#🍥|anime
1 messages · Page 201 of 1
i'm using forge, a a1111 branch
those i've been posting are 1.5. switching to xl now 😄
forge is a bit better for smaller gpus
i've got 8gb gpu so i'll install it
hmm xl is just.. shrugs
isnt xl better at hands and stuff like that?
shrugs
it's completely subjective and based on specific models and prompts
i have 8gb vram gpu and am using forge. For XL generation, I'm done in about 1 minute
most of the time 50seconds
20 - 30
with high resolutions?
for me it allows me to do the higher res, tho i think that might be because it goes into lowvram mode automatically
for 1.5, most of the time im doing 512x768 at 20 steps, then hires fix at 2x at 10 steps
i'm currently installing it anyways
for XL, most of the time im doing 768x1024 at 35 steps, then hires fix at 1.5x at 10 steps
hmm I wonder , tries at 2048 x 2048
oh boy this will take a bit
10 -15. keeps changing
modal?
A model I merged for my own use, with a character lora

sadge
17 mins
but why 2048 by 2048
a test
oh a test I see
directly generating 2048x2048 is nightmare fuel.. but yeah it's slow, 34.5s
😄
gah
i was gonna post the same seed rendered normally but.. uh.. let's just say i figured out the real reason people like ponyXL..
bad choice of name tbh, it's not specifically a furry model
it's an anime model, which includes furry
i've been playing with autismMix (another bad name) and was having trouble getting anything particularly interesting or "better" from it
then i tried some nsfw prompting and now i see
hmm wonder what NOMD_F would do with this prompt.. tries
what's the prompt
ooh I l iike
kinda
Fully body portrait, far perspective, medium view, extra detailed, HDR
BREAK
lora:unagedown_V20:.5 lora:more_details:.8 (KimberlyNobodySD15:.8), (80's anime girl child) wearing random color jumpsuit,
BREAK
(dissolving particles effect), (holographic lighting effects:.5)
BREAK
((small pieces, tiny, Nano, shattered/fractured glass image , neon sharp edges:1.3)) plasma
with some saved stuff added
😄
tho not sure the unagedown lora is doing anything . shrugs
removes it
if i remove all the breaks in the prompt
interesting
i also don't have any of those loras, so i removed that
tyyyy
ok guess it likes that image, did it again
takes out breaks and saved prompt helpers
ok it didn't like that hmmms
it go crash
i'm generating the same prompt on both autismMix and my own pony mix and they are vastly different for this kind of thing
nitemare
unfortunately I can't use it right now because of some bugs it has, but hopefully they fix it soon
higher sampling helps
and.. back to miku hahah
standing at the center of a multicolored explosion of glass,```
I think my model is a little more creative with the prompts, ponyXL seems to need more specifics instead of 'thinking' for itself
oh boy
standing at the center of a multicolored explosion of glass shards, fractals, broken glass, glittering shrapnel,```
stops doing art and goes back to watching some anime
autismmix_Confetti
well, autismmix is better at smol characters than my model is at least


@gloomy scroll where is dark magician yue
Or did you mean something like this?
Better take few steps back
1girl, (small breasts, open hair, long hair, straight hair, white hair, blunt bangs, azure eyes:1.3), pointy ears, long eyelashes, choker, blush, fangs,


Handle with care
😄
lol, i dunno where this came from
sleepy time
mummy?
Miku loves us little bit too much?
T-there’s no bondage in prompt
how
well there's straitjacket
Yes
The best model came out
Pony diffusion
It's SDXL
Or autism mix (it's based on pony but has more of the generic 1.5 anime look, instead of artsy look)
And forge new ui with speed and other improvements 
but they can produce more characters and more artists styles only difference is that its less coherent but that can be fixed with controlnet
real characters
Not fun :p
Pony knows a ton of artist styles
That's definitely not a problem
tons but not all
1.5 doesn't know all of them
That's impossible
There's literally hundreds of styles on pony
It that's not enough then I'm not sure why you need more
nai base is bigger by at least twice (5million images in nai 1.5 vs 2million in pony) so it has more data (characters and styles) embed into it
plus the author of pony already said he fked up during training and damaged the sdxl base so its not a perfect model either compared to the polished training that nai has
i want to come back to using sd but idk how to use sdxl lol
pony answered very tastefully to so many exotic and absurd situations where 1.5 failed miserably
the only point in favor of 1.5 is that it's way faster
so much faster
It's just a model, you need the sdxl vae
and more styles and characters
is the prompting the same tags as 1.5? Also idk what is refiner lol
i'm not an expert for this, but you don't use lora for styles?
If you're going by tags with the auto complete extension you should be fine using pony
yes the only reason we dont have all those styles for xl is because its too expensive to train them with lots of images and u want a lot to train a specific artist style
Danbooru tags are very well trained on it
No need to use refiner
Recently eface made a list with hundreds of artist styles
little example, prompting, a charming lady lazily sitting on a stool with crossed legs, without stuff like hd, absurdress, on pony, shows an absurd difference
really? where?
I summon @wispy canopy
The images are peculiar
how many anime artists are in that list
real anime artists like the ones found on nai not the ones that never worked on any anime
it's primarily trained on booru stuff, i haven't been able to get it to do actual artist like hiromu arakawa or yuzuke murata
@eface you are a big bro >.<

