🍥｜anime

[keen hollow]

Frankly, I just need to buy some gum. Does she have spearmint?

[sharp imp]

Probably, might contain a spear though

i made a new recursive character creator prompt that should spit out completely random characters that still kinda make sense. its supposed to have a character concept background, or simple gradient background, but AI be doing AI shit sometimes lol

[river shale]

👀

[sharp imp]

interesting take on the uv tattoo idea i guess lol

[river shale]

How do you get multiple poses liek that

[sharp imp]

character concept art or character sheet with multiple views as a tag

its finiicky sometimes, and will sometimes render one of the views in lower quality unless you use a really good lora for it or controlnet/regional prompting. illustrious v2.0 seems to handle multiple characters pretty well so if you have a checkpoint with that as a base it should be fine

[river shale]

thank yaaa

[sharp imp]

anytime

[solar herald]

hello there, im looking for a anime LORA for Illustrious that is modern SFW and NSFW and works perfetcly for male and female characters. if some knows such a LORA please let me know, i find only Female or hardcore sexualized Hunk/muclemen LORAS. thanks in advance. and please no naruto or other mainstream shit

[keen hollow]

Vertlain, looks like Ilya

[gleaming flame]

which Ilya?

[gleaming flame]

now it looks like Ilya

[keen hollow]

Ilya Kuvshinov

[sharp imp]

"its not a phase mom"

[sharp imp]

thats quite a creative use of foreshortening lol

[gleaming flame]

;p

[sharp imp]

i like your art style. its very aesthetically pleasing without being overly sharp lol

[gleaming flame]

not "my" style, but thanks

[sharp imp]

cute but ready for war lol

[gleaming flame]

she going to go boom in 3... 2.. 1..

[sharp imp]

hahaha if she goes to grab a cake and grabs the grenade, probably

[gleaming flame]

she holds the grenade pin...

[sharp imp]

omg i just noticed that hahaha

[wooden sail]

So young already a mother of an aspiring photographer 👍

[river shale]

Location I've created for my D&D campaign :3

[sharp imp]

lol i created a custom campaign that ive never run. its still not totally finished, since im still learning what exactly a custom campaign needs, but most of the bones are there lol

thats really pretty though!

[river shale]

yep that sounds like me 😂

thank you :3

Rosebane Castle

[sharp imp]

shesssss meelllltttiiinnngg

[gleaming flame]

what about her?

[sharp imp]

ok that ones just cool lol

[glacial coral]

i need text that says "No mames guey"

[glacial coral]

What are some good models that’s trained on Illustrious 2.0?

[sharp imp]

uhhh if you search civitai for models and filter by newest, a lot of those are being created/merged or updated with 2.0 as a base

make sure you filter for illustrious as well of course

[glacial coral]

owkay

[sharp imp]

it depends on what you are looking for aesthetically. some models are merged to get a specific look, so what i like might not be what you like lol

[glacial coral]

im quite the more darker side of stuff like in fight scenes or the image above made with SIH that or very dreamy lighting

[glacial coral]

like this also

[sharp imp]

lol this was made with a custom merge and someone elses prompt that i tweaked lol, but the model itself doesnt especially do dark images. i think a lot of models can if you prompt it for it (color pallette, lighting, etc)

some just do it more readily than others for sure

i merged SinGlo into this, and Plant Milk Walnut, and i think those two are whats giving my model some darker stuff

[glacial coral]

ah i see image example wise ive seen noob ai v pred and sih look the best wit "dark" images

[sharp imp]

this was 100% illustrious

[glacial coral]

i just wanted to know how much dies illus 2 improve on fine tunes

[sharp imp]

illustrious v2.0 is more stable, better trained, and understands natural language prompts better. Thats why i used it as a base for my merge.

[glacial coral]

ah i see

well i want wai or hassle to use it

but its gonna be a while

so im struggling finding 2.0 fine tunes XD

[sharp imp]

just get the v2.0 base model and merge it with whatever other model you have in mind. i dont know how to get it to merge with NAI, thats above my IQ though

[glacial coral]

ok i should, standard SDXL VAE?

as for bias do i put it illus v2 bias or the original model i wanna merge it with bias?

[sharp imp]

i did weighted merges with Forge, so i can only speak from that experience. And yeah most any SDXL VAE should work with illustrious models, it did for me. Depends on which one your gonna use but Neptunia has a high contrast one you might like.

I did 0.7 weight for my first merge, then 0.5 for the next, 0.3 for the next and 0.2 for the final. Lol merged 4 models into Illustrious v2.0 to give them all even influence (not approximate, but as close as i could get)

since illustrious 2.0 is decent base but not refined stylistically, you wanna lean your weight in favor of the model you are merging to give it style

[glacial coral]

i use forge also

[glacial coral]

so if im only doing 3 models say illus 2.0, sih and wai

what values can u recommend?

[sharp imp]

and you want equal influence across the board at the end?

or do you lean more heavily toward one that you want to have stand out in the end result?

the order you merge them also matters

i didnt do a multi (3 way) merge, i did each merge individually so i could test the results

[glacial coral]

i mainly want as much benefits of 2.0 without yk getting the untrained parts if that makes sense and i mainly use SIH more than wai

[sharp imp]

youll get the benefits of 2.0 regardless, because its the base.

So if you use SIH more than Wai, merge wai first at 0.5 or 0.7 weight

then merge SIH at 0.5 or 0.3 weight to the new model you made.

[glacial coral]

owkay

[sharp imp]

the reason you merge wai first is because when you merge SIH into it, it will dilute the wai effect

and by going 0.5-0.7 on your first merge, you add more of the style you like to 2.0

[glacial coral]

so goes model a 2.0 then b model wai at 0.7 or 0.5 then new merge as model a then sih as model b with 0.5?

[sharp imp]

yep

label or name the models with what was merged and the weight it was merged, and test a few prompts youve used before to see how it turns out

if you find it lacking or want more or less influence, you can remerge with different weights

[glacial coral]

i see

ty

i shall come back with results

[sharp imp]

awesome! i look forward to it 😄

[glacial coral]

yes and thanks once again for the help

[sharp imp]

haha no problem, i had help getting to where im at with SD, its the least i can do to pass it on

[glacial coral]

yes :3 as u can see i have improved on gens tho

[sharp imp]

you have haha keep it up!

