#🍥|anime
1 messages · Page 51 of 1
webui-user.bat
i prompted 'the gap' too... think the AI understood? 😄
I think you need to use thigh gap
and, i didn't even prompt for mine 
i don't want thigh gap. i want 'the gap' which i actually got 😄
wait what gap are we talking about? 
thigh gap -> full gap between legs
ahhh
'the gap' -> only the top part gap
that small little gap
'THE GAP'
holy ganimede, i got a 1600x1600 on my old gc
lmao, is that the official name for that
so cool
the 'official name', i suppose? XD
it even tried here
@lyric zealot yeah, my model knows what 'the gap' means
try adding absolute territory
uahahah
i think i'll have to change image dimensions for that tho 🙂
zettai ryouiki
i kinda like her yellow eyes with pink hair, it's kinda like my OC

can anyone help please
best place to ask for tech help will be here > #🤝|tech-support
probably need to include thighhigh

this image needs a microphone lol

holy moly
sussy baka
Saint Seiyaaaaa!!!
haha not intentional but it does look similar
too lazy to fix hand 

no issue. bad hands is the SD way of life ❤️
that one is cool, but why muchrooms? lol
yee, bad hands 4life
sd is smoking or what? XD
she loves mushroom 
ohhhhh... that explains a lot
she's high - up in the sky
i have no idea what i'm prompting tbh
sd. way. of. life.
I'm looking at your shirt, i promise!
god, why i can't see her in her eyes?
added thighhighs, thicc, and still. THE GAP

my eyes falls every time elsewere :/
Legs too big in comparison to her body

Shroom trips be like 
that's good legs
Still using expline?
yall wanna solve puzzle together?
Pervert
Yeee i oughtta try it now im on rocm
Im able to hires x4
To 2048
Takes 20mins but it works lol
let me know if you need the model file, pretty sure it's taken down from civitai or something
Hey guys, my video production company and I are experimenting with AI tools, and recently made an entire music video with what I call an "AI assisted workflow".
Basically we wanted to tie Stable Diffusion and some other AI tools into a conventional video production pipeline. So we made a music video for our own band which is loosely based on the video of the guys from Corridor Digital, but we had to come up with a lot of new practices to efficiently integrate the ai tools in their current form into our workflow. it was a lot of work, but definitely less than hand-drawing the frames by hand.
After all, I think it turned out a great experience and a nice little end product. Here's a trailer of it, let me know if you'd like to see the entire video as well.
Is it possible to create anime using only AI?
Yee it has been but i has it on that rentry i sent you
Awhile.ago
ahhhh, then you're all good 
this is looking mighty impressive
Yis
the bigger they are, heaviest the weight they can sustain!!!
Working on my oc now atm
guilty. I'm prompting on my hentai model.
Thanks 🙌
There's 2 types of people in this world: those who admit being perverts, and the rest.
wow, that's big, well done
Character movement is a little off,heads are too big otherwise looks pretty good
thickness is the true way
I rly wish people wouldnt classify sensual as perverted. Allot of classical art I like to recreate is tasteful but cant share
it's not what we feel the true sin, but how we express it!
If it makes you feel, i consider it art
This is supposed to be an anime channel but everyone's talking about art interesting
anime art
we talk about everything here
This is technically the general channel since the actual general channel is dead
but mostly share anime art
Always miku
@ocean reef did you see my blacksmiths and later on some warriors? that was some mighty thirsty husbandos earlier this day 😄
?
dayum no! Lemme check it!
i'm working on doing tales of symphonia art for example and than extremely experimental lain next
no, it;'s not
thank you ❤️

