AI Waifus and You 101

[Watamate]

Proctor's edit: This technology is synonymous with degeneracy and pornography at the moment. It has incredible potential and is worth discussing, but if you only want to use it to live out your fetishes of making albums full of Gura shark vore or shit like that, keep it to yourself.

@Clem the Gem has made an updated and more in-depth and up-to-date Guide deeper in the thread.

Part 1/Part 2

I wanted to start a thread centered on generating AI images where we can share tips, tricks, good prompts or even good generations.

The biggest help on getting you started is the Voldy Retard Guide.

--GUIDE--

The definitive Stable Diffusion experience ™ (Special thanks to all anons who contributed) ---FEATURE SHOWCASE & HOWTO--- SD News Japanese guide here 日本語ガイド (JP Resources) NovelAI FAQ Interested in LoRA training? Try the Quickstart LoRA Guide --GUIDE-- Official Stable Diffusion Alternate Mod...

rentry.org

I will emphasize that the current requirements to get started are as follow:

• 16gb of RAM
• Nvidia Maxwell (7xx) or newer with at least 2gb VRAM
• Linux or Windows 7/8/10+
• 20gb disk space

If you meet all these requirements you should have no to minimal issues following the guide I linked. I will say, if your main goal is to recreate images featuring Hololive members or other potentially well know vtubers, I have had the most success with NovelAI specifically. Which the guide also shows you how to install and use it. Keep in mind however that not all AI vtubers are created equal. There are some members that I've been able to replicate incredibly well and others that don't. These are my personal experiences and depending on how much work you put in, you could very well have different results.

If you do go the NovelAI route, a common problem that I ran into myself was not applying the settings after changing the "Stop At last layers of CLIP Model" to 2.

Some interesting resources to help with thinking of which tags work are basically any image board along the lines of Danbooru, Gelbooru, etc. To help you with that, you can use the following Javascript along with either tampermonkey or greasemonkey.

GitHub - JetBoom/boorutagparser: Copy current post tags and rating on boorus and illustration2vec in to the clipboard for easy import in to a program or another booru.

Copy current post tags and rating on boorus and illustration2vec in to the clipboard for easy import in to a program or another booru. - JetBoom/boorutagparser

github.com

And I have also added a simple zip file that has a lot of potential tags, fairly well categorized. (Not mine)

I'll add some of my personal examples now. I will also reiterate, some members work 99% of the time if you use tags that accurately describe them. Botan, Fubuki and Mio are members that I've had the most success with. Others are hit in miss. Members that I've tried that fit in this category are people like Subaru, Towa, Suisei, Nene. And some probably won't and will just give you a fairly bland anime lady, for example, Achan, Nodoka, etc.

You're not limited to SFW stuff I should add if you're into that. Don't let your degenerate dreams be dreams.

Spoiler: Suisei

Spoiler: Botan

Spoiler: Aqua

Spoiler: Random Members

Spoiler: Degeneracy (Mildly 18+) AKA I like Botan a lot

 

[The Proctor]

I was literally just thinking yesterday of making this thread myself. Get out of my head.

 

[Jan K. Hater]

Another good resource in addition to the Voldy Guide:

SD RESOURCE GOLDMINE

character limit hit, move here for the next version, this will no longer be updated: https://rentry.org/sdupdates2 Warnings: Ckpts/hypernetworks/embeddings are not interently safe as of right now. They can be pickled/contain malicious code. Use your common sense and protect yourself as you would...

rentry.org

That's news updates and basically a trove of resource links to shit like models, hypernetworks, etc. Use that stuff at your own risk.

I've been...well, I actually fucked up my install and it took me days to figure out a brute force workaround...but I've been messing with Umi Ai, a wildcard generator that seems to be put together by what seems like some smarmy otherkin schizo. It gives you options to generate random waifus by filling in some wildcard blanks for you, but might actually provide an easy way to wrap up a bunch of prompt descriptions of a vtuber into a single tag. I'm not sure though.

 

[Fucking YTs]

While I'm not getting great results yet, this is a lot of fun. I think sharing tags would be nice as well, so I've added mine.

