically? Is that why you couldn't get them to change?


I've read that one can train some AI schemes using image prompts. Sort of like one would do with image fakes, but that's really a whole different thing. (I've looked into it as well, for funsies. Neat, but dang the local training time if you want to use your own. Aint nobody got time for that...)

Condensed for brevity:

I went down a 3D rabbit hole last week, trying to create controls for a control mesh that could deform a shape in a certain way using real-world measurements. The problem is solvable, just not quite yet. Decent results could be achievable, but not with the precision I would think would be worthy of the attempt, hence why I am not doin' 3D stuffs atm and posting on Steam, instead. :)

But, here's the rub:

I don't know the GPT dataset and what it was trained on, but at the end of the day a point-cloud in 3D space is simply another language that describes a solid shape... Many filetypes are easily capable of being edited with any text editor. I have no idea if such an AI has already read those or not. It's not English, but it is a sort of "language." It's the language of "shapes." (kinda...I'm not a topologist.) With the revolution in photogrammetry and athropometry, data is out there waiting to be read. An AI like the GPT stuff should be able to eventually use it effectively if properly trained.

Personally, I think much of the conventional use of 3D modeling today is eventually doomed. I've long held the opinion that revolutions in 2D will replace it. That is proving to be true, or will be. (Predicted long before image fakes or any image generation was ever in the news. Predict enough stuff, eventually you'll get some right... ;) )

[Morkonan]

Originally posted by ᲼⁧⁧Smelly Fish:

..
Okay I maybe should have clarified that I meant the playable parts of the game and not actual inner workings of it :P

:)

Fair enough.

But, that's like saying "want car, go fast" and getting a fast car that goes fast, but has no brakes or sets itself on fire... :) Somebody's gotta think of that stuff to make the thing "do fast" without 'sploding.

I bet you could probably stuff a couple of z-levels into Rimworld and still make it "work." I don't know how many that old mod used, but that seems about right.

[VoiD]

Originally posted by Morkonan:

Originally posted by VoiD:

...
That's what the big, general checkpoints are, trained in very specific styles (or not, you can train it to do a bit of everything too)....
yeah, it's a good hobby for sure, real pain in the ass to add any extensions or anything else to it, but other than that it's pretty cool.

Thanks for that demo pic! Nice!

@$^@^ I'm old... The first thing i looked at in detail were the hands. :)


I don't know the characters and/or may not have been able to correctly pick their prompts from your example. But, from what I interpret and have seen in various walkthroughs/demos, I understand that prompts can be weighted and a number of parameters set. What I've seen detailed demos of usually included a GUI, but command-prompt-shenanigans are also popular. :)


These characters used - The AI was trained on these specifically? Is that why you couldn't get them to change?


I've read that one can train some AI schemes using image prompts. Sort of like one would do with image fakes, but that's really a whole different thing. (I've looked into it as well, for funsies. Neat, but dang the local training time if you want to use your own. Aint nobody got time for that...)

Condensed for brevity:

I went down a 3D rabbit hole last week, trying to create controls for a control mesh that could deform a shape in a certain way using real-world measurements. The problem is solvable, just not quite yet. Decent results could be achievable, but not with the precision I would think would be worthy of the attempt, hence why I am not doin' 3D stuffs atm and posting on Steam, instead. :)

But, here's the rub:

I don't know the GPT dataset and what it was trained on, but at the end of the day a point-cloud in 3D space is simply another language that describes a solid shape... Many filetypes are easily capable of being edited with any text editor. I have no idea if such an AI has already read those or not. It's not English, but it is a sort of "language." It's the language of "shapes." (kinda...I'm not a topologist.) With the revolution in photogrammetry and athropometry, data is out there waiting to be read. An AI like the GPT stuff should be able to eventually use it effectively if properly trained.

Personally, I think much of the conventional use of 3D modeling today is eventually doomed. I've long held the opinion that revolutions in 2D will replace it. That is proving to be true, or will be. (Predicted long before image fakes or any image generation was ever in the news. Predict enough stuff, eventually you'll get some right... ;) )

I don't think that said character exists anywhere, it's often harder to tell with stylized art like anime or cartoons as they don't have as many details as a photograph (which can also be generated), but I remember adding some specific information about clothes, the situation, the general mood of the picture, etc.. so I doubt I'd find it anywhere, but it's possible since I wasn't nearly as specific as a single character picture where you often detail every single part of the body, outfit, measures, positions, etc...

The reason why the faces are often the same, or very close to the same is has probably something to do with the seed.