we try our best 😄
you should go to 4chan.... TAHT's perverted
the true perversion is not accepting our feelings!
especially the degenerate section...
at least I can upload male nipples 
2chan? my god, what i saw on 2chan!
this is probably the only AI art server that's not filled with nsfw stuff
4chan is an open platform where you can post whatever you want makes sense
Wrong reply oops
Looks like Brad Pitt on 2nd one xD
was testing my LLM prototype, very pleased with those results
Anyone watching death mount death play?
i think i only asked for "blacksmith" and my llm did the rest
no, but looks like i should
not watching any anime of this season
yeah, i'm watching it. that one seems SO fucking good
Not even hell's paradise?
The opening hits hard
pretty cool
nope 😄
it's so unexpected too... reverse isekai. LETS GO
Tf
It's hard to find interesting anime, everything is just repeating patterns
that's a body that can sustain th world
the times of paranoia agent ended years ago
OH. SHIT. feeding my prompt into LLM and then feed that back into SD... this is... illegal
i wanted a train 
she is so lazy, just an egg roll and what's that in the sachet?
sorry, clyde will block this probably, if not. not sorry.
she's a kid, she don't need to eat much 
she need more meat, she need to get tick!

i don't think my girl can get more meat...
she should became like her!
Don't make nsfw jokes about kid, i think the mods here won't like it
I don't think you are making one, but just letting you know in advance before things happen 
my god, these forms! there are enough proteins in this world to sustain her XD
wtf is she doing in my kitchen! making a mess 
man come on, let's not sour the mod lol
it's sef explanatory i'm talking about growing up
no no no, im just trying to let you know before things get too far
in good health
k ty
don't want you to get banned 
yeah, that would really sour the modXD
so warm
she's burning herself 
omg STOP she's so cute what 
no words needed

the last image to miss tick
boys being boys 😏
wait. everyone is a man on the internet...
wait, no!
@ocean reef, you are a man in my imagination... don't shatter it!

I'm gay then!
❤️
guess it's time to make some 
at least your thirst for husbandos is still good!
duuuh
yeah, lemme see what my LLM can whip up
gimme those muscles!
best cook evah
there's a certain charm to guys having long hair
now that's just vulgar... who would hold titties?!
smashing the eggies?
shhhh dont say that... the prepubescent boys can't handle that information!
balding long haired guy reporting in. nothing sexy about it.
anybody here knows a bit of philiosophy (metaphysics and epistemology and consciousness specifically)?
how is that balding 
this looks like a guy that will scam all my money away
i'm balding. i'm the real life experience of a long haired guy. have been long-haired longer than short haired
agree 😄
yeah I know I already put my thingy there
just wanna make people here come over there
and again, we have Nubby here just coming in, dropping knowledge and leaves 
Don't need to know where you're putting your thingy. Thanks anyways though. 
right? 
bru
I am not good with words am I
guys are bad hands embeddings working for you? cause mine most of the time dont work...
I'm sry but I can't accept this as waifu
just use like 10 of em
and like (:1.5)
or something
but tbh I just upscale and then inpaint only masked
They won't fix all hands, but both bad-hands5 and negative_hand improve my success rate. Good prompting and a good model are still very important though.
well, what about him?
@ocean reef
nope
surely this?
dayum, way too ripped!
looks like eface fr
does it? 
no
if i could look like that, i would.... too little hair remains on my head 
@ocean reef these better?
I remembered first time he sent ai generated image of himself (with trained model or soemthing)
I said he looked pedo
a bit 🤔
I'm already over making guys 
controlnet canny
I need more tact fr
be glad you said it to me: you'd need a nuke to break my subtelty filter and have me like "you're rude"
how can i please the husbando gods then?!
add groin
It's a rly hard question
this is not a nsfw channel, i shall not add bulges...
hehehe

