This is gud :D

Cute! <3

richiebre said:
Wait... it says ai generated. No fucking way seriously??

Yeah I've seen quite a few similar styles by Novel AI

I see no evidence of this being AI generated. You can make just about any style with AI, since it learns from existing images.

Alright. As someone somewhat experienced with AI generation, let's go through the signs.

-FIrst of all, the image resolution is 512x768. This is VERY obviously a default resolution for ai-generation for programs like stable diffusion. It's very strange that such a high-quality image was made at such a low resolution, with no appearance of higher res copies anywhere. Obviously, maybe the artist is keeping the higher-res image to himself for potential copyright / patreon reasons. But choosing to upload his image at exactly 512x768 makes this so suspicious.

-Ai-generation currently has a problem with background elements aligning on the other side of frontal objects. For example, as it currently stands it's nearly impossible to have a perfect straight line go past the main character.
With this in mind, you can clearly see that the background doesn't continue behind the eevee. While this doesn't necessarily mean it's AI-generated, it's definitely a key feature to point out. The window and the ledge on the left are clearly unable to be 'generated' on the right side of the eevee. But maybe the artist simply decided to draw an eevee in a corner, that's just as likely.

-You can see that the window ceil in the bar appears to follow the same curve as the hair just infront of the eevee's right ear (our left). These lines that seem to pass through objects are definitely the marking of an AI. However, it could've just been another thing the artist chose to do.

-AI generation shouldn't be able to generate an eevee. At least not with this much complex detail. Unless he has a model specifically trained on thousands of eevee images, which is unlikely no matter how you put it (especially at this quality where details such as fur are insane here), this seems almost impossible. The only exception would be if the creator carefully fine-tuned a single image over and over until he got the character he wanted. Which, to be honest, is possible. However, based on my experience with fine-tuning images for increased quality, this easily would've taken hours to get right.

-Inconsistency with facial expression and the events happening in the image are definitely odd. I'm not sure if that's the face one would make if they got caught mid-orgasm. But once again, this could've just been an artists' choice. This for me however was definitely the first sign that something was a tad off.

-You can see in the dark shadow part of the scarf that there's this certain fabric-like layout going on. The scarf has no reason to bend downards like that into the tip, and this seems like a simple thing for ai to have generated. Random forms with no comprehension are incredibly easy for an ai to pull off- So despite it seeming somewhat normal, you can for example...

-Look closely at the top of the tail. The white stroke that blurs/blends into the background is 100% a dead giveaway of ai generation. I can't tell you how many times I've seen this happening in ai, where a "line"/stroke of paint blends into an object it shouldn't blend into. And while it looks amazing when an artist does this on purpose, AI can never seem to get it right which results in the blending happening in a sort of reversed way. Here's an example:
'Real': https://i.imgur.com/mUtGEjC.png | AI generated: https://i.imgur.com/JAusGAr.png

-AI can't really do straight lines. Once again, maybe it's easily an artistic choice, but being able to do such complicated fur patterns while still having a messy non-straight lines in the background (window-ceil and edge of ledge) just scream ai generation to me.

Unless the artist confirms whether or not this is AI-generated, I'd put my money towards it being ai. Which.. complicates things, since it is a really good image. But if AI generation has become so good that it's starting to become impossible to discern real artists' images and easily creatable mass-produced images, I can't tell if this is a good thing or a bad thing. And that's coming from somebody who loves this type of stuff.

leafeon said:
Alright. As someone somewhat experienced with AI generation, let's go through the signs.

-FIrst of all, the image resolution is 512x768. This is VERY obviously a default resolution for ai-generation for programs like stable diffusion. It's very strange that such a high-quality image was made at such a low resolution, with no appearance of higher res copies anywhere. Obviously, maybe the artist is keeping the higher-res image to himself for potential copyright / patreon reasons. But choosing to upload his image at exactly 512x768 makes this so suspicious.

-Ai-generation currently has a problem with background elements aligning on the other side of frontal objects. For example, as it currently stands it's nearly impossible to have a perfect straight line go past the main character.
With this in mind, you can clearly see that the background doesn't continue behind the eevee. While this doesn't necessarily mean it's AI-generated, it's definitely a key feature to point out. The window and the ledge on the left are clearly unable to be 'generated' on the right side of the eevee. But maybe the artist simply decided to draw an eevee in a corner, that's just as likely.

-You can see that the window ceil in the bar appears to follow the same curve as the hair just infront of the eevee's right ear (our left). These lines that seem to pass through objects are definitely the marking of an AI. However, it could've just been another thing the artist chose to do.

