#๐ฅ๏ฝanime
1 messages ยท Page 57 of 1
how to get one
usually, you buy them with the complete server, those things don't come as a normal pci device
@acoustic agate this is your prompt on alfahentai, i LOVE #1
the DGX H100 stack is probably like $400k 
yeah, i think our AI division 4x a100 machine was $150k+
uhh.. my next image generation is called "mistakes were made" lol
lol, can post 0 ๐ฎ
@rain valve U be a tasty looking blonde dude โค๏ธ
33,000 pesos right :)
Excuse me Eface, I have a question for you. Which LLM model are you using on your machine anyway? I'm guessing Tavern or Ooba is your UI?
Ooba, the model is https://huggingface.co/TheBloke/OpenAssistant-SFT-7-Llama-30B-GPTQ
Good old Ooba.
Sure buddy
Also isn't it like 3am? You should sleep soon.
ugh. openassistant failed on trashpanda, going back to wizard vicuna. (consistant reasoning error errors)
I am in a weird state of procrastinating on work. I am exercising, watching youtube, listening to music, and chatting on discord. All that instead of doing my audio engineering work.
oh. my. i can see the previews ๐ฎ this is... THIS IS... TRASHPANDA LEVELUP COMING IN!
Fair enough then. Wait until you start cleaning too
ooooooooo so fucking cool

what kind of music do you engineer on?
anime jpop
@autumn violet LOOK AT THIS!
gear 4th panda
We get all kinds of generes coming through the studio. I record pretty much everything. Then I do audio editing and basic mixing work before passing it off to a genre-specific engineer.
Though right now my work is just editing a euphonium performance and processing it to sound live.
this... seems kinda uhh.. i know this is like 100% sfw... but my brain combines slime + miku in... certain ways
I need my trash can
We need @finite herald's opinion 
ohhhhhh fuck you make me fucking handsome
Careful with the excitement there friend 
guys, I know you need more blood in the morning but hold your horses down ๐
She is going to ruin the floor :O
@ocean reef i was about to call you ๐
I already saw pretty faces ๐
Not anatomically possible 
no1 said it's gonna be easy!
That's what they get for having carpeted floors.
the other 2 were cursed
i see thats pretty nice! do you know i also wanted to get into audo engineering for some time because i made my own music. but felt like its pretty niche and maybe not so easy to get into and find actual workplace later.
so keep it more as a hobby. ๐
how can i use this image assistant
I guess the world is my trash can
That's nice. I went to school for music production / composition, never really thought I'd go the engineering route. But then I got work at a studio to teach beginner engineering techniques, and I've learned a lot along the way. I'm a musicion/engineer combo myself 
I think that is philosophical or something
or something
I mean we are on the same timezone so can you blame me?
It's a local installation of the AI software oobabooga i'm running, unless you have a powerful gpu (12gb+), LLM's (large language models) are impossible
We're both wasting our time chatting on discord instead of making real world connections and contributions, so no, I cannot blame you.
blame the Sun

i have 24g vram๏ผ
3090/4090? ๐
yes!
iternet is a great way to make connections! You are not limited to your area, we have whole World!
in that case, you surely can!
3090
redacted to avoid controversy mhm mhm

grab oobabooga https://github.com/oobabooga/text-generation-webui (be sure to use the one-click installers, don't just clone the repo) , and grab this model https://huggingface.co/TheBloke/OpenAssistant-SFT-7-Llama-30B-GPTQ
and then you need the character "ImageAssistant", that's a simple yaml i can share
just 30B?
that 'just 30b' barely fits on your 24gb card
no i m just have a 70b model i found in Internet

if i'm not mistaken, a 24gb card can run a 30b 4bit quantized model just barely
fuck !
experimental workflow for faster comic creation
i need a 100 H100 cards!
so I'm not gonna even try it with my 8GB xD
Controlnet dosnt listen to me lately.... or I'm doing sth wrong xD
dang nice ^^
which sampler are you using? when i use UniPC controlnet goes wonky
it ignore my sticky pose!
it was working fine with previous ver
I will just watch some tutorials to educate myself on it xD
sticky pose? openpose?
the stick pose from openpose, yeah
you might want to look at your console output (where the webui is running), controlnet might be giving you an error
yeah, gonna educate myself more ๐
welcome back 
yea took a beak came back and saw not much has changed
There are a few folks pushing the boundaries, but they're few and far between :V
a slime fairy.... lol
Yesh
@wispy canopy updating my oobabooga borked everything 
reinstall windows!
got it chief
uhmm... is it giving the errors on gptq?