pretty sure becaues their stuff is all protected by IP, it's not smart training directly with tags on those

yeah, about that one

i dont think nai cares about that considering they use the artists names on the new nai3
i think pony did, so they removed some artist tags

for example, range murata doesn't work either
and that dude has got some pretty notable works
again, on pony, i can't speak for nai
never used it
oh dang 
well, pony got so popular, that civitai has their own filter for that
@iron star @steady grail this is the list that @cold talon made on the artist names that i have ffound, posting this version here because my version has nsfw in it.
eheheh
or maybe because stuff made trained on pony will only be usable on pony/pony mixed stuff
nsfw authors always gives nsfw results?

Nsfw do be nsfw
what do you mean by that?
rating_explicit?
There are 3 levels of 
rating safe
rating questionable 
rating explicit

not always, you can try to combat it by using rating_safe etc. but depending on the author it will lead the image towards nsfw
If you use the tags I said, it forces into it
i was expecting something like this
Otherwise it's pretty random if you don't prompt much
i use everything i find appropiate, tuning down nsfw, with good results
with most of the artist tags you can avoid nsfw just by using rating_safe and a negative with rating_explicit for example
yeah i agree
no, not perse, but you do push the latents towards making horny gen
i mean, if i use melkor mancin by itself: that's nsfw 100% of the time
but prompting away from it with either rating_safe or stuff that defines clothing and the such, not an issue, and i keep his lovely style

(from behind, from below:1.4)

I dun think dat will be a good one with dragon tail
IT'S THE BEST WITH DRAGON TAIL
I mean with dragon tail its kinda wierd or smthing?
I see I see
I'm the local expert around my area
who me?
wings are kinda eh...
It's leaf wings
these wings gooood

Maybe it's time to start Forge
Stupid tag
anyone tried RTX chat?
what you mean?
not yet, but it doesnt comes with memory right?
Create a personalized chatbot with the Chat with RTX tech demo. Accelerated by TensorRT-LLM and Tensor Cores, you can quickly get tailored info from your files and content. Just connect your data to an LLM on RTX-Powered PCs for local, fast, generative AI.
Try it now. Available to download now for GeForce RTX 30 and 40 Series owners: https://w...
wut memory?
their webui, if im not wrong, don't really have memory storing
so you can't load up a conversation
oh deym thats sadge
from what i read, it's using tensorflow or something, so it's kinda fast
well, you can get the webui of textgen
called oobabooga, more steps to get it working tho
i'm interested in a bot able to translate well
something that runs on 4gb vram?
gemini ultra is a paid service, so not local
the current best local model will probably be mistral or something
You think 2 is enough?
You should play guilty gear to really know

i'm actually listening to guilty gear ost atm 😄

I should listen to it again
Provided to YouTube by TuneCore Japan
Sheep Will Sleep ~, if you become fatigued~ · Arc System Works
ギルティギア サウンドコンプリートBOX (7)
℗ 2005 ARC SYSTEM WORKS
Released on: 2005-11-23
Composer: Daisuke Ishiwatari
Auto-generated by YouTube.
such a banger
2 bangers I see
why is the gear guilty 
I'll tell you
exosuit cyberpunk girl?
well... not here...

gimme prompt
danger groin lines 
I don't know what you're talking about
They are just shadows
I'll dm you
kay
sexy man
#1047610792226340935 sexy man
a little late to the conversation, but @wispy canopy can you send me the full list? 😄
#🍥|anime message this one?
nah, the full list 😏
@vital raptor dm'ed u
watch the OVA
What
what?
small horn
it was bundled with guilty gear x2 https://www.youtube.com/watch?v=T_Cio7iVb7s 5min promotional ova
An original video anime which briefly details upon the origins of the Guilty Gear storyline, as well as introduces the characters.

