#218 xl
1 messages · Page 2 of 1
it's annoying
used photoshop autofill but thats still slow
I'll try firefly later maybe its better
where
np
it has healing brush tool there and other stuff
but remove tool is just like lama cleaner
1016 messages :o
we are chatting
wtf happened
everything
If anyone finds this useful
by rolua by akizero
kinda generic style
kinda yea
looks cool
loss keeps going up 
oh I fucked it up
supposed to be 2 repeats
ok 64 32 32 16 batch 2 1024x uses 15gb vram
2s/it
ok this lora is shit, either .0002 te too high or the parameters
holy
and I trained it on the merge of aidxl

fuck
.6 beta just came out
says better than alpha but not as good as .52

better is better
i wonder if it's same tier as .51
bc that was pretty good once merged
all of these w .51 merge
Train LoRAs on SDXL1.0 itself
wanted to try it
AID XL updates faster than I try the models
Imagine if the dude got a cluster of A100 running
where do they get the money for this
China probably has different prices for compute
Linaqruf has Yodayo and Patreon money
And the other trainers on civitai have also too many patreons
SDv1 made them rich enough
Transmog 6 beta
Wish I got rich too so I could help with training ><
Also probably bad idea to train on mixes , better on finetunes or base xl
Linaqruf has patreon?
Maybe not patreon but they have a donation platform
Yea I saw they have kofi
yeah
I’m Honestly wondering what kind of compute the chinese dude has
As for lynn im kinda disappointed he finetuned on Genshin and starrail
They may have put that info somewhere lemme see

You can tell them that on the degen- camen duru server
Hm I’m not there xD Dunno what that server is
Feel free to give me a link if you have tho xD
it's the server of that dude that made the most popular colab notebook
it's really not a good server
and I'm fearful that it may get banned soon enough which could put a warning on my account
Damn
Well I guess the more servers I am at for ai the better
Also I’m not sure what to say to lynn 
You can talk about your concern with training mainly on gacha stuff
Well Uh gacha on it’s own is fine there is a ton of good art there but focussing on mihoyo is definetely not a good idea
They seem to be doing Grandblue Fantasy too for the next version
The only good thing about camen server is that you can check on Lynn's progress
runpod A100 or H100 I think?
Nice
okay scroll up a bit and I think 2xA100
time to look at the dictionary
hah
ultimate comparison time
v51, v52, v51 merge
v6 beta merge, imi merge, v6beta
this might be a bad prompt though
now i'm generating a x graph for the 50 green artists
but it might take a bit 
i'll also upload the merge if anyone wants to try it for fun (nothing special used the stable diffusion built in merger)
~~and bc i have work today and won't be able to do it later
~~
I like .6 beta merge and .52
i think the 'patchwork' word in the prompt prob was a bad idea
wait is it my imagination or do they all look similar 
.
this civitai post showcases a lot of styles available for the v6 https://civitai.com/images/4371606
by bebefate,
by teketo,
by snhiyori,
by otenkiya,
by gyoku,
by goldcan,
by ask,

but at least it looks different
the reason they look the same is cuz its the same dataset
other than when he adds new artists
by ningen mame,
ah 
same seed + exactly same prompt + same dataset
= similarity

who dis
by bebefate,
clean style, by antifreeze3,
animeillust creator giving us too many options 
by glutton,
by buerq,
@wide beacon
energy prices are pretty cheap in Mongolia. Lotta dudes were mining crypto there before China banned it
Well those are actually pretty inefficient for training
Oh that's good I like the eyes especially
Zumizumi
HOLY SHIT WE FKEY NOW
so pretty 
this is w the model
does it even look fkey 
now using pure v6beta
The problem with v6 beta is that a lot of time anatomy is all over the place , at least as far as I tested
yes kinda
been a recurring thing for me in prev versions too but maybe not as much
omg so cute
this one looks prety close
as109 doesnt work at all
for me at least
can u try that and mignon
okay maybe this one is too much
like not rly
but maybe
oh that gen i posted
what artist

i also added ulzzang
idk how much it helps though
jesus
Ah that’s from the bayesian process?
ye

