Uploaded by Background Pony #9435
 1920x1280 PNG 2.11 MB
Interested in advertising on Derpibooru? Click here for information!
Furry Body Pillows - Preset and Custom Designs

Help fund the $15 daily operational cost of Derpibooru - support us financially!

Description

Thanks to Mr. Bat for the use of his wonderful Daybreaker Dreambooth.
This is something I’ve been thinking about doing for some time, combining the results of multiple specialized models into one coherent image. Naturally I’m being a little bit conservative with this first outing as the characters are not truly interacting and this more merely demonstrates the models ability to not immediately explode any lesser understood characters who happen to overlap with a primary subject. Woo, small victories. In the future I’d like to explore more dynamic interactions like hugging, etc but in the meantime, this will do.
I started by generating two 1:1 images of Daybreaker and NMM each, using Mr. Bat’s Daybreaker model, my own NMM model and a few of my own embeddings. In photoshop I quite crudely stitched the images together and began img2img runs until they had a coherent shared background across the width of the image. Following that I began hotswapping between models to rebuild and redesign both characters before taking the image back together to paint in Daybreakers mane, the mist effect and both ‘glowing horn’ effects, all of which are largely original painting with the exception of daybreakers mane which was largely lost over further iterations and error corrections slowly eroded any original work.
once I had a near final 512x768 image, I used the lollypop upscale model to bring the image up to its current resolution, after that I split the image into separate 1280x1280 chunks and began error correcting some of the more egregious ‘ai upscale’ artifacts, either by hand or successive low denoise inpainting. This largely took the form of correcting blurred lines, shapes that dissolved into nothingless, coloring artifacts and one very troublesome feather.
After that, I reconstructed the image and did some simple denoising in photoshop plus /sharpen/minor color correction/etc.

safe2264457 ai assisted1978 ai content32885 editor:craft70 generator:purplesmart.ai4211 generator:stable diffusion18568 daybreaker4435 nightmare moon21406 princess celestia116862 princess luna121587 alicorn335246 pony1696597 g42124239 chest fluff72203 folded wings24024 helmet16918 moon33655 nightmare luna435 peytral9067 slender11363 slit pupils8725 the quality of ai art is frightening282 thin16733 white sclera144 wings248663
Source

Comments

Syntax quick reference: **bold** *italic* ||hide text|| `code` __underline__ ~~strike~~ ^sup^ %sub%

Detailed syntax guide

Auroriia
A Really Classy Artist - 250+ images under their artist tag
An Artist Who Rocks - 100+ images under their artist tag
Artist -

Look As Much as I dislike AI, I’m sorry I clogged your Derpibooru feed.
I’ll help you out.
Luna’s left eye,
Luna’s wing. (Celestia’s wing are different styles)
Luna has 3 Legs
Celestia’s helmet at the top duplicates.
But hey, You do AI art, Power to you!
Auroriia
A Really Classy Artist - 250+ images under their artist tag
An Artist Who Rocks - 100+ images under their artist tag
Artist -

Recapping:
  1. As of right now. The only Authenticity I have is that I have a Brand new watermark/signature to sepreate
    Myself from AI art. Because (I*) got hit the hardest out of all artists here because of how evenly matched the machine does shading/rendering (Please check out my shading). People say that I’m the same as a machine in the community, which makes me really depressed.
  2. As for art theft. I don’t understand considering I have multiple people saying different things here on both sides.
    To me at least when I look at the art. It has poor wonky signature and patreon labels. So that obliviously came from artist works. Perhaps It’s not the AI. But it’s the AI users using artist’s content without consideration. Maybe people here Like lemon have a point.
People in these comments are not understanding. It’s not about the Pony IP. It’s about the piece of artwork.
As far as I understand (Artstation community/discord) told me it’s not about the style(You can’t copyright style) but the finished work, The way that it is. People can’t take that away from you. But they are. This is like A whole grey area because people sell fan art at conventions.

