Stable Diffusion, NovelAI, Machine Learning Art - AI art generation discussion and image dump

[Glowie]

Protip Diffusion AI can replicate artists and studios
Text to image
Cell shaded spider robot monitor for head by Madhouse


Once this AI learns more and spreads DeviantArt and others will be shitting their pants and cry: BAN ALL AI ART!

Due to the fact AI draws Better sonic fetish artists and tracers there, Joe schmos RTX 2080 and it learns slowly.

Think about that for a minute.

 

[purpleflurp]

I can see a few reasons why people would still prefer commissioning artists over using AI art.

Commercial use: Weirdly enough, a PETA lawsuit regarding an ape selfie determined that a work is only copyrightable if a human created it. This is further expanded upon with regards to an attempt at copyrighting procedurally generated music, wherein it was determined that algorithmically generated content can't be copyrighted, even if it's the creator of the software that's attempting to exercise ownership over the outputted content.

If you're making a game, visual novel, or comic, the work generated by programs like this technically fall into the public domain. Which sucks if you make a dope CG and want to make sure nobody else can use it in their projects, but $60-100 isn't that big a deal if you're making a commercial product.

Unique styles: Say you want a painting done in the style of Kaja Foglio or Quinton Hoover. Both of these artists have very distinct art styles, which would require a new model built off of their respective works. However, you may run into a roadblock where there aren't enough samples to train with, or their portfolio consists of low-res imagery.

Modifying/correcting AI Art: At the moment, there's a lot of tinkering that's required to get AI art looking just right.

Pixel Art: Current models for AI art are lackluster when it comes to pixel art. There's workarounds, sure, however for pixel art that replicates console limitations (resolution, palette, tilesets) this can be a big ask for AI art.

2D animation rigs: A bit more niche, but 2D game designers that rely on puppet animation (think Plants vs Zombies) will need characters that can be broken up into dozens of components to allow for fluid animation in their games.

Tangible artwork: Traditional artists will likely have a longer lasting stance in the market thanks to clients that prefer having a hand painted canvas. I know I'm personally a huge fan of Inkwash Watercolor art, and still pay $200+ for original works.

Consistency on fine details: Having an AI replicate things like a character's tattoo, logos and text on clothes, or the placement of freckles and birthmarks is still a bit wishy-washy. For smaller projects and indie commissions, I could see clients hiring artists to place these types of details on AI generated artwork. I'm sure this'll be made redundant sometime in the next few years, but for now I can see some issues.

 

[Faggitory]

Commercial use: Weirdly enough, a PETA lawsuit regarding an ape selfie determined that a work is only copyrightable if a human created it. This is further expanded upon with regards to an attempt at copyrighting procedurally generated music, wherein it was determined that algorithmically generated content can't be copyrighted, even if it's the creator of the software that's attempting to exercise ownership over the outputted content.

I'd imagine this is beatable by using an AI generated image as a base, and then editing it and expanding on it to gain a level of ownership out of the remixed product, right?

 

[Sausage Link Korean]

It had been settled law for ages that a photo's copyright is triggered by clicking the button to the take the photo, and copyright is vested to whoever clicked the button. That PETA lawsuit only made the news because A lol monkey and B PETA being menaces again.

I'm not familiar with the circumstances of the algorithmically produced music but for this AI image shit its possible the same copyright trigger of pressing the generate button along with the prompts and settings etc would be considered enough to create a copyright in the work. Maybe.

 

[SomethingProblematic]

Something I find curious is how everybody cries about artists being replaced by AI, but nobody seems to care about writers being replaced by AI storytellers. Or someone taking an AI story and doing a bit of editing and releasing it as their own book.

 

[74164978231]

I've got nai.ckpt

Do you have that model selected?

 

[Blue Miaplacidus]

https://docs.novelai.net/ has been incredibly helpful. I've been playing with this since it's been featured, as someone who does a lot of video editing and can use gimp but cannot create art (at least anything high quality that looks good) this is a game changer. I'm thoroughly enjoying this, and it's going to help with my little youtube hobby immensely.

 

[Banana Hammock]

Do you have that model selected?

Yup, you guessed it. Somehow I missed that thing right up at the top of the page, until モモスモモス mentioned it.

 

[purpleflurp]

It had been settled law for ages that a photo's copyright is triggered by clicking the button to the take the photo, and copyright is vested to whoever clicked the button. That PETA lawsuit only made the news because A lol monkey and B PETA being menaces again.