I only played and watched the in-game story
And read the rest cause it really made me curious the fuck is going on
In the end I have no idea, just random bullshit
wow, compared to gg:xrd, that looks so bad, actually
well it was made in 2003 so cant compare it to today's anime style
i really like how did made full 2d in 3d
ever generate an image and the preview image looks a little blurry but otherwise perfect, and then it finishes and it's completely ruined?
it's so annoying
@native halo wait
who is your favorite character/main character if you played the game
try to play with the Always discard next-to-last sigma in sample params settings to see if u can fix it
the girl who has two angels on her back

and the guy with the paperbag
elphelt best waifu
best ult in the game tbh
finally managed it..
Maybe I should try doing some Yue chibis with Autismmix
most of mine looked like that but much more suggestive
it's pretty good at it
so close, too many fingers 
i was making those to not be lewd 😄
I rly would like to use XL models but style I get with soushiki is just so much better
she has fingers?
gen a bunch of images and then train XL lora, problem solved 😄
using what as a base?
I heard aoi is making a SoushikiXL? 
That's plan but kinda too lazy to get good dataset
or am i 
i'd probably use autismmix for a base if i was doing it, it probably already has a style close enough to easily train
This is last answer I got
probably animage or pony or u could even merge animage/pony/autism and then use that as a base
autism is already a pony merge
oh right then have u tried to merge animage + autism?
here with proper fingers, tho again, you guys probably won't notice that
i just started playing with it, not sure i'd need to merge anything for my uses

now i remember theres a lora for 1.5 that does this + the other pose from behind 
w-what do you mean other pose from behind 
Also I'm using 3 loras on top of Soushiki for my style 
the fanny press against glass 
are any of the lora self trained? :or are any of those style available on ponyXL? 
perfection
when i was using 1.5 i had a couple loras and a handful of negative embeddings i had to use to get the right style and look, trained an xl lora then merged it with a few models and now i just have a model for it
but, i only trained a particular face style, much easier
Actually it's 5 loras and I think 1 has been trained for pony and pony already knows one style
hold on a minute, there's a new extension that claims to be able to port 1.5 lora to xl
let me check
That came for a1111 already?
no
I only know that it has some Comfyui wrapper
i can't remember
i read too many stuff about these type of things, and that's why i mixed up

hard to do, they don't have similar weights, i ran into that myself
yup i think that's the one
pony weights are vastly changed from regular SDXL, making merging pretty hard
looks like loras dont work
and by extension: lora's that are NOT trained on pony won't work either
i merged pony with my custom xl model, it mostly works pretty well, but i also merged at a pretty low amount, like ~10% pony, so it's hard to tell how much was retained without a lot of testing
the higher blocks (OUT) are fine, mostly, it's the IN and M blocks where fuckups happen
yeah, ultimately i decided if I want the overall look of my model with the pony capabilities i'm just better off training a new lora based on autism or something
a simple character lora is maybe a couple hours of training if it proves difficult, if i remember right it was like 20-30 minutes for a 10-20 image dataset
of course, it took me around 10 tries to figure out settings and what data i needed
Angel descends 
people that loves elden ring
i still need to go higher difficulty 
who is dat other dude? master chief?
they are both helldivers
ouh
lol, i changed angel to devil in my prompt and it's generating the base character 😄
FOR DEMOCRACY
@lyric zealot you also playing it?
hell yeah brother
angel vs devil..
it's friggin awesome!
it surely is 
How about a cup of LIBER-TEA
can't wait for mech/vehicle to be added
butt-wings
and i'm pretty sure they're working on some extra factions too
i really love how the game feels totally different between the automatons and Tyrianids
yup, if you ever played helldivers 1, there's another faction
yup, loved that one too
but no-one played that one with me
"cuz is topdown shooter"
i only have that one other friend that played this top down with me. But he joined my on helldivers 2 as well

i already have more friends playing with me than party slots, so i'm happy ^^
Imagine not making friends on demand to hell dive with
oh for sure

what, you mean that dropping a 500kg bomb on someone is NOT a good way to make friends?
i've been doing this ALL wrong?!