I tried to understand how it works but im not exactly sure how to set it up and Shit xF
But I probably don’t have any need of it
its al lwritten down on the github page if u ever feel like trying again
- some in #1064802771716554792 when I complained about shit not working


might have killed too much of the style
first try might be fail
not always tho
I guess sthats why fkey is red (not trained well)
on the github page of bayesian process?
bit better
yeah all the steps are there
rest you can ask from bing/chatgpt
thats what I did a tleast
I want to try it too
I could but im not exactly sure what i would do with it
how does animeillustdiff handles multiple angles?
in one image?
got an example?
yea looks sick
but hey it looks cool
cute kinda
hmm yea
uhh

i might have to split this into 2 loras
mhm
i love doing weapons
seems like just weapons themselves are nicer than I had in 1.5
do they have meta?
like these
i have some good weapons occasionally
close to perfection on 1.5
but its my watercolour lora carrying
it has good weapons occasionally
would be sick af on xl if we had a model to train on easily
apparently scythe=shovel
lmao
yeah there's not enough quality tags and variety in the dateset
ye
but like if i try on nai i will get poor results i feel like
its too damn intricate
¯_(ツ)_/¯
ok rella and ziyun together are pog
glad I randomly decide to try it
and I'm happy with how the merge looks from today
last one
try by sannso by caaaaarrot
okey
and by sho by dsmile9
good ass combinations
how does it look?
good af
ye I'm just that good
just find the milf artists in the list
yeah i found some already
i dont mind both
just was curious if you found some nice combinations
I believe in duality kek 
i wanna try this artist on the finetune , maybe i can get something good cooked
its heavily stylized and just straight up epic
I prefer medium and flat
based
model can do medium on most artists
just wholesome
noice
by as109 @floral mulch
ehhh
yeah saw since then
it looks okayish here but not really that similar to the sauce and scuffed bg I think
true
I'll upload the merge from today to hf
the guy probably struggles with compute
yeah
yeah defo

but this is just much much longer
same here bro
im afraid that even if i have the compute the whole dataset prep will be too much
What we would need to do is scrape all of danbooru with the right tags
yeah or you could tag them with a script but this would take some time xd
i hope so
last year december was scuffed compared to now
ye ye
if you give him cleaned up datasets for a good artist I'd assume he'd probably train it in the next release
less work for him
maybe I'll do that
oooh
youre damn right
i could give him too
we should make contact kek
hmm that kinda sounded like encountering alien lifeforms lmao
ok I did hires fix latent nearest exact with restart sampler, before the restart it looked perfect, after it looks ruined, what do
change sampler I guess 
restart
wanted to see how much detail I'd get
interesting
alright if anyone wanna try the beta merge
the txt is the wildcards for the new artists
Style merging: The stylization effect of trigger words can merge with each other to produce new styles. For example, [by A | by B | by C]. However, directly connection, such as by A, by B, by A and B, or by A and by B is unable to take effect and only the first triggered style will take effect."

have you guys not been using [ | ] with multiple artists?
I think he's wrong about this even though it's his own model, by x, by z definitely worked for me
Sent a ton of pics like that in the past few hours as well
I tried that but didn't like the results I got at the time so just stuck with comma
I'll try later

Also I tried merging the tokens directly and that did something too
But i didn't test it much
Just fucked around
Like rellagm etc
Also there is iwzry I typod his name and it still kinda gave his artstyle unless I'm completely tripping
yeah it also worked for me separating artists by comma
he might be tripping maybe hes sleep schedule also dog
hahah i will just gonna put the model in the oven
its kinda scuffed cuz i intended to nap and my nap changed to a 6 hour sleep earlier so im not sleepy ;-;
from like 4pm to 10 pm
Unlicky
hard to get rid of dog
bober
@wide beacon

Putting a bunch of artist names does stuff, idk if it looks like a merge of their styles but you can get ok results
Yeah he's wrong but no one is the master of their creations
Or it could also be a translation error
@floral mulch #commands message the glitter
never ask what doing this cost me

lmao

real very xl
yes
true


the 2 loras seem to help
this is w no loras
let me link both of them
i put them at strenght 1
https://civitai.com/models/131706/breakcore-style?modelVersionId=144786
https://civitai.com/models/169170/xl-outline?modelVersionId=190141
如图所示,这个lora的作用是给角色加上轮廓线,可以使用±权重达成不同效果。 As shown in the image, this Lora is to add an outline to the character, and different effects by using ±weig...
by akizero, 1girl alternate_costume closed_mouth flower hair_between_eyes long_hair official_alternate_costume red_eyes red_flower rose simple_background solo suit vampire white_background
just to try as loras separately 
okay gonna try this prompt w diff artists
by goldcan
by neg101
sure let me try 
i did try some yesterday but not on the new merge
i also have the x graph
try this by sho, by dsmile9 and by sannso, by caaaaarrot, and by rella, by ziyun 
should prob turn off the loras
but i'll leave them on for now
funnest part is combining artist

