#🍥|anime
1 messages · Page 135 of 1
lmao
still coming from hell
@wind girder can it even do like cosplaying as a computer or random stuff like that ?
car cosplay 🔥
does anybody want a 480 image+tags dataset of this year's toplist on wallhaven?
They got a pretty cute style
(Of course you didn't, why would there be cosplays like that
)
Was gonna train a LoRA of their style once
😮 @wind girder 🤗
bruh
wwdym cute style
So that LoRA gonna come out cooked af
bruh
BROOOOOO Yes !
so cursed lol
@wispy canopy 
Cute artstyle? Good shading? Good linework? :o
doing not much, not going that well, but working on it.
and dog sitting 4 dogs for the last month
get wwell soon ❤️
but the context is important
Good work
I mean, not really for a LoRA
what are they drawing with that style

If they had enough variety it wouldn't matter for a LoRA
as an apple
have some testing on my new model merge!
is it gonna be an nsfw lora
And it might still work, but it would look like ass
@wispy canopy have thighs
or like ddid you put nsfw images in dataset
I mean, no, my Chelodoy LoRA is like 95% NSFW, but it didn't come out NSFW
It just depends on the variety in the dataset
And luck
XD
u on sdxl or 1.5?
trained from 1.5. this was one of the first models I trained, like 14 months ago
14 months ago 😮 that's an ancient model by today's standards
i found sd1.4 hidden on my drive 😮
it's what I call a "party model". it's not for quality, but for fun :p
it should have fun results?
I need to clean my drive one day
agreed...
right noww in the testing phase of my 1.2 mix, hoping it's actually going to be finally better than the 1.1 mix
did SO many fucking recipe's hoping they'd be an improvement to visual clarity, and reducing jaggies
and each time, at some point during testing i'd end up with a failure, examine where it failed, tried some more prompts that go in a similar direction and be like 
so now i'm again holding my breath
just use loras
😄 darkalfa 😄
oh wait.
that's just self-promotion
it's a decent anime model nonetheless
you just can't use booru prroperly
oh
(Dunno what this is just searched anime on gifs) 
there are models for that
screencap
something
I think the style was called
because it's like screenshots from anime
@brittle jay "AAM - AnyLoRA Anime Mix - Anime Screencap Style Model" ? 
idk
only found this and another one for like 90s stuff
one dataset I never managed to train correctly, "button eyes"
Some would say "thank god", I'm still sad 😢
becomes a LOT easier with regional prompting
YOU are the regional prompting ! racist !
sorry
outburst of dumbness
if alcohol is involved, all is good 😛
I need to get up to date on lots of things
I haven't tried SD XL since it released yet
if none is involved, still is good 🙂
some was involved but some hours ago
i lke that style
makes me wanna play stalker
looks so like Grave of the Fireflies

damn, my kokoro, that anime still hurts...
looks like a really old model 😮

booba
but
hand
@wispy canopy found a fun little thing to do, put embeddings in and_salt
lmao
i have 0 😄
I downloaded some for specific colors pretty fun
(fishnet, thick thighs:1.5) 
(Goth waifu, fishnets, thick thighs:15) 🤤
are you reminiscing the good old days of "sd ancient edition"?
somthing like this?
I haven't downloaded models since I came back. all I have is old
well, time to plunder civitai!
Download the latest 
04/2023 is my most recent model
time to game a little and go to bed I think x)
well, game... I'm jumping from one idle game to the next
but there's also a new Kairosoft game so... might go deep into that xD
relaxing
I need to get back the tiny bit of skill I had at sd before


this desk seems comfy
holy fire angel
wide range of noise
that's not admissible
just testing my model
I really want to post something but it might be way too sussy
i'll put that as a pass on thicc goth mommy

old me being boobs police would say, if you feel it's sus, then it is certainly sus.
/spoiler is a good friend in case of doubt too
I'm not going to boob police though
put that as one of the pros of the model on civitai
lmao
has Miku science gone further by the way ? are we able to have virtual Miku real time already ?
I need to get my LLMs going 👀
It's an old gen
Just let me know if it's too sussy
and i'll delete it
yeah, like I said, not gonna do the boob police.
but yeah, now that I have it in my PMs for future personal analysis, I can say, lots too sus 😉
Most of my models tend to make them really big for some reason
some reason
haha
i mean, it doesn't beat my biggest generated boobs
"some reason" being the training maybe ?
xD
I meant the hat, you horny bastards 
her leg tho