[glacial coral]

thank you i shall

im normally a flux merchant lmao

[sharp imp]

aint nothin wrong with that lol i cant run flux locally yet, so i make do

[glacial coral]

cant on forge or comfy but macos specific optimised applications are crazy good for keeping it stable and fast

if i had 24gb ram or more i can run on forge or comfy

but i have 16gb

so anything that isnt flux i can run on forge

[sharp imp]

yeah, im still rocking a 3080 rtx right now, so i just stick with SD for now haha

[glacial coral]

fair enough for Neptunia vae u got a link cant find it

[sharp imp]

one sec

https://civitai.com/models/1290283/neptunia-xlilnai-vae

[glacial coral]

ty

[glacial coral]

@sharp imp after i merge with sih i messed up results but the mix with illus 2 and wai is fine

[sharp imp]

how did you mess up?

[glacial coral]

i did what u said i cant actually mess it up as the first worked

second didnt

i didnt even change anything

yh idk whats wrong

[sharp imp]

the second didnt merge or the second merge didnt produce images?

[glacial coral]

it did produce images but its really messed up

lol

[sharp imp]

did you remerge the VAE?

[glacial coral]

remerge as in?

[sharp imp]

as in make sure the second merge also had the right vae selected.

[glacial coral]

ive kept the vae part as the sdxl vae for both i set once then didnt witch it i just switched the mdoel after the first merge was done

[sharp imp]

mmm. interesting. ive not had this problem lol

i wonder if the second model is a merge with something else and the VAE is incompatible, but that VAE should work with several model bases

including illustrious

[glacial coral]

hmm maybe sih is done diff

let me try hassle as second model

[sharp imp]

im not familiar with Sih

ahhh its NAI vpred. yeah a simple merge wont work for this, the model metadata is too different i think. It might be why the second merge was unstable.

[glacial coral]

ahhhh

ty for telling me

[sharp imp]

youd have to manually finetune it i believe, which is what i meant by above my IQ level lol

i merged all illustrious exclusive models for mine, because it rejected merging with any other model. might be able to force pony or sdxl into it, but it probably wont be as stable then.

[glacial coral]

got u

[glacial coral]

@sharp imp one more question if i want 2 characters doing things like say hu tao giving a piggyback ride to lisa how would i prompt that?

on flux its obv natural language but with tags idk how to do certain things like that

[sharp imp]

with illustrious 2.0 it can understand natural language, but its still gonna be finicky to get it to render multiple characters accurately without using regional prompting. sometimes it will, sometimes it wont.

Flux has significantly improved prompt understanding and adherence compared to other models. You could turn your CFG up to force it to align more closely with the prompt, but got too high and youll overbake the image

[glacial coral]

i usually use 7 cfg

[sharp imp]

and thats typically fine, but if your using a complex prompt or the model isnt getting it right, you can bump it to 8 or 9 to get it to more closely follow it

anything over 10 is risky

though if you use one of the CFG++ samplers/schedulers, their designed for high CFG parameters

[glacial coral]

ah how far can i push on that

[sharp imp]

i dont really know lol depends on the model i guess

[glacial coral]

With CFG++

[sharp imp]

i havent tested it yet

[glacial coral]

I keep tagging wrong messages sorry

[river shale]

[river shale]

I love those so much

What is the width and length you're using?

[sharp imp]

1536x576 lol but i upscale 2x

so in reality its rendered at 3072x1152

[river shale]

Gonna run my image though with that size, out of curiosity of what it does, if my PC will let me 💀

[sharp imp]

lol good luck! also, it helps to include "ultra wide angle" in the prompt, so it doesnt try to stretch the image or duplicate the subject

[toxic lotus]

Fire shit ret

Been a while man

[sharp imp]

heyyyy long time no see lol

[river shale]

Am tryina make tyranical queen, but I can't think of what else to add to the photo @_@

[sharp imp]

kneeling subjects?

[river shale]

oo yes

it keeps making her kneel

lol

Never been good at making images with multiple people in them

[sharp imp]

lmao yeah its tricky, its even harder when there isnt room in the composition to render it

the framing of her as the majority of the image makes it hard to add kneeling subjects, if it were a wider aspect ratio it might be easier

[river shale]

I'll tinker with it

[toxic lotus]

Ong I plan on being around a bit more sometime soon

Maybe

[dusky linden]

I think you haven't posted this in enough channels, i recommend doing it again /s

[woven gust]

that video sucks ngl

[tawny wasp]

What do you think of my latest creation Lora Noobai: I present to you Miko Kubota (Glitch Techs) 🙂

[wooden sail]

✌️

[tawny wasp]

🙂

[river shale]

Cute!

kinda reminds me of splatoon

[sharp imp]

oooohhh magical

[river shale]

🤩 You'd never imagine a tyrant lives there

[sharp imp]

hahaha

[tawny wasp]

Kris (11 22) Illustrious Apex_Ai
https://civitai.com/models/1622741?modelVersionId=1836566

[river shale]

This is SO cool!!!

[sharp imp]

hehehehe

its from my "Paint the World" wildcard

fun fact, according to chat gpt, given the number of wildcard files ive made, it would take me longer than the universe has been alive even if i were able to generate a billion images per second before i ever see a duplicate image

[echo pagoda]

Nice. Is that NTR mix by any chance?

[near elbow]

neko chrome angels of neon tokyo 🙂

[sharp imp]

nah its a personal merge

[echo pagoda]

It's very nice!

[sharp imp]

thanks!

[woven gust]

the first one has a cool style

[glacial coral]

@sharp imp RTX 6000 cloud system bought

i shall enjoy my time now

[sharp imp]

lol dope

[glacial coral]

41gb ram and 28 core CPU

gives me full windows 10 to do whatever

1gbps up and down speed

so im happy

[sharp imp]

sounds fast lol time to see if its worth it!

[glacial coral]

ill record my speed

like 5 years ago i was with same compay for their gaming package XD

until i got a local gaming pc

[sharp imp]

id be worried about them collecting my data and selling it, but thats cause im suspicious of eeeeevvvveeeerrryyything

[glacial coral]

ah fair

i trust this company im very much u but not that high of scale yk

[sharp imp]

yeah

[glacial coral]

said my stuff will be ready under 60 min its 20 min now

cant wait

[glacial coral]

@sharp imp Wai V14 2nd run 4-5 loras 1024x1360 with hi-res fix on 1.5x 20 steps 0.4 denoising and with Adetailer at custom res 1024x1024 and separate detail 50 steps and detail daimon way faster than my Mac quite happy

(I’m using —share)

[sharp imp]

Not too shabby

[glacial coral]

Ye

[bold ether]

(remember to set a password with it using --gradio-auth)

otherwise you re pretty much open to anyone.