I’m sorry, but I don’t want to apologize for calling it fake. Hey maybe other people will enjoy it.
I did not spend a decade asking the pony art community for advice on how to improve just to only have every rule book fundamental fall apart and be broken and to be thrown out of the window.
Background Pony #DB99
@Cosmicscribbles
OK, let’s recap the damage of AI generated image and see if I have got it right.
  1. They are taking away the jobs of actual artists.
    That sounds possible, but AI generated images are usually random and in very similar styles if they actually look good(you can notice the shading and the lining).
    Real artists have their own styles, and they can do work accurately. If it is just one art piece for visual pleasure then its possible that AI is more effective, but if you want to make a full video animatic, design your own OC, or make a scene for a story, then real artists would certainly do better immensely.
  2. People are using AI as an excuse to bother artists and say AI is better.
    Really, no one deserves to be harassed, but there are always haters, with or without AI, no one can do much about them either.
3.They are taking the skills of established art and is art theft.
That sounds true, but how is AI doing that? Taking someone’s one art piece, let it go through an AI and use the outcome as themselves is art theft, but I’d say that it’s the same when real artists do that, because it’s essentially taking the original art’s idea inside. I don’t think AI makes a difference here.
Then there’s using lots of art to train an AI to do their own thing, but how does it work? Based on my knowledge, I think the AI is learning keywords according to the art pieces. They look at art and know what do human bodies usually look like and randomize the size’s outcome, same for animal faces, objects, lighting on objects and things like that. But don’t real people work the same way? It’s just faster, is still random, and except for some certain symbols that isn’t common enough, most of it looks nice.
Then there’s the copyright problem. Most AI’s so far forbidden copies of a certain artist’s art style to avoid problems, and remember real artists can can copy styles, too. There are so many artists in the world and lots of their art look similar.There’s also the fact that AI belong to us and has no power to hold copyright to its art. If some people use it and sell its art? I’d say it’s an idea of finders keepers, as long as the idea inside the art didn’t obviously copy anyone’s ideas, because it is free for all to use so far. But like I said it is still random. It is only going to be fine to use for those who don’t take it seriously. If someone wanted to use their idea of art creation seriously, they aren’t going to rely on AI art.
I hope I’ve got it right, I’ve been thinking about it for a while.
Hazy Skies