PC is trash. Send it to me and buy a new one.
it was before I updated, then I updated, now I launch it but cannot access the local host

@runic vault

ok?
clean install with the oneclick doesn't work ?
it missed the mark? ๐ฎ
well i don't really know what to say
^^ that's a valid answer too
doing that now
he does it with everyone active here kether so dont worry ๐
oh yeah, i've been turning everyone into prompts... ๐
you enter the chat when i'm here, you get prompt-ized
did you miss me or something :V
I left because I didn't want to keep staring at stuff like this
And I will probably leave again for a while. Anime chat is gross.
sadge
Huh? What's gross about a clothed slime girl fairy?
If we're a good match I'm down
scantily clad slime girl is nothing more than a sex symbol
I will bless you marriage
use the cursed lyrics! (Meganeko - Milkshake)
Furries and anomalies
The hyperactive scatterbrains
Whit milkshake running though your veins
Unlimited sleep deprivation
Dankest memes and vaporwave
Procastinate at five am
Forever, that shit is my jam!
I come from the future and the
Future will be beautiful
With pizza at the speed of light
And hugs multidimensional
Don't hold anything back, dude your
Fantasies are real to me
So taste'em like a lollipop
Yeah! Go in hard and never stop!```
@magic wave
An actual rifle ๐ฎ
sorta yeah
lots of inpainting?
single shot
Nice
Clean AF
Fair XD
I'll pass on any promo that includes furries XD
aight, that just calls for the vengeance of the gods, i'll prompt it myself!
how doesn't it make sense?
i can ask the LLM that you know ๐
I mean, I don't know how they got that from "Smile โญ" xD
oh... i already replaced the chat... it missed 'diez nutz' though, got nice images instead
O, fair :O
i can rebuild it i think ๐ so just replace the brain back ๐
@magic wave
oh no... i've been restricting it too much ๐ฎ
this is way better ๐ฎ
Ohhh, thats fair then :O
@ocean reef threw in your name again, and somehow it got me this. these are gorgeous ๐ฎ
looks cool! Like a tribe woman! ๐
A figure stands tall against a backdrop of snow-covered mountains. Their skin is dark and weathered, marked by years spent under the harsh sun. They wear thick furs and leathers, adorned with intricate patterns and symbols that speak of ancient traditions. In their hands, they hold a bow made of polished wood, strung tightly with a single arrow ready to fly. Their gaze is steady and unwavering, fixed firmly on some distant horizon. Despite the cold wind whipping around them, they stand resolute, a testament to the resilience and strength of those who have lived in these lands for generations.
well, it failed on the bow part it seems
it's fine by me ๐
Prompt ?
โค๏ธ
oh right, i was busy doing the forbidden lyrics too...
its... horrible
blursed. very blursed. they're not nsfw by the definition of the word but... i'm not sure i should post stuff like this here xD
๐ค
aiska, don't tell me you want evena single one of those outputs here ?!
good evening
cool
im still doing more mecha waifu
trying to think of a new waifu to make tho
or at least a new style of the pink hair girl
in other news, this character (idk how much cheek is too much cheek)
actually ive decided that if i dont know i probably shouldnt post it
Anya in a mecha suit?
lol nice
how do you mean?
if it's too revealing?
it had some side boob that i missed, and the mecha suit wrapped quite snug around he butt
smart move ๐
๐ซก
any promopt tips to make the eyes look the same directions
mine just do it without extra prompting
maybe try stuff like "looking at camera" or something
i tried both looking at viewer and looking at camera but still doesnt help i dont know why mabye its the model or mabye ill make a lora that will fix the issue
those eyes look fine to me
they are but i just want them a bit better
dont mind her broken finger
maybe add beautiful eyes
Yo is there a nsfw channel in this server
no, the devs are trying to keep their creation sfw
Iโve lost my previous account and Iโm trying to find my old servers
Thanks for the answer I appreciate it
yeah, this server has no nsfw.
we should tho 
at times, i hate it i can't post art which has nude elements in it (but is very obviously not sexual in intent), but, rules are dem rules
WHAT DID I DO TO THIS MODEL?!