If you pick the exact same prompt, the exact same settings, with the same seed and using the same hardware (the GPU can change the results) then you'd get the exact same generation unless you add a variation setting, so that one seed is generating a character that fits the description, it just does it twice, I've managed to create some sort of different characters, but they are always very close because of that seed, that's why I'd need to generate one, change the seed, and generate another to have a real difference.

Ofc this is subject to change in the future, this technology is quite young and trained on small pic sizes right now.

And yes, you can train the AI in several ways to create different kinds of files, some checkpoints affecting everything including the style, or more specific inputs meant to simply create a specific character in any way you want, which you can teach it, but I haven't messed too much with it.

Here are a few examples:
https://pbs.twimg.com/media/FsA6vFwWcAIQp1N?format=jpg&name=medium
https://pbs.twimg.com/media/FsAinEdWwAA75pe?format=jpg&name=medium
https://pbs.twimg.com/media/FsAf3DMWYAgO6iA?format=jpg&name=medium

Here's with a different checkpoint and a heavy handed lora meant to create gesugao pictures, very clearly inspired by ebiblue:
https://pbs.twimg.com/media/FsASL0FXoAAZt9e?format=jpg&name=medium
https://pbs.twimg.com/media/FsASNpVXwAAS3mO?format=jpg&name=medium
https://pbs.twimg.com/media/FsAbw_fWABoyWWr?format=jpg&name=medium

Of course the general style exists, the expression exists, the character exists, but there was never a picture like those last 3 before

Of course if you wanted to create something nonsensical just to make a point you could create a new background completely unrelated to nier automata, use a stickman to set a slavic squatting pose, throw in a gesugao lora with the 2B lora to make a squatting gesugao 2B over the ocean holding a frying pan or something.

Edit: More pictures as an example using the same seed, with the same checkpoint:
https://imgur.com/a/NO4CDKC
The only thing I changed was an imput for "Rias Gremory" from Highschool DXD, and a lora file that contains all of the characters from that anime, it changed the picture slightly, which could have been prevented if I was a lot more descriptive (or used inpainting to keep everything but the character the exact same), but the AI clearly goes by the general idea of a picture, I could change up the character again and again, and even without a stickman or a sketch to guide it, the composition would probably be about the same with this exact same seed.
3rd pic adds Rias Gremory and Red Hair without the lora (low weight), I'm fairly confident whoever trained this file probably did it without many/any clothes on the characters, which makes the AI more likely to create lighter armor and show more skin.
4th and 5th pics are rias with then without that same lora, but having a 2.0 weight on medieval armor, it ended up changing the composition a bit, but they are still both similar to each other.
Oddly enough she gets a closer body type to the original without the low weight dxd characters lora.
Last is a picture of the exact same prompt from the 5th picture, in a completely different checkpoint trained in other things, with other styles.

Edit2: Added yet another repeat of the last prompt on a different prompt, this one isn't trained in anime at all, somehow it knows how Rias should look like, and it's accepting the lora.

The prompts, if you're curious: (masterpiece), (extremely intricate:1.3), (realistic), portrait of a girl, the most beautiful artwork in the world, (medieval armor:2), metal reflections, upper body, outdoors, intense sunlight, far away castle, professional oil painting of a stunning woman by Ed Blinkey and Atey Ghailan and Jeremy Mann and Greg Rutkowski, detailed, sharp focus, dramatic, award winning, matte drawing, cinematic lighting, octane render, unreal engine, volumetrics dtx, rias gremory, red hair, <lora:dxdAllGirlsGasperRiasSonaAkenoAsiaTsubaki_dxd:0.4> full body,

Took them from some picture review on civitai at random.

[corisai]

Originally posted by VoiD:

Yeah, no, that's false, mess around with it for a little bit and you'll realize

I know a bit about how it's really working instead of rely on feelings like you :)

Originally posted by VoiD:

and if you do it right it all ends up perfectly coherently

YOU are acting creatively here and YOU are a source of inspiration. Midjorney only drawing under your command - just a tool and nothing more.

Originally posted by VoiD:

Ai generation is just a tool for artists, and it enables people without physical skills to create something nice, something new, something unlike anything ever done before.

SIgh. Maybe you should learn first how neutal nets are working? They're not AI (even weak AI) now. Yes, they are a great tool in skilled hands - but nothing more now.

And no, Midjorney like any kind of NN is unable to draw anything new, anything in style she never saw an example in learning data set.

P.S. And should I mention that real art have a some idea behind? That's currently absolutely inaccessible to machine. That's why always to fun to watch on @#$% trying to "discuss" things with GPT-bots and not understanding that is exactly like speaking with mirror.