reject husbando, there is only gobbo girl now.
@ocean reef this better for the imagination?
lemme render it myself xD
but I need to change a model
OH. MY. F. GOD. THIS... ... LOL
A muscular man stands confidently in a modern kitchen, wearing only a pair of tight black briefs. He grins mischievously as he flips a pancake with ease in a sizzling hot skillet. His bulging biceps ripple under his golden skin as he stirs the food with a spatula, steam wafting upwards in plumes. The countertop beside him is cluttered with various ingredients, but there’s also room for some playful items- like a bottle of hot sauce shaped like a pepperoncinni. Behind him, a framed photo of a beautiful woman hangs on the wall, perhaps his beloved. She looks happy and adoring as she watches him cook from across the room.
prompt at your own discretion
Wanted a man, this model gave me that!
might be 🤔
@supple raptor you've been gobboposting a lot lately
Thanks for cnet
Yesh. There's a reason for it.
?
i might get banned for this, or clyde intervenes. Can't post the rest, i don't think sexy speedos are allowed...
why would you get banned?
this amazing
project I'm working on.
project? you maknig me excited now
👆
How do I prompt just straight hair without any buns or accessories?
also this one looks amazing
you have a gpu?
yes
try bangs, long straight hair
I'm using this for my character (open hair, long hair, straight hair, white hair, hime cut, blunt bangs, azure eyes:1.3)
might need to consult sensei nubby
aurora with just prompting and Horrid's twitch emote LoRA.
aurora is so good man
actually better than kenshi
and better than my model which is better than kenshi
It's the only model I've been able to get a fairly consistent gobbo girl from.
straight hair should do it pretty easily. Could be a lora or something in the model causing the double buns and hair accessory.
Just using Aurora atm
This one is still my favorite one of Kiki. pinches incoming
why has lottalewds been quiet since second of may?
hmm.... maybe part of your chibifying prompt is doing it? Maybe try removing all of those parts to see if it works right then start adding them back to try to figure out what it is.
i mean, everyone got their own life outside of this server, so i guess life happened?
yeah
What is life?

"Hello darkness my old friend"
what do you mean there's life outside of waifus?
I just noticed, what is the giveaways role and how do you have it
actually, you are right! Imma take back my sentence. 
wait you all have it
umm it was in annoucments? I think?
charming
lmao is it gone

maybe it was only for like stable society?
nah, i can't find the roles channel as well
does this link to no access for you as well? #1072240143521554592 message
wait. why does my model do well on that xD
dont ask me! xD
I asked her to hide it so that you couldn't get the role
wtf is going on?
forgot my model is hentai - > "a figure" becomes woman. obviously
@finite herald how get giveaways role? am I too late? in this message #1072240143521554592 message it links to No Access
omg fruit is online?
#👥|roles is locked for now. It'll reopen back up soon.
everything. everything is going on. at the same time.
I see thanks
does anyone here make loras?
💖
yes me
no. i prompt.
you're kinda mean but i'll tolerate you this time ❤️
lmao
I am mean?? you are mean
calling everyone bitches
he mean you mean something to him

lol
A man stands confidently, hands casually stuffed into pockets as they lean against a wall. Their eyes crinkle ever-so slightly, hinting at the smirk that lingers just beneath the surface. Dressed in dark clothing, they emanate an air of mystery and intrigue - a person who isn't afraid to stand apart from the crowd but also knows how to have fun. With a subtle smirk, they convey intelligence and wit without saying a word.
she is my ideal 🥰
ngl, my model does GREAT on natural language
changed model, kinda better now 😄
still looks evil 😄
evil, but.. damn.... soooo handsome
What is this No Access your are talking about?
wow, this remind me so much about this https://www.youtube.com/watch?v=XfTWgMgknpY
🎵 LISTEN BY STREAMING
https://linkco.re/FTUMav06
For the upcoming game "Limbus Company" https://twitter.com/liberarelimbus
Follow Mili
Spotify: https://open.spotify.com/artist/0K05TDnN7xPwIHDOwD2YYs
Apple Music: https://itunes.apple.com/jp/artist/mili/906934266
Twitter: https://twitter.com/ProjectMili
Facebook: https://www.facebook.com/Project...
i KNOW i heard this song before. but where?
i think i heard it without myth&roid.. is that possible?
or is it because mili occisionally enters my playlists?
dayum! Make them younger at least! xD
after 50!
LOL that's after 50?!
all i prompted was
just a regular dude, working hard, natural body, sweaty, working the field
oh it does have the same energy.
llm takes care of the rest: `
An illustration of a man working on a farm under the hot sun. He is depicted as a hardworking individual, with dirt smudges on his shirt and a sweat-stained hat covering his forehead. His muscular physique is evident as he wields a large hoe, digging into the soil with determination. The background shows a sprawling farmland with fields of crops swaying gently in the breeze. There is a sense of ruggedness and simplicity about the scene, emphasizing the laborious nature of agricultural work.'
`
chibi version
best one yet?!
Welp
Yeah I remembered that I stole one of Nubby's emotes and stole quality tags from it
then my stomach stopped working 🤣
wait. how does that work?
Jesus, at least tell me that you drink with really low sugar content