i just want some good hands
we should have a rule that just makes a warning before and allow it imo
nah, they gotta draw the line ๐ฆ
use control net hands
or... uhh... shit, this model is not public yet ๐
yeah, i should
Yeah same, my account got banned and I lost my whole ass server with all my ai art
negative hand embeddings
Iโm pretty sure midjourney does that no ?
share im good at keepin secrets
nah, first gotta get the ok from my collaborator ๐
idk i dont like midjourny much never really used it
Ight Ight
if u do let me know i have some amazing prompots that i would like to test
is a merger of alfahentai and darkgemini ๐
I got you
i can never get control net to do hands still lol
has anyone tried out runway
good hands?! what is this? blasphemy?!
i have no idea why my model can generate this crisp font ๐ฎ but... what she says man
this is inner you
it fou!
what model are you using?
this is the unreleased darkalfa prototype
DAFAQ IS UP WITH THIS MODEL HANDS?!.. proper holding a microphone, are you SHITTING me?!
goodnight
nah i am hungry and its nearly midnight
Yes embrace the ((1mechanical girl))
no 1mechanical girls were harmed in this production ๐
A small, delicate figure flits among the branches of a tree, its wings a blur of motion as it darts here and there. Its skin is a warm, golden tone, and its hair falls in wild curls down its back. But where its human-like features end, mechanical components begin - gears and wires and tiny, intricate machines whirring away inside its body. Despite the contrast between organic and artificial elements, the two somehow come together seamlessly, creating a harmonious whole that is both beautiful and strange. -> this was the prompt
Are such prompts better than keywords?
i've only recently started prompting like that
it usually does results in more natural images (better interaction between subject and background)
You use your llm for that right?
Yeah can't really do that on my 3070ti
but i've been doing all kinds of random things to my prmopts, throwing in full lyrics, BREAK lyrics, keywords, short descriptions, all kinds of stuff ๐
FUCK. this model is GOOD
Haha yeah I'm also experimenting right now. Started learning SD like 1 week ago
experimenting is the most fun!
and its addicting... i started like 2 weeks ago and i just cant stop, my gpu never worked so hard 
i've never upgraded my pc that bad... jstu for SD and AI, i've picked up a 3090, 32gb memory extra, and a seperate disk pure for AI
I already have a 3080 but, so its fine for now.. but rest of PC could def use an upgrade 
it was fine for me. until i wanted to run LLM's too
llm?
Large Language Models
(Like chatgpt)
ahhhhh
It's making my prompts ๐
i kinda like experimenting by myself, and see where it goes
of course, that's the best kind of fun!
I am also using mostly the same prompt with a few deviations atm.
You can still generate so much different stuff especially with controlnets
controlnets? 
i've just found a setup i really like and am satisfied with, so ima probly create lotta stuff with it before i try something else 
waiit.. you can control poses??
๐ best way!
yes, controlnet can control poses!
This was done with the lineart_anime controlnet
(First is base, second is my generation)
holy fuck i got soo much stuff to learn still
Don't worry I'm also new
everyday i've been sitting at the PC my renders just been getting better, does it ever stop?
the addiction is too powerful
it is.
It never stops.
What model is this?
the two model weightings i made, which one is the best model? ๐ฆ
counterfeitv3
Depends but I'd say first
problem is, i like them both ๐ฆ
but i think #1 is retains a bit more of it's generic model heritage
for some reason, i get super good Upper body shots, but if i go full body the faces look like dog water
Ahh ok, glad to hear its not only me
i have add_detail lora
Like the only times i get good full body is if i dont ask for upper body and AI decides to give me full body, then generally itll be good
but if i try to force it, look like a coffee stain as a face 
Can I get prompts?
which one... all of them? XD
All
Never

100% Stable Diffusion
whats a good prompt to get the subject to always look at camera?
"looking at viewer"
me too
amen 

I want to print this design on tshirt
do u guys know how I can do that in inpaint ๐
I want to create mockup designs
Which one did it better?! 
she's not the design though ๐
for my work?
yup
cool how you got her to appear tho, is she a lora?
I'm only using style lora
yes
can't get her into the blueprint?