thinking of biased training, I had some crazy ass requests when I was training lots of models though... Some people PM you with personal project and you ask yourself how they think it could be OK x)
I once had a guy ask me to img2img images of his gf
bruh
like fucking PM'd me
i did do it tho
kind of the same but a full model tuned for revenge nudes was the craziest I got
holy shit
had some of those too, kindly rejected all of them
wtf
some people really are wired weird
the people contact and say that almost first thing, almost not sugarcoating it
oh and also a girl who wanted images of her and her gf (the closest I could)
man
you making me rember
model really focusing on the good stuff 
I wonder why

(yeah man....)
On this sun-filled day in September, Jacob invited a group of fans to meet him at the Heritage & Arts Centre in Bow, London, for a special, Audience-Choir-inspired rendition of his new single, Little Blue. Thanks to all who came, in their bluest attire, learned the song that day and sang so beautifully – and thank you to Mahogany Blog for captur...
prompt is lyrics to this
time to get the rick roll model out
hehe
thighs and close up on prompt 
I used his video clip as regularisation lots of times
man I made a gatto model
you remember
?
yeah ^^
I like doing requests but its so slow to do anything 
I'll try it 
send proof after (in dms if they are too powerful)
but still good
noice
makes me wanna prompt my old [redacted] prompts and see what i get
really you should!
singing [redacted] underwater
makes you appreciate how hard the tech evolved in such a shor ttime
man, the sheer amount of {{}} and [] is making me mad already
i live by those 😮
[token1|token2]
{token1|token2}
BREAK
old prompts revisted
[__mediums__ by __artists__ ,, {1-2$$abstract|surrealism|absurdism|realistic|photo} (2::__elements__|__nouns__) concept art, __movements__, , [{male|man|masculine|girl|woman|feminine}::0.3]|__songmaster__] -> [a screenprint by Eva Švankmajerová ,, surrealism (2::Dark|evidence) concept art, rayonism, , [girl::0.3]|It only takes one lone soul It only takes one in a thousand The absence of fear in your eyes No, that's not bravery As children we learn it's wrong To put out the light of another Our innocence lost over time Means to an end]
wtf
lol
I like both tbh
they both have their charm
I remember having a bad time when prompting slime girl on [redacted]
I'll give it a try

... don't ever use my cursed slime lyrics 😛


I mean a really bad time
hello again

slime girl 👍
I wish it was controlnet
please send it to me
that's putting a face on the expression "dick head"
do you have dynamic prompts? the og prompt has a 20% chance of not infecting your eyes
no :(
My favorite [redacted] result by far
this one got little arms for a chin
well, it's way less cursed now
but not quite there
need lora
would
would²
would to the power of 10
slime girls is a slipery subject, literally, and we love it for it
ambatulick
I'll prompt this
man I just realized I have used sd for over a year
Dankest memes and vaporwave
Procastinate at five am
Forever, that shit is my jam!
BREAK 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!
BREAK This is for my slime boys out there
Furries and anomalies
The hyperactive scatterbrains
Whit milkshake running though your veins
[BREAK anime cell shading:0.2]``` @unreal ridge
yeah, that goes fast !
I joined (first account before hack) around september, 2 years ago now... I realised yesterday
this isn't the cursed lyrics, i suppose
still cursed though
still cursed
let's go everyone, what's your cursiest generation ever ? worse cursed image that still makes it for the NSFW rule ?
Those measurements are realistic.