Aura Dust (AURAequine)
Not gonna lie, I knew I was being lazy with the wing but what’s wrong with the eye?
I think Cosmic was referring to Luna’s eye on the right-most side (her left eye). Her right one does seem to have a similar slit pupil like Daybreaker, but the other one is a bit wonky and doesn’t share that same style.
Classic AI telltale; eye inconsistencies. I’m torn on whether this flaw should remain to help better identify those who have put further effort into AI generated works, or if the technology should be refined towards that to not have as many instances of the flaw in order to to minimize artist corrections.
PS: I think it might be a good idea to also share an image of the base AI-generated result in the description so that others can see more clearly how much work you put into refining it. This will help emphasize that this is an edited AI generated art piece and not solely AI generated.
Keep Letting people mention that they are going to Mass Spam my art style, and devalue it 100X fold?
Most of the fault should be placed on the people that would abuse AI technology in order to do these acts, like sabotage and bullying; using a technology that’s shown to have positive uses instead be used for something very much negative. Of course, I still can understand why the technology that enables them could be seen negatively as a result, but that does seem unfair. The quiet majority use it to create their own works and enables fascination, inspiration and works of beauty and wonder. Don’t allow the loud minority of those who would steal and belittle others to distract you from the good that AI can allow people to achieve.
I believe the issues with AI are less technological and legal, and rather societal. Where it’s people that would act to perform illegal acts and not the fault of the AI (or developers of) themselves. An AI can spit out tons of a particular artist’s work, yes, but in order to do that someone must first:
  • Isolate all the works of that artist for use in training
  • Finetune/train an AI on those works for a long time, carefully so as not to overfit or otherwise.
  • Tell the AI specifically to create works using tags made from the training
  • Repeat the process if the works aren’t up to par or don’t match the style
All actions of a person and the AI does not do any of which on it’s own. And so the individual/group is at fault. This also takes considerable time and resources, and unless you have thousands of completed works, most finetuning on limited data looks awful and is easy to tell apart from the original; also it also inherits a lot of style from the base model it was finetuned from, making it unlikely to match your style completely. Your work is unlikely to be affected at all.
Background Pony #FBC7
If anyone wants me to change to the pro AI side. Help me Learn how to not get my signature scalped.
For currently existing models, there’s not much to do. Eventually those will be considered unwieldy or archaic, and fall out of use. For anything currently in the works, I imagine you can try the “opt-out” on LAION’s site, as that’s what SD and others pull from.
Future commercial models (ala whatever Adobe are clearly already working on) are likely going to avoid using art not in the public domain or that they didn’t license.
For models that might still include future works (noncommercial/“homebrew”, in general, I would recommend an obvious signature (like a logo or print text) in a corner, ala >>2844058
This is because the way AIs “know” what’s in an image is through a textual description of the image, and subtle signatures are likely to not be part of that description, while a more clearly existing one is likely to be included in the description.
Also, I don’t think people want you (or anyone else) to be “pro AI”, just not hate the average user of the tech. There are definitely concerns about how companies will continue cutting costs with automation, (as they have been since the mechanical loom…), but outside of a few obnoxious and loud psychos, people playing with these things just want to try to create something they think looks nice and aren’t trying to ruin anyone’s lives or diminish their work.
I’m on the optimistic side of believing those that stick around with these tools and don’t just lose interest will learn a lot; not just how to better use AI, but multiple art tools and techniques, to understand why something “looks good” and to improve on what they do.
Auroriia
A Really Classy Artist - 250+ images under their artist tag
An Artist Who Rocks - 100+ images under their artist tag
Artist -

@Background Pony #5435
There a massive difference between Learning from art and Working on top of art, My dude.
If anyone wants me to change to the pro AI side. Help me Learn how to not get my signature scalped.
(I’ve been updating my signature for days now)
@Background Pony #9435
Not gonna lie, I knew I was being lazy with the wing but what’s wrong with the eye?
They are clearly different.
Background Pony #5435
Artists bitching that their work is used “for AI” is making a joke of themselves.
Do people who looks at your art, learn from it, gets inspired to make art needs to obtain consent from you to make art? Because this is what AI does. It looks at your art and learns from it. Exactly the same way (conceptually) as a human would. When you post something online publicly, people will look at it. People will unconsciously learn from it.
Do you want to prevent people using your work? Do not post it online at all, its that easy. Keep it a secret, destroy it once you are done making it.
How did you learn to draw? Have you looked at ANY prior piece of work before your first drawing? If you have, you just stole from the people who made those! Because they made something and you learned from it, even if you don’t realized it.
Also, there is the fact that whomever is drawing anything resembling MLP, is straight stealing from the intellectual property owner of MLP (who is Hasbro). Stop drawing anything related to any copyrighted material, invent your own characters, own creatures never seen before (ponies are already copyright protected). Let’s see how quick you get traction.
And I see literally 0 resemblance to the art what the loudest artist here makes.
Background Pony #9435
Well, it’s hard to say. I started using the AI in october but a lot of that early time was wasted on poorly understanding/poorly documented parts of the program, on the purplesmart discord for example there were long threads about how exactly mask blurs work [and I maintain they’re misnamed and should be called a feather mask!] and as a community we don’t want people to have to redo all that, so someone starting today would probably improve a lot faster than I did. We’ve also had plugins developed to help people with prompting like the ponymaster project.
I would like to think a few weeks could probably get you up to speed if you were taking it as a fun side hobby, a week if you really pigeonholed yourself into it and asked around for help. The real problem is the lack of documentation and not really having organized tutorials yet. Most of us who have been using it for awhile still lack the confidence to speak with authority on some of the subjects, like for example there’s this understanding that ‘DDIM sampler is more conservative when inpainting solid lines’ but if you asked us to prove it, oof.
I also cheat a bit, I’m a huge advocate for inpainting as a process because I’ve been using photoshop for 17 years so I can get better results by way of dragging iterations in and making changes by hand. that fire effect on Daybreakers horn and the embers for example, basically entirely me because the AI sucks at that.
for reference, this is what it looked like when I started the inpainting process. just two individual generations crudely slapped together.
Fleetfoot
Fleetfoot - For Patreon supporters
Perfect Fleetfoot Plot Provider - Uploader of 10+ saucy Fleetfoot images with 350+ upvotes (Perfect Pony Plot Provider)
Woona - To my most special somepony~
PonyCon Holland -
Pixel Perfection - Hot Pockets Spotted
Solar Supporter - Fought against the New Lunar Republic rebellion on the side of the Solar Deity (April Fools 2023).
Tree of Harmony -
Non-Fungible Trixie -
Preenhub - We all know what you were up to this evening~
Twinkling Balloon - Took part in the 2021 community collab.