I drink the sweetest ones! Always!
europe.exe reporting in: stomach.exe SHOULD be empty at this moment
Lmao
sth like that 😄
I like that!
Also I had this one which is also decent
boba... is that an actual drink?!
Boba = bubble tea
this one is applicable whenever internet-drama unfolds
Or 
i'll settle for the bubble tea context, when consumption is involved
high sugar content other context bobas are... not good. (physically)
I mean
all boobas are good!
boba consumption = good regardless
Not literally tho
literally
high sugar boba consumption = only good in 1 case...
high sugar in the other case is indicating things that are not so good (in general)
but that's a discussion for another time and place
@ocean reef this better?
Hardworking chad 
i need this prompt for refining science!
That's your prompt, dummy! xD
it is!?
then my prmpt is good!
amount of messages: cryonicus > me > eface

I shall now commence slepy times
more worker chads . disregard hand potatoes 
Goodnight 
I so wanna just watch this one episode though

working on it!
you guys might have jobs but I have an important test tomorrow (today)
I;m my own boss (yay!)
what you do?
as if I study lol
🤣
I just guess and get good scores
I'm 'just a sysadmin', that can make his own hours
so if i get up late, or early, boss says ok after 8 hours 😄
my step mom (whom I hate) is also an interior designer and my mom is a graphics designer
Our emotes now
GODDAMMIT LLM -> HUSBANDO MEANS MAN
define it for the LLM
But llm knew you wanted waifu?
Look what I've got!
this channel needs more chadposting though
@wispy canopy were you one who liked cutoff a lot?
I agree
yeah, cutoff is really good
cutoff barely works for me
it's... weird in how it works...
Do I just but white hair, azure eyes, white hoodie in cutoff?
white, azure
ai animation ?
you want adjectives in the cutoff
i don't know why it works that way, but it seems to work that way best
the optimism in this guy
I dont think so. Just a .gif from tenor
oh I thought
oof, bwm_nubby bringing the hurt.