OG dynamic prompt lyrics have a chance of not being cursed
not much of a difference lol
I got something weird
yes. that's what the cursed lyrics do
that's why they're the cursed lyrics
no matter the model, this prompt usually makes it spazz out in weirdness
damn
yes. slime is good for you
I thought she was snorting at first
LOL
slime = good
uhh.... slime...globe waifu?
weird
I wish I could read AI language 😭
very VERY overtuned model 😮
uhhh....
my slime is trying to do science...it'snot going well
could a slime take a sunbath on a boat ?
I'll just throw existentials questions, we never know what science can bring
Looks like she was attacked by a slime 
I would take her
okay next one too 
can a slime-teacher teach to human teenagers without the slime token going over to the full classroom ?

yeah, with regional prompting, easy -> but that's a whole differrent... uhh... slime to tackle
hmmmmm tackling slime can be slipery
we never know anything
to be really existential about it
or rather epistemological
I'll just throw existentials questions, we never know what science can bring```
wow that's cute
clown
the foot man xD
nice one

damn, you managed to get it working even witht the second part of my comment ?
nah, just responding my bad x=)
Gboard チームからの新しいご提案、Gboard 帽バージョンをご紹介します。
Gboard 帽バージョンは、頭に乗せて運べるキーボードです。
昨年の Gboard 棒バージョンと名前が『かぶって』しまいましたが、帽子なだけに良いかと。
詳しくはこちらをご確認ください。
Gboard 帽バージョン:https://g.co/CAPS
Google Japan ブログ:https://japan.googleblog.com/2023/10/caps.html
#Google #Gboard
i think they are onto something
this already aclassic
@wind girder want regional prompting horror?
the area where it's applied is contained tho
not safe for eyes
The leaning tower of Pizza.
send to me so I can analyze it 

i might have deleted it tho

https://youtube.com/shorts/dwwBvDE37L8?si=OLc5Q5ebo1qEXbqy
A friendly hello to my other ai enthusiasts here. I am
currently making a short film with stable diffusion, I
nave already made a small teaser, in which you get
glimpse of what it's about and already get to see
some of the protagonists
Let me know what you think about it.
Voiced by the ai voice of Morgan Freeman

meow 
- the animation is not the best that is currently possible, you should investigate what other people use to make animations. there are way too many techniques (here are some of them) and I am not the best at animation myself so I don't have a particular one to recommend. if you insist on using this technique (seems to be just doing img2img on each frame), at least use Ebsynth or something.
- the voice generation is not the best. I would recommend using tortoise/vits + RVC (make text2speech using tortoise and then take the output and pass it through RVC) for better results. (this is a good video for this)
so many edits 
ye man
cuz the formatting for links is hard for me to get used to
@wind girder what do I need to do to get the devs role?
https://youtube.com/shorts/dwwBvDE37L8?si=OLc5Q5ebo1qEXbqy
A friendly hello to my other anime enthusiasts here. I am currently making an anime film with stable diffusion. I Have already made a small teaser, in which you get
glimpse of what it's about and already get to see
some of the protagonists
Let me know what you think about it.
Voiced by the ai voice of Morgan Freeman
man we saw it already
no need to spam

