#🍥|anime
1 messages · Page 74 of 1
Ye, cause it hijacks the built in LoRA loading, the LyCoris one I think doesn't, so I use the LoCon extension, which is the same, but better 
yeah fair
LOL
is 4 months that old?
anyone got tips for getting a plain simple background if "simple background" isnt enough
🤔
idk feels old to me
the showcase pics are so bad too 😭
i really gotta update those too
I mean...
the only one thats slightly updated is the first one
Models 4 months ago was pretty meh too
There's no way this is worse than my first lora
that shit was only usable at :0.3
WHAT??
Why is my model like this
It's either good hand or straight up put the hand in the blender
is BREAK new tech?
I've seen it used and sometimes try it in prompts. I'm not sure if it's tied to an extension or not but it really does seem to help some of my prompts become more dynamic.
She spilled ink all over
cursed and nsfw at the same time
How cursed
bannable cursed
Deformed body 😨
it's deformed, it's semi-nude, it's underage, but the prompt is : A small creature with large eyes, floppy ears, and a tiny nose. Its round belly is covered in spots, and its tail is short and fluffy. The character is wearing a colorful outfit with bright colors like pink, blue, and green.
ugh. these are all horrible. don't prompt that on an ex-hentai model lol
wait. how da F does it almost get this right!?

@wooden coral quite a difference in image categories here XD
im getting a lot of lewd monsters i cant post or even wanna keep 
Stay hydrated
so pretty! ❤️

is this guy jelly of the others?
fireworks trees?
I TOLD YOU TO STAY INSIDE THE LINES! 
holding staffs is still hard
magical floating staff
"it's free real estate" -> don't trust this guy with his wicked smile
He sells you this as NFT
"DID I HEAR NFT?!"
NTR
my dude needs to chill somewhere cozy
preparing for the relaxation trip
damn, gotta remove the wine glass prompt... that kinda destroys images
nice rock
doing photoshoots with rocks. dis dude know's where it's at
ok, vacation over, back to doing vigilante stuff
"I uhh... i brought a colleague"
so villain theme ok
just going with ehh... whatever comes out 😄
injecting some danger husbando's in chat to offset waifu ratio
If you use GGML you can run 13b models, maybe even 30b(will be slow as hell tho)
And I recommend guanaco and manticore pyg
intergalactic super villian
is that a tree growing out of her, or is it just a tree disguised as a waifu?
smol
Elf Queen, Selfie, now it's my Kingdom

xD

perfect hands jumpscare for no reason

join the battle
serl dropped a new banger, as always https://www.youtube.com/watch?v=HYRo4A1D8tE
For my first collab with No Hero, we thought we'd make a cliché song about a cliché Shōjo.
🌸 Out now https://fanlink.to/FeelThatWay
No Hero https://linktr.ee/officialnohero
S3RL - https://withkoji.com/@S3RL
Video by Walras https://www.youtube.com/WALRAS_official
ok sorry man, chill 😔
Results from a new style i trained. Dataset was 255 images of strictly monster girls
Uh, eyes burn warning.
Male (♀️)
It's still up unfortunately
Epik
aw wholesome
smol family
space station propaganda?
prob associating children with space future or idk
and also getting some monarch portriat
casual isekai traveler
gotta dig, gotta dig. in the nostril lies gold
i stopped using it when it just breaks down every other update 😦
what would the update be?
Uhh lemme grab the post
Havent used it myself yet
use vlad if you're ok with stuff breaking down all the time
@wispy canopy ur model works really well with this monster girl lora i trained
😄 nice to hear ^^ i suppose the ex-horny weightings help with monster girls 😄
Diffusers integration is interesting. If he gets that stable and working with extensions and normal optimizations that could make switching worthwhile. Diffusers is supposed to be significantly faster from what I've heard.
Im amd ofc im ok with it
Comes with the territory
Somewhat
I am getting a few private partless girls but other than that it mixes well
Diffusers is another tech?
It only 'recalls' what the privates are though, it forgot a lot about them though, but I'm OK with that
diffusers is decompressed models. When you see models on HF that have this stuff, that is all the diffusers library.
Something else from compiled models then? Which... Is honestly is not that great a speedup
i should try out vlad today
didnt mean to reply
it supports linux/rocm right?
I think it does, I recall seeing some options related to that
I'm not sure how much of a speed up it is exactly. I've just seen others talk about it. Never ran Diffusers myself since it wasn't supported in any of the WebUIs I've used.
Ima try it
anyone tried this update for anime stuff? https://github.com/Mikubill/sd-webui-controlnet/discussions/1464
You already downloaded this?