[VoiD]

Originally posted by corisai:

Originally posted by VoiD:

Yeah, no, that's false, mess around with it for a little bit and you'll realize

I know a bit about how it's really working instead of rely on feelings like you :)

Originally posted by VoiD:

and if you do it right it all ends up perfectly coherently

YOU are acting creatively here and YOU are a source of inspiration. Midjorney only drawing under your command - just a tool and nothing more.

Originally posted by VoiD:

Ai generation is just a tool for artists, and it enables people without physical skills to create something nice, something new, something unlike anything ever done before.

SIgh. Maybe you should learn first how neutal nets are working? They're not AI (even weak AI) now. Yes, they are a great tool in skilled hands - but nothing more now.

And no, Midjorney like any kind of NN is unable to draw anything new, anything in style she never saw an example in learning data set.

P.S. And should I mention that real art have a some idea behind? That's currently absolutely inaccessible to machine. That's why always to fun to watch on @#$% trying to "discuss" things with GPT-bots and not understanding that is exactly like speaking with mirror.

tell me when you have something to say that I haven't already covered

[Lord Freedom]

Originally posted by Nyx (She/Her):

you're very clearly poisoned already
AI "art" is a tool of art theft and cannot generate anything New

Art experts would likely disagree with you that there is anything in human history that is truly original. There's a reason why our laws defining what is 'art' are so vague and require the bare minimum input on behalf of the artist. Everything you do is a copy, because you cannot escape the exterior influences. Art history has been a progressive history built upon the past.

Although for example, sometimes nature forms into something beautiful, but we don't call that 'art.' Do we then agree to define 'art' as anything made by humans?

If we compare the people who created the AI tools to people who make brushes, paint, etc (not the actual artists), then is AI just a 'tool' that does not create art... WE still create the art.

What then, if AI creates the art without our input? Long after the creators of the tool have passed... Is that still art?

...Or a monstrosity?

[corisai]

Originally posted by Lord Freedom:

What then, if AI creates the art without our input?

Strong AI now is a fairy* tale.

*actually a nightmare. Because people don't understand that to make a strong AI we need to know exactly how our mind is working and that will open a pandora box: mind control, brainwashing, digital upload, etc... So waaay before we will see a strong AI we will encounter a trully horrible things.

[Lord Freedom]

Originally posted by corisai:

Originally posted by Lord Freedom:

What then, if AI creates the art without our input?

Strong AI now is a fairy* tale.

*actually a nightmare. Because people don't understand that to make a strong AI we need to know exactly how our mind is working and that will open a pandora box: mind control, brainwashing, digital upload, etc... So waaay before we will see a strong AI we will encounter a trully horrible things.

We thought AI would do all the menial tasks, making our lives easier and able to spend more time on art and hobbies. Turns out it's the opposite. It will take away our creativity, expression, thoughts, and our bodies will be still stuck doing the hard work struggling to survive and pay the bills.

[whatamidoing]

Originally posted by Lord Freedom:

Originally posted by corisai:

Strong AI now is a fairy* tale.

*actually a nightmare. Because people don't understand that to make a strong AI we need to know exactly how our mind is working and that will open a pandora box: mind control, brainwashing, digital upload, etc... So waaay before we will see a strong AI we will encounter a trully horrible things.

We thought AI would do all the menial tasks, making our lives easier and able to spend more time on art and hobbies. Turns out it's the opposite. It will take away our creativity, expression, thoughts, and our bodies will be still stuck doing the hard work struggling to survive and pay the bills.

Welcome to capitalism, where automation becomes the enemy of workers instead of being a tool to let people work less.

[corisai]

Originally posted by Lord Freedom:

We thought AI would do all the menial tasks, making our lives easier and able to spend more time on art and hobbies.

So basically viewed AI as Slavery 2.0? Huh. Ironic in that case & west people are never change.

[Morkonan]

PS: Gettin' pretty OT, so...

Any chance of a lora or other prompt/guide/style set that can generate "Rimworld Characters?"

Like... what if xxCellebrityxx was a Rimworld character? A genre of them? Warhammer Rimworld promps? Family Guy Rimworld? :)

IOW - You could have fun with your hobby and produce OT material at the same time! Woot.

(Thanks a ton for all the info and examples! It's muchly appreciatecated! :))

Originally posted by VoiD:

...
Ofc this is subject to change in the future, this technology is quite young and trained on small pic sizes right now....