let us pray instead
Sorry, wasn't sure
Can anybody tell me what the heck is going on here? It's maddening, and only seems to be getting worse with my system.
you left the COW unattended so it wandered off. it's all your fault 
always happening at face region, could it be your adetailer?
or vae?
I don't have any adetailer on for that. That's just raw Aurora, with the standard anime2 vae.
"restore faces"
Oh, but with negative_hand on.
Lol. Yeah, sure.
Something's up with the gen process. The glitches are far more frequent with more complicated prompting, but it still happens on the simpler prompts.
something to do with restore faces in settings
definitely looks like adetailer with too few steps in the inpaint pass. do you have this extension?
I do, and it's off.
I have nothing on associated with that.
just for testing purposes, try changing vae?
Yeah, I did that. Still happened with Nubby's version of anime2
check in settings- face restoration
just checked png info. face restoration isn't on so that's not the problem. maybe the CFG Rescale thing is messing with it. try updating extensions and if that one is already up to date, try disabling the extension from the extensions menu.
@untold glacier
tried opening sd without any extension
just pure sd
all default idk I think theres an option to do that
Alright, I'll try y'all's suggestions. It'll take a moment. I'm not entirely sure what the CFG Rescale you're referring to is, Nubby. But I'll try and track it down.
Thanks for the help btw. Y'all are awesome, as usual.
I'm not sure what exactly it is either. I just saw it in the PNG info and it was the only thing there that wasn't common.
Hmm
Wow. Okay, this is with no extensions.
Lemme try rebooting my PC.
...but before that, lemme try a fresh start and manually putting in the settings, in case it's taking something from the previous gen and changing a setting that I can't see.
negative BREAK ? 
@untold glacier try without any negatives just a girl prompt or something
better to just turn on Pad prompt/negative prompt to be same length instead but there's nothing inherently wrong with using BREAK in negatives.
After full restart... This is my life now. 😭
Such crisp. So clean. I love it
did you try without anything just a simple prompt ?
Yeah, that's with no extensions. The only thing I'm using is negative_hand embed, but considering others have had no problems with it I haven't factored it into my testing. I can try and take it off and see if further gens don't have the same issue... but I doubt it'll change things.
thats very weird
try with unipc or euler a sampler
You know, I didn't think to change the sampler. I believe I've tried that before in the past when I've had the same issue, but I can't remember. I'll try that now.
.
By the way, can anyone who has Aurora try this gen with the same settings? I'd love to see if anyone else has the same issue:
im purposely trying to get your effect
lmao
cfg 80 🤠
your webui have that function bulti-in 
i can only set it to cfg30, and it's already at that
u can increase the limit by editing the ui json file

Lol
Do you have Aurora? You can just copy my prompt/settings from #🍥|anime message
Then I'd be able to tell if it's just me or not
changing model
Why go through such pain

Do you not have space on anything else
So much goodness. So little space
passport huh
Yeah, might be.
maybe cos someone was trying to gen as well 


something is wrong
after hires fix
It's just a new fad. Them kids and their chaotic face tattoos.
Wow. So it's not just me after all.
Yeah, hires fix helps with the denoise and adetailer after that as well, but often it's just too much and their char ends up with a strange mark on the face or something.
i did
I've often had it happen with other models, like dreamshaper or others, but 🤷♂️
i have no idea why

i can mimic all your pic
with same problem
those with lora, i can't tho
since i don't have the lora
i thought it might be just some broken seed, but tested same seed with different prompt, and it works
It seems I am adept at finding obscure problems with SD.
cow girl action move lmao
That's likely a part of it. With such problems as these, it's probably a combination of things.
a cyber punk prompt?
then it might be the sampler, not the model

too many implants?
Stablesychosis.
I'm doing a batch of 16 with different samplers now to see what difference it makes.
she's in witness protection 
i wanted to censor the eyes only
qr code cosplay
There's a new adetailer model??
lol
ya i wish theyd stop making the cute hand pose
heres another hand pose
You're reminding me of the time I genned eating a candy bar
Would cooperate with her
👍
Squish
(squish:0.5)
Those eyesssss
That's the right place. Not sure where it stores the included ones.
yup, i realised all the models that came with it, is inside venv by default
so, maybe external stuff is supposed to be in that folder then

Ahhhh my heart
nubby, is your models inside the folder?? 
Yep, the note I posted about additional models is from the ADetailer repo
nope mine are hiding somewhere else. I didn't look too hard for them, but they aren't in there or in the extension's directory.
With a model loading time like yours, I'd be the same way
then should be in venv like mine
i might actually put some commonly used models to SSD, so that i can change with my discord bot

That sounds like a great idea
ah, makes sense I think i do remember seeing them get installed by pip when I installed the extension.
I'm too used to SSD load times for everything
people read that? 
Even my regular HD is a hybrid
how fast is it for most models changing?
wanna see if it's worth the swap
Maybe 10-20 seconds tops.
jesus