you were wrong, you need more bombs
when you do something, you must do it well
"380mm orbital barrage, for when you hate bugs and your team alike"
look at America, after 2 atomic bombs, we are all friends with them
even japan loves America
looks like a fun game, reminds me of destiny or warframe kind of play, but i don't have friends and if it's too grindy i'll get bored fast
What are friends?
🤷♂️
what game?
as stated earlier: things you can call in an airstrike to have them feel the weight of 500kg of pure friendship
helldivers 2
i love dbd for that exact reason. what's more friendly than tormenting and trolling?

i rewatched starship troopers a couple days ago too
have a friend who likes to troll a lot too: he quickly learned that trolling in this game just makes you not able to finish the mission. xD
ehehe
i can't find the name of the game in chat
what's the game you are talking about?
you can only troll at lower difficulty
helldivers 2
helldivers 2
you can always troll, you just need more skill
true, until you get overun by the whole planet of bugs
helldivers 2 is carrying the corpse of starship troopers xD
lol
wut is this?
google trends
such an underrated movie tbh, some of the acting was a bit terrible but the rest is a masterpiece
nice
basically, with helldivers 2 becoming as popluar as it is: starship troopers is getting some love too
same way i love the day after tomorrow
the movie is a mess, but goddamn do i love it xD
this game iks so full of love
love for freedom and democracy
it looks really good, i imagine not so great for a solo player though
I mean, read their ceo statements after release, this dude gets it
You can solo it, it's just extremely difficult lol
Just kiting and ammo management
you can solo some low difficulty stuff, up to maybe 4? But you can also matchmake
fyi, there's 9 difficulty levels
i'll have to watch a few more videos on it, maybe i'll get it just to burn some time since i was going to get skull and bones until they ruined it
skull and bones 
i remember seeing some stuff about it a few years ago and it was going to be amazing, then they cut the budget and made a totally different game 😄
I managed 5 solo, but that was not a fun experience lol
yup, no idea why they did that
yeah, they didnt even bother making it as good as black flag
definitely a lot of sneaking around i assumed
the ship boarding is literally non-existent

Nah, you run like hell xD kill bile titan with orbital drop pods
I don't have the rail gun or 500kg so I have to use... Alternative methods
it takes so many shots of railgun on bile titan
that you will probably get surrounded by bugs
U already lvl 20?
Wow, mad respect

I'm only 13 or smth
i didn't play much now tho
just doing personal for the day and done
wanna slowly let the game cook
Yeah, I'm sure they're cooking up great stuff
it's definitely gonna be great 
I mean, this has all the makings for a very long, healthy and fun live services game
yup, and the dev seeing the major success, their roadmap totally changed
I think they were aiming for 50k players or so?
i guess so
Mad critical success
i means it's reasonable since they didn't do too too much promotion before this, and the previous game was just about 9k peak players on steam
but now it's 400k
im wrong, it was 6744 peak for helldivers 1
for helldivers 2, it's current peak is 457649
How is there that many people playing a game filled with bugs and crashes?


if you want the real answer
it's that not all of them are playing
a lot of them realised there's a max server capacity now, so they just afk in their ship
Are you seeing that on stream charts
and there's no afk auto kick atm

I've seen it from friends who play
tbh, i didn't have any issue joining tho
all games need a player count
steam charts is stoopid cos it's only tracking steam?
from someone's point of view.. yes

there's actually an in game counter kinda
to show how many helldivers there are on this planet
how many shots fired
how many kia
yup, that's also awesome
i mean, the game is full of these small details
yup
my brother found out, if you got a flamethrower, and you empty a full can and hit enough mobs: your character will do maniacal laughter
s-stoopid game

my computer plays all games, stoopid games 
it didn't trigger all the time, but it was with the flamethrower that he managed to do it
i love how when using those with crazy amount of ammo, like machine guns, and you just keep spraying down the bugs, your character will shout those democracy speech
I justed finished the recent epdition of NMS last night. was good

