AI Waifus and You 101

[Alchemyth]

Here's Irys to complete the promise girls:

AIRyS by @minimalisticmasteringengineer875 | Suno

female singer, city pop, japanese rock,, catchy, upbeat, bass, melodic, drum, pop rock song. Listen and make your own with Suno.

suno.com

 

[Panopticon]

I used an Ai generator to try and create an image of Pekora themed currency using a picture of her model. What I got, instead, was Joe Biden.

 

[Clem the Gem]

Just wanted to announce that I released a Lui LORA and the first thing someone did with it was use it to make furry art. I'll take what I can get I guess.

Takane Lui (Hololive) - 4 Outfits - v1.0 | Stable Diffusion LoRA | Civitai

Because no one else has done it for some reason. I still don't 100% know what I'm doing with multi-outfit LORAs, so until someone else makes a bett...

civitai.com

Also someone else released a probably better one than mine mere hours later, while mine is not even showing in the search results, so that's cool.
Maybe a bad time to make i t anyway, since what with the recent new outfit reveals for other Holox members, it's going to be obsolete soon anyway.

 

[God's Strongest Dragoon]

Just wanted to announce that I released a Lui LORA and the first thing someone did with it was use it to make furry art. I'll take what I can get I guess.

Takane Lui (Hololive) - 4 Outfits - v1.0 | Stable Diffusion LoRA | Civitai

Because no one else has done it for some reason. I still don't 100% know what I'm doing with multi-outfit LORAs, so until someone else makes a bett...

civitai.com

Also someone else released a probably better one than mine mere hours later, while mine is not even showing in the search results, so that's cool.
Maybe a bad time to make i t anyway, since what with the recent new outfit reveals for other Holox members, it's going to be obsolete soon anyway.

It could be worse, it could have been futa furry porn.

 

[Clem the Gem]

It could be worse, it could have been futa furry porn.

Spoiler

Probably

 

[Koronesuki]

She's already a hawk girl. Why on earth would someone want to turn her into a furry?

 

[Clem the Gem]

She's already a hawk girl. Why on earth would someone want to turn her into a furry?

Liked the outfit without knowing the character maybe? Undertandable I guess.

 

[Alchemyth]

From the guys who left Stable Diffusion because of various forms of fuckery, Flux: https://blog.fal.ai/flux-the-largest-open-sourced-text2img-model-now-available-on-fal/
12B model, so far anons on /lmg/ think it's better then DALL-E, close to Midjourney tier. I don't really bother with image generation stuff so I have no idea if they're right or not.
FP8 version for vramlets: https://huggingface.co/Kijai/flux-fp8

 

[Clem the Gem]

From the guys who left Stable Diffusion because of various forms of fuckery, Flux: https://blog.fal.ai/flux-the-largest-open-sourced-text2img-model-now-available-on-fal/
12B model, so far anons on /lmg/ think it's better then DALL-E, close to Midjourney tier. I don't really bother with image generation stuff so I have no idea if they're right or not.
FP8 version for vramlets: https://huggingface.co/Kijai/flux-fp8

Interesting, this looks promising. I haven't read everything yet, but are you able to run it locally?

My first and only prompt so far using their website:


"an elf girl wearing japanese clothes leaning on a balcony overlooking a rural town at night."
Yup, that's pretty much exactly what I asked for. Good job.

 

[Alchemyth]

Interesting, this looks promising. I haven't read everything yet, but are you able to run it locally?

My first and only prompt so far using their website:

View attachment 75022
"an elf girl wearing japanese clothes leaning on a balcony overlooking a rural town at night."
Yup, that's pretty much exactly what I asked for. Good job.

The huggingface safetensor I linked should run fine in comfyUI or whatever backend you use locally. It's a quant though, so expect slightly worse performance and results compared to the full weights.

 

[Clem the Gem]

I can report it handles text perfectly. I was never able to give the disastrous Stable Diffusion v3 a try since it didn't work on Automatic1111, but now it looks like I don't need to.
Back when I did the pepeloni Suno songs I wanted to give them some proper album covers, but I couldn't do it with SD1.5, XL or Pony. Every prompt I've run through Flux has been perfect.

Spoiler

(I swear I'm not obsessed with the pepeloni)

Spoiler: Edit..

I decided now is the time to launch my AI-generated Youtube music career. Behold:

 

[Azehara]

From the guys who left Stable Diffusion because of various forms of fuckery, Flux: https://blog.fal.ai/flux-the-largest-open-sourced-text2img-model-now-available-on-fal/
12B model, so far anons on /lmg/ think it's better then DALL-E, close to Midjourney tier. I don't really bother with image generation stuff so I have no idea if they're right or not.
FP8 version for vramlets: https://huggingface.co/Kijai/flux-fp8

Not bad but it seems to be cucked. I wanted to make a big titty elf warrior but it didnt show anything. Might be good for overall background photos.

Interesting, this looks promising. I haven't read everything yet, but are you able to run it locally?

My first and only prompt so far using their website:

View attachment 75022
"an elf girl wearing japanese clothes leaning on a balcony overlooking a rural town at night."
Yup, that's pretty much exactly what I asked for. Good job.

i would assume oyu would have to download a multi GB model like it was with Stable Diffusion.