Spoiler: Cyberpunk Pekora-style

---

Spoiler: Tags used for the above

Prompt:
cinematic lighting, unity, detailed neon operator, cyberpunk futuristic, neon, reflective puffy coat, two small bunny ears, blue hair, braids, carrots, brown eyes, usada pekora, standing with a gun, bokeh

Negative Prompt:
lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts,signature, watermark, username, blurry, artist name, no aliasing

 

[Watamate]

While I'm not getting great results yet, this is a lot of fun.

Spoiler: Cyberpunk Pekora-style

View attachment 3869
View attachment 3870

These are my prompts for my few Pekora generations, but hopefully it gives everyone a good idea what to look for and how to describe them.

((Usada Pekora)), (Hololive), (Rabbit girl), virtual youtuber, rabbit ears, brown eyes, blue hair, white hair, twintails braid, bangs, crossed bangs, ((Yuuki Hagure)),

This is just Pekora, no setting, camera angle, clothes, etc.

Something I like to do as well is to have a hard stop between the things I'm describing. Starting with how I want the picture to look, who I want to picture to be of, what I want them to wear and finally setting, lighting and camera settings or stuff I forget. So for example:

cinematic lighting, highres, best quality, masterpiece, (These are always safe bets)
((Usada Pekora)), (Hololive), (Rabbit girl), virtual youtuber, rabbit ears, brown eyes, blue hair, white hair, twintails braid, bangs, crossed bangs, multicolored hair, ((Yuuki Hagure)),
small breasts, denim shorts, bomber jacket, white leotard, leotard under clothes, belt, black gloves, single thighhigh, holster, waving, fingerless gloves, sunglasses,
1girl, solo, official art, cowboy shot, outdoors, rooftop, rain, dutch angle, contrapposto

 

[Fucking YTs]

These are my prompts for my few Pekora generations, but hopefully it gives everyone a good idea what to look for and how to describe them.

((Usada Pekora)), (Hololive), (Rabbit girl), virtual youtuber, rabbit ears, brown eyes, blue hair, white hair, twintails braid, bangs, crossed bangs, ((Yuuki Hagure)),

This is just Pekora, no setting, camera angle, clothes, etc.

Something I like to do as well is to have a hard stop between the things I'm describing. Starting with how I want the picture to look, who I want to picture to be of, what I want them to wear and finally setting, lighting and camera settings or stuff I forget. So for example:

cinematic lighting, highres, best quality, masterpiece, (These are always safe bets)
((Usada Pekora)), (Hololive), (Rabbit girl), virtual youtuber, rabbit ears, brown eyes, blue hair, white hair, twintails braid, bangs, crossed bangs, multicolored hair, ((Yuuki Hagure)),
small breasts, denim shorts, bomber jacket, white leotard, leotard under clothes, belt, black gloves, single thighhigh, holster, waving, fingerless gloves, sunglasses,
1girl, solo, official art, cowboy shot, outdoors, rooftop, rain, dutch angle, contrapposto

That's awesome stuff.

This is made for horror:

 

[The Proctor]

One question I have about this stuff; how good it is for OC's and portraits? I used to play a lot of TTRPG's and more informal roleplaying stuff, and it was standard practice in my group at the time when we were doing stuff long-term over IM to emphasize character actions with an appropriate expression. Naturally this led to an awful lot of trawling safebooru, etc, looking for images that vaguely matched the characters we were playing. Here's an example;



Obviously these are all cropped from different images. The lighting is different, even the artists are different. How hard would it be to get one of these A.I to crank out the same character in different expressions, like this?

 

[Watamate]

One question I have about this stuff; how good it is for OC's and portraits? I used to play a lot of TTRPG's and more informal roleplaying stuff, and it was standard practice in my group at the time when we were doing stuff long-term over IM to emphasize character actions with an appropriate expression. Naturally this led to an awful lot of trawling safebooru, etc, looking for images that vaguely matched the characters we were playing. Here's an example;

View attachment 3873

Obviously these are all cropped from different images. The lighting is different, even the artists are different. How hard would it be to get one of these A.I to crank out the same character in different expressions, like this?