the new one? how was it?
i stopped playing nms but probably cos i spoilt it for myself. Went to someone base, and took all the stuff to sell
and now im so rich, it's not even fun anymore
nah cant be
impossibAl
nothings impossible
13900K, 4090, 64GB ram, 8TB of NVME.. I play all the things
I've been avoiding that one after how much time i spent on the first one 😄
can you play spider man 2 (Marvel)
it's so buggy and unoptimized, that you might even have problem running that with 120fps
nah you cant
or maybe even 60fps
hmm I should put some of these pirate anime designs on a shirt...
thought of doing that make anime great again, someone beat me to it
hmm 3 legs?
love it when I put random on style pile comes out with interesing art
yeah, i play a lot with style wild cards

u stupid
yeah they made it to easy to make money
there are custom games settings now where you can make things more expensive
it's a good game
It doesn't look too interesting to me, I like games with something cool about them
and nms just looks like a spaceship game galaxy thing 
im interested in the next game that they are making tho
but please don't over promise this time 
yup, but if it works, it's gonna be good
would love to see an anime style game like nms tho
"don't worry sisters, I got this"
character sheet?
what could ever go wrong
@lyric zealot tier 6 solo, down ^^
She looks like she would slap my ass.
that's actually impressive
you only pick fights when needed, and basically do everything by strategems...
bomb the everloving shit out of everything
120mm is NOT your friend
i always try to make it a point to complete every single thing
but can only do that on 4 and lower
5 with at least 1 friend, possible
duck duck .. grenade
5 is harder with friends, if they run around like a headless chicken
oh wait. nm
"oh i like the machine gun" -> no, it's a useless weapon
he played helldivers 1 with me as well, and we are both in tune of what needs to be done 

bliss ❤️
FOR DEMOCRACY! (too bad it doesn't know helldivers)
miku is not subscribing to this idea of democracy? she's looking very peaceful
this is the super earth before the invasion of the bug 
they destroyed the miku?!
No! Sweet liberty... NOOOOOOOOOOOO
pony and guns aren't good friends 😦
for autism, you can get some decent one
as in lora?
nope, im just prompting ||assault rifle||
yup, most of the time is bad as well
bpn did it better, but i can't port shit from blocks 7-8 😦
or i'd have to figure otu another way to get the weights working
nope, it won't respond well to the word helldivers 😛
it definitely isn't trained on that tag 
various miku operative variants ^^
last of my miku warriors!
im not a brony
If you scroll up this chat, most of whats been posted is using that model 
thats allot of msgs i dont wanna read kek
Its not a pony model, just people from the pony community who made it, thats why the name is pony diffusion
nubby 
anyone else notice that pony/autism is slower than other sdxl models or is it just me?
adetailer now has hand fixing that at least sort of works (the technique works better in manual inpainting than it does in adetailer) - https://github.com/Bing-su/adetailer/pull/460
faster fork of a1111's webui. most extensions are compatible. many of the big ones are built in - https://github.com/lllyasviel/stable-diffusion-webui-forge
not really an extension, but someone made a thing for x-adapter in a UI (allows using 1.5 controlnets and loras with xl models) - https://github.com/kijai/ComfyUI-Diffusers-X-Adapter
oooo bless its nub
lemme take a looksie
ooooooo this new fork
looks nicee
got tired of fighting linux so im slowly migrating all my stuff back on windows
linux
You have nvidia gpu now right? I remember you had amd at one point.
Did you test that hand adetailer thingy nubby
yeee still rockiong my 3060
and oc'd my mobo to 4.2
mainly so i can play star cit
lol
is xl still better used on comfy?
Yeah, I've used it a little bit. I don't feel like adetailer is reliable enough for it to use all the time like i do the face model. You can use the same controlnet preprocessor in inpainting though. Usually I just fix hands manually though. It's good practice.
Forge basically is some of Comfy's code wrapped in A1111 UI. It works well for XL.
very nice
steals your cookie
If you like Comfy, but want an actual UI there's also StableSwarm - https://github.com/Stability-AI/StableSwarmUI
Uses Comfy's API for the backend and still has a node graph that you can use if you want, but the main tab is a real UI that will build the nodegraph automatically for you.
You can have it Yue has still many cookies left to eat from my Chrome userdata
is it the same speeds as forge