What? You don't open loras you plan to download to new tabs and then be too lazy to actually download anything and repeat this for days
If it's something i wanna download, i actually download it immediately
or bookmarked it, and close the tab

^
weird
???
the fish people!
Do u mean to tell me u are racist towards human fishes?

these are great. they remind me of Rikuo
Whats that
Dark Stalkers Rikuo
Oh is that his name?

I am. As a demonic elf hybrid. I am born with a hatred for fish and humans
Humans I understand but the fish are just trying to swim along and live their best lives
I used to be an adventurer once, but then I took a fish to the knee. Ever since they're my sworn enemy. Humans? Well I just like killing em
Scored an 11.3 rating on this one!
A wild story to be sure
Yes and a tragic one. My wife was fishing on that boat and when the fish jumped out and decapitated her, I ran in to kill it...but then it rammed into my knee and took all my godpowers
I heard theyre called kneefish
Im sorry for your loss
Thank you...they weren't even done there. One of the fishes, jetted itself with her head still on its horn...It flew all the way to my house, crashed through my window and framed her to the wall before flipping me off and then starting to squirm and die on the ground

That's not all...the next day I woke up and they had written "I heard you kneed some help walking", and that was it...I cut my leg off and allowed my demon part to take over more and regrew it and then I boiled the lake away until there was nothing but dead fish on the bottom and a few hundred dead tourists that were swimming on the beaches. Worth it imo. Anyway it's time to go to bed. Thank you for listening to my story of my past. It's a painful moment but...At least I got back at them. Now I'm being hunted by bounty hunters for some reason, something about murder? When the fu did I murder anyone? FISHES!? Damn
good night ;D
what are the best anime quality tags for general use??
i think it really depends on the model
Legg
Easynegative vs simplenegative, what's better?
as with all embeddings, it will depend heavily on the model and somewhat on the prompt. Use an XYZ plot to compare and see what you like the best. I don't personally use either of those ones any more.
hmm, are there ones for just general use? also what models do you use?
i mean, if it's general use, then maybe like masterpiece, best quality, detailed background, cinematic lighting and stuff like this?
obviously, it's not gonna work on all models tho
and sometimes, less is more
im trying "new" model, but usually expmix_livev2 and holokukiv2
I think the bat is a little big for her...
hows everybodi
lmao, why are you still using that nickname
it's been awhile buddy

but imma go sleep now tho

goodnight
2b figurine
isnt it noon
how are you sleeping
Once a smolsusser always a smol susser
tyy
"Are humans even real? Stay tuned for our complete investigation. Now, a word from our sponsors."
Strike witches?
dunno what spell is cast here, but i'm pretty sure being on the receiving end isn't a good thing
ketchup shirt, lets go with that
upper right dude tho ❤️
blood dragon? 😮 uhh.. ketchup dragon
really love this prompt tho ❤️
MACSHER!
I went with something more friendly
nah, this is wholesome -> it's very anthromorph unfriendly just to assume they're evil, i mean, that guy on the bottom right looks evil, but for his kind, that's nice smile
good point

made this guy an emoji
BNAK of OTE
its getting a bit... odd
this seems pretty wholesome, girl raised by marshfolk?
pretty cool as unstable prompt, just change the actor name to get a character concept change 😄
of course, lets have an LLM help me out in that case 😄
can you guess the actors that have been referenced? (without checking the prompts)
His ball is showing 