Yoink
its guuuud
i love how Sleepti always comes and drop cute generation 

@lyric zealot Do you not like the cat lick gif its so cute 
Very proud of it lol
https://music.youtube.com/watch?v=s1YZEjod5ZQ&list=OLAK5uy_nIEWr3yuL2Xf-lr0imKNIq87--W_FGemE damn, this slapts like burning magma
Casey and Adrienne’s voices compliment each other so well! What a great duo.
So that everybody knows, these are the official lyrics provided to me by Jeff Williams himself, so please don’t try to correct me on any of them.
Adrienne Cowan = Yellow Text = Yang
Casey Lee Williams = Black Text = Blake
Both = White Text
Thank you so much to my fri...
a master piece
smol thighs 
Alrighty. Performed some tests on Aurora, BreakdownAnime, and SakushimiV2 with a few different samplers: Euler A, UniPC, DPM++ 2M Karras, 2M SDE, 3M SDE, Restart.
It looks like there is indeed an issue using the 2M/3M SDE samplers with these models. A lot of problems with the face generation for some reason. Take a look if you want, and do note that the comparison images are quite large:
SamplerTest-Aurora:
SamplerTest-BreakdownAnime
SamplerTest-SakushimiV2
I still don't like Restart based on these tests, lol.
I know many of you enjoy it, but I just don't see the appeal.
it's great when there's hard to generate concept that you might want it to "restart" to get a proper generation
since it will first generate a "layout" then using that as "base" to "infer" better
from what I looked in comparisons is basically that
for restart, do you mean BREAK ?
it fixes small stuff well
Nah, the sampler Restart.
oh, still not updated to it
Yep, since 1.5 I think.
1.6 if im not wrong
I could be wrong.
wait that might be refiner
It was somewhere around there.
What if we're both wrong?

Powerful frame in there.
rvc didn't wwork on my compuiter
for some unknown reason
so I finally managed to make it run on docker
but it's fucking 5:25am
I have school in like 2 days
or rather in a dayy
so anyways, here is ariana grande saying goodnight to y'all
LOL
cute
we live in a society
oh yeah so GN
😭 this gotta be the most useful emoji out there rn
Societhighs
i wanna be inside that societhigh
👍
latent without enough denoising? but it looks cool af
Meow human, I am cat
this cat looks like he's probably smarter than me
yea it gives that glitch effect
mfw i live in a society
I recommend beaches
Will calm you down
@lyric zealot 

Lewd

I'm just a cat
U changed models?
aurora with a little expmixlinev3 as refiner 
Is refiner a XL thing?
well, people usually think of it that way. But in auto1111, the refiner setting basically let you choose a certain model to switch to in the middle of generation, depending on the ratio
Wait what? I've tried to do that before and it told me it wasn't a refiner model.
so example, if ratio is at 0.5, then it will generate half the allocated steps with MODEL1 then switch to refiner model
time to update buddy 
maybe
not sure

same with me
time to learn something new I guess
I'm latest in main, but I may not have tried it since the upgrade to 1.6. 
remember to set your model the max number of model loaded
this your fav hand gesture?
no just a lora 
what the fuck do you even call it tho
found it
I would like to learn more about that in private with her