I'm not familiar with the circumstances of the algorithmically produced music but for this AI image shit its possible the same copyright trigger of pressing the generate button along with the prompts and settings etc would be considered enough to create a copyright in the work. Maybe.

The issue here is that any individual can use the same prompt and settings to generate the exact same image.

Let's say the two of us generate the same image of a Triple-Breasted Whore from Eroticon 6 without realizing it. Who owns the copyright in this case? If I generate the image a week after you did, does that mean I have less of a right to use it than you do? Alternatively, if you generated the image a week before me but I release my book using the image first, would I be able to stop you from releasing your game that uses the image?

 

[Glowie]

I can see a few reasons why people would still prefer commissioning artists over using AI art.

Commercial use: Weirdly enough, a PETA lawsuit regarding an ape selfie determined that a work is only copyrightable if a human created it. This is further expanded upon with regards to an attempt at copyrighting procedurally generated music, wherein it was determined that algorithmically generated content can't be copyrighted, even if it's the creator of the software that's attempting to exercise ownership over the outputted content.

If you're making a game, visual novel, or comic, the work generated by programs like this technically fall into the public domain. Which sucks if you make a dope CG and want to make sure nobody else can use it in their projects, but $60-100 isn't that big a deal if you're making a commercial product.

Unique styles: Say you want a painting done in the style of Kaja Foglio or Quinton Hoover. Both of these artists have very distinct art styles, which would require a new model built off of their respective works. However, you may run into a roadblock where there aren't enough samples to train with, or their portfolio consists of low-res imagery.

Modifying/correcting AI Art: At the moment, there's a lot of tinkering that's required to get AI art looking just right.

Pixel Art: Current models for AI art are lackluster when it comes to pixel art. There's workarounds, sure, however for pixel art that replicates console limitations (resolution, palette, tilesets) this can be a big ask for AI art.

2D animation rigs: A bit more niche, but 2D game designers that rely on puppet animation (think Plants vs Zombies) will need characters that can be broken up into dozens of components to allow for fluid animation in their games.

Tangible artwork: Traditional artists will likely have a longer lasting stance in the market thanks to clients that prefer having a hand painted canvas. I know I'm personally a huge fan of Inkwash Watercolor art, and still pay $200+ for original works.

Consistency on fine details: Having an AI replicate things like a character's tattoo, logos and text on clothes, or the placement of freckles and birthmarks is still a bit wishy-washy. For smaller projects and indie commissions, I could see clients hiring artists to place these types of details on AI generated artwork. I'm sure this'll be made redundant sometime in the next few years, but for now I can see some issues.

AI in this case is self learning after being taught just like human. This technology is in it's infancy what IBM virtual workstations were to full body tracking and facial reckongnition in VR and what virtual boy was to 3DS and Augmented reality.

Like any fine product it just needs tech savvy people and people who want to see the art world burn while teaching a AI.

Private boorus, anatomy instructions, double checking for major mistake, extra limbs, misaligned clothes, desired watermarks and other human quirks.

Soon enough art generating AIs creating art like individuals for people who keep data and algorithm that's stored locally.

Nothing is stopping human human intervention to make more unique art that can't be detected easily once the AI is trained and given more room to grow.

Can you imagine the ramifications?

People sieze means of artworks and memesmithing

For example
It took a MONTH for /pol/ to train Tay and same for 2ch for the Japanese AI who learned to script and make its own jump scare page which even works on way back machine.

I doubt a normie Japanese would go their way teach bot love Hitler and programming languages.

Never underestimate power of autism.

 

[purpleflurp]

Nothing is stopping human human intervention to make more unique art that can't be detected easily once the AI is trained and given more room to grow.

Can you imagine the ramifications?

I'm actually working on that right now. One of my computers is currently training a model with all 7000+ strips of Sinfest.

My goal is to randomly post AI-generated strips on their forum and see how long it takes people to notice Tatsuya didn't make it. I mean the comic's been running every day for 20+ years. Who's to say there wasn't a sunday strip of Monique taking bong rips and playing Mario romhacks?

People sieze means of artworks and memesmithing

I'm seeing this as the ultimate form of gaslighting once it's been refined. Imagine training the AI with the original 25 MTG artists, printing an entire "lost" print run of Spectral Chaos, and leaving a pallet of it in a public storage unit before it goes to auction.