or is she the engineer?
looks like she's trying to hold a gun, but can only get to a blueprint ๐
I believe AI is trying to write "Hime" since that's what it does quite often
@wispy canopy mecha[:0.3] what is this doing?
lol xD
that must be an error
any pro knows if rainbow hair is doable at all?
sure ๐
easy
im listening/reading 
yeah, i rarely use lora's at all
rainbow hair didnt work, lets try multicolor hair 
i don't think i went with rainbow hair though
thanks for the Looking at viewers btw, works wonder
and the problem is, is that rainbow gets everywhere ๐
i rarely, if ever use it
So far what happened
got red heads and blondes but rainbows everywhere
is there a way also to randomise between a set amout of colored eyes?
you need the extension *dynamic prompts * for that
Thanks, you are like a god at this! ill look into that, but now its 10am, i must go sleep 
go to sleep man ๐
it do be like: JUST ONE MORE SET 
you from gmt+12 or smth?
Est
Vayne from LoL
this is the gauche by vhils?
an abstract drawing by Christoph Amberger , sorceress, echoes, imagination, solar and magic themed, surrealism, , orphism, yellow background, (anime, waifu, exceptional, best aesthetic, new, newest, best quality, masterpiece:1.2), (best quality:1.3), (highres:1.1), (best quality:1.3), (highres:1.1) (chromatic aberration, glitch:1.3), (jpg artifacts, jpeg artifacts:1.1), to8contrast style, 80's, 90's, screencap 1980s \(style\)
That's just what I copied from prompt you sent
๐ฎ oh is that that last purple one ๐ฎ
It was the 3rd last
๐
I like the lil string of hair
New discord update woo
I hope it won't break the app
Or this
(it will break the app)
It's not an update
It's secret thingy
Well ofc you have to update if you didn't
I got this like last month
But they reverted it I guess?
But you have to activate it yourself
Oh
delicious
Dang that's one heck of a secret
hello
I made a mistake in updating discord mobile
AI waifu
I blame this
wait... more updates?
when comes the point in life the updates can stop... furiously keeps clicking huggingface for the next best LLM model
what is up with this model, why is it making such consistent hands?!
i blame nerfgun!
blame yourself!
i think even that left hand IS correct (look at that specific hand posture in real life and get freaked out)
ummm true
but... but... i only wanted a better story model
โค๏ธ getting tatsumaki vibes from #1
yup, thought about her too ๐ haha
CLYDE WHY
i've bettered my ways, i stopped posting stuff from a H-Model
i just wanted to show this model fails at hands too
๐คฃ
dat tummy tho โค๏ธ
no1 perfect!
it got the wrist and thumb right though
Are we blaming nerf now?
I prefer good ass over good hand ๐
are you challenging me to make pretty asses? ๐
do it!
i think she is drawing attention with that backside...
dat see-through dress tho

is it only unipc/ddim that get transparancy right? i don't think i've ever seen transparancy happen on dpm 2m+
2m karras -> no transparancy
hey
heey
unipc -> gets it right again
I got some see-though but I can see details since it was made on my old UI
check your dm please
well, this one is... damn... only this hand xD
Does it count into nsfw? xD
and I got see-though cape!
this is sfw! it should be... technically
see, no-one on the streets is watching her
lol
my eyes are up here... no up here!
girl, portrait, protagonist, multicolored hair
Got a really cool blend of the axe and the background on this one
looks amazing!
i love this color choice
You used a controlnet on irl images for these right?
nope, pure prompt
no lora, no controlnet
Damn then how did the hands turn out that good
name?
Oh it's your own?
You gonna release it on civitai?
most probably ๐
Ping me once you release it then 
damn, I've just got sth adorable as hell!
ghost in a mirror!
you have prompt for this colors? or are part of the model?
chibi, sticker, solo, eagle suit, rawr, thick outline
seriously. this is it
uehh. wait that was the dinosaur suit
here is the eagle suit
Can I ask you to test out a prompt for me?
sure
((1mechanical girl)),(blood vessels connected to tubes),(mechanical vertebra attaching to back),((mechanical cervial attaching to neck)),(sitting),smug face,(wires and cables attaching to neck:1.4),(wires and cables on head:1.4),(blood:1.5)
Add some standard negative prompt to it
at least, you can always ask ๐
@ocean reef i feel this fashion is right up your alley?
oooh, that's some horror allright, first one, can't post it... imma do a set of 4
oooo looks rly fashionable
Yeah these are my standard prompts 
aight chief, i'm pretty sure i can do 1000 generations, i'm not gonna be able to post those here
imma dm you
Alright
mama's turn now ๐
Wow I didn't know MAL got hacked.
wait, mal got hacked?
ye, they have announced it yesterday?
that's a LOT of kechup.