I think @Jan K. Hater was trying out something new that seemed to have features hinting towards stuff that you're looking for.

Get more from Umi, the Cryoverse AI on Patreon

creating Building Virtual Universes

www.patreon.com

It's free and seems to work on Stable Diffusion as well. Again full credit to @Jan K. Hater.

Personally it seems a bit too complex for me, but I might try it at some point.

 

[Jan K. Hater]

One question I have about this stuff; how good it is for OC's and portraits? I used to play a lot of TTRPG's and more informal roleplaying stuff, and it was standard practice in my group at the time when we were doing stuff long-term over IM to emphasize character actions with an appropriate expression. Naturally this led to an awful lot of trawling safebooru, etc, looking for images that vaguely matched the characters we were playing. Here's an example;

View attachment 3873

Obviously these are all cropped from different images. The lighting is different, even the artists are different. How hard would it be to get one of these A.I to crank out the same character in different expressions, like this?

Yeah, you can do stuff like this pretty easily. If you've trawled existing images for stuff that fits for TTRPG tokens and junk, you have the right expectations going in. The basic way these models work is essentially like search engines that look up results for your prompts and then deliver the results in a collage that it stitches together. So if you were generating the character in your example, that'd be pretty easy, but you're gonna get an amount of random weirdness thrown in. Here, I'll try it now:

this was first 4 results based on this prompt: masterpiece, best quality, hires, detailed face, face focus, beautiful girl with green hair, glowing yellow eyes.

Change the prompt a little: masterpiece, best quality, hires, detailed face, face focus, giggling, :3, beautiful girl with green hair, glowing yellow eyes. art by frank frazetta

and you can see I'd need to work on the prompt and do some more generating to get that cat smile

It'd be some extra work to mess with hairstyle, outfit and junk. Would probably want to tone down the eyes too, but this should give you an idea of what 10 minutes of fucking around can do if you have a little experience.

Oh and @Tatsunoko is talking about a thing that'll add in random chargen elements, so going from the prompt I was using:

The Umi stuff is based on danbooru training though, so you can see the Frazetta influence gets really toned down.

 

[Nenélove]

MFW vtuber forum becomes virtual A.I. generated hentai sweatshop.

 

[The Proctor]

MFW vtuber forum becomes virtual A.I. generated hentai sweatshop.

I've inserted a mod edit to warn off degeneracy. I don't really care about stuff like bikini pics or whatever, but if it gets more explicit than that I will commence with the bonks. However, I am fascinated by the practical potential of this technology and encourage productive discussion/experimentation.

 

[Fucking YTs]

I've inserted a mod edit to warn off degeneracy. I don't really care about stuff like bikini pics or whatever, but if it gets more explicit than that I will commence with the bonks. However, I am fascinated by the practical potential of this technology and encourage productive discussion/experimentation.

I do like making horror. Is it OK to post stuff like that as long as there is a spoiler tag? I have no interest in generating porn.

 

[The Proctor]

I do like making horror. Is it OK to post stuff like that as long as there is a spoiler tag? I have no interest in generating porn.

If it's not obvious fetishbait then I don't really care.

@Tatsunoko In your example images, what's the 'failure rate', I.E prompts that made something obviously bad/broken/finished with amateur mistakes an actual artist would've solved immediately?

 

[Watamate]

If it's not obvious fetishbait then I don't really care.

@Tatsunoko In your example images, what's the 'failure rate', I.E prompts that made something obviously bad/broken/finished with amateur mistakes an actual artist would've solved immediately?

Fingers, hands, feet, etc have a very high failure rate. Even with the negative prompts. But in my opinion they're fairly easily overlooked. If you disregard that the success rate for my usual members like Botan is exceedingly high.

The only exception is when you add very specific poses. In which case it can be 50%, 20% or even lower. You'll get your subject with multiple bodies, heads, legs, arms. Butts morphing into boobs, etc. Very strange things. But if you brute force it, and generate enough images. You'll get something that you're satisfied with sooner or later.