-AI generation shouldn't be able to generate an eevee. At least not with this much complex detail. Unless he has a model specifically trained on thousands of eevee images, which is unlikely no matter how you put it (especially at this quality where details such as fur are insane here), this seems almost impossible. The only exception would be if the creator carefully fine-tuned a single image over and over until he got the character he wanted. Which, to be honest, is possible. However, based on my experience with fine-tuning images for increased quality, this easily would've taken hours to get right.

-Inconsistency with facial expression and the events happening in the image are definitely odd. I'm not sure if that's the face one would make if they got caught mid-orgasm. But once again, this could've just been an artists' choice. This for me however was definitely the first sign that something was a tad off.

-You can see in the dark shadow part of the scarf that there's this certain fabric-like layout going on. The scarf has no reason to bend downards like that into the tip, and this seems like a simple thing for ai to have generated. Random forms with no comprehension are incredibly easy for an ai to pull off- So despite it seeming somewhat normal, you can for example...

-Look closely at the top of the tail. The white stroke that blurs/blends into the background is 100% a dead giveaway of ai generation. I can't tell you how many times I've seen this happening in ai, where a "line"/stroke of paint blends into an object it shouldn't blend into. And while it looks amazing when an artist does this on purpose, AI can never seem to get it right which results in the blending happening in a sort of reversed way. Here's an example:
'Real': https://i.imgur.com/mUtGEjC.png | AI generated: https://i.imgur.com/JAusGAr.png

-AI can't really do straight lines. Once again, maybe it's easily an artistic choice, but being able to do such complicated fur patterns while still having a messy non-straight lines in the background (window-ceil and edge of ledge) just scream ai generation to me.

Unless the artist confirms whether or not this is AI-generated, I'd put my money towards it being ai. Which.. complicates things, since it is a really good image. But if AI generation has become so good that it's starting to become impossible to discern real artists' images and easily creatable mass-produced images, I can't tell if this is a good thing or a bad thing. And that's coming from somebody who loves this type of stuff.

Is it possible they just used AI for the background? That would explain a lot of things, including the oddly specific resolution and the fact that the Eevee is far more coherent than I'd ever seen an AI come close to producing. And I really doubt they spent that much time training an AI model and fiddling around to get the perfect Eevee in that unusual style, considering this is their first and only Eevee, after putting out completely different pieces not long before it.

As for the rest, that seems a bit nitpicky. It's possible the artist is using unorthodox programs and methods, such as GIMP and special effect brushes (fur brushes, etc.). I've seen a few other small-time artists long ago making beautifully rendered (albeit somewhat off-looking) drawings this way.

fishyvap said:
Is it possible they just used AI for the background?

Entirely possible, that would make a lot of sense actually.

leafeon said:
Entirely possible, that would make a lot of sense actually.

Furries have been training stable diffusion on e621 datasets for a while now, and it can make very coherent results. People have probably gotten away with posting untagged ai shit here already

kitaroking said:
Furries have been training stable diffusion on e621 datasets for a while now, and it can make very coherent results. People have probably gotten away with posting untagged ai shit here already

I've been in the stable diffusion Discord and I've yet to see anyone come close to generating coherent Pokemon, unless like 70% of their body is covered by a cloak or something.

I do paint over AI images as a base for my backgrounds, sorry for not clarifying that
＞︿＜

Quote from the uploader at https://inkbunny.net/s/2850878

fishyvap said:
I've been in the stable diffusion Discord and I've yet to see anyone come close to generating coherent Pokemon, unless like 70% of their body is covered by a cloak or something.

I'm in the Furry Diffusion server and know there's a particular prompt people have used to generate very similar looking pictures with Eevee and a few other pokemon, all of them at very similar low-resolutions.

https://i.imgur.com/89Mgprl.png
https://i.imgur.com/PYQ3Mib.png
https://i.imgur.com/F9PD9Mp.png

There's no doubt AI generation played a part in the making of this piece. I think it deserves the AI_Generated tag.

constellation said:
Quote from the uploader at https://inkbunny.net/s/2850878

Well that settles that.

constellation said:
I'm in the Furry Diffusion server and know there's a particular prompt people have used to generate very similar looking pictures with Eevee and a few other pokemon, all of them at very similar low-resolutions.

https://i.imgur.com/89Mgprl.png
https://i.imgur.com/PYQ3Mib.png
https://i.imgur.com/F9PD9Mp.png

There's no doubt AI generation played a part in the making of this piece. I think it deserves the AI_Generated tag.

Those are definitely way better than most I've seen, but they're still far off from matching the quality seen here. The style of those don't really match this one's, and they all have glaring issues that this one lacks.

I know NMNY himself said that AI-generated backgrounds are allowed, but I'm not sure what the tagging policy is for that. The wiki page for the ai_generated tag isn't at all helpful, either.