Senior Moderator
Fletch the Enfield
Damn this looks like Yakovlev and Evehly collaboration like it can totally fool me.
Out of curiosity how long does it even take to get it to this point?
Background Pony #9435
What I understand is that It takes information from those datasets and Generates random noise to create something brand new.
Close but you’ve got your order reversed, it generates random noise and then uses information from the dataset to ‘clean’ the noise. because you start from pure noise and your initial changes are largely random [any effort to clean pure noise is ultimately a crapshoot since there is nothing to clean] the end results can be drastically different each time, this is essentially how it all works. it’s a cloud of ever narrowing possibilities that is seeded with random noise to keep each new generation fresh, you can actually slow it down and watch it happen. do you see how it starts with essentially a blot of color and slowly ‘finds’ shape? This is also how you get stuff like double horns and extra limbs, the ai isn’t working towards building anything specific it just reduces noise until shapes emerge, where those shapes are it doesn’t care. if it thinks there is a pointy black bit on lunas skull and there are three blurry blackish bits you get three horns.
Same with the discord servers That have “SPECIFIC” artist tags names to get art that generates art specifically like the OC Artist. That’s not okay. Especially when those AI users pay money for that. Both Art and AI Users get hugely cheesed. And Mostly Art Users because AI users Demand money for the Generations they have paid for. (Building from Artist’s data)
I just generate ponies so its possible all that is true elsewhere, but I can promise you none of that happens in the mlpAI discord, which is free to join, free to use, has scrubbed fandom artist tags from its data [to the point where some useful words don’t work because those words are also artist names] and I’ve never seen anyone sell anything.
No, Idea how to do this.
My fault, I thought that site referenced multiple AI generators. Not Just LAION-5B.
I’m referring to Novelai/Midjourney/Nightcafe/every AI service out there.
It’s not right that alot of them are making huge profit off our creations.
This is not about the Ai users either, because they are also getting cheesed too.
Unfortunately there will never be a way to see what is inside the private generators like midjourney/novelai/etc, although presumably they all started with some version of LAION they have almost certainly moved to custom models that could contain anything and everything and we wouldn’t know until they overfit.
And now I’m the bad guy because this AI artist can’t bother fixing Luna’s left eye, Or wing that merge’s with her body?
Not gonna lie, I knew I was being lazy with the wing but what’s wrong with the eye?
Background Pony #FBC7
And now I’m the bad guy because this AI artist can’t bother fixing Luna’s left eye, Or wing that merge’s with her body?
I think these are absolutely valid things to point out, and maybe it’ll help people improve and be better (or “actual”) artists.
When I first started publishing my own things years ago I was ecstatic because I didn’t even know all these different digital drawing tools even existed, like layers or pen stabilization or vector lines (I’d honestly not drawn anything since I was a toddler), and I rushed in and brute forced my way into making “pretty things” and ignored all the errors and fundamental problems because I was too excited about putting something I had in my head onto a screen I could look at again and show others.
But over time I learned how to do things better, learned perspective and anatomy and color theory and a whole lot of other things I didn’t know about when I started and actually matter. And there’s loads more I need to learn and enjoy learning and improving!
Right now we’re seeing a lot of people dabbling with new tools that let them express themselves, and the majority are too excited and happy that “it looks good!” to care about the all the errors or even why something looks right or wrong. Most will move on after finally scratching that “itch”, but a lot will actually want to improve and learn and understand what all the art fundamentals are. It’s happened with so many other things; just in this fandom, there are hundreds of people that started using crappy DA “bases” just to be able to see their OCs, but over time learned to actually draw things on their own and can be rightfully proud of what they create.
But aside from all that… please don’t let a few rabid, loud or headstrong idiots get to you, they don’t deserve the attention.
LemonDrop
Duckinator - Same nonsensical quacks in every pond
Nightmare in the Moon - Had their OC in the 2024 Derpibooru Collab.
Pixel Perfection - I still call her Lightning Bolt
Lunar Guardian - Earned a place among the ranks of the most loyal New Lunar Republic soldiers (April Fools 2023).
Crystal Roseluck - Had their OC in the 2023 Derpibooru Collab.
Elements of Harmony - Had an OC in the 2022 Community Collab
Non-Fungible Trixie -
Twinkling Balloon - Took part in the 2021 community collab.
Ten years of changes - Celebrated the 10th anniversary of MLP:FiM!
My Little Pony - 1992 Edition