[glacial coral]

By time it’s on hi-res my Mac would be 15 steps on sampling XD

[glacial coral]

Thank you so I just use that command in the webUI file and then it will ask for a password?

[bold ether]

--gradio-auth username:password

[glacial coral]

Ah ok ty

First time I’m using Parameters

[river shale]

Fae girl~

[tawny wasp]

hi guys,in the meantime I created a new Lora
I present to you: Miko Kubota (Glitch Techs)
in super saiyan 3 version 🙂

SSJ4 🙂

[tawny wasp]

https://civitai.com/models/1628861?modelVersionId=1843579

Miko Kubota From (Glitch Techs) NoobAI Apex_Ai

[native cave]

Can someone tell me how I can create images with this quality? I want to make backgrounds and regenerate them so they look like they’re taken straight out of an anime.

[near elbow]

use a decent model, e.g. this is "detailed empty office background art 1980s anime style" with flux schnell

[green quest]

have you tried anime screencap lora

[green quest]

yeah? tell that to my 4060 card

[sharp imp]

lil spooky gens for the morningbefore i leave for work

[native cave]

The problem I see is that it doesn’t generate an anime-style background like the examples I shared. You can tell it was made by AI. However, there are many AI-generated anime characters that don’t look like they were made by AI.

My goal is to achieve backgrounds with consistency similar to this image, but without the girl.

[wooden sail]

WAI-Shuffle-NoobAI with background, empty room, game cg, shopping shopping district, street, day, trees, anime screencap, sidewalk

[wooden sail]

Don't forget masterpiece tag

[wooden sail]

Why on Earth Reve produces images of SD1.5 style

[acoustic agate]

https://tenor.com/view/chrome-spongebob-future-fuuuuutuuurreee-gif-19192606

[keen hollow]

evenin'

[near elbow]

huh?

[wild charm]

After tens of tries, finally something where I say "I like it"

[sharp imp]

Nice! It's really good!

[glacial coral]

Oooo tell me the spooky prompt names

[sharp imp]

Their from my wildcards. Call one Echoes of Flesh and the other Error302: Body Not Recovered. Used a horrorcore lora and a dramatic lighting lora with them.

[river shale]

Your wildcards are op uwu

[sharp imp]

All with the same illustrious model I've used for most of my other stuff haha.

The horror core wildcards get way darker and more nsfw than most of my others. Couldn't share a lot of the gens here lol

[glacial coral]

Seems like wildcard on forge doesn’t show even on the new system lmao

I’ll check if I even have it installed one more time

I’m getting confused with Mac and pc installs lmao

[sharp imp]

I have wildcards on forge. It should be there. If not you may need the dynamic prompts extension. I can't remember if it was baked in or not

[glacial coral]

Alr can u help me set it up if it does show and installed

I know last time I had to grab some txt file or sum and put it in wild card directory

[sharp imp]

I can once I'm off work. Bunni just did it so she might be able to help as well.

[river shale]

^

👋

[glacial coral]

Ah ok

[glacial coral]

Hallo

[river shale]

Henlo ok so you wanna go to extensions > Avaliabe > Load from

Then look up: sd-dynamic-prompts

apply it, and reboot stable diffusion 🙂

THEN

you want to go to whatever folder you have forge in and go

FORGE > web ui > extensions > sd-dynamic-promts

And create a wildcard folder

[glacial coral]

Alr hold up let me boot into forge real quick

[river shale]

yep take your time 🙂

[glacial coral]

Bruh I left my AirPods case upstairs 😭

[glacial coral]

Alr so named wildcard or wildcards?

[river shale]

Whatever you prefer tbh

either one

[glacial coral]

Alr I just kept to wildcards

[river shale]

mmk then you create a text file

and you boop into it whatever it is you want it to randomize.

I can send you somthing to boop into one, gimme a sec.

if you'd like:

This is some anime style clothings

cyberpunk, bodysuit, neon accents
traditional, miko shrine maiden, robes
steampunk adventurer gear, goggles, belts
idol outfit, frills, sparkles
kimono, floral patterns, obi
battle-ready, magical girl costume, glowing gems
ninja outfit, scarf, kunai holsters
sci-fi military uniform, armor plating
maid outfit, lace and apron
gothic lolita dress, parasol, lace gloves
fantasy mage robe, oversized sleeves
catgirl outfit, ears, tail, bell collar
samurai armor, flowing hakama
punk streetwear, torn jeans, patches
princess ballgown, ribbons, tiara
demon hunter outfit, runic tattoos
healer’s robes, glowing staff
royal knight armor, cape, insignia
beach yukata, summer patterns
elegant vampire attire, high collar, cravat
hooded rogue outfit, daggers and belts
android battle suit, holographic details
school blazer, skirt, striped stockings
witch outfit, pointy hat, broom
post-apocalyptic scavenger gear, goggles
sailor uniform, cosmic accessories
maid cafe uniform, bunny ears
neon street dancer outfit, LED accents
forest elf tunic, leaf embroidery```

you can put the codes you want it to refrence into different text files then when you want to use them you go to forge and refrence the FILE NAME

so this I'd name "random anime clothes"

so in forge you'd refrence it like __random anime clothes__

and it will pull something random from that text file

you dont have to use that list if you dont want to btw ‼️ it's just an example of how the codes go into the text file 🙂

up to you n_n

[glacial coral]

So in wildcards folder I make a txt and name random anime clothes and then copy and paste? Or I name it with the “__”?

[river shale]

you make a text and name random anime clothes then copy paste and save

[glacial coral]

Understood

[river shale]

then you go to FORGE and go to make your image and refrence the file name __random anime clothes__ in your coding 🙂

[glacial coral]

I see

I’ll give it ago now and come back if I have issues or success

[river shale]

yes! Ping me in #🤝｜tech-support ❤️

[glacial coral]

Understood my friend

[glacial coral]

super cool

ty

[river shale]

yay!

You're welcome!! 😄

[glacial coral]

what model do u use for ur gens?