androgenous prince?
too feminine!
probably, would pass as "feminine prince"
for sure
my llm+sd thinks i wanted hobos... uhh...
Mine. Shes mine.
If i was a hobo id be depressed too
yeah, lemme steer the llm
and... the suits are back lol. XD oh well
lets let this generation run (60s)
🤣
I literally cannot see the names of anyone with stable society role, this is funny lol
totaly dark theme?
Yes lol
Interesting how u turn on it too
Almost like its an easter egg
2nd one! 
@ocean reef i cant -> disabled avatars...
this with the cutoff?
That's one I made yesterday
This one is nice!
This is with cutoff
not sure if this it the anti-hero protagonist, or the backstabbing 'ally'
choose one 😏
I would put a lid on my drink near him frfr
Its the hair. He looks more trustworthy
@ocean reef seems hoodies were popular, needs more brooding anti-heroes?
3rd one!
not just the hair... the expression too
Yeah went to look at the first one again, i see what yall saying
strange how just a few pixels of difference can make sucha change in first impression
Pls AI just give me straight hair without accessories
nope
u can hear the sound effect
are you prompting twin tails?>
Yeah I noticed it
Just added
did we just find @ocean reef's anime husbando?!
grieves in agony -> if i untied my hair 20 years ago... oh wait... 20 years ago long hair wasn't popular lol
ok I finished the episode
now I slep fr
GN 😄
gn @brittle jay
love the colors
Yeah, me too 😄
looks good!
we incognito 🕵️
if output = random -> user is eface
sir incognito, would u happen to know why my lora is giving me unsharp results like this?
An illustration of two characters standing together in a barren wasteland, facing each other with determined expressions on their faces. Behind them, a bright star shines in the sky, casting light onto their faces. One character is holding a sword, signifying readiness for battle, while the other has their arm wrapped around the first, symbolizing unity and support. The background is filled with swirling dust and debris, indicating the struggle and conflict taking place. The colors used are muted browns and grays, contrasted against the vibrant yellow of the star, emphasizing the determination and strength of the duo.
upscale/highres fix?
yep
Will try it out tomorrow 😄 I mean later today
too low denoise
Its at .5
Then .5 cant be terribly bad from good right?
I'm not that good with LoRA training, but I would assume you could get blurry results from using too large of training images. Even with bucketing enabled, it has to downscale the images so that they have the same number of pixels as the training resolution which could result in blurriness.
0.5 is very bad.
0.54 is borderline unacceptable for 2m+ karras imho
.58 denoise result
i heard euler a doesn't care about it tho
My images are almost always done at 0.51 denoising for Hires fix. Like probably less than 20 of the last 4-5k images I've made have used any different denoising level.
i downscaled the dataset to like 768x768 cropped
Did you check the downscaled versions to make sure they don't have downscaling artifacts/blurriness?
I'm assuming you probably did. Just making sure
I mean i used that birme website which has an option to keep the quality.
After checking, they do seem slightly less quality, but i wouldnt think it was to the point of causing this
lol. how did a natural language prompt actually fail this hard? AWESOME
Should i upscale them and try again?
Because CLIP is a pretty bad language model.
Last guy for today's night session 😄
was doing puzzle on your pic 
i know... just lemme savour the moment ❤️
i really wanna see what sdxl can do with my prompts though
scared and interested at the same time -> scared because it won't glitch as much -> interested who wouldn't be?
I might want to try this downscaling tool https://github.com/Anime4000/sd_dreambooth_extension/wiki/Anime-Training#downscale
I want to share image but I'm saving it for Nubby's contest
try it out in #1100170312106127410
i'm kinda like given up on that contest, my waifu is just a really generic pink hair girl 
I've been sharing all of mine that I made for the contest 😅
Oh boy. Lots of reading
token limit reached, not even finished my prmopt xD
Fine here we go @lyric zealot @supple raptor
Ah its only one section tho
rip in peace my friend
well, inpaint is a good function exactly because 75 tokens are not enough
i'll be honest though. those results are MIGHTY impressive
Yeah, you probably don't need to read the rest of the guide, since it's primarily for DB training.
4g of vram is enough to make and train a model?
no, but you could train on colab there's lots of guides on Youtube for how to do that.