Truly Tartar'd.
TERD MURTTHC
OH FUCK ๐ฎ
so this model has pretty stable hands, stable microphone, stable books?! what's next, an actual working gun?!
WHAT. THE. F. IS. THIS?!?!?!
i am ... kinda refusing to believe what i'm seeing
Looking pretty great. You testing a model?
yeah, the darkalfa model
alfa hentai already had clean AF lines, but it's context awareness was lacking at times
so i tried merging with darkgemini which actually has GREAT context awareness
but... the art is a bit blotchy imho (don't take my word for it, it's still great art)
so... my first attempt at merger failed
and then today i had a brilliant flash of non-drunk idiocracy, and ... well, this happened
3/4 generations ... it's just plain absurd
I'd say these are remarkably coherent for the prompting method, lol.
looking real nice, when do you realese it?
first i wanna hand it over to cryo, as his model was key in creating this, but haven't been able to reach him
You want some help in testing?
nice try ๐
I was actually serious, but alright lol.
Ability to follow the prompt, changes at different CFG levels, ability to gen at different resolutions, etc.
the ability to follow the prompt is uncanny ๐ฎ that's what i was actually gunning for
with unipc, you get this AMAZING TRANSPARANT cloth
How does it do with different ratios/ultrawide, work with LoRAs/networks, no negative prompting? Also can it reproduce a variety of styles easily, or does it take a fair amount of prompt engineering?
got ultrawide in the oven ๐
512x932 upscale x2
i think... it might just fckn smash that resultion with this prompt ๐ฎ
it. Fucking. SLAYED
HOLY F
Alright, well if you ever want some help in mapping it's capabilities let me know. I've had a fair amount of experience with detailed testing and actual numbers.
@untold glacier pm
"Bow before me, peasants"
"so, you are a Maidenless"
reimu is now slime
Wow. Look at those clear legs.
we should call her, Slaimu
slimegirl11 LoRA is very good.
yeah, it can be use in many ways
Someone in the Touhou server named her Suraimu Hakurei
masterpiece of the day (for me ofc) ๐
lelouch de lamperouge
Omg toho
miss tuwho herself
I loved it when she said "it's towing time!" And proceeded to toe all over the hoe.
Y no ping? Also, totally your fault fr fr no cap
Ping is broken bro
Target Destruction Unit (TDU) 451 at your service!
Also this was my first work where I realised that you can delete your positive prompt during inpainting and just write what you want in the inpainted area....
@autumn violet UPDATED TRASHPANDA (Good fortune edition) PICTURES!!!!
pic 1. is the most indecisive raccoon panda ive ever seen he looks like hes about to apologize to a friend or something or thinking about calling that one workplace he had an interview at last week
i just want to help him
poor fella
hey all, can someone help me getting good crossed arms with hands ?
only the hands are bad :/
used controlnet for the pose
but even with openpose normal, full, hands the hands get bad
the hell of AI - bad hands
yea ๐
inpainting didnt help?
find a photo of crossed arms, img2img it with similar style
then move them to this one
and img2img again?
runaround but should work
hmmm
can inpaint only edited area, so whole img doesnt change
This... is not what I asked for... 
She's a beaut.
maybe this works for inpaint
changing the pose was easier
still massive hands tho ๐
i totally understand that
I like
big hands, big โค๏ธ
Does anyone know what model/ lora i should use to get these kind of results?
Here's an ultrawide, mostly-natural-language prompt comparison of a few of the great anime models out there. Interesting stuff.
You can try flat2d. But that type of style is largely up to prompting and knowing how to reach that look with the model you're working with. Unless there's a LoRA for the artist.
sd is trying to say something to you
happy birthday mango ๐
ty
the one on the right reminds me of the helltaker lora
first row 7thC is top tier

Heh, yeah. But it's one of the furthest from the actual prompt.
i got these from a dude on tiktok, he has a lot of imgs posted but he doesnt say what model he uses
i have been trying to replicate it for some time now
I think I've seen something similar somewhere
Imma look it up and let you know if I find it
tysm, i really want to try to get that style
if you want to i can send you the user on tiktok so you get an better idea
yeah go for it 
It's pretty likely that they're edited after generating. They probably generate to get something close to what they're going for then manually edit to get it just how they want.