[river shale]

let me find it for you ❤️

[glacial coral]

cheers

[river shale]

https://civitai.com/user/A1kodesu/models?sort=Highest Rated

I use a mix of all of these 🙂 I alternate alot, I've even mixed some together 🤷‍♂️

Take your pick n_n I love them all

melon is my #1 favorite ✨

[glacial coral]

ah ty

[river shale]

youuuur welcome 🙂

[glacial coral]

Such a small change like a wildcards makes it so much fun to create images

[river shale]

I agree! I'm glad Ret showed us these 😄

simple install too 🔥

[glacial coral]

lookie

[glacial coral]

agreed

[sharp imp]

Hehe your just getting to see the tip of the iceberg

[river shale]

beautiful! 😄

^

Ret's stuff is op uwu i can't keep up

[glacial coral]

I would love to know ur one that u did for that image I liked

love it

This is with SIH model and Resmulti step with KL optimal as sampler and scheduler combo

[river shale]

ill see if I can find one I can post 💀

[glacial coral]

Oh I’ll assume ur on the “lack of clothes” maker XD

[river shale]

na shes just in lingerie alot

and often has cleavage showing >_>

[glacial coral]

Im very much the same but take it up level a notch hence why u dont see much of my stuff posted here lmao

[river shale]

but ive been making her in gowns alot lately n_n

also exactly 🤭 rated R at best

[glacial coral]

Fair XD I’m more on XX and XXX rating

[river shale]

fuuuuuuuuuuuuuun

[glacial coral]

Veeeeryyyy

I shall wait for some of @sharp imp stuff when he gets home from work

I grab some juice brb

With more gens I’ll show

[sharp imp]

Lol my personal creations (the files) i don't share, but I have a massive repository i can share once I'm home that I grabbed from github when I started playing with wildcards. Alternatively, you can always plug any image i generate into chatgpt and ask it to tell you what prompt likely made it (cause my prompts don't make sense a lot of the time lol) and that will get you better results. A lot of my stuff i spend hours testing with style loras and a couple of personal merges to get just the look i want lol.

[glacial coral]

ah ok thats fair

[river shale]

yeah the file I shared with you for the clothing, I generated on ChatGPT 🙂

[sharp imp]

Also, the personal files contain a fair bit of nsfw stuff, and since I can't verify anyone's age I don't share as a rule. But the repository stuff anyone can look up and I didn't make it, so I feel no ownership or responsibility for.

[river shale]

wish discord age verified uwu

[glacial coral]

understood ❤️

[glacial coral]

10 days ill be 20 xd

[sharp imp]

Lol 4 months I'll be 35.

[river shale]

👵 Im 33

most people just say im "old"

💀

[glacial coral]

nah u guys are chill i mean im young as shi but u guys aint that old in gen

[river shale]

happy early bday btw~

[glacial coral]

thank youuu

[sharp imp]

Aye to be 20 again lol

[river shale]

maaaan the things id do differently 😂

[glacial coral]

many people i speak to around ur ages tell me to enjoy life (i have no idea how to lmao)

[sharp imp]

I mean enjoying it is all a matter of perspective. Now having an easy life or living a profitable one, that's a different story lol.

[dusky linden]

enjoy being young, adulthood isnt easy, skibidi or whatever the young uns say nowadays

[river shale]

my #1 advice is to focus on yourself and get your bag and not worry about relationships uwu

20s imo should be for setting a foundation for yourself.

[glacial coral]

man said skibidi 💀

[river shale]

100%

[glacial coral]

ive only had 2 relationships but im prolly not gonna have anymore anytime soon

[river shale]

Relationships HARD

[glacial coral]

yep

i do have the multi option pick with genders tho by being bi lmao

[sharp imp]

Up until I turned 30 I was constantly looking for ways to escape life. But since then I've cut all that out and I am in a better mindset just to wake up. Makes a hell of a difference

[glacial coral]

Fair enough I thank all of u for that information means a lot

[river shale]

really does; escapism is hard to... escape too

ima be back later; fam and I are heading to the pool ❤️

yall have a great day

[glacial coral]

U to enjoy ur time there

[wooden sail]

Sometimes it's hard to explain

[sharp imp]

From my "its not a phase, MOM!" wildcard pack lol

[river shale]

forever OP

HER FREAKIN' EYES

REEEEEEEEEEEEEEEEEEEEE

i love them- uwu

[sharp imp]

lmao

[river shale]

love the dark sclera eyes

[sharp imp]

yeah they are neat. i should really do more color combinations with that

the teal/ cyan is one of my favorites aesthetically though

[river shale]

The shine on that first girls haiiiir

its so golden!

[sharp imp]

hahaha

the concept for that one was paint that matches an emotion, in color at least. of course, the wildcards add a bit of random flair so its doesnt perfectly match, but its a fun one. i like paint and body paint stuff. i made another called paint the world where the characters all have comically oversized paint brushes and are either trying to paint the environment or paint creatures into existence

this is from one i called "The Wanderers Lantern". Its all wide angle shots and ultra wide landscape shots, with a simple cloaked character and vibrant landscapes and backgrounds

[river shale]

My favorite gen I've ever made 🤩

[keen hollow]

wow lotsa posts yesterday

[sharp imp]

I've created a movement

[keen hollow]

about to blow away this instance so if anyone wants prompts, speak now before they're gone forever 😛

[keen hollow]

you didn't like the attacking ninja?

[wooden sail]

Imperfect recreation

[wooden sail]

And this one from Flux.1 Kontext - new image editing model

[keen hollow]

try an Aerith...

or a Terra..

[wooden sail]

SDXL and Flux

[river shale]

I LOVE HER SO MUCH

[river shale]

Good Morning 😄

[main quest]

yo

[tawny wasp]

SolFlare From Kanchigai no Atelier Meister NoobAI Apex_Ai
https://civitai.com/models/1637296?modelVersionId=1853307

[near elbow]

need more cyber

or neko?

[sharp imp]

why not both?

gimme one sec, i got a wildcard for this lol

in the meantime while i look, have this

there we go, plenty of cyberneko

[river shale]

lovelovelove

I cant wait to get good at wildcards like you!

[sharp imp]

lol the only part of this that was wildcards was the hair and the pose 😛

[river shale]

noice!

[tawny wasp]

Jessie demon girl is here

[river shale]

uh YES

She's really freakin' cool!

[tawny wasp]

My hekapoo from star vs the forces of evil 🙂

[sharp imp]

tifa gone emo

well, more grunge than emo i guess but meh

[indigo oyster]

holy! what did you use for this? This is sooo coool

[sharp imp]

i think @glacial coral uses primarily an NoobAI model or flux right now, maybe when they get online they can tell you exactly which one

[indigo oyster]

I guess I'll throw a couple of mine in too, they're beginner level though

[sharp imp]

still pretty good, especially the character framing and camera positioning in the second one. and of course, gorgeous backgrounds

im just having fun with a new sampler/scheduler combo im trying with one of my wildcards i made XD

[glacial coral]

I’m here

Hello

Me here

How can I assist

[sharp imp]

someone had asked about an image youd shared

[glacial coral]

Ah ok for that one was a mixture of a couple loras at diff values and I used SIH for that one if I recall the loras are on Mac but I currently switched for now to a windows machine so it’s kinda a struggle grabbing those loras and prompt I’m sorry if I couldn’t be much help in that regard

[glacial coral]

Ye ❤️ I like when people ask about stuff I do regardless of what it is makes me happy

Ty for the ping

They didn’t noti ping me so I didn’t check

[river shale]

That random middle finger in the corner of #5 🖕 😂

[wooden sail]

Shiny gloss is overrated 😤

[keen hollow]

I tend to gravitate toward more muted, also.