like rella has nice lighting so when u combine it it gives nice results
classic

like that
okay they are all in queue
i accidentally did by sho, by dsmile9 and by sannso, by caaaaarrot, and by rella, by ziyun
by sho, by dsmile9
by sannso, by caaaaarrot
by rella, by ziyun
cool style
I thought the merge went okay, but I didn't test all the new artists and if they worked decently

I swear hires fix makes images worse on these models
sometimes at least
if it's denoise ur used to from 1.5 or other models
Heads up it’s using a LoRA
yepp
the devil one
i think
so then it inspired me to try loras as well 
which lora?
got it
merge the outline one imo
nah dont wanna change the base styles too much
merges already change it
and I havent tried it
gonna try it
@floral mulch Have Blue Archive char lora for SDXL
https://civitai.com/user/kitarz/models
Nice, seems like they have almost all of them
I waited one hour for the rotate shit and it errored out at the also 20 steps
Go agane tmrw 
oh nice

oh it worked
if u hav etime sometime can u put some more pics on the civit page, all of the ones I put are similar 
oh sure 
i forgot to update my civitai in so long 
upload them all individually so they are all showcased

wtf
hmm ?
renaming to what
output
idk i was thinking of naming the file names w the artist from now on
totally did not spam imi's civitai page w like 20 new images
😇

https://civitai.com/models/118406 bruh I just got done
Support☕ https://ko-fi.com/sfa837348 hugging face https://huggingface.co/gsdf/CounterfeitXL-V2.0 Negative Embedding https://civitai.com/models/1184...
dunno but its counterfeit so
oh can u check if these work in the tete model I asked you to try? if u have time sometime
`masterpiece: Beautifully and intricately drawn illustrations.
cinematic style: Stunning illustrations with detailed backgrounds and lighting.
aesthetic style: Illustrations featuring beautiful color schemes.
cute style: Cute.
flat color: Illustrations with strong lines and a minimal color palette.
anime style: Anime.
comic style: Illustrations characterized by bold color choices and abstract representations.`
counterfeit 2+ has style tags like these
surez
base image
oh i should prob remove the artist
1girl :d ahoge blonde_hair blue_dress blue_ribbon brown_eyes collared_dress day dress feet_out_of_frame field flower flower_field frilled_sleeves frills green_ribbon hair_tubes long_hair looking_at_viewer
no artist is already cute
masterpiece: Beautifully and intricately drawn illustrations.
oh
so masterpiece etc
ah

okay that makes a lot more sense lol
masterpiece
cinematic style
aesthetic style
cute style
doesnt seem to have that much of an impact
comic style
dogecry
by limuli ceey, by glutton, by poire
I'll try counterfeit 2.5
omg awesome 
I'm merging 218xl and tete, so reproductionxl87xcounterfeit1.0xaid52 x reproductionv2xcounterfeit2xaid62beta


介绍(Chinese Ver.) 1. 简介 本模型用于 以 SDXL 为底的动漫模型 ,帮助优化所使用模型生成图像的: 基础质量。包括增加 清晰度 、 对比度 和 饱和度 至普遍认可的美学水平,(较强地)缓解严重内容 崩坏 、 扭曲 、 杂乱 ; 人物肢体。包括改善 人体结构 、 面部 和 ...