C++ Crazed
@Cosmicscribbles
Again AI is just recognizing signatures incorrectly as important features of images, it doesn’t have the intelligence to really realize they are not something it should be adding to art (yet).
Also I wouldn’t worry about what other people say, as others have said people will always complain and you shouldn’t let that discourage you from producing art, just do what makes you happy etc etc.
Auroriia
A Really Classy Artist - 250+ images under their artist tag
An Artist Who Rocks - 100+ images under their artist tag
Artist -

What am I apposed to Do?
Keep Letting people mention that they are going to Mass Spam my art style, and devalue it 100X fold?
Keep hearing How my art is AI art and Not my real art, when I’ve been doing it for decades?
I’m just apposed to accept that Art now Doesn’t have No common courtesy to fundamentals anymore
When I’ve been chewed out by artist’s for decades to focus on the fundamentals to improve as an artist?
And now I’m the bad guy because this AI artist can’t bother fixing Luna’s left eye, Or wing that merge’s with her body?
@Lemondrop,
Respectfully, I’m not just a fan artist. It’s directed at all* my content I make. Including my original content.
I don’t think you understand what I’m saying. This is not about the content of the art. It’s about the final product.
“You need to actually explain why in your mind a human is allowed to look through a gallery of public-facing images and learn to create art based on study of this information and practice whereas an AI is not allowed to do such a thing, otherwise this argument does not make sense.”
Ai is clearly Building from memory, and not creating anything new. If it was New, It would not be trying to place down artist signatures.
I’m withdrawing. This is making me horribly sad.
Paracompact
Terramar - For supporting the site
Happy Derpy! - For site supporters
Gold Supporter - Gold Supporter
Artist -