It is now, the AI decided it. Hence the cake in your previous image.
i usually do the same thing
it's rare to get what do you want directly from sd
wow, dark gemini it's pretty compatible with me
๏ผจapppy ๏ผขirrhhdaa๏ผ
who is the big girl? are you the big girl? it's you? it's you? ahhhhhm lol
She's just disappointed in #๐ฅ๏ฝanime's lack of activity.
is the AI craze dying down?
Lol
where'd the constant influx of ai generated waifus go u_u
They're here in spurts now. The AI craze hasn't ended by a long shot, though this particular group of anime-fans using SD may have slacked off in their waifu genning frenzy.
I don't have any ideas I would like to do and not feeling of just stealing borrowing others prompts and doing in different style
yah I feel that I have 0 ideas rattling in the ol brain cage lol
I do think that the general interest in AI image generation has declined a bit. I think another part of it is that models have gotten good enough that people are more picky with what they share now too. LLMs have had another surge in popularity which has pulled some attention away from image gen.
24gb brothers unite
lol
Definitely, I think it's sort of been that when the models were new any output was amazing but as the cultural space has reached saturation with AI images, it's something less people are interested in.
"maybe it's about the friends you made along the way"
I don't even have any interest in llm
To express in words, it went from "I can't believe an AI made this" to "I can believe an AI made this"
it will become "ugh, when will we see something made by an actual human"
it's already sort of there lol. I want AI quality rendering and lighting but with human levels of design damnit
this good enough?
actually yeah, sort of
the only thing is that it looks off
the trees for example
don't follow perspective
they're all just random sizes
like the tree in the back left, is just a miniature version
it's not a younger, smaller tree, it's just the same age tree but smaller
there's also a wooden platform, with a moss glob on it? or not? idk it's confusing perspective
yeah, for exact control i'd need controlnet
it also feels artificially confining
the wooden fence locks you in along with the thick foliage, and nothing really gives the illusion that there's something 'beyond' you know?
I mean all that said, it's a damn good image lol
thx, i'm uploading that model today or tomorrow ๐
rendering this now ๐
the zen garden you saw was also an LLM prompt
ngl prompting seems like a lot of the detail in the prompt doesn't actually matter
whats ur experience with this?
this specific image... it doesn't work well, it's hard to translate to actual canvas
like whenever people describe stuff in an overly complex way, it's sort of like the AI was not trained on elaborate explanation and would have to be way phatter if it was
i explicitly instruct it to do so! these are all llm prompts
I think its 50-50
Some models do others dont
i made sure mine does ๐
all very stunning
really have to just get a feel for it probably
this new model mix i made is absolutely stunning, even if i say so myself
You're exactly right. CLIP is not a very good model to use as our text encoder, but it's relatively small which is what allows us to run these models on 4GB GPUs. Using a larger and better text encoder would allow for better understanding of language (which is part of the magic behind SDXL and Deep-Floyd) but will make AI considerably less accessible, especially when it comes to training it.
they came to conquer earth!
thanks for the affirmation, I'm sort of just making guesses lol. Wouldn't using CLIP basically bind the model to only be able to represent or recreate what CLIP is capable of recognizing?
i keep reading that vladmandic should have interchangable text encoders...
guess I should read up on text encoders
Sort of, yes. With some training methods we can introduce new concepts to the CLIP model, though this will always come with sacrifices in other areas which can sometimes be unpredictable. This makes large scale finetuning like the Waifu Diffusion team does important to create a good base model for other more specialized models to be trained on. CLIP is what SD 1.4 and 1.5 based models (like pretty much everything we use in the Anime community use). SD2.x based models use OpenCLIP which is considerably larger and, in theory should be much better with the right training, though the base training on SD2.x is widely considered to be much worse than 1.4 or 1.5 so getting to that level will be more difficult.
Deepfloyd and its million GB Encoder

Maybe it's just my prompting for it, but I could never get it to work properly with my testing. Tbf I only tried it very briefly.