i should look in that
i have so many ideas, i would love so much to play with lights and fluids
Whats dreambooth imagebucket? I dont see that anywhere
this week, my heater has not been on. the outside temperature is certainly below my set temperature
pressure? air in the tubes?
my 3090 is heating my room!
uahaah
It's just the term they used for aspect ratio bucketing. Instead of needing 1:1 aspect ratio images, you can use basically any aspect ratio and it just resizes the images to fit within "buckets" of similar aspect ratios which at a resolution that has the same number of pixels as the training resolution. TLDR: You don't need to crop images if you have it turned on.
So should i run the manually cropped images in the extra tab then or take the original big ones and downscale them again
omg..... why i couldn't think about it before?
i could use poesy in prompts
i could make William Blake poesies real!
By "manually cropped" do you just mean ones that you had to crop to remove parts of the image you didn't want?
More or less, yeah. They all were cropped to the same size
If I could preseve the quality of the original images while also downscaling them that would preferable
so sweet lol
i need more evil in her eyes ❤️
I would recommend only keeping the cropped versions of any that needed parts cropped out of them, like if there was a second character that you cropped out or the character was too far away. Then try with the full resolution versions of all of the rest of them. The trainer will handle resizing the images so you shouldn't need to do that at all.
the white 1?
yeah, still alfahentai
i need that model
it's on civitai
i love his sharp colors
vil, what's your model?
why i'm still downloading more models?!?!!!! i have still finisced to eliminate the old once >.<
yeah, that's how i'm feeling
I'm using Aurora atm
But I also use these often. ||Since you asked my model
||
https://huggingface.co/Surficialruby/RandomMixes/tree/main
That latter part is what scares me. See, Ive trained this style before, and i had done it successfully, but during some cleaning i deleted them and now i cant figure out what my solution had been before. I do remember having to downscale the images but idk if that truly was what fixed it.
I would think even higher res images would be great for training, cuz like u said, it automatically crops them.
Yet for some reason thats not the case here, or at least seems like that isnt working
can never have too many models 😅
i only have 200 gb of models i constantly delete them to make new space
Same
The issue is probably that it downscales which is always a lossy process. If you start with too large of images they can become blurry when they are downscaled, but somewhere around 2k pixels for the longer side should be fine for letting bucketing handle the resizing for you.
You've got me by a bit on LoRAs.
zzz time, byebye
So from 5K to 3K pixels sounds better?
why? lol
it makes me laugh so much XD
i don't know why i find this image so funny lol
so sweet lol
5k is probably higher than you want to feed to it. I would manually downscale any that are that large to make sure they still look good. Between 2-3k is usually what I've tried to aim for. Smaller is fine too as long as they have at least as many total pixels as the training resolution.
wow, that's tifa, used a lora?
Guys I want to upscale 8x but I don't have enough good gpu
is there a soluation ?
I can't upgrade my gpu rn
@steady grail
no lol
Imagine getting isekai’d by the scooby doo gang. Id be pissed
HAHAHAHA
so cursed XD
Yus
yeah
first time i seen the image, yep thats whats gonna happen
are model that you apply on the princiipal models to alter an guide the images in determinate directions
They gonna make it into a conspiracy and go on a mystery acting like they dunno how it happened
so it's like poses ?
imagine your model makes a blonde girl with a black shirt and long black pants, you set a lora that alter the image so that she have in any result a girl with black hairs and a kimono
Manually? Oof, thats almost all of them 
there are loras of anime characters so that any result resemble such characters
lora that corrects fingers
is there a video explains this more ?
That type of upscaling, doesn't really add quality. It just increases size and tries to smooth things out to make it look as good as it can. If you want increased quality with upscaling then you want to use either Tiled Diffusion or SD Upscale script in the img2img tab which both upscales and runs img2img on the upscaled result to try to improve the level of detail and quality.
I think I need it to make better poses since I want to make animation
wow