[sharp imp]

i just gravitate toward high quality, but as you can see my model can do both lol

[river shale]

Great gens today guys 😄

[river shale]

This set is nuts tbh

[river shale]

playin' around with my wildcards some more n_n

Edited them to add more depth~

[indigo oyster]

"Throughout Heaven and Earth, I Alone Am The Honored One" ahh lora💀

[tawny wasp]

Jentry Chau Vs The Underworld V2 (Adult) Apex_Ai Illustrious
https://civitai.com/models/1646498

[river shale]

^ the yellow 3rd eye, and the dark full moon demon ones are cool as hecc

[river shale]

Im freaking out- this is so cool!

I forgot I added a 'bloodmagic' lora 🥲

[gusty palm]

everything is coming out just a bit overbaked looking and im not sure why

ive been trying to turn cfg down and samples down but it seems inconsistent

if i turn on live preview they look pretty good at like 50% but then they kinda look shittier as they get less noisy

[sharp imp]

whats your denoise set to?

and are you using hires fix at all

[gusty palm]

not using hires fix usually just doing a bunch of 512ish t2i til i get one i like then using img2img to upscale

denoise around .32-.48 depending on how good it looks

its getting better the more i tweak it and stuff dang i didnt notice that arm

[sharp imp]

Hires will sharpen the image if you use the right upscaler and set the denoise between 0.35 - 0.4. If you are genning low res, you can 2x upscale using hires, then upscale further in extras or i2i. Thats what i usually do, but hardware can be a constraint.

Your model and what sampler/scheduler can also play a part, at least in quality. Not so much things like wonky hands lol. Thats entirely on the model.

[indigo oyster]

This proved quite hard, but I'm happy with the result
-# p.s uploaded an unfinished version on accident earlier xD

[river shale]

👀 how did you do that p.s.

[indigo oyster]

that's subtext, "-#"

[river shale]

nice!

thanks 😄

[wooden sail]

This is Flux

[tawny wasp]

Otaku MIX illustrious XL Apex_Ai - My First CHeckpoint Merge 🙂
https://civitai.com/models/1647578/otaku-mix-illustrious-xl-apexai

[sharp imp]

annnndreported

[thick sonnet]

wow, all these images are amazing

um, i just now got automatic1111 to work with zluda, what is everybody using here?

[river shale]

I just love it when it spits out something so perfect

so satisfying

https://tenor.com/view/adhd-goku-dopamine-powerup-gif-7470948201429011900

[modern sandal]

bro how to make this photo？

[indigo oyster]

Generator: plantmilk walnut
steps: 50
Sampler: DPM++ 2M, Schedule type: Karras,

CFG scale: 7, Seed: 2276656332, Size: 1320x720 + HiRes fix 2x, latent upscale
+extension called "ADetailer", default settings for face_yolov8n.pt

prompt: oil painting, (fisheye lens), uv lighting, wide shot, full body with environment, character in scene context, high contrast, masterpiece, best quality, amazing quality, very aesthetic, absurdres, highres, newest, HDR, 8K, high detail, depth color, colorful lighting, vivid deep blacks, dark field, beautiful glossy environment silhouettes, (solo:4), water puddles on the ground, reflective, (stunning night sky, milky way), falling stars, mountains, trees, fireflies, barely visible dawn

negative prompt: bad quality, worst quality, lowres, jpeg artifacts, bad anatomy, bad hands, multiple views, signature, watermark, censored, ugly, messy, nudity, long body, multiple characters, mountains too far up in the sky, repetitive patterns

+character-specific lora

P.S. it was also edited in photoshop to remove artefacts

[sharp imp]

the character was rimuru tempest, in case your using an anime model that has been trained on newer anime characters

[indigo oyster]

Rimuruchug

[sharp imp]

😛

done with your prompt and my model, no inpaint, no lora, no img2img lol

2560x1440 resolution

[indigo oyster]

that was fast too

and the galaxy sleeve looks nicer but also a bit blurrier, maybe because of "oil painting"

[sharp imp]

yeah well, im using forge and have a 3080 rtx lol could have been faster if id quit being cheap and upgrade my hardware

yeah oil painting and the sampler/scheduler i used maybe. i also only set it to 30 steps

and 10 steps of hires to push it to 2560x1440

[indigo oyster]

oh also the fireflies are green for some reason xD

but looks really sick

[sharp imp]

also looks like the forest is on fire lol but meh

[indigo oyster]

ohhh ahahahah just noticed lol

but imo it look better than walnut

I just learned prompt matrix and xyz plot yesterday, idk when I'll get to a point of training my own model or merging

[sharp imp]

merging is easier than you think, only becomes tedious when you get to the testing stages lol cause then you find out if you merged too much of model B or too little, and you gotta test a range of styles if thats what you were going for.

[indigo oyster]

also, how do you go about loras? I noticed that if you have many, they get unintentionally triggered all the time and affect the image, but always moving them back and forth in folders is a hassle too

[sharp imp]

ive never had that issue with them getting unintentionally triggered, though i also havent tested it. could be a setting under the loras section of your settings.