 

[74164978231]

Interesting how it can get a complicated body part visually like an ear right but then fail so much on the eyes.

 

[Glowie]

FYI:
The novelAI torrent is ~56GB.

Good I didn't like having Fallout 4 or Skyrim cluttering my computer anyway.

Something productive takes up that space now.

I've been fucking around with text to image and results from that alone look better than a coomer artist on Twitter.

Your wallet and your GPU is thankful for your proper contribution.

I gave myself my own rating, it's hot accurate but that's what you get for 2 minutes of effort.

 

[cybertoaster]

Remember the whole "AI will do the shitty work so you can focus on your art" and all? that backfired badly didn't it? You can't get an AI that can reliably do stuff like OCR data entry with no errors but you get AI that can make art that's almost on par with the pros, let alone the wannabes from DA and such.

NGL the ones that actually worry me are the ones that can code a website from just a description, that alone could leave a lot of people out of work.

Motherfucker i don't mean destroy it beyond repair, hell i don't even mean ban the regular porn,but certainly there should have been a hardcoded way to make it recognize kids and not make naked pictures of them.

You can't tell me there's not a way because its been done before on other things

No it hasn't, if you mean that bullshit apple said it was an excuse to scan your phone constantly, they don't care about child porn just want to mine the crap out of your life. Meanwhile there has been a number of false-positives for child porn on phones and companies like google have permabanned people because of these "mistakes".

I can see a few reasons why people would still prefer commissioning artists over using AI art.

Commercial use: Weirdly enough, a PETA lawsuit regarding an ape selfie determined that a work is only copyrightable if a human created it. This is further expanded upon with regards to an attempt at copyrighting procedurally generated music, wherein it was determined that algorithmically generated content can't be copyrighted, even if it's the creator of the software that's attempting to exercise ownership over the outputted content.

If you're making a game, visual novel, or comic, the work generated by programs like this technically fall into the public domain. Which sucks if you make a dope CG and want to make sure nobody else can use it in their projects, but $60-100 isn't that big a deal if you're making a commercial product.

So code written by an AI would also be automatically in the public domain? that could limit the use of this tech drastically/never move from the amateur space since nobody living off it would want their work to be essentially unmarketable and given away for free.

On the other hand what if the work is a hybrid? like for example a visual novel where the graphics are from an AI but the script was made by an actual person, what happens then?

2ch for the Japanese AI who learned to script and make its own jump scare page which even works on way back machine.

Never heard of that before, got more?

 

[Banana Hammock]

Good I didn't like having Fallout 4 or Skyrim cluttering my computer anyway.

Something productive takes up that space now.

I've been fucking around with text to image and results from that alone look better than a coomer artist on Twitter.

Your wallet and your GPU is thankful for your proper contribution.

You only need animefull-final-pruned (or animesfw-final-pruned), so it's more like 14GB total.

 

[purpleflurp]

So code written by an AI would also be automatically in the public domain? that could limit the use of this tech drastically/never move from the amateur space since nobody living off it would want their work to be essentially unmarketable and given away for free.

On the other hand what if the work is a hybrid? like for example a visual novel where the graphics are from an AI but the script was made by an actual person, what happens then?

Obligatory "I am a retard, not a copyright lawyer" disclaimer, but my understanding based on existing copyright laws is that you would own the rights to the game, the characters, and story. However, establishing ownership of any unaltered assets wouldn't be possible due to anyone being able to generate the same image by using the same settings and prompt that you did.

 

[Mr.Miyagi]

Something I find curious is how everybody cries about artists being replaced by AI, but nobody seems to care about writers being replaced by AI storytellers. Or someone taking an AI story and doing a bit of editing and releasing it as their own book.

While there might have been some doomposting/journo click-baiting about "The End of Literature" when OpenAI unveiled GPT-3, I think that it (and the open-source EleutherAI models) ultimately caused less hand-wringing for two main reasons:

1. The overall level of coherence is lower. While you can certainly get some impressive results, having the AI run on its own for the entire length of a short story, let alone a novel, would quickly have things spiral into nonsense. It's definitely a lot more coherent than it used to be (the older stuff has a very surreal, dream-like quality), but it can't really "hold itself together" long enough to approach the long of a commercially viable product.