for sure for sure 

Now that I've sorted the sampler issue and have done some comparison on what is actually best to use, I feel like good gens come so much easier now.
That and using the refiner for more checkpoints is great. Now I'm using at least three different checkpoints per image 😅
Starting checkpoint > Refiner > Hires checkpoint > Refiner > Adetailer checkpoint.
It would take so long to set this up in comfyui.
👀
They're gonna play monopoly
he bought boardwalk and park place, and she just landed there and is out of cash
Very true
When do we invite the Rinner, for dinner?
evenin'
how long did those gens take? 😄
IKR? Her blouse could fall right off!
oh geez
each one takes maybe a minute, minute and a half?
I often only upscale to 1920x800, and then take those to tiled diffusion up to 3840x1600. But went a little higher for these initially.
these aren't upscales... they're hi-res'd at that size native.
try with this (5120 x 2048)
I have a weird ultrawide. (3840x1600)
tiled vae surely?
no
or you have a god GPU?
ah I have 10gb vram
I can't quite generate 3840x1600 so I'll use tiled to get there.
but you're still producing those quite fast
this crap drives me insane.
Same
oh you just don't like it?
yeah it's annoying
so I'm doing a 512x512 tiled diff 3x with tiled vae, it's still at 0% for like a minute now
starting up
When they work well, they work well.
which upscaler?
latent here
it's like a little dance haha
something simple like ultra4x or lollypop
That was the prompt. 😄
nice!
So you know what would be handy? If I had a program that would look at my huge list of prompts and find ones that are like 70% similar.
A tensor with all NaNs was produced in VAE.
dang
where are they stored?
do you maintain a file of them?
If only I could get fantastic hands from said prompt
the more ppl u get the more deformed hands gets
It's very easy to run a scan of all my saved images and generate a list of every prompt.
There's something strangely creepy about this image.
but many are very similar... just differ by maybe a few words or so.
would be nice to get a nice trimmed list
chatgpt can prob make you a script
nothing creepy is just my sisters forest cult
eh maybe
so a 512x512 x2 highres fix is 16sec, a 512x512 x3 highresfix is 1 min. 7.8 sec.
I'll try your res now
i only made a few of those 5k res images
Lol
are you just randoming the highres pics then? because latent needs such high denoising
I'm finding these results to not be as good as 2x results. Seems like a waste, so probably won't run large 3x batches like this at this ratio again.
I've got some dalle credits tonight so gonna have to try some stuff. 🙂
960x400 at 2x is 1 min. 46.8 sec. for me
maybe there's something wrong with my GPU???
once i get the perfect prompt and denoising strenght i do a 100 img batch
What in the name of generation

Why is my GPU so bad
my recent approach had been finding a good base gen, then upscaling it
going to try some raw luck latent highresfixs
same man
I'll try some controlled timing tests after this generation run
latent really brings out the details
btw, did yall see my new program?
I have a new image viewer that's like a companion to my other one. A simple thumbnail/prompt browser for quick viewing.
available here: https://github.com/GarlicCookie/SD-Quick-View
If only hands... Also, RIP to the girl on the right.
looks like a bunch of luggage 🤣
👍
idk whats happenin in background
I would like to take place of whatever is in that background


if you get it, though, hang on, cause im uploading a new release as we speak 🙂
help her with the lotion
captain tessa reportin for duty

oh this brings back memories lol
honestly forgot the storyline entirely

Who is this one?
what i remember its about a rogue mercenary unit that tries to stop wars,also the mc is an orphan that meets a girl that has some sort of power for their mechas
i dont even know if they adapted the whole mange prob not 😔
Idk, just made them up
Also anyone got tips for better hands, I took a massive break from SD and can't seem to get the results I used to
i dont understand what this book says can u help me?
I'd like to check the book from behind just to make sure it's a book

I am suddenly reminded of a certain scene in Amadeus.
Where can I see that? 👀
that was something my discord bot did 
not from this server tho

remember to pray 🙏
I'd be worried too if I had this strangeness.
lemme fix her with 🫳
... can I not save a style from the Auto1111 interface anymore?
they jacked with this and made it unnecessarily complicated...
It's insane at the moment, yeah. I for one have no idea how to save anything.
press the pencil button, in here, type the name of the style, then a save button should appear
then type your prompt and neg
ah, didn't realize a magic save button would appear
yup
from what i heard, they already revert back to the old style, in the dev branch
good
put me in coach
hey at least it's not satanic... (holy hell that was stupid as hell)
that home depot sign is perfect
dalle
dont have many credits left though so gotta use 'em wisely
or tokens or boosts or whatever they are
Isn't that 浮世絵 (ukiyoe) or something?
it can do quite a few things
cyber bunny
Welcome! Start by heading over to #1072220168534642768 to get yourself situated and help find the channels you are looking for! Please make sure you review our #✍🏼|rules-and-tos and feel free to assign yourself some #👥|roles as well! Answer any questions your may have at our #1072229020520947753. There are many ways of accessing Stable Diffusion, take a look at #1080946152318443610 to start your journey!
yes
dalle3?
stable diffuion
oh
How i generate images.???
i gotchu
Omnia can u help.?
hot girl right?
yeah she looks hot
Fuck you what is this
Forgot the fire
that grandmother has fire coming out of a guitar!
the beach is pretty hot ngl
HOT LADY IN YOUR AREA
i think he left