[indigo oyster]

I've noticed this when I installed the add-detail lora and my prompt included "detailed image", and it got f'd up beyond recognition xD

[sharp imp]

ah yeah, i stay away from quality loras with trigger words. ive got a detail slider lora for illustrious models that doesnt have one, so its only active when you actually add the lora to the prompt. but even then, ive never had any of the others randomly activated from a trigger word like that...at least i dont think i have lol

[indigo oyster]

HOLY THIS LOOKS ABSOLUTELY MAJESTIC

May I steal your prompt good sir?🧐

[sharp imp]

Thanks, i was testing a concept that i wanna turn into a wildcard

uhhhh id have to find it, one moment

masterpiece,best quality,amazing quality,very aesthetic,absurdres,intricate lineart,vibrant colors,professional composition,(high detail:1.3),ultra HD,8k,detailed eyes,detailed hair,solo,1girl, inaba tewi, touhou,exhaling smoke,abstract,vector art style,flowing lines,glowing eyes,elemental energy,smoke trail,surreal style,black background,UV dragon coil fading until lit tattoo motif,glowing outlines,spiritual aesthetic,neon gradient,mystical,dark silhouette,Complementary Colors Palette line work,(inverted color,chromatic aberration,high contrast,color contrast:1.4),minimalism,surrealism,eye trail,flowing hair,fractals,glitch effects,blacklight, uv light,

there were a couple of wildcards in it, so some of it might not make complete sense lol

[indigo oyster]

Thank you kindly my lord

[indigo oyster]

A gag gen from a meme I found

[near elbow]

"wholesome neko tea ceremony" 🙂

more neko less chaos

with "exquisite intricate silhouette neon lines backlight"

[indigo oyster]

Someone's about to get it (another meme gen)

https://tenor.com/view/xenoverse-goku-super-saiyan-angry-dbz-gif-1416275111944307575

[uneven hemlock]

卡通风格，小熊站在方格图右下角终点，周围围绕着代表“最大和”的金色大数字（比如样例里的 9、-10 ），身后方格图亮起胜利光环，小熊开心欢呼，背景用彩虹和花朵点缀，突出“求最大值”的最终目标

[tawny wasp]

Hekapoo From Star vs The Forces of Evil (Illustrious) Apex_Ai
https://civitai.com/models/1652731?modelVersionId=1870712

[glacial coral]

Seems like I’ll need a Ken Kaneki Lora since only model that can do it is Midjouney and mor any open source models I’ve tried time to go hunting now I guess

[glacial coral]

Challenge but it’s going to a fake website and putting a code in 💀💀💀 ok then

Seems like that’s what the challenge is…. Unless the challenge is after putting in ur details u have to protect ur details as much as possible

[sharp imp]

Lol it's very clearly a scam

[keen hollow]

evenin'

[river shale]

👋

[keen hollow]

what's shakin'?

[river shale]

🤷‍♂️ Not much- running some image batches, messing with wildcards tryina get things just right. Whats up with you? 🙂

[keen hollow]

nothin' here

[near elbow]

anime claymation, anyone? 🙂

[indigo oyster]

I can't decide if the hand looks f'd up or not, can I get some opinions please?

[sharp imp]

from what i can see of it it looks fine

[near elbow]

not much to see of them anyway, ok for me

[indigo oyster]

I might've misused "hand" actually, what I meant was the entire limb, arm included😅 It's just that the part between the shoulder and the elbow seems too long and just long enough at the same time for me

[near elbow]

how to notice that your AI model is messing with you 🤪

[near elbow]

another one (just noticed that i had the upscaler active, unnecessarily big 🙂 )

and the right hand is really f'ed up, two tumbs or something like that

[tawny wasp]

Hekapoo SSJ3 🙂

[sharp imp]

spam, dont click

[pale mural]

Damn they feisty today

[sharp imp]

must be close to quota time

[slate remnant]

i found this.. what should I do with it?

[heady apex]

@slate remnant

☝🏽☝🏽☝🏽

[slate remnant]

I run away

[sharp imp]

lol

[slate remnant]

I return

uhh oh

oop's

slow today

[slate remnant]

Guess it's just us two

[sharp imp]

probably

[slate remnant]

almost

[sharp imp]

waiting on an image from this wildcard prompt that i can actually share to pop up lol

[slate remnant]

combines singing and the bar

[slate remnant]

finally

ok hulk woman was freaking me out

goes to play games

[glacial coral]

I thank my friend for making switch from forge to swarm Ui I like it a lot better after getting over the starting hurdle so here’s 3 images I made with 3 diff aspect ratios 2:3, 16:9 and 21:9 (Wai V14 was the model)

[sharp imp]

sick!

[glacial coral]

Yes so far loving swarm although forge is great and easy compared to swarm although I expected swarm to be a bit more harder cuz of Comfy Backend

[sharp imp]

lol i think their all easy once you figure it out, but yeah swarm and comfy have a steeper learning curve at the start for sure

[sharp imp]

there has been a very noticeable improvement in your stuff. keep it up!

gonna make me have to work harder to keep my polish lol

[river shale]

Appreciate the compliments Ret!

[tawny wasp]

Hekapoo From Star vs The Forces of Evil V2 (Illustrious) Apex_Ai
https://civitai.com/models/1652731

[sharp imp]

that is exquisite!

[wispy canopy]

not hard to make:
A) make a model mix that responds very well to artist tags.
B) just generate 24k artist tag previews,
C) Also use Deepseek to make a browser gallery for easy browsing
D) Make a script that sorts the previews generated into strong and weak activations
E) finetune controlnet promax tile upscale
F) Figure out a way to send different prompts to different unet layers
G) Also some text encoder bullshit XD

... all so i don't have to use lora's (this is prompt only) 😄

[sharp imp]

sounds like a lot more work than making a bunch of wildcards lol

cant argue with the quality though

[wispy canopy]

... oh yeah, forgot. assemble a list of artist tags (all on danbooru) and then feed that to preview generation workflow xD

[sharp imp]

too much work for me haha but thats dope!

[river shale]

Love the horns holding the halo

:D

[wispy canopy]

wel, that was a fun attempt ^^ thx for the inspiration

[sharp imp]

any time!

[rough oxide]

if i designed arcane jinx color palette.

gud thing i r not designer.

[sharp imp]

lol

[indigo oyster]

redid this since it didn't feel threatening enough xD

[sharp imp]

Hmmm rimuru tempest, cinematic, dramatic foreshortening, eye trails, lightning effects, intense bokeh effect. Sound pretty close to the actual prompt? Or am I way off lol

Made with an online generator so not as polished as my local gen work

[indigo oyster]

I was kinda assembling the prompt based on what kept the style I was looking for, so it basically sucks xD

uv lighting, masterpiece, best quality, amazing quality, very aesthetic, absurdres, highres, newest, HDR, 8K, high detail, smooth shading, glossy surfaces, high contrast, depth color, dak depth, depth, colorful lighting, dark field, clear, vivid deep blacks, Rimuru Tempest, furious expression, looking at viewer, platinum blue hair, light golden eyes, upper body, (eye trail), dutch angle, dynamic pose, katana mid strike, swinging at viewer, (katana glowing golden), golden electricity sparking from katana, battle night, chromatic aberration, cinematic composition, blurry foreground, blurry background, golden-toned background

[sharp imp]

Ah I see.

masterpiece, best quality, amazing quality, 1girl, rimuru tempest, glowing eyes, eye trails, lightning effects, dramatic foreshortening, arm forward, glowing hand, wielding sword, lightning blade, cinematic, heavy bokeh effect, yellow lightning, yellow glow, was what I used lol

I could have added a lighting tag to get better shadows. Probably a proper pose tag or camera angle to frame it more like yours.