it's pretty easy to use lora. look for them in civitai
Might need to delve into controlnet tech for that
yeah, loras are a must for consistence
With that tool in the rentry link I sent, you can do it by batches. You would just have to sort them into portrait and landscape first.
Upscaling won't help with poses in any way.
controlnet is confuused
okay so 4 it's enough
Oh u mean the xnconvert tool?
And most of them are landscape so i can maybe get off the hook only doing those
yep
512x512px is plenty to get the pose you want. From there you can use img2img or controlnet to change the image. In fact, somewhere between 512x512 and 768x768 you will get far more consistent and coherent poses than trying to generate at higher resolutions, which is why we have Hires fix.
yeah, models go crazy in resolutions bigger than 512x512
you should try this https://github.com/Bing-su/adetailer.git
It automatically detects and inpaints faces for you when you generate the image.
they are not trained for bigger resolutions
thanks man
Anime models usually work well up to 768x768 or a single dimension up to around 960 (example: 512x960) before they start getting weird.
This one used the adetailer extension. Check out the quality of that face, even at such a small size in the image.
What kind of settings are good for ControlNet tile upscale? I've tried a few things but there's some strange stuff going on, probably because my settings are off
without long body parts?
In general, yeah. It's somewhat prompt dependent though.
well, yeah
but usually, with big resolutions, i got weird artefacts, 4 legs, giraffe legs
960x512 with hires fix for 2x upscaling. No cherry picking, these are just the first batch that I made with the prompt I had loaded up.
I could probably even push this one to 1024 wide without getting too many duplicate characters.
Feel free to pull PNG info if you want to see the rest of my parameters and how I prompt.
No its ((((((((masterpiece))))))))
Is it better to make a small image and upscale, or make a large initial image and then upscale it less (but the end result is the same resolution)? The initial images from a blender render, not AI
omg, that's face detection, the extension i mean
they are pushing stable diffusion very far
scary
Typically images are trained in 512x512 so youre better off leaving it there and turning on hires, but doesnt mean native 1024x1024 will look bad too.
yeah, that's what i knew too
chwtoy, what the theme of this images? post apocalyptical hell?
doom scene,
Yep, I always try to keep the smaller dimension at 512 and only increase the other dimension, then use Hires fix to upscale that result by 2x. If I want the final image even larger I throw the Hires Fix result into img2img and pass it through Tiled Diffusion upscaling for another 2-3x upscale.
1 and 4 are spettacular
nothing is good is good as making images and listening mili
i like 2
Thats good workflow, ive been testing my limits onr resolution today im able to hires x4 to 2048 and now im trying to ultimate upscsle that x3
Almost done
i tried making a lora, it's unusable.
We'll see if it works
1024 96 is best for me
it's the same as the maximum 1:1 aspect ratio gen you can do
how much vram?
8
for me it's 768 64
accidentally made an artstyle lora instead of the character i wanted
I find I get more of the typical high resolution type artifacts if I use over 2x upscaling on Hires fix. That's already pushing my typical gens to 1024x1536 or 1536x1024. Things like fingers getting messed up during the high res pass or backgrounds becoming less coherent.
that's a blueprint of a bomb
Steps: 25, Sampler: DPM++ 2S a Karras, CFG scale: 7, Seed: 1906559651, Size: 5120x5120, Model hash: 841515f5b6, Model: MixedDownv1.2fp16, Denoising strength: 0.35, Ultimate SD upscale upscaler: 8x_HugePaint_v1, Ultimate SD upscale tile_width: 512, Ultimate SD upscale tile_height: 512, Ultimate SD upscale mask_blur: 8, Ultimate SD upscale padding: 32, Noise multiplier: 1.05
Time taken: 25m 1.37s
Torch active/reserved: 5010/5886 MiB, Sys VRAM: 5950/8176 MiB (72.77%)
weee
Finally, I got something good.
I mean, open original, it's a real high res
but by no means good
stop making better arts than mine, i literally teached you 😭
you can use it for making slime girls lol
oh, i made something like the white one when i was testing
Do you have a particular use in mind?
1920x1080
What resolutions are they?
Ugh, I was reading forums and experimenting for 4 hours.
And I turned off Tiling VAE, cuz I couldn't manage to fix problems with it.
Anyway, thanks for helping.
right
both 1080 still
why i didn't do a wallpaper for my pc?
i'm actually an idiot?
i wanted to get better at using SD first lol
try 504x1792, that will match the aspect ratio of the two combined.
hires x2?
nemu!!!!!!!!!!!!!
or jus native
Anybody uses inpaint? I can't get good results, it seems I don't understand the mechanic of it.
2x might go OoM but it's worth a try. If it does, try like 1.5x.
aight bet ty ill be back
I like wide gens
Damn it, how do you even do that? Outpainting?
Nope, that was just txt2img with 2048x512 resolution, Hires fix 2x upscaling and Adetailer extension to improve the quality of the faces.
The main problem with going that wide is you'll often get 2 or more characters.
tol boi
Yeah, sometimes the tall ones can come out pretty cool. They're usually more abstract though.
Oh I like it
been playing an about to get on elden ring with a friend on the seamless coop, i wanna make an erdtree
someone posted a new model https://civitai.com/models/60638/model-ex-v10, i wander how good is it 😛 FOR SCIENCE!
a quick look at their doc they posted and i hate it lol the styles are so bloated
the negs are sooooooooo long
a few ng inversions and words is enough
yep
ultimate sd upscale VS multidiffusion upscale
This MF really recommending a 9584 token negative?!
WHAT
???
:3.0 ????????????
as much as saying my model just does random 💩
With that many tokens 3.0 emphasis is probably necessary for those ones to even have any effect XD
but the prompt will just pick them randomly