Yeah. No one else does either lol
lets see the star trek vessel attempts tho lol, now I'm curious to see how it messed up
Oh I didn't keep them. Such monstrosities don't deserve to be seen by human eyes.
how bad could they have been, at least it wasn't spitting out meat nuggets
True. I've had enough of those to last 100 lifetimes.
vampire fangs?
jesus what have u seen
I saw exactly one meat nugget and thought "I'm never using Dall-e again"
i was just thinking, if you feed to an ai all the tokes of a model, with their weight, you can use it to look for synonyms of your prompts, with more weight, and correct you your prompt with one with more relevance
I want winter back so there would be colder in my room so gpu fan would need to run around so fast
So true. It's way too hot in my room atm.
i NEED winter back because there's no way i can keep my room cool this way... my heater hasn't been on, and i'm actually opening windows to vent heat XD
cooling setup in my PC working too good
I bought a portable air conditioner
climate control is worth any cost
I'm just sad that it fks up the environment
THIS MODEL MERGE IS CRAZY. IT KNOWS HOW TO HANDLE ITEMS
rivals for 2b when it comes to androids? XD
summer afternoon project...
wow cool
For a split second I thought she was at Disney lmao
UCH
I like that second pic
LOl that's what it says on the picture
lol no worries ^^
50% of the time that's true so it was a good guess
wantin to try your mix so bad!
i think she depicts your wish ๐
(purely accident)
too tired to make the whole description, i've tried with cryo earlier this evening, and found out it's mostly the negative prompt doing it's job
๐
that said... i'm just running all kinds of prompts right now, which i did on alfahentai, and there's only one prompt which got worse
somehow, i feel accepting a drink is a bad idea...
Who said she's offering one?
in that case, even a worse idea!
last one feels cruel tho... a pixie who has long lost the way home, and is resorting to painting to remind her of her old home
Anyone tried comparing using depth preprocessor vs depth map from blender?
Fuck. she is cool โค๏ธ
Before she was LLM'ed. I think i like the OG (this) prompt more (in terms of individuality)
Nah, she's fine XD
i think she smoked 1 weed too many... i dunno man, her eyes are a bit red... common symptom...y'know
This one really reminds me of Ilya Kuvshinov's style
you shouldn't be here
lips not pouty enough lol
not a care in the world
not exactly sure what this is depicting...
whats the best embedding for hands?
warrior deity stands atop a mountain peak, overlooking a vast landscape. Their armor is adorned with symbols of strength and courage, and their weapon of choice reflects their unique powers and abilities.
Both of these are pretty good. I haven't directly compared a lot, but from the few tests I did, the 1st one seemed to work a little better and didn't change the style or composition of the image as much as the 2nd one.
https://civitai.com/models/56519/negativehand-negative-embedding
There's no image generation bot in here we pretty much all have stable diffusion installed on our own PCs and generate using our own GPUs. If you have access to #1100170312106127410 you can use the bot there. If not, check #1072220168534642768 and #1072229020520947753 for some information on running it locally.
The Orisha Shango appears before you, with lightning bolts crackling around him and flames dancing at his feet. He wears a red tunic and a crown of brass, and his eyes blaze with power and passion.
ty, can i use the two at the same time?
Yeah, the bot is currently a beta, and isn't usually active but if you can access the channel then you can use it and the local installation method is free and open sourced.
are you a bot?
hello
I put this vae in the correct place but still doesn't show here
I don't find it in stable diffusion
VAEs go in .\models\VAE. You have it in .\models\Stable-diffusion
Then to change your selected VAE, the best way is to go to settings in the WebUI find the "Quicksettings list" under User interface and add ,sd_vae to the end of that list. Click Apply Settings then Reload UI. If a dropdown at does not appear at the top called SD VAE then close SD completely and reopen it. Once you have the dropdown menu, you'll just need to select it in the SD VAE dropdown the same way you do with models in the Stable Diffusion checkpoint dropdown.
Yes I did change
Loras go in .\models\Lora
Yep, that looks right.
I assume you clicked the refresh button in the Lora menu after adding it there?
what are these
extention
Is add_detail a locon?
how can be they useuful?
its normal lora i think
how can I get better negetive promots ?
That's what I thought. I'm usually good at checking to see if they are when I download and putting them in the LyCORIS directory if so, but that one is in my Lora directory. Neither of those extensions would be required to use it if it's regular LoRA.