@Cosmicscribbles
All matters of AI aside, I think it’s sad if you wound up harassed by anyone for any reason, or if you were ever made to feel ashamed or obsolete for your artstyle. You’re a great artist who’s done great work, and I’ve never heard of you doing anyone wrong. If you had offered those free art lessons a couple years ago (back when I had more time and interest), I certainly would have taken you up on them. You can ask Rocket Lawn Chair about that if you think I’m just humoring you.
However, you should understand that AI artists/editors/enthusiasts are not a monolith. Some of them are trolls, some of them are zealots. Some of them are narcissists who feel empowered by a tool that gives them art superpowers, and they just want to show off. However, having spent a lot of time in and around these communities, I think the great majority are just ordinary people who are innocently entertained by the new tech. Others, like craft (the artist of this image, and who I know internet-personally), devote hours into each piece and are constantly trying to do new and unique things with the medium. People like these, as well as their friends, are going to wind up hurt in just the same way you are now if you go out of your way to denigrate their art or paint them as simple villains.
Anyway. In the end, I just want to wish you the best. I hope you can find a way to reconcile with the AI and human art communities and continue to produce the unique work you’re known for.
LemonDrop
Duckinator - Same nonsensical quacks in every pond
Nightmare in the Moon - Had their OC in the 2024 Derpibooru Collab.
Pixel Perfection - I still call her Lightning Bolt
Lunar Guardian - Earned a place among the ranks of the most loyal New Lunar Republic soldiers (April Fools 2023).
Crystal Roseluck - Had their OC in the 2023 Derpibooru Collab.
Elements of Harmony - Had an OC in the 2022 Community Collab
Non-Fungible Trixie -
Twinkling Balloon - Took part in the 2021 community collab.
Ten years of changes - Celebrated the 10th anniversary of MLP:FiM!
My Little Pony - 1992 Edition

C++ Crazed
@Cosmicscribbles
That may not be wrong but that’s also an apt description for the majority of human artists, so it’s not a very effective criticism of AI unless you think you’re stealing other people’s art by drawing inspiration and learning from things you see on Derpibooru.
You need to actually explain why in your mind a human is allowed to look through a gallery of public-facing images and learn to create art based on study of this information and practice whereas an AI is not allowed to do such a thing, otherwise this argument does not make sense.
Furthermore you have not explained why you think you’re allowed to commit trademark infringement against Hasbro by drawing trademarked characters from MLP, this is not even fair use and something they can sue you for, but apparently you’re fine with this direct exploitation of their work too for some unexplained reason. Note I am not defending copyright law here, I think it is a bad idea, but this sort of trademark infringement is far more of something you could equate to theft than pretty much fair use usage of images to train an AI, so I am not sure why you think it is acceptable given your heavy support of IP rights.
Auroriia
A Really Classy Artist - 250+ images under their artist tag
An Artist Who Rocks - 100+ images under their artist tag
Artist -

Appreciation? For what? All AI does is spit out new concepts and ideas out from a visual Library that artist’s didn’t consent to building. You can’t tell me I’m wrong. All that Noise Generation comes from somewhere.
I’m apposed to appreciate Having Bad fundamentals?
You understand I also have the Art community Getting nasty with me too, right?
LemonDrop
Duckinator - Same nonsensical quacks in every pond
Nightmare in the Moon - Had their OC in the 2024 Derpibooru Collab.
Pixel Perfection - I still call her Lightning Bolt
Lunar Guardian - Earned a place among the ranks of the most loyal New Lunar Republic soldiers (April Fools 2023).
Crystal Roseluck - Had their OC in the 2023 Derpibooru Collab.
Elements of Harmony - Had an OC in the 2022 Community Collab
Non-Fungible Trixie -
Twinkling Balloon - Took part in the 2021 community collab.
Ten years of changes - Celebrated the 10th anniversary of MLP:FiM!
My Little Pony - 1992 Edition