If anything it's a tiny bit faster.
me and my 5 addons will be happy

if you have channels here hidden, you should set this one to visible. lots of useful info for Swarm in there and the developer (mcmonkey) is very responsive if you have any problems/questions https://discord.com/channels/1002292111942635562/1134546054872842290
big cookie 
Cookie party
nice
hewo lewwds
anime basketball poster
don't press it
I love this style
you could get a 50$ steam gift card, too bad lol (disclaimer: don't take me seriously, plz)

today i played with a chatgpt model
what a mess
translating with it is harder than it looks like
i need something to load a certain portion of text, feed on the model and than load the result on an excell sheet
line by line. now i'm thinking if it's better to complete my study on python and do it myself, or just learn japanese or chinese XD
a worthy challenge
Can't you tell it in the settings to always translate stuff to English automatically?
Learn Russian much cooler
👍
Is it like hundreds of lines?
I'll never financially recover from missing this opportunity 
@steady grail have you tried Bard/Gemini
gpt4?
i'm using gpt for all in local and looks like we still don't have bard/gemini
i'm playing with orca
you might have better luck translating with the best models
service online aren't free or tokens are very limited
entire books?
yeah
the japanese classical blue collection
I think you can get it pretty low if you split it smartly
how many tokens may be? 100000?
IDK that book collection
if you like deep and tragic literature, it's very suggested
I’ve developed a publishing tool that handles large text files, utilizing the GPT API for both editing and translation tasks. The system is generally functioning smoothly, barring a significant problem: the API operates at a painfully slow pace. For instance, processing a book with 500 pages and 90,000 words takes approximately four hours. I’m w...
it's very close to kafka style
do you have a link to the book or isbn?
I'd try it out on like, 2 pages and verify how good the translation was
but I don't think I'd ever want to read an auto-translated novel lol
how would you ever know if it got fucked? unless you could read the original too (side by side)
yeah, that's why i wanted to do something automatic
and than proof read while learning the language
like i did with english
it'd be nice to have a tool that you could use to verify your understanding
read sentence -> think -> click button to translate sentence in context of paragraph -> you can see if you understood
i learned the english with google translator and a big a** vocabulary
the grammar comes from school
so now the process would be, an ai translator for the stranslation and a big a** japanese vocabulary
for what i don't understand, i ask chat gpt
yeah so you don't really need to translate the entire novel, just pieces
yep
this book it's mind blowing
i read it so long ago, and i was so deeply hurt but at the same time, it was all so true
That's just tragedy in a book
i know
right?
where is the boundary between nobilty and weakness?
i'm still trying to answer to this question

Google recently announced Gemini 1.5, with up to 1 million context length. It’s open to certain developers, and some people have tried it.
that's cool
The close build of that was said to have up to 10 millions context
wow
And to visualise how big that is, you can fit the whole Harry Potter book, all of them into it, and more
Don’t quote me on this, I heard someone said that
ever make a prompt that gives you a piece that makes you go.. wait how did it get to this? what
?
Read the manga, very good 
@steady grail they made 2 lol
i love it
💕I love this one
i attempted to cheat asking the chat if it could code a translator using a local model
useless to say i failed miserably
it used again and again outdated libraries lol
Are LLMs even decent at translating in the first place?
you still need some proofreading, but it's a good start, imposing the right conditions to the model
for example, specifying the gender of the various characters, or the context of various names
for what i tested, translations are more appropiate than google translation
perfect lightning at the cost of hands
dreamshaper lightning is shockingly realistic
whats the best exntension for like vid2vid for like short gifs or mp4s . i just wanna be able to apply like the ai filter type thing
How do I get rid of Shiny plastic skin in the Anime styled pics I make?
It's probably either the model you're using or something in the prompt (possibly negative embeddings)

use flat colors???
maybe
[hatsune miku::0.7]
prompt and neg?
1girl, and whatever tag you wanna add,
source_anime```
try this format
(worst quality, low quality:1.2), 3d, censored, abdominal muscle, artistic glow,
this is my negative ^
you can add score_4_up and score_5_up in negatives too, works pretty well too
in both positive and negative? or just in negative?
just neg
oh, and another one that's pretty good in negatives: expresionless
I need expressionless

pout =/= expressionless
you can get more pout if you put in expressionless in negatives
why is the shirt clippin tru her skirt
hmm who knows
ou shes here without a lora

❣️