[near elbow]

with hidream full, added "anime style" - also interesting but of course not at the detail level as a real anime model

ah, ugly miss on "however"... rest is nice though 🙂

[near elbow]

also with highdream and "anime style" for fun

[rough oxide]

if y'all are using danbooru-based models, rimuru is usually trained under "1other" not "1girl"

can matter for composition

[near elbow]

none of the models i'm using has booru tags (flux, hidream mostly) 😇
they do seem to understand some aspects though

[rough oxide]

yeh. more towards retsubu i think

[near elbow]

trying "1other" instead of "1girl" with the first image and hidream, for a direct comparison. interesting... does make a difference, more than i thought

[rough oxide]

hidream has like no anime knowledge but when i tried it the anime output for local models is probly one of the better ones

on par with imagen 3

like the actual aesthetic

[near elbow]

just takes so looooong... i mean, 90 seconds on a 4090? when a 20 step flux dev image takes 10s and sdxl 1s?

[rough oxide]

sounds fast.

i was hearing a100 took 120s

for the full unquantized ver

(i never quantize xD)

[near elbow]

yeah, just the fp8 version

otherwise vram becomes an issue as well

[sharp imp]

It understood who I meant regardless. I honestly think the name as a tag is enough guidance for newer models. My personal merge knows most characters, and would know more if I merged any character loras into it, without the need for a 1girl or 1boy tag.

[near elbow]

just for some imagery in between all the text 🙂

[rough oxide]

when it's trained with like 1000+ data points all pointing at 1other, sometimes 1girl can cause unexpected differences

the ones that are 1girl with rimuru usually have what you'd expect.

nsfw compositions xD

and larger chest size

etc

[sharp imp]

Interesting. I haven't had that problem but then I don't use the usual degenerate models lol.

[rough oxide]

also wanted to say, online anime tools are the best ;D open source can't keep up lately

[rough oxide]

but specifically novelai diffusion and maybe if nijijourney ever updates their thing again, niji still higher quality than noob/illus tho just less knowledge

[sharp imp]

Mmm debatable. But to each their own I suppose haha I primarily gen open source with local setup and I've had no complaints about the quality of my images.

[rough oxide]

to put it one way

[sharp imp]

But I've got a solid comfy pipeline that carries the load.

[rough oxide]

novelai can do anything on danbooru down to 10-15 tags, also extremely adheres to artist styles/etc.

no need for loras

[sharp imp]

I use only quality or lighting loras, no character loras in any of my images. I also very heavily use wildcards to gen random characters lol I don't need 1000 images of Rias Gremory in every possible situation clogging my outputs lol

[rough oxide]

i use wildcards too

34k+ chars work in nai

xD

and 50k+ artists

i still update my char wildcard almost every day with new stuff i find that works

[sharp imp]

As for artist tags, style loras achieve similar results if trained well. Finding well trained ones can be tough. But I'm not trying to rip off a specific artists style either. I feel like that's the area that real artists are up in arms about, and is likely the only leg they have to stand on when it comes to "fair use" or copyrighted work.

I've heard NAI is really good for that style of images though, where character and artist tags are relied on. It's got its niche and it's followers, same as SDXL and Illustrious. And Flux. Their all tools with different approaches.

[rough oxide]

good for oc stuff too, but with artists it's usually not like you're just using one

it's combination of like 2-6 that make a really unique composition

then controlling their weights

etc

at any rate open source anime models hit like a brick wall of compute basically

no open source is training on like 256 h100s

[rough oxide]

noobai is the closest that came to that but that's pretty much done now

and then in terms of conceptual understanding where style doesn't matter as much (almost no style control) stuff like chatgpt/imagen 4 is in the lead

os models can't touch chatgpt in terms of knowledge just due to the sheer like

[sharp imp]

Maybe. But I've got good enough models to make the kind of art I like. And that's enough for me. Every couple weeks there's a new open source model using new architecture or math that's constantly nipping at the heels of or passing closed source models. It's a matter of time before that bleeds into image gen as well.

[rough oxide]

yea, when you find something you like that's always a good thing. i'm like a pure wildcard user

so i want everything to work and to be surprised by things basically

xD

or if i come up with an idea, no matter what char/artists i think of i'd want them to work

like this combo i jsut got i like

[sharp imp]

Lol even with illustrious I'm constantly surprised by my wildcards. But I've got 87 wildcard folders with 4-8 wildcard files each, and each file has a minimum of 30 lines in them. According to chatgpt, I could generate a billion images a second for longer than the universe has been alive and still not be close to seeing a duplicate image. And I'm constantly building new wildcard packs.

[rough oxide]

oh, i always come up with location/scene details myself. my wildcards are just for char/artists but i have a special script i made to randomize stuff within that and even fetch things from booru

[sharp imp]

Nah I randomize the entire image. Every aspect of the character, the mood, the setting, the outfits, the camera angles. All of it. And each pack is thematic. So I can either run the pack as a prompt itself, or mix and match themes to get even more chaos.

[rough oxide]

my booru randomize function kind of does that since gelb has like 11mil+ images

[sharp imp]

I could wildcard my loras and even get crazier with it, but then id have to set weights before hand and that kind of testing makes my brain hurt

[rough oxide]

that's why i use a custom python script xD

i can add stuff like that after the fact

purely made by AI

text ai is stronk at coding stuff like that

also

makes it transferable anywhere

since it auto copies to clipboard & then i just paste it anywhere, comfyui, nai

etc

and can modify it too

[sharp imp]

Obligatory images to break up the text

[rough oxide]

when you make it a script you have more control over multichar scenes too

[wispy canopy]

can you believe i made 24k preview images of the danbooru artist list?