Try coffee 😄
1 coffeegirl coming right up 
I think she's about to drop her iPad in that last one 
Very nice 😄
Keanu
yup ^^
Leo?
ryan gosling 😮
Aaaaa okay
i mean, even i guessed that one 😮
Its 4am im getting to go in bed lol been messing with the deck all night again 
Making some starry ohjou
how did you get animal legs?
Special lora im releasing shortly, but u don’t necessarily need it for that
insect girl, insect wings, arthropod limbs, carapace, joints
just try that with some models
any ideas on how to fix this
this is the prompt
a painting of a decrepit cathedral engulfed by mist with the face of a single girl with a 1750 hair carved stone, sculpture, thick fog, looking toward infinity, Grim, Hopelessness,Despair,Loss, Weeping statues, Impasto, Golden ratio, Triadic, in the style of (Beksinski:1.3) and Flix Vallotton, Gold, ultra detailed, intricate, surrealism,
Neg: white border
Maybe
Enklish is hard 
it helped ❤️
dont worry english is very hard its my 3rd language try out better discord
spellcheck
another figurine

prob one of the most successful tomboi although the hands destroyed
Evening chat
well, cos i sleep at noon, and spend my time at night 
Same


simple prompt ```detailed background, masterpiece, best quality, highly detailed, high res, cinematic lighting, realistic, hyper realistic, photo-realistic,
2girls, (kiss:1.2), cute, closed eyes,
Can u teach me llm
What's the best one to dl now
imma be honest, i don't really know much about llm now

Smol sus u should try to monetize this I swear
how can i stop other views of the character,( like a close up of the hands or smth) from appearing? i just want the full body and the background
what checkpoint?
monetize what 
this should have metadata
thanks bro

Allow me to introduce, Marshal Akane Kurokawa.
hello everyone
damn I didn't message for like a whole 2 days or something
feels way longer
anyone has a very high bpm song that is very "happy"
something with preferably a lot of drums
I really need a song like that too 😔
ありがとう!このチャンネルで日本人が何か言うことは珍しいから、嬉しいよ!
Artist: Photek
Album: Ni-Ten-Ichi-Ryu
Label: Science
Released: 1997
Track: 1 Ni Ten Ichi Ryu 4:36
(from MTV's AMP 1997)
Finally a decent quality video to make justice to what may be Photeks' best work.
Ni Ten Ichi Ryu is the name of a technique of using two swords (longsword and shortsword) simultaneously. It was invented by Japan's most famo...
Is there any extension that utilizes a prompt LLM? But sort of narrowed down so it's smaller and more specific and easier on the vram and only loaded temporarily
silly tavern??
why would you use that to generate prompts?
literally just another frontend
No extension like that, but if you have a decent amount system RAM you could run a GGML version of a model off your CPU and system ram, or GPT4 API + Silly Tavern like 川上 mentioned is another solid option if you don't mind paying for GPT4.
I mean well if you are using the api then might as well
Uh, does that GPT4 run off some online server then?
It's a fork of TavernAI, but group chat alone makes it worth using even if you're running your own LLM locally.
cost money
then that's not helpful :p
GPT4 runs on OpenAI's hardware. It's the better version of ChatGPT.
I know what sillytavern is, I just meant that if they are only using it to create prompts for sd there is not much reason to use it
I have 32gb ddr4 ram and a amd 5600x
just run on cpu
you run on ooba?
yes
https://github.com/oobabooga/text-generation-webui use this and pick a GGML model to use.
hmmm any recommendations on a good prompt ggml model?
or like llama cpp. at this point I cannot even differentiate
I've tried llama's and on cpu it's so so slow
I think ggml is a certain type of model that works on llama cpp. idk
depends on settings and stuff
amd cpu
Here are 126 models to look though @cyan burrow any of the ones that end in GGML can be run on CPU easily. https://huggingface.co/TheBloke
if you include koboldcpp
occam's fork i think