Update: According to a janitor, it should not be tagged with anything relating to AI.

The art style of this looks similar to NovelAI generated images.
https://twitter.com/rubai42/status/1577618268671483906
https://twitter.com/rubai42/status/1577251921693286401 (notice the arm on the 2nd image being similar in shape and shading.)
The soft airbrushed fur that is shown here for example. The eye style is also common with the kemono images being generated with NovelAI.

asfjkhkhjfsajaf said:
The art style of this looks similar to NovelAI generated images.
https://twitter.com/rubai42/status/1577618268671483906
https://twitter.com/rubai42/status/1577251921693286401 (notice the arm on the 2nd image being similar in shape and shading.)
The soft airbrushed fur that is shown here for example. The eye style is also common with the kemono images being generated with NovelAI.

I'm not surprised that an AI model trained on a Kemono dataset is going to produce eyes made in a kemono style. Now try producing a decent Eevee with that same dataset.

Regardless, the artist has already clarified that only the background uses AI elements, so this is a pointless debate to have.

fishyvap said:
Now try producing a decent Eevee with that same dataset.

By now there are probably plenty more AI-generated images that have been passed off as original work, but I found an image on the Furry Diffusion discord that not only has the same dimensions as this image but is also depicting an Eevee with a similar face and ears.
https://cdn.discordapp.com/attachments/1029418536675856425/1029655380617334794/unknown.png

Also I'm not sure if this character is popular at the moment, but it's worth pointing out that this guide features a slug cat with similar facial proportions and style as the slug cat this person had posted. Both the guide and this person have a sfw and nsfw version of their respective images too. NovelAI is specifically mentioned.
https://i.imgur.com/0hRMsVQ.png
https://www.furaffinity.net/view/49482918/

While people can get away with it at this point I believe dishonesty is not going to contribute to the acceptance of this medium. Seeing that img2img takes some amount of drawing, it is odd that the author feels that they have to lie about their method.

asfjkhkhjfsajaf said:
By now there are probably plenty more AI-generated images that have been passed off as original work, but I found an image on the Furry Diffusion discord that not only has the same dimensions as this image but is also depicting an Eevee with a similar face and ears.
https://cdn.discordapp.com/attachments/1029418536675856425/1029655380617334794/unknown.png

That's supposed to be an Eevee? 💀
I think the dimensions thing was already explained, btw

asfjkhkhjfsajaf said:
Also I'm not sure if this character is popular at the moment, but it's worth pointing out that this guide features a slug cat with similar facial proportions and style as the slug cat this person had posted. Both the guide and this person have a sfw and nsfw version of their respective images too. NovelAI is specifically mentioned.

https://i.imgur.com/0hRMsVQ.png
https://www.furaffinity.net/view/49482918/

While people can get away with it at this point I believe dishonesty is not going to contribute to the acceptance of this medium. Seeing that img2img takes some amount of drawing, it is odd that the author feels that they have to lie about their method.

That's an interesting process and I can see someone maybe pulling off something like this with a ton more inpainting, compositing, and redrawing... But at that point, why put in all that effort? You could hardly call it AI-generated by then

Also, the creator of that image is Weblure. They're an actual artist that to my knowledge was just messing around with the AI

abscondler said:
That's supposed to be an Eevee? 💀

You're being intentionally obtuse if you do not see the similarities between the two images.
I understand that you have an interest in AI-generated images, and that methods such as the one Weblure use can be a good example for how AI can potentially aid artists, but lying about the thing you're trying to destigmatize is counterproductive.

asfjkhkhjfsajaf said:
By now there are probably plenty more AI-generated images that have been passed off as original work, but I found an image on the Furry Diffusion discord that not only has the same dimensions as this image but is also depicting an Eevee with a similar face and ears.
https://cdn.discordapp.com/attachments/1029418536675856425/1029655380617334794/unknown.png

LOL WHAT IS THAT

asfjkhkhjfsajaf said:
You're being intentionally obtuse if you do not see the similarities between the two images.
I understand that you have an interest in AI-generated images, and that methods such as the one Weblure use can be a good example for how AI can potentially aid artists, but lying about the thing you're trying to destigmatize is counterproductive.

What are you even accusing them of lying of? I asked for a decent Eevee in that art style, and you came up with that abomination. As I'd already said, it's not surprising that a kemono dataset produces art in common kemono styles. The fact that the best you could come up with was some obese anthro Eevee look-alike says a lot, but it doesn't say what you're hoping for. Most kemono artwork is based around anthros, so getting a perfect on-model Eevee in that style from this dataset is near impossible.

fishyvap said:
LOL WHAT IS THAT

What are you even accusing them of lying of? I asked for a decent Eevee in that art style, and you came up with that abomination. As I'd already said, it's not surprising that a kemono dataset produces art in common kemono styles. The fact that the best you could come up with was some obese anthro Eevee look-alike says a lot, but it doesn't say what you're hoping for. Most kemono artwork is based around anthros, so getting a perfect on-model Eevee in that style from this dataset is near impossible.

