Home Artists Posts Import Register
The Offical Matrix Groupchat is online! >>CLICK HERE<<

Content

So, apparently AI image generation is now a thing that's readily available on the consumer level. Furthermore, NovelAI has an AI image generation function that specifically makes anime characters. Also, since I messed around with their service a few years ago, even as a free user, I had 3k generation tokens lying around from the subscription, asking me to play with it. So I did.

This was my very first try on Snowy, and what convinced me that this actually kind of works.

Here she is being a neat, dignified meido. She's a really nice, innocent girl, so don't lewd her.

Hey! What did I just tell you!? Stop it!

Celestial outfit Angie also turned out pretty well, and it even managed to get the "That's not an outfit, that's a freaking bed sheet!" vibe across.

Ammy needed a couple of tries, but turned out pretty well. Don't ask me where the bubbles came from, only the AI knows.

This was the best Elly I could manage. The AI apparently doesn't like horns. On the other hand, for some reason every Elly image I generated had dummy thick hips and gag boobs.

Seriously.

Every single one of them.

Since I used Elly's banquet outfit, I decided to do the same for Judy, and she turned out pretty cute as well.

Unfortunately, the AI apparently can't parse girls having bows in their hair, and put a bow on her head (as in, the kind you use to shoot arrows), which was hilarious, but unfortunately I forgot to save it. Oh well.

Let's also have a casual Judy as well. I really wanted to have her either eat a sandwich, or use her pink phone to take notes, but the AI is kind of bad with fingers, and it tends to result is all kinds of body horror.

Out of the 3k tokens, getting a semi-decent Bel took me close to 900, and while I'm still not fully satisfied, it's okay, I suppose.

Penny's images made me realize I might have a thing for girls with freckles.

I mean, come on! That's cute as heck!

Let's not ignore the menfolk. This one is hilarious because the only instructions I gave were "Old man, sits at desk, robe, panama hat, sean connery", and this was the first result. Say what you will, Lord Grandpa is a dapper guy.

Tried as I might, this is the closest I could get to what a Faun is supposed to look like.

This one is Brang-ish, I suppose.

Ideally, the Fauns are supposed to be more straightforwardly ram-like, like this, but I couldn't really generate any good matches without either going full furry, or the portraits above. Oh well, it's AI, not magic.

Last, but not least, a wild Josh in his natural habitat, consuming his prey that he absolutely didn't mooch off someone else.


I hope you find these amusing. I still have a bunch of tokens left, so if you guys like this, say so in the comments, and I might try to generate a few other characters for shits and giggles. 

That's all for now. See you soon-ish with the next part, and till then, ciao!

Comments

Jacob Roy

Penny is cute as heck and Judy definitely pulls of background character as well as expected

CaptainZM

The real question is what it thinks of the goldfish poop gang. (I honestly can't remember if that was the right name.)

egathentale

Tried it. The AI can't really parse the pompadour hairstyle, and if I try to generate a heavy-set/fat delinquent, I get a bunch of bara art style weirdness. As I said above, while this works astonishingly well, it's not magic.

Javidom

so, the pictures are REALLY GREAT, thank you very much regarding the bow in hair - try "ribbon" instead? And yeah, redheads with freckles are best; even better if they have bright green or blue eyes ^^

Steven

Cute, cute, cute, cute, cute!. Also have a thing for freckle redheads.

cris

This is how they look like now it’s cannon in my head thanks !

Danielle Warvel

What is this thing called again? Do you have a link?

closeded

Novel AI's image gen is amazing. I can't wait for them to add integration with their adventure mode. If you didn't know, getting stable diffusion running locally is trivial if you've got an rtx card. Even then, Novel AI's models are trained specifically on tagged anime images, so nothing matches their quality for that. Specifically, any of the tags on danbooru work, even if they don't show up in Novel AI's tag list; they censored some of the tags from showing, but not from affecting the image generation. On that note... for your Elly images, try adding the 'small breasts' tag or something similar to offset what happened to her. Edit: I got a few ok images for Elly with "blonde hair, twin drills, green eyes, dragon horns, dragon girl, school uniform, dragon wings, standing in front of the school gate," on both of their anime models. Oddly, the lighting was far better on the full model than the curated one.

egathentale

Oh, so it's based on the booru's image base? That explains a LOT, especially the mountains of suggestive images in a distinct bara style the AI threw at me while I was trying to get the Fauns right.

closeded

Yep. They stripped down a lot/most of the images available, and trained the ai on comma separated lists of tags. One thing I've been doing is finding something I like on one of the boards loading the image into img2img and copying the tags into the prompt.

Andrew Fox

Snowy got me feeling some things... 🥵