thebloke madlad
fr fr. Dude even makes model cards for every model so we can actually know how to run them.
https://huggingface.co/TheBloke/Wizard-Vicuna-13B-Uncensored-GGML ah this one perhaps
Yeah Wizard-Vicuna models are good. Wizard-Mega models may even be slightly better.
yeah 30b might be tight depending on how much of your RAM Windows is using. You could definitely do it, but might end up using page file which would be very bad for performance.
I would rather have an okay performance on 13b but with perhaps less power behind the model, than a superior model that will give me 0.2 words per second
characters*
I think my ooba is broken
it cant install quant cuda or anything
or something
trying manual update instead of using the cmd one
Why do you like Wizard-Mega better?
what did she mean by this
statistically, it's very slightly more accurate.
and more natural language capability
My step brother just called me but as it turns out it was his friends who told me he was shit faced drunk and that they will take care of them, then passed him the phone and he sounded like he was gonna collapse at any moment, told me he drank 32 shots. very funny.
If you speak you dedge
Here's few noodle finger images
Veruy cute
Heh... bun eating a bun 
always gotta make sure you get the uncensored/unfiltered/unlocked LLM
just downloading it now but shouldn't take long
hmmm should I take the offer for 1000/100mbps internet at 60 dollars a month for 12 months, bound for those 12 months...still cheapest there is where I live :p
What are you on now?
500/100 but offer running out this month (30 dollars a month) and will now cost 65 dollars
1000/1000 is 75 dollars
same kind of offer
You're saying that your current plan will increase to 65 for the same speeds, vs this new plan you're looking at for 60?
well what I did was switch ISP to get 500/100 for around 30 dollars a month for 12 months, now I can switch again to another one and take another offer
how do I do 1 step this 1 step that?
[x|y]
thanks
but if I don't switch, I will have to pay 70 dollars a month for 500/100
or 65
depends on currency exchange
Then I'd switch for sure. Sounds like a decent deal--but man 1000 mbits down is a lot to my eyes.
she's becoming a unicorn
60 dollars vs 75 dollars on 1000/100 vs 1000/1000
You plan on uploading tons of stuff? Backing up things, etc? That's the only thing in my mind that'd make it worth it.
I don't know
Downloading every model in existence
I might end up having a server of sorts in a separate computer if I can get someone to fund me a high vram gpu and stuff
where I will then provide access to a limited number of people to share it
@supple raptor do you know what to add to the webui.py to start without a model loaded?
(on ooba)
on ooba it automatically starts without a model
I don't know how to start it any other way
I have start-webui.bat
may I have it?
mine is from a fork that uses miniconda instead of I think it was micromamba
so won't work
how did you install?
If you used the installer then you should have start-webui.bat.
I don't
Dunno what to tell yah
I can try installing it again...third times a charm?
oh
and there still isn't a start-webui.bat created in ooba
not from the installer anyway
I guess it's actually called start_windows.bat, but I just downloaded a fresh copy of the zip and it's definitely there.
@cyan burrow what exactly are you trying to do anyways?
I have start_windows yes
Im trying to start it without forcing me to load a model and I had to reinstall for it to allow me that
Oh ok. Then according to crazykazoo, the start_windows.bat should do that. I installed manually in WSL so I don't have any of the bat files. I just launch everything through CLI and have never had a reason to want to launch without a model.
why through WSL?
It's the recommended method for Windows installation. Not sure if it's better now, but when I installed there were tons of problems with native windows installation.
May I ask which model you are using?

argh well, I took them sitting next to each other, photoshopped them and moved them closer, shifted perspective, skew etc, played with layers to make their mouths touch and then ran it in img2img
I peeped at wsl. Looked like there too many steps for my lazy gremlin brain. Think ill stick to dual boot
Attempt was made
It was a lot easier than it sounds. Literally just wsl --install in PowerShell to install WSL, then download Ubuntu from the Microsoft Store. Then launch the Ubuntu app which will take you through setup the first time and leave you in a Ubuntu CLI inside Windows.
oh i was looking at manjaro, something im already familiar with lol
twins
There is an unofficial version of Arch built for WSL on the Windows store and github, not sure how good it is. Probably not worth it if you're happy with dual booting unless there's something that you need to be able to use Windows at the same time for. https://github.com/VSWSL/Arch-WSL
yeaaa was onkly thinking about it to merge both ssd's for abit more space but not rlly worth it in the end
he just got home mostly unconscious, his father washed him a bit and now he sleeping. do not drink too much alcohol my friends
i just had a colleagues evening, and had a drink. or 2. or 3. counts are arbitrary at this point
might have had a drink.
😄
I mean if my step brother really did drink 32 shots he'd probably be dead by now or something idk
in any case he drank way too much
32 shots would kill me too. unless you really like having him around.
32 shots will kill any full grown man
what denoising you using when upscaling?
and full grown man + increased body weight is pretty much the best defence you can get against alcohol poisoining
latent (any varient) 0.58.
I mean my step brother turned 18 like a month ago
upscale (any esrgan variant) 0.5
talking like sd upscale script
if upscaling not in original res AND NOT using an upscaler: 0.55 is the lowest you can go on dpm2++ (any type)
else you'll end up with blurry/hazy images
if u use upscaler -> even 0 is fair game, but then it's just upscaler being at work
yeah
upscaler + 0.2-0.3 -> you just want your model to fix minor details
in that case, try 0.4-0.5
0.58 would be too high in that case
but 0.3 would nearly be "just upscaler"
is this just adetailer?
what do you mean by just adetailer? 
adetailer is active here, yes
like hires + adetailer?