---

NovelAI added a few extra tools. One turns an image into a sketch, the other turns it into lineart. There is also a background remover which can be nice to get art for a thumbnail without having to manually delete the background which can take a few minutes. There is also a tool that lets you change the color or things (or fully color a B&W image) and enother one that will "de-clutter" things. Basically if you have a manga panel, it gets rid of all of the text and effects in the page.

 

[Clem the Gem]

Good news, Flux is now available to run locally on Forge. For those that don't know (like me a few days ago), Forge is basically a better version of the Automatic1111 webUI with some nice performance enhancements. Previously Flux was only working on ComfyUI, which I also installed for the first time and hated.

https://www.patreon.com/posts/flux-in-forge-to-110007661

Image generation is slower than SD1.5 / XL / Pony, but that's to be expected. More importantly I haven't had any issues with running out of memory (on a 12GB card).

(I've run out of time now, but I'll add some example pictures later)

Edit: Some example pictures. I deliberately kept the prompts very short and simple without using any detail tags. I might have got better quality images if I had - the anime style ones are a bit lacking, but I wanted to demonstrate how well it can interpret plain English prompts.
The same propmps are used, tagged with either "realistic" or "anime style"

"a fox on a cliff overlooking a beautiful fantasy landscape with waterfalls and floating islands"


"girl sitting by a fountain in a crowded medieval fantasy town square, smile, looking at viewer"


"a dog in an abandoned overgrown city"


"anime school girl with pink ponytail hair and blue eyes aiming a rifle in a ruined city, one eye closed, dramatic, cinematic lighting, fire, smoke, backlighting"


"a fox girl running along thhe roof of a moving freight train, leaping between carriages, desert background"

 

[Clem the Gem]

Can't believe it's been almost half a year. I will now attempt to revive this dead thread with another episode of,

"Showcasing Months-Old Technology That I've Only Just Found Out About"​

This time we have something called IllustriousXL and its spin-off, NoobAI-XL. From what I understand, NoobAI is to Illustrious what Pony is to SDXL, and has itself become a base for further models. In my examples I've mostly been using a mix called KonPaEvo.

Features:​

• Able to reproduce characters with amazing accuracy without any help from LORAs. Of course the more well-known the character, the better results you get.
• Very good at separating said characters. If you ask for two characters, you're going to get them, and not a mashup of them merged into one person.
• Seems to understand your prompt much better, while still using simple booru style tags.
• Much better than Pony at dark scenes, and lighting in general.
• Also more detailed backgrounds compared to Pony
• No need to upscale. Preview nicely detailed results before commiting to a time-consuming upscale.
• No need for loads of detail tags and huge negative prompt. I often used no negative at all *
• Reasonable system requirements. Unlike Flux, you can get images without needing to go away for a tea break each time you hit generate.

* This does not seem to be the case for the NoobAI-XL base model. From what I've read, it is very powerful, but you have to prompt in a particular way, with negatives, artist names, and upscaling required before you can get a good image. The model mix I'm using does works well without any of this.

And of course. here are a few examples.
Once again, not a single LORA was used in any of these. Nor was regional prompting (apart from one image), or any other extension. Just the most basic of prompts.

Spoiler: Multi-Character

The 3rd one is the one image I used regional prompting on, just so I could split it into 4 sections. Still, the only thing prompted was "ookami mio \(1st costume\)", "ookami mio \(casual\)", etc.

Spoiler: Lihgting, more multi-character

Spoiler: Holox

Lastly, I had a lot of fun with these images and couldn't decide on just one to post, so have a bunch of them

 

[Clem the Gem]

Adding on to the last post, I just had to share this.
Having accurate outfits on a character and swapping clothes around is old new now, but I was kinda blown away by how it can also figure out how to draw clothes that aren't being worn, while matching them to the character:

 

[Koronesuki]

For some reason these images aren't showing when I view the post, but I see them while editing... anyone else? They link to the full size images correctly at least

I can't see them unless I click on them, except for the last image, which I can't see at all.

 

[Faceless Waifu]

Speaking about these NoobAi-XL thingie, do they have the same way of prompting things like PonyXL (needing score_9, etc) or is it similar like most ai art generator prompt these days? Part of the reason why I don't want to touch Flux is because the way it goes (needing to describe instead of just tags), so I'm curious about it.

I think I've see those stuff already available on online AI art generation site, which makes me keen to give it a try.

 

[Clem the Gem]

OK I think catbox is having a moment.. thought I'd try re-uploading the images, and now my supposedly uploaded images aren't even showing up in my files. I'll give it a while then check back. I'll just use imgbb instead. Should be showing now.

Speaking about these NoobAi-XL thingie, do they have the same way of prompting things like PonyXL (needing score_9, etc) or is it similar like most ai art generator prompt these days? Part of the reason why I don't want to touch Flux is because the way it goes (needing to describe instead of just tags), so I'm curious about it.

I think I've see those stuff already available on online AI art generation site, which makes me keen to give it a try.

I've been using the same few quality tags appended to the start of every prompt, but now that I actually just tested a few images, it doesn't really seem to make any difference if left out.

For the prompt itself, it's purely tags. 1girl, theirname, sitting, outdoors, whatever. You can always download my images and load them into the PNG info viewer (even though they're jpgs - another thing I learned not that long ago).