Thanks a ton for all those examples. I enjoyed looking through them for certain sorts of tell-tale oddities and the like. I'm not familiar with any of them or anime in general, though I do know a little bit about the style and can recognize, but maybe not name, popular characters.

An interesting bit is where symmetry breaks down. We are very keen to notice symmetry/asymmetry when it is "wrong." Trees are not symmetrical, people are more-or-less symmetrical, etc.. Clothing is "generally" symmetrical in its basic style, since we are bilaterally symmetrical. Asymmetrical clothing is usually a demonstration of expression or artistic intent, not strictly practical. Expressive designs, as well, follow similar basic rules.

I noticed several examples of broken/nonsensical symmetry, mostly in certain clothing details, a couple of times in proportion and at least once, which was striking, in a face of one of the characters. (Eyes were badly distorted)

And yes, you can train the AI in several ways to create different kinds of files, some checkpoints affecting everything including the style, or more specific inputs meant to simply create a specific character in any way you want, which you can teach it, but I haven't messed too much with it.

Had to look up and read a bit about "lora." So, they're basically custom "style" sorts of guides, in a way. They could contain subject prompts/trained frameworks and the like or style or basically some general kind of "weight" that says "I want it like this," kinda. (AFAIK atm :)) They're also low-impact in size/processing as they're designed for a more general guide ... creative template sort of thing. (I assume other more intense forms exist, but they're not typically used in a home setup.)

..Oddly enough she gets a closer body type to the original without the low weight dxd characters lora...

Perhaps due to a larger training set that generally reflects ideals in human aesthetics. There are plenty of "rules" we use when deciding whether or not something should look a certain way and whether or not its attempt at looking like that actually "works" or not. (Golden Ratio, symmetry, etc) It is no mistake that Pixar characters and Disney characters and Anime characters look the way they do. None of them widely diverge from basic principles, though they all take certain paths determined by their genre/style. We can tell they are different from each other because of those genre differences, but many of the principles used in creating them are nearly fundamental human aesthetic principles. (Though, once may argue that certain cultures may have different aesthetic cues, they're largely isolated ones. )

..Took them from some picture review on civitai at random.

This was all good stuff and very informative! Thanks!

So... what about a lora or some other training/prompt to generate "Rimworld Characters?" I'm wondering what the AI would come up with, there. :) Would it default to a 2d billboard sort of character like we have, now, or would it make them look like 3D Rimworld characters... like Lego figures?

Serious question, though - The most difficult "practical" issue in photogrammetry/anthropometry is successfully locating joints in a limb with precision. This is very difficult to accurately do and it's very necessary for a goodly number of worthwhile applications, from diagnosing developmental diseases/afflictions to sports medicine and physical therapy. In the past, schemes for doing so have relied on manually edited markers/points and motion tracking. Some systems have successfully used just motion capture and velocity, with some physics, but all systems are limited by the date they contain. To do it well, one needs the application to do the heavy lifting - One can't force grandma to do jumping-jacks in order to determine, at a glance, if she's truly suffering from potentiallly dangerous osteoporosis, but only if she does jumping-jacks for three hours... (Physicians often use palpable landmarks, where bone/etc is detectable via touching/palpating the surface of the skin, to locate bones leading to approximations they can then derive conclusions from without using any invasive means or radiation/etc. This is, as of right now, not easily possible to determine to any exactitude, either.)

I wonder... can such a graphically minded tool locate a human joint with any precision?? Even an approximation in a cartoon-like figure might yield some worthwhile results.

[Morkonan]

Originally posted by corisai:

Originally posted by Lord Freedom:

What then, if AI creates the art without our input?

Strong AI now is a fairy* tale.

This would be more likely called "General A.I."
("Strong" can be used, but it's inherently misleading, IMO.)

*actually a nightmare. Because people don't understand that to make a strong AI we need to know exactly how our mind is working and that will open a pandora box: mind control, brainwashing, digital upload, etc... So waaay before we will see a strong AI we will encounter a trully horrible things.

Consciousness is not determined by the workings of the human mind. Neither is "sentience." A squirrel is both conscious and sentient. Though, very limited in the latter...

Mind control, to a certain extent, is already possible via direct manipulation. Indirectly, it's also arguable that it's already possible - That's why marketing, propaganda, and gas-lighting exist. :)

I don't discount the potential of "Bad Things" happening, here. But, predictions need to modified by gained insights and fact we can use as effective predictors, even if they're only weakly interacting. (ex: There's no evidence atm to even potentially suggest that "uploading" is possible at all, ever... And, if so, there's about fifty-eleven philsopophers ready to murderize everyone in arguments about it. :))