no upscaling
yup yup
adetailer OP for certain
metadata if you want
you just reminded me of when I check metadata on images from projectai
and then I proceed to get a headache
how tf they doing it without having brain overload
lmao why
they doing latent couple + adetailer + lora mbw + weird model mix + 20 embeds + dynamic thresholding or whatever it's called (the cfg thingy) + 20 BREAKs + etc
not human
just try looking at one of ctd's images
my pnginfo stuff is broken here
These aren't even that complicated though 
yeah I know just wanted to grab ones fast
What does BREAK do?
I always see this shit in a prompt but never understood what it does
.
DAFAQ IS THIS COMIC?! cursed i tell you
lmao
as nubby's quote -> BREAK is a special keyword
I would also like to know the answer to this
Oooh
your prompt is broken up in sets of 75 tokens
so you can have a bigger prompt?
but, with BREAK you can go "early" to the next set of 75 tokens
you always have been
very early on, A1111 or someone, figured how to go beyond the spec of 75 tokens
by just submitting all 75 groups of tokens, and just weighting them average
so 150 tokens is in all reality : 75+ 75 tokens, and each set of 75 has the same weight
but, for example, you have 50+2 tokens, which you want weighted equally
and here is where BREAK comes in -> by using BREAK you can make sure those 50 get BROKEN early, so they become 50(+25) + 2 (+73)
SD only understands prompts that are 75 tokens long. Automatic1111 added a feature that allows it to split longer prompts into blocks that are 75 tokens long, then average those together. BREAK simply fills the current block of 75 tokens with blank ones so that the next token will be the start of the next block of 75. This is useful because the closer the tokens are to the beginning of a block of 75 tokens are considered more important by the AI, and also because sometimes a single word might ocupy both token 75 and 76.
(the empty space is just padded with 'nothing')
@supple raptor how many times you figure we'll have to explain this? 😄
I don't have enough fingers to count that high. 
@unreal ridge so don't worry about it, it's pretty obtuse, the 75+ method was figured pretty early on, BREAK came later, and it's just magic abacadabra -> if you don't know, or are not certain, it's ok to just ignore it for now
I pretty much just prompt whatever I want and that's about it. There's a shit ton of stuff I don't know
you get my upvote, just do what you want.
same
Thanks for the explanation you two ❤️
more upvotes. just do whatever DA F you want
Nothing wrong with that. If you want to learn more, there are lots of neat features that you can explore here https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features
yes. thanks for the explanation you both
rip gif
Might not have been a problem with the gif. My link didn't get an embed either.
gif broken 😮
rip embeds