[rough oxide]

;P only 26k+ more to go xD

but yeh

that's the same kinda insanity i do with maintaining my wildcard lists lmao

find new anime series add like all the chars that work from it after testing that they worked

if i was maintaining a site or smth too i'd probably do a base example of everything xD

at this point with modern models danbooru/e621 aren't even the only sources anymore

so much stuff with 0 booru entries works nowadays 🤣

[rough oxide]

noobai had like... 35.5k that worked, maybe, ish?

i think the top 35.5k of the one i posted generally work

[wispy canopy]

there's also a lot of duds in there tho

[rough oxide]

yeh, that'd make sense

[wispy canopy]

the duds vs the strong reactions

[rough oxide]

i have trouble discovering new artists more than new chars

but some of the last ones i discovered are pretty good

[rough oxide]

rei (sss35 ss3)

only 38 entries so doesn't work in things like noob

nice style tho

rgreywind , distinct lineart-ish style

not lineart but uh

very heavily emphasized lines i mean

[rough oxide]

but i've only added a handful of artists, with chars i've added thousands more xD

[near elbow]

"an anime neko hacker in a dark room in front of the computer, walls plastered covered with anime art thumbnail prints"

[wispy canopy]

i'm only on illustrious, doesn't do natural text xD

[rough oxide]

nai can do some level of it

an awful crayon drawing made by a child of saber alter, she is outside at night under the full moon, the full moon has a frowny face on it

[wispy canopy]

lineart is hard tho, really easy to see faults

[rough oxide]

lineart's gud in nai but ineed to random something not in a bikini xD

[rough oxide]

even in line art danbooru finds a way to make most things lewd xD

in other news no line art is gud too

xD

[wispy canopy]

i have a nice way to get around the lewd issue tho, i mean, this model is perfectly fine for smut

[rough oxide]

oh i mean

i'm using a func of my script

to grab random tags from danbooru off the tag i give it

in this case "line art" xD

[wispy canopy]

yeah, lineart doesn't work that well, i nmy model, i think

[rough oxide]

let's just say

most aren't as seiso as this

xD

i get like 3 nsfw

before i get something badass

xD

[wispy canopy]

can't neg prompt?

[rough oxide]

oh i could but

i'm literally just

it's the same as going to danbooru, going to line art page

and pressing random button

where it randoms within lineart

then pulls a tags from a random image

finally a non monochrome xD

my bad

it's still monochrome

but a diff color

xD

lol "too many watermarks"

ogey that's enuff lineart testing, probly xD

not all artists are created equal

this one has lots of detail, i like that

[opaque mason]

what model is this?

[opaque mason]

oh lawd what model is this

[sharp imp]

finally home

[rough oxide]

novelai diffusion

[indigo oyster]

the flower field and the graffiti look so good🥵

[indigo oyster]

She must've tried playing Paganini and became depressed, relatable

[glacial coral]

👀

[tawny wasp]

cosmos girl 🙂

Recluse (Elden Ring Nightreign) illustrious Apex_Ai
https://civitai.com/models/1668488?modelVersionId=1888522

[misty violet]

sexy witch🔥 btw, are there any tips for balancing character details with background elements to keep a cohesive look?

[tawny wasp]

do you mean during the Lora creation phase? or in general when you create a photo?

[terse ore]

Anyone try this? Framepack

Image of a 3d generic pose+ image of a character= character in that pose

Oh. Looks like it'll sometimes mix the stuff that's on the poser onto the output ie a barefoot 3d doll will give the office lady barefoot even though the original pic she had shoes

[stable crane]

thought framepack was about making vids

I have used it before, it's not bad

[terse ore]

Trying an new inpaint workflow, not working. Input:

Output

Mask

workflow

[wooden sail]

Close enough?

[lyric zealot]

Scam^

[indigo oyster]

https://tenor.com/view/omni-man-omni-man-are-you-sure-are-you-sure-invincible-gif-3935116808772397515

https://tenor.com/view/low-tier-god-banned-mods-smite-gif-4866200577432974147

[river shale]

BORGER

[sharp imp]

Omnomnom

[river shale]

Loving makeing swords rn

[sharp imp]

scam dont click

[indigo oyster]

https://tenor.com/view/low-tier-god-banned-mods-smite-gif-4866200577432974147

[river shale]

Im on a roll today 😄

[near elbow]

whoever falls for these stupid scams anyway?

probably hacked accounts... anyone who doesn't have 2FA enabled do it NOW

[near elbow]

red menace

[opaque mason]

what models did you use for this?

i wanna try this style

[woven gust]

send me 2500 USD

you idiot!

[charred venture]

/help

[indigo oyster]

Does anyone know how to reliably do fighting poses? Because I just spent an entire day trying to do a back kick pose, even with control net openpose, and the result sucked

[indigo oyster]

Also having trouble with making multiple characters in on render too

[near elbow]

"anime style ninja warrior doing a back kick" with hidream pro

discord upload trouble

[rough oxide]

novelai diffusion

[opaque mason]

what was the prompt?

[near elbow]

anime style ninja warrior doing a back kick, next to a robotic neko attacking with a sword

my message is - with an advanced model, all that should be not much of an issue. with sd 1.5 you're out of luck, with sdxl it's unlikely

[rough oxide]

i keep metadata on most of the pics i post; that one was: 1girl, animalia (anime), shangri-la frontier, rumie, 0.4::watanabe tomari::, purple eyes, location, grin, peace sign

the style was from "rumie, 0.4::watanabe tomari::"

[opaque mason]

did you use civitai for this?

[rough oxide]

no

[opaque mason]

what did you use

[rough oxide]

novelai diffusion, it's a paid/subscription model

[opaque mason]

ah

[river shale]

novelai's pretty cool- you can even write stories with it, and program it to remember plots, plothooks, characters and such things, it's a cool ai.

[rough oxide]

yea; i used to plug it into ai roguelite which makes a full on rpg with text/images; doesn't seem like airl is updated for the current models though

[river shale]

👀 Ai roguelite sounds cool

do you have to signup/pay for it?

oh wow its a thing on steam~

Looks interesting tbh

[rough oxide]

ye; sorry, went to get breakfast. you can use your own local models or apis from other models (or they offer their own api service for a price) and it's basically a more advanced ai dungeon that keeps track of stats & lots of things, they added a bunch of stuff since i last played, still waiting for them to update their nai support b4 i try to get back into it tho ;_;

[near elbow]

with gemma 3 27b (abliterated, i.e. uncensored) you can have really nice LLM based local roleplaying sessions. not sure how that compares to novelAI models, and of course chatGPT and the big commercial stuff is much more advanced, but also very much... restricted in scenarios and output

[river shale]

Cuuute~

[sharp imp]

hehehe

[lyric zealot]

^scam

[rough oxide]

@autumn roost phisher alert

(2 in a row cuz one is above that too as a "likely spammer" hidden)

[green temple]

i like 5th 😍

[rough oxide]

was experimenting with artist combo there, one of those artists is overly stronk tho

[misty violet]

bro ur art has mesmerizing details and composition...btw what was the original idea behind her?

[river shale]

Oh I didn't have a main idea, I just used wildcards. 🙂
It's completly random tbh;