Merge looks good
ig it's time to start finishing my nurse y graph
cuz i forgot to finish it
and it was missing a lot
just use the same model i was using then for consistency
oh ig someone on civitai posted a full excell file w every single example
Yes, it's neg emb for the upcoming AIDXLv0.6 release. It has less impact on the original style of the model than v0.5.
wildcard list variant - alphabetized
ie use with dynamic prompts {__aid-artists__|__aid-artists__}
did anyone try the new embed yet if its alright
I thought it was

I think the rurudo looking pic in ai art is not gen either
if iti s its a copy pasta gen which is pointless
it doesn't look good w v52
but it works for v6 beta
well i only tried w one prompt
hk?
sky?
its on the same page as sky and jug
aiy yiy or whateverthefuck
that one is pretty strong
9 times weaker neg tho
interesting
soem styles are already overwhelming
いろいろあるので好みのやつを使ってください Of course, since there are various options, please use your preferred one. 強弱や組み合わせもありです It can be strong or weak or a combin...
does it actually work though
oh iots new
in the example showcaws
I prefer the colors of sky
I like less strong colors I guess
but sky was already pretty strong tbh
this seems even stronger

fuck
I shouldve done the other way around
the style got weaker than I wanted in this direction

so many 🥲🥲
yeah
ok confirmed that I killed aidv62
but I Know what to do
old tags kinda an improvement tho

Cute
Model Introduction (英文部分) I Contents In this introduction, you'll learn about: Model information (see Section II ); Instructions for use (see Secti...
FUCK v6 is out
I didnt even finish doing my shit with beta

Hundreds of new painting styles have been added.
I spent a lot of time and money trying new training parameters and training strategies. Here are some developments:
(i) I developed a simple and unique model fusion algorithm to stabilize model composition, and applied it to the production of v0.52 and v0.6 versions;
(ii) I developed a training strategy to solve problems such as data imbalance in multi-category data training, you can find details about it in the English part of the model introduction;
The most based Chinese dude ever
omg hype

if only i didn't have work all day

ig could use stableswarm and gen on phone
If only I was home today

animeillust creator rly be dropping updates too quickly
He did say that he spent a lot of money
It can’t play games
who the fuck plays games
You?
no
Blob?
i own games but never play them
why isthe dpo model 10gb and cant be pruned
Have no idea
cuz its diffusers
Pretty sure the model is only unet
Doesn’t explain anything tho
Just convert to ldm
Also dumbass Huggingface stan
Probably more because you have to buy them off the black market in china
if u want me to try any artist
@heady echo @wide beacon nodders
v6 seems weaker?
but maybe thats why he weakened the embed
tete clears this
Try mixing Tete and 0.6
It’s still a bayesian mbw merge that is more consistent

which one looks like it
3 middle ones ig
idk what the original artist looks like though
by goldcan
German eyes
by shino sto
i put gun in the prompt but no gun though
oh nvm my prompt was messed up
v6 has char tags
char tags ? 
ye
Is tete the pre release vrr
Just an experiment from a few days ago
What was the weight
merge with beta version of last aid
weight 2
Mygod
My combination’s are just that gud
he doesnt know u lil bro
why is everything i'm genning so clean 
like idk
v6 vs 217
i love it though
by tenoimu
by iamuu
by yu2maru
i guess so 
the artists in the red section seem to be more interesting than the ones in the green section



genshin 💀
has become cringe since they never did anything with it
each patch I slowly stopped playing, then they just kept doing the stuff I didn't like so it's dead now
😔
genshit 
best gambling game out there
To test the tags since he trained the characters
Ah yes the game where you either pull a character or 2 star weapons
I'll do one more day of merging after that I'm doing 218 ahn after that I'm trying finetune after that I'm doing loras
Wtf???? Why is there so much shit to do
Not even close to best , I’d say best is arknights
There is no gameplay in Genshin , plus gacha is very scummy
Atknights is the game you play during your preparation for the overnight stay at the zoo
Gacha frogs when someone tells them there are no good gacha games
🫵
(Other than blur archive)
And Al I guess
Coomer lane is also good but I got bored of it after playing pretty long
Arknights has both gameplay and great characters
Al has characters and lewd skins 
Blue archive got good characters , gameplay was boring to me too
Fgo has very bad rates but story , characters are awesome , gameplay is repeating but managable i guess if you can stay for the chars and story
Limbus company is a great niche gacha but they don’t do much to keep it alive ;-; it’s a shame really , their past games are also amazing af , lobotomy corporation and library of ruina
You play shit games
Imagine saying that when you play Gacha
The only thing that matters in gachas us char design and how easy it is to get them
Uve been moaning about reverse for months
So I guess unlucky
Lmao
imi you do know that you call out gacha players but exclude some games just because you like them?
Not all of them are made equal
I can call out gacha players because I tried all of them
Spoiler alert
No you didn't
Most are shit
Yeah I didn't try the ones that look the same as the one that came before eot
What makes a good gacha game:
You don't have to play it
Do you play games as a job?
I think I kinda did that when I was 16
Eventually you play enough where you only care about having fun and won't spend time on stuff if they aren't
No?
Then stop being a bitch
what?!
You literally said this
Not everything has to be an argument for your adhd brain
It's 8 fucking am and I'm I'm bed
Fuck you
what the hell
Let's recap
You say every gacha games are bad except BA and AL
I include them back with my comment
and you tell me that I play shit games
I didn't say at any moment that Reverse was good here btw
I'm just confused by why would you be triggered by that
Kinda weird take also liking al but not arknights
But aight
I do agree tho majority of gacha is shit
cute
great
imi, I greatly enjoy the 218xl model and I'm grateful that you made it.
oh god
i can just fork lynn's demo and then run imi's model in colab?
and i didnt even think about that?
noway