And obviously the more powerful your GPU the faster you can brute force it till you get what you want.

 

[The Proctor]

Fingers, hands, feet, etc have a very high failure rate. Even with the negative prompts. But in my opinion they're fairly easily overlooked. If you disregard that the success rate for my usual members like Botan is exceedingly high.

The only exception is when you add very specific poses. In which case it can be 50%, 20% or even lower. You'll get your subject with multiple bodies, heads, legs, arms. Butts morphing into boobs, etc. Very strange things. But if you brute force it, and generate enough images. You'll get something that you're satisfied with sooner or later.

And obviously the more powerful your GPU the faster you can brute force it till you get what you want.

Interesting. I suppose it illustrates how hard it is to convey artistic concepts to an A.I that even a brain-dead human has an instinctive grasp on. Is text the only way to interact with this stuff? I have a limited knowledge of code and 3D/2D design, and I can think of multiple interesting ways this could go.

For example, you could solve the weird poses issue by providing more direct guidelines. Body artists are taught to break down the human figure into simple shapes and geometric patterns, which follow specific ratios in order to define the length of limbs and proportions. Instead of leaving it to chance, an artist could provide a gesture drawing template for an A.I to then construct a body and features around. Of course, I have no idea how that could be done, but it's obviously already correlating multiple things together from templates anyway, so more direct interaction seems a logical step forwards.

 

[Fucking YTs]

As someone who can't draw for shit, but who used to be decent in 3D (I used to be a pro at 3D modeling, but I hung it up for photography) this is a concept artists dream. I do wish there was a way to control the f-stop (beyond just typing "depth of field"), as the image composition prompts are a little lacking. It's nothing 5 minutes in Photoshop can't fix, though.

This was the only good IRyS I could generate on my first day messing with this.

The heterochromia should be swapped. No horns, as those are insanely random.

Also, Splatooning:

 

[ecol4]

For example, you could solve the weird poses issue by providing more direct guidelines. Body artists are taught to break down the human figure into simple shapes and geometric patterns, which follow specific ratios in order to define the length of limbs and proportions. Instead of leaving it to chance, an artist could provide a gesture drawing template for an A.I to then construct a body and features around. Of course, I have no idea how that could be done, but it's obviously already correlating multiple things together from templates anyway, so more direct interaction seems a logical step forwards.

You’ve somewhat got an idea of one of the potential next steps for this technology. The “ai” we have doesn’t have enough input data or alternatively doesn’t have the hard coded approaches like what you mentioned about giving them a sort of blueprint for anatomy, so they end up doing well on simpler body parts like arms, legs, torso which look similar from different perspectives relative to hands or feet(think about the amount of different overlaps between digits and how fingers could be hidden). Of course the issue with any sort of “blueprint” is making them abstract enough to be able to be used across different art styles and creatures(probably very far off for the latter for obvious reasons). Which loops back to the reason an ai was made(first/successfully) for digital art instead of a hard coded approach, it’s a lot easier to slap a bunch of labels onto pictures and data(the math+code wasn’t TOO hard in the grand scheme of things).

 

[Watamate]

MFW vtuber forum becomes virtual A.I. generated hentai sweatshop.

Noone would ever think to do that to your wife though.

Spoiler: sus

 

[Nenélove]

Noone would ever think to do that to your wife though.

Spoiler: sus

View attachment 3919View attachment 3920View attachment 3921View attachment 3922View attachment 3923

Damn, why is Sana cosplaying as my wife? And since when does she have white skin?
Edit: Alien abduction and replacement confirmed, though it seems that the aliens forgot that nene isn't a double H or whatever those things are, orrrr that humans only have 5 toes on each foot...

 

[Fucking YTs]

Damn, why is Sana cosplaying as my wife? And since when does she have white skin?
Edit: Alien abduction and replacement confirmed, though it seems that the aliens forgot that nene isn't a double H or whatever those things are, orrrr that humans only have 5 toes on each foot...
View attachment 3930

Obviously she lives in the Gattaca universe and she needs those to play the piano. Honestly, so far, if you only get six toes generating these images you're doing something right.