C++ Crazed
@Background Pony #E544
Limited meaning within the confines of fair use. If the usage is fair use the artist has no right over it and cannot complain about copyright infringement or submit a DMCA etc. Copyright is not an infinitely powerful weapon you can use to silence or suppress anything you dislike. People have the right to criticize you using your work, people have a right to parody your work, people have a right to use it in various forms of research, and people likely have a right to use it in large generative AI models like this.
LemonDrop
Duckinator - Same nonsensical quacks in every pond
Nightmare in the Moon - Had their OC in the 2024 Derpibooru Collab.
Pixel Perfection - I still call her Lightning Bolt
Lunar Guardian - Earned a place among the ranks of the most loyal New Lunar Republic soldiers (April Fools 2023).
Crystal Roseluck - Had their OC in the 2023 Derpibooru Collab.
Elements of Harmony - Had an OC in the 2022 Community Collab
Non-Fungible Trixie -
Twinkling Balloon - Took part in the 2021 community collab.
Ten years of changes - Celebrated the 10th anniversary of MLP:FiM!
My Little Pony - 1992 Edition

C++ Crazed
@Cosmicscribbles
That is not how AI works. Learn how AI works first and maybe you’ll have a better appreciation for what it is doing. In short though, AI learns to generalize various constructs of an image on every level, from the overall composition (e.g. what arrangement of different high level concepts will create a “pony”) down to the finest detail (stuff like texture, shading, lines, curves, etc). It is quite literally learning in a way very similar to how humans conceptualize things which is one of the one of the goals of modern machine learning in a way as things that see the world in a fuzzy way similar to how we do are quite useful for automating various workflows (e.g. an AI that can recognize images can take the place of a human content moderator or something like that). AI obviously is different from us and especially for art it generates art in a much different way (though do note stroke-based AI art is a thing, it’s just a really stupid way for a computer to go about making art when brushes and strokes like that are more a necessity due to limited human/computer interaction technology like drawing tablets vs hooking your brain up to a computer directly), but it is not just taking images and pasting them together, it is learning the fundamental aspects of what make up images and art and that is provable in its generalization ability. You can say its inspiration is a tad lacking however since once learning these concepts you are in a way just interpolating around in a huge high-dimensional space of all the things that make up “art”, but it still is able to be creative despite that as much of that space is unexplored, and humans do the same thing a lot of the time as well to discover new combinations of existing ideas that work well (it’s hard to say how much human inspiration truly comes from some magical complex brain BS vs how much is us just mimicking things in the environment and other content we’ve seen).
As for signatures that’s really just a limitation of the model. It is finite sized and can only encode so much information in it, and a full understanding of language and writing and even a concept of an “identity” are beyond its capabilities (when its primarily focused on artwork not writing), so you’ll either get just jumbled attempts at things that look like a mishmash of signature styles (since there’s virtually no stylistic correlation between most dataset labels and the style or content of a signature) or reproduction of a specific artist’s signature (when asking for something in a specific artist’s style since the signature if drawn in the same way on each image will be highly correlated with that label and its near-identical reproduction usually will train the AI to focus on memorizing it as it doesn’t really understand its purpose and thinks its just a critical feature of the art).
Tldr weird signature stuff an undesirable artifact of the training/generation process but also a minor one not worth a ton of time fixing when there’s bigger things to be working on (like actually developing the model’s artistic skills). You can expect future models though to remove signatures from their datasets to prevent the AI from thinking they have any importance in art (this is already done for watermarks for similar reasons).
Auroriia
A Really Classy Artist - 250+ images under their artist tag
An Artist Who Rocks - 100+ images under their artist tag
Artist -

No way the AI works in that same manner.
Ai learns by taking Images in the dataset, then reapplying that As “Hey”, Thats a Head, or Thats a chest, and Needs to be there. It’s also why theres signatures that don’t make sense either. It’s trying to imitate that. It doesn’t build a brand new style. It only applies what it Sees in the dataset.
As with an artist. They have the freedom to learn from Artist Art. But Apply techniques to their own works.
Not the same anatomy/lighting/features/creatures/landscapes.
It’s like Dead/soulless. Like if Frankenstein were a piece of art.