2. The difficulty of use is much higher. Writing something alongside GPT-3 means that it'll try to predict what follows based on your input, which naturally means that writing total garbage will get you garbage. While this is fun enough to be shown off on Twitch streams, it ultimately means that to get the best quality writing (whatever someone might think that is), you have to give it high quality writing to not only start with, and not just in your inputs, but actively edit any errant outputs to maintain a standard. So you have to play both writer and editor simultaneously.

Then you compare that to Stable Diffusion which, for all of its faults and nightmarish generations, is capable of generating some surprisingly passable pictures with nothing required of the user more than a good sense of what prompts to put in and patience. It's understandable that some people are viewing this as more likely to mess with artists, though I think it'll be much more of a tool than an outright replacement. In the next ten or twenty years, though, when some tech company releases a language model with 10 gorillion hyperparameters? I expect quite a few seething articles about AI writers spreading "fake news" or dissuading hobbyists from writing on their own. But it'll still just be a tool to be used.

 

[Sausage Link Korean]

The issue here is that any individual can use the same prompt and settings to generate the exact same image.

Let's say the two of us generate the same image of a Triple-Breasted Whore from Eroticon 6 without realizing it. Who owns the copyright in this case? If I generate the image a week after you did, does that mean I have less of a right to use it than you do? Alternatively, if you generated the image a week before me but I release my book using the image first, would I be able to stop you from releasing your game that uses the image?

If I followed you around with the same camera that you had, set to the same configurations, tripod in same spot, same angle, same lighting, taking pictures of the same inanimate objects resulting in a photo identical in all respects other than the time it was taken, I would probably have trouble establishing copyright based on the fact that your image existed earlier.
So yes, if copyright extends to these images, which I can see being possible, I can also see the situation of two people with identical images created at different times occuring, going to court and being decided based on the opinion of a decrepit 80 year old whose understanding of AI generation extends to "so the picture.. is in the computer?", an opinion that will probably stand unchallenged for years.

 

[Glowie]

Remember the whole "AI will do the shitty work so you can focus on your art" and all? that backfired badly didn't it? You can't get an AI that can reliably do stuff like OCR data entry with no errors but you get AI that can make art that's almost on par with the pros, let alone the wannabes from DA and such.

NGL the ones that actually worry me are the ones that can code a website from just a description, that alone could leave a lot of people out of work.


No it hasn't, if you mean that bullshit apple said it was an excuse to scan your phone constantly, they don't care about child porn just want to mine the crap out of your life. Meanwhile there has been a number of false-positives for child porn on phones and companies like google have permabanned people because of these "mistakes".

So code written by an AI would also be automatically in the public domain? that could limit the use of this tech drastically/never move from the amateur space since nobody living off it would want their work to be essentially unmarketable and given away for free.

On the other hand what if the work is a hybrid? like for example a visual novel where the graphics are from an AI but the script was made by an actual person, what happens then?

Never heard of that before, got more?

Note Jannies Yes I know that I am a fag

On art, computer generated art can be made yours with help of Photoshop, filters and such.
Anyway with computer generated art you NEED to know exact seed and values, data which isn't stored in meta data, EXIF and such that is shown only picture generation and batch file if I'm not mistaken.
One you work on a template that you made with a tool that work is essentially yours, not public domain. Which is why you don't see any artists given flak using doll and posing programs and tracing for animators who in this day and age make 8 frame animations, porn of course.

Work on a reference and it will be yours, simple as.

I won't bother with morality arguements, I'm in this to witness art die and prove that human creativity is a finite resource.

Large of chunk of professional art(cows) who do commission work made their bed and now they will lie in it.

I'm just waiting for the day some unfortunate bastard catalogs e621, FA, F-List and others to see random Furry art and porn generated to come crashing down like the world economy.


Embedded links are broken, I'm stuck on a mobile embedded links are broken desktop view or not. Rinna's blog which now dead like Tay had this

https://web.archive.org/web/20170608002413/http://www.fujitv.co.jp/kimyo/rinna_blog/

No flash or otherwise obvious scripts. Rinna will ring you and blast your headphones.

 

[Rise]

img2img is rad as hell



I haven't seen this posted anywhere so far, but you'll want to periodically clear out your output folder. It can be found by going to: {wherever you set up the program}\stable-diffusion-webui\outputs. Everything you render gets dumped into a folder in there. I was running several batches to find the right picture and added 10+ gbs of images over a relatively short amount of time.