without inspecitng prompt, who can guess which actor was referenced? (continuing from this afternoon/morning)
FYI, on the topic of BREAK, prompts are always padded to make exactly 75 token blocks so adding BREAK at the end of a prompt does nothing. First grid is 1girl second is 1girl BREAK
that's why i always say (in specific case of BREAK) "just ignore until it becomes relevant"
you're absolutely right!
You guys use positive tags at the beginning of the prompt or at the end?
Beginning (usually)
morning all, it's been a while
👋
I usually throw them at the end, but it depends on the prompt. Sometimes I throw a BREAK right before them. I don't use many quality tags these days though.
the new Evil Dead movie is really good
How is version 1.3 of auto1111? I haven't updated yet
I love the trapped claustrophobic feeling and how it takes both the gritty and frightening and the black comedy from evil dead 2
I totally lied lmao now that I'm looking through my styles all of my positive prompts are at the end. When did that happen lmao
and some Sam Raimi inspired gore
I haven't found any problems with it so far and it loads a little faster for me.
Are there any newer Korean or Japanese horror movies that introduce new concepts but maintain the slow burn terror of the past? Maybe even apocalyptic horror
Ok, hope all goes well for me 
You may want to delete your venv and rebuild it once you update because pytorch was upgraded to 2.0.1 and I've seen some people not get updated otherwise. If you use xformers, you can install the latest stable version after rebuilding venv by activating your venv and running pip install -U xformers
guys is there a fire costume name in anime or a diamond clothes ?
We are now in the age where we can ask our AI characters what they want to look like. This is one of those mind-blowing realization moments for me, heh. Crazy times of innovation and chaos.
All hail Light Yagami from the greatest anime ever, Sipiennote. LMAO
𝐒 𝐈 𝐏 𝐈 𝐄 𝐍 𝐍 𝐎 𝐓 𝐄
Haha
Or this one....Sith Note, or whatever it says down there.
More like Hell Note....
I never realized that someone threw BREAK at the end of the prompt, what was the process or theory behind this?
anyone know what causes all this mess to appear in the background
cring

I did the same thing to mine a few days ago
.
What model is that
One of them was efaces model, another was anythingv5, and one was breakdomain m2150
Nice good stuff
Yet another chain of Akane.
wow this channel is now like completely inactive lol
you guys remember the times were you couldnt even read all messages because it scrolled so fast?
Momento when an image goes 85 MB
Tru, I rember
I mean on the other hand, almost everyone that is active constantly knows each other like almost personally

Yeah, I've been playing games again recently and messing around with chat bots. For a while SD was taking up pretty much all of my free time.
i suppose we're all waiiting for the next big tech? 😄
Generating images isn't as much fun when it takes almost no effort to get a good image. There's a reason gacha games keep drop rates low on the stuff you want.
what's that?
Basically automatic inpainting at full resolution. It detects the faces in your image then inpaints them to get more details. It's especially good when the subject is too far away to normally get a detailed face https://github.com/Bing-su/adetailer
I see
Same, since getting the deck for my birthday thats been taking all my time and hardly been on sd.
I resubbed to FFXIV
i made a fake manga cover 
Same with wow but i got both psnow and xcloud working on my deck, playing ghost of tshusima on it rn lol


Personally waiting for WD.
But that'd be forever

Waiting for WD be like: https://tenor.com/view/waiting-for-you-waiting-waiting-for-text-gif-14699092
I'm more suprised to see how there's nothing news over a month, when we used to miss QOL extensions every 10 hours.
Atleast catching my eyes.
Yeah, everything has really slowed down in image gen.
Not sure why, I also don't wanna browse CivitAI; their marketing don't target me at all.
I surely miss those days
I really love back then where we all exploring SD models through huggingface and just smashing stick n stones.
Fear me for I have signed up for theater lessons
n i c e
Waifu Diffusion
Wd40
I love corrupting gpt bots and llms etc into saying the most profane disgusting things
like how it would be "non consensual" about stuff. Or just straight up murder
ok merging Exalted Exarch 0.3
I'll take, "Things a serial killer would say for 300, Alex."
I'm fairly pleased with the block settings
X/Y/Z plot will create 24 images on 1 12x2 grid. (Total steps to process: 360)
testing the new model with 12 prompts on 2 vae's
did not expect this
Incubus and Succubus
I think it's doing well so far
but that's the bell ringing, I must go sleep
good night everyone
before I go, here's two tests with a few prompts on early alpha version of Exalted
In which chat can I ask for help? i'm new here
hi, you can get help at #🤝|tech-support
it is not tech related, it is a bit specific, basically i wanted to know the best way to remove the background of the image and leave only the character without messing up the antialiasing
you can try using abg extension
ty i will check that