NovelAI can generate ferals in a kemono style, too, as I've already shown you. You can play dumb and downvote all you want but you're starting to look emotional. There's no use in engaging with you people further.

asfjkhkhjfsajaf said:
NovelAI can generate ferals in a kemono style, too, as I've already shown you. You can play dumb and downvote all you want but you're starting to look emotional. There's no use in engaging with you people further.

"You're starting to look emotional!!!" he says, running away from their calm debate 😂

asfjkhkhjfsajaf said:
You can play dumb and downvote all you want but you're starting to look emotional.

abscondler said:
>Instantly gets 2 downvotes in less than a minute

my guy is straight up botting in a furry porn debate 💀

Irony.

asfjkhkhjfsajaf said:
There's no use in engaging with you people further.

Agreed.

abscondler said:
>Instantly gets 2 downvotes in less than a minute

my guy is straight up botting in a furry porn debate 💀

Project harder my dude

knotty_curls said:
Project harder my dude

How is pointing out that my post instantly got 2 downvotes right after I posted it projecting?

And what am I even reading here?? 💀

What's any of this got to do with me? Did you ask them where they got the screenshot? How was I 'trolling' any more than anyone else was? Did you also leave a note on all the accounts 'trolling' and instantly downvoting me? Because I'm sure "asfjkhkhjfsajaf" is totally not an alt account 😂

And what's the reason for suddenly deleting this when it already got approved after the artist confirmed it wasn't AI-generated outside of the background? Shit even if it wasn't just the background you guys allowed a self-admitted AI touch-up to stay here, so why pick on this image specifically? Even the other side agreed that if it was AI-generated this would've taken a bunch of repainting and shit (much more than that pony headshot), so why give one special treatment over the other?

Do y'all ever talk to each other to get on the same page? Posting on e621 is like a box of chocolates you never know what you're gonna get 😔

abscondler said:
How is pointing out that my post instantly got 2 downvotes right after I posted it projecting?

And what am I even reading here?? 💀

What's any of this got to do with me? Did you ask them where they got the screenshot? How was I 'trolling' any more than anyone else was? Did you also leave a note on all the accounts 'trolling' and instantly downvoting me? Because I'm sure "asfjkhkhjfsajaf" is totally not an alt account 😂

And what's the reason for suddenly deleting this when it already got approved after the artist confirmed it wasn't AI-generated outside of the background? Shit even if it wasn't just the background you guys allowed a self-admitted AI touch-up to stay here, so why pick on this image specifically? Even the other side agreed that if it was AI-generated this would've taken a bunch of repainting and shit (much more than that pony headshot), so why give one special treatment over the other?

Do y'all ever talk to each other to get on the same page? Posting on e621 is like a box of chocolates you never know what you're gonna get 😔

Hey, just so I don't sound like a backtracking cunt - I did bring it up. It was just conveniently forgotten while we looked a bit deeper.

Anyways, I dunno why you kept your head down so long yet decided to get so invested in this. The post is restored, but at what cost

fishyvap said:
I see no evidence of this being AI generated. You can make just about any style with AI, since it learns from existing images.

you can tell from the curtain and light on the table

This image wasn’t AI assisted, the entire image is generated by AI. Idk what that leafeon was talking about (no offense to him). This is clearly completely AI generated using some form of FluffyRock stable diffusion offshoot. AI can definitely replicate an eevee effortlessly with the right base model/checkpoints, and if you have a model that can’t, there are Loras for that. The dead giveaway is definitely the art style, the low abstract resolution, and the overall graininess of the photo overall. I’ve seen billions (hyperbole) of eevee generations from many different people that use the same exact “art” style.

I’m starting to really dislike the fact that AI has come to the point where most people cannot distinguish between artificial images and images that are made by humans. Of course you could always just run the image through an ai detector to verify that the image in question has or hasn’t been artificially generated.

looks a lot like https://e6ai.net/posts/638

ottmutt said:
AI can definitely replicate an eevee effortlessly with the right base model/checkpoints, and if you have a model that can’t, there are Loras for that.

At the time of my post, there wasn't anything even remotely close to getting proper Pokemon. Loras didn't exist.
Nowadays, you can find some incredibly accurate eeveelutions in the furry diffusion server. My old post seems incredibly outdated now, and that's because it is.

vv7 said:
looks a lot like https://e6ai.net/posts/638

They could have just used this image as a base.