9000 token is as much as asking you to find your position using the map of the milky way!!
and it's slow as... my god!
he rated his own model and gave it 4 stars KEKW
LOL
ehy guys, in how much time you can elaborate 9000 tokens with your gc?
i don't want even try, it will take 3 days minimium
hello
ehy
I made ai animation . and now I will aprrove that ai animation is also good
show it to us, man
I mean, I've run a 20k token prompt before for the luls, but it's pointless.
mefi, you just made a meme material, shame on you
this image will curse us for the rest of our lives
praise the goddess of eepy
I need more frames to make it look pose fight
I want to make anime animations for my channel
I used 6 images
mass produce animations pretty easily
I need to use more to make it look more good
i'm gonna be pretty honest, when you first joined the discord, i thought you were gonna quit soon. glad to be proven wrong, this is good!
yes I thought I can't use stable diffusuion
but 3 men helped me
pretty sure this image will crush
i pushed tiled vae too much this time
holy christ don't explode the chat
she cute so I forgive her mangeled fingers
@subtle crown and this was one of the 3 men
@supple raptor and this also
thank you guys
aww
fix her, you beast
it's so good, looking like this breaks my ❤️ lol
my, the upscaling messed my image lol
ultimate sd upscale might be better than multidiffusion
i shouldn't have used hires fix upscaler
i always tell people to not use highres fix
you are so right, bro -.-'
Can I make animation with hard poses ?
like the character moves a lot
left and right
Nah, ultimate is barely better than regular SD Upscale, and only if you're trying to push denoising a little bit too high for regular SD Upscale.
or fight sence
now i want to see adetailer fix this face XD
i prefer the second
the last verison with editing the speed
if I use flowframes it will make it better
Your creativity and level of patience is the limit. Using controlnet to basically use the AI for rotoscoping means you can make basically anything you want.
This is a bit old now, but it still has many of the basics. https://www.reddit.com/r/StableDiffusion/comments/119o71b/a1111_controlnet_extension_explained_like_youre_5/
isn't there a video?
did you already read the explanation on github?
Probably, search youtube.
I like video cuz I can understand it better
i'm arabian
and my english not that good
so video is easier to get it
oh
I can see what he does and get
you write it pretty well
yes but I hate reading english articles
let me see what i find
Yeah, I can see how a video would be easier to follow if english isn't your first language. I learn better personally from written guides so that's just what I have on hand.
most of english video I don't get what the video saying exactly but I get from the video
sooo if you guys have a video it would be very nice
https://www.youtube.com/@sebastiankamph this channel is pretty good
Professional Art Director helping you learn AI Art and Generative AI.
- Dad jokes.
Support my work on Patreon to help me and get unique Discord perks: https://www.patreon.com/sebastiankamph
This channel is about artificial intelligence and machine learning news and content related to the creative arts.
Tutorials in Stable diffusion, Midjour...
i learned many things from him
especially how to do prompts
She big mad... Time for cheek pinches
she is a yandere?