Mostly by stealing them from other people. Personally I prefer a short, simple negative and only add to it if a specific prompt needs it. This is mine: negative_hand, (KHFB, AuroraNegative), (Worst Quality, Low Quality:1.2), border, nsfw, skimpy, grayscale, multiple_girls, (watermark:1.2)
is there for boys ?
civi has many diffrent types of loras, if you dont have those extentions you cant load them, directory dose not matter
KHFB and AuroraNegative are available here: https://huggingface.co/SweetLuna/Aurora/tree/main
I don't run the Locon extension, I run Lycoris instead because it replaced it. Lycoris requires them to be in the lycoris directory.
You shouldn't need a separate negative prompt for boys. The multiple_girls is only there because wide images are prone to generating multiple girls. If you're having the same problem but with boys, you could easily just switch the word "girls" to "boys" and it will help a little.
or just download a boy model and you always get boys
How do I find good anime boy model?
Can't say I see any need for a model specifically for making guys. Just don't use "sexy_anime_grils_6168731.safetensors" or some shit, and you should be able to make guys no problem.
sure you can make guys but most models are like 90% girls lol, booba will always slip in
True, but some dreambooth model trained on probably no more than 200 images of guys isn't a very good solution to that problem.
Guys are honestly pretty easy to get in my experience
Yeah, I've never had problem getting them unless it was just a badly overtrained model.
Thats my main ckpt
nothing wrong with that as long as you're not trying to make guys that don't look like girls with it XD
use a femboy lora for that 
Is the new version out?
Can you help me
Its saying i have to reinstall a new version of torch
2.0.0.0
Check ds support 
Oh that's pretty easy. I assume you just recently git pulled to update WebUI?
Ima be honest i was just tryna generate and it happened 
I just meant the fix is easy to do once you know how. The simplest method is to just delete your venv and repositories folders. Make sure SD is completely closed, then launch it again. It will take a little while depending on your internet speed, but it will reinstall everything you need to have for the new version.
Can u guide me through it
Oh and ye internet speed aint a problem 300mbps because my uncle did something with the wifi
Just delete those two folders then completely close SD and launch it again. Everything else will happen automatically
Ill update on what happends
If i dont then it was fixed
Btw i really appreciate everyone's help including you 
why? just adding --reinstall-torch on the bat is faster
I have no idea yet it seems more efficient in my perspective
edit the .bat file and after set COMMANDLINE_ARGS= add that line, open sd, let it update and than edit again the bat and delete that line
Not sure how old his venv is and there have been some other dependency changes along the way. He said he wasn't technical so didn't want to have to get into any potential cmd work to update other things.
ohh
i have just this bat with -reinstall-torch --reinstall-xformers and i never had problems with updates. yeah, i had that time i had to reinstal pyton, but that have nothing to do with venv.
oh right, the git pull.... forgot
To put in to perspective i was just activating SD and when i tried to generate something it didnt work
The rest was here finding help๐ซ
just delete the veng and restart sd, he is right
Sweet ima do that as soon im done thinking about the prompt i wrote
after this, i suggest to learn how to update weekly sd, how it will be a pain, in future
sd is still a work in progress and is often updated.
i have a working copy of the venv directory, so in case of a faulty update
than to update i click right button on the directory stable-diffusion-webui
and click on git brash here
in the new window opened, i write git pull
my operative system is not in english
when you click with the right button on a directory, do you have a command git brush here?
damned derpy faces!
lol
Or maybe she's just surprised that she has one flawless hand and another that didn't take much fixing to make it good enough.
yeah, that's surprising enough
These days i wonder how people get perfect hands
What's your model?
Aurora: https://huggingface.co/SweetLuna/Aurora
negative_hand embedding: https://civitai.com/models/56519/negativehand-negative-embedding
Thx
hello, what is the best upscaler option ?
Btw update
It froze
Do i still have to wait even though the download is already completed
Yet its stuck
yep, it need time to update
that's why i prefer git pull and the update from bat
I don't particularly like any of the included upscalers. I prefer 4x-AnimeSharp and 4x_fatal_Anime_500000_G which are both available from here: https://upscale.wiki/wiki/Model_Database
You just have to download the .pth file and put it in .\models\ESRGAN. You may need to restart SD in order for it to show up.
than, he have to install it
Click in the CMD window and hit the enter key. CMD might just be paused.
what's the difference between shard and fatal?
i knew about shard and i agree that's good, but for fatal, it's the first time
There we go thank you again man
thanks
do you know a fix to these red and blue lines ?
if it's not a vae problem. it's the base resolution too high
the model is not predisposed for such high resolution
it was 512x512