it's gen time ? 
i forgot 
will try to configure tomorrow

good thinking
wish u da bess
real
I have a fairly simple outlook on what makes a good gacha game
do you get fucked when rolling for a character you want? Yes=shit game, no=good game
gachas are all about characters
if you can't even reasonably get them
then there's no point
just because both are published by yostar it doesn't mean both are the same
which is why genshin is not even near the bottom of the list
even though it's boring as fuck
btw in blue archive I have 86k gems and I have rolled for every single char I wanted so far (and got them)
86k gems is enough for 700 rolls
meanwhile in arknights I was
for pretty much the entirety of the game even when I spent like 200 bucks
dual rateup banners are aids
you can't talk about games you have never played, and you're new to gachas anyways
so you can't have a valid opinion

people like us have seen how shit they are and have more ground to stand on other than "lol gacha sucks" from reading it from other people

Read this
semantics are rtarded as an argument
no it's just dumb, I don't think gacha games are inherently shit, they just exist in a shit way most of the time
so gacha games are shit
because its true
if you take 50 gacha games
I'm not gonna find more than 5 that are good
that ratio is
shit

like I eat meat, but if the majroity of options was like beef, I'd say meat sucks even though I liked chicken
holy
v6 struggles hard without a specific prompt
best quality, aesthetic, detailed, by fkey, solo, 1girl, young girl
but the important parts are from aid so be thankful to the chinese guy that madei t
i agree with you
genshin isnt from yostar nani
nah I just brought it up cuz a lot of ppl think it has a bad gacha
but it's far from the worst
true that
also I get that in arknights a lot of ppl don't care about limited characters and there aren't that many of them anyways etc but still, if I wanted a limited char I'd be almost always getting fucked unless I saved for months
at least that was how it was when I still played during the first year
like
just make skins buyable and make money on that 
nah its really generous system if you spend responsibly and dont roll on every single banner , tho having a monthly does help saving
but well if you played since day 1 rn you would have much easier time saving
pretty sure you'd need something dumb like 3+ months of saving up to hit pity on limited banners
unless they changed the system
yeah pity is high but usually you dont need to hit pity
most of the time pity is there so you can get units you missed on
every time i had to pity was for the limited units i was missing cuz i didnt play back then
you need pity if the banner has 2 6 stars
nah
yah
also keep in mind only one is limited
yeah thats the issue
I want the limited one but keep fucking getting the other one
cuz the system sucks
like i said i never had to pity for the limited on banner and usually get both of 6 stars
imagine getting this hoe 4 times in a row when you're rolling for the othe rchar
holy fuck
the soft pity for 6 star is 60 rolls
I'm triggered
this hoe is beyond broken
XD
also did you see her on e2?
but yea either way she is broken af
but doesnt change the fact that I wasn't fucking rolling for her 
that's about it 

This post is 2k messages about gacha

yessir
but yea i think overall the gacha in arknights is generous
and as you get more 5 star dupes you will find yourself saving very easily
as you will be getting more gold certs
I get the feeling that it's great if you don't particularly care about specific chars
nah its not even that , i dont remember what the state of the game was but usually you roll on normal banners when you want a character , the pity also carries over so you can plan ahead and the rest goes for limited banners
tete-218xl is mix between tete and 218xl?
dunno man, during the first year it feels like everyone was broke, or maybe it was just me but I did everything and I still barely even saved up rolls when I didn't roll for shit
like I Got to like 100 rolls from the pity
maybe
and I bought out everything
ye






