#🍥|anime
1 messages · Page 3 of 1
Fr
the space is international D
aahhhh i gues its cause my prompt has following: wearing a (space long uniform, long sleeve, trousers, gloves:1.5)
try that
Could give that a try
i is cosmonaut, vodka is fuel
This is a very weird and rather unintended outcome. Female astronaut walking through a strange portal door onto an alien planet
God damn what happened to yellow crewmate
oi, konosuba is out
Ok where do I prompt
also,
Boku ga Aishita Subete no Kimi e(blue movie)
Kimi wo Aishita Hitori no Boku e(purple/red movie)
-->
blue movie to purple = sad ending
purple to blue = good ending
choose wisely
Where do I prompt
you need to get automatic1111 to make pics, and after that you're gonna need to download a model
Can’t we just imagine
got any ideas how much video card(VRAM) memory do you have?
can't do it here anymore
but you can use dreamstudio for that, 25 first generations free
afterwards it costs money, like 1 buck every 100 200 pics dunno
ok?...
I think they're saying Midjourney is better because they don't have to install anything...?
No bcz it has better images
Pirating
It has better images when you don't know what you're doing in stable diffusion, sure. When you do, SD can blow MJ out of the water
anime pfp
What does that even mean?
Just pretend he doesn't exist
@desert arch "Yellow Crewmate" after doing some post-gen work XD
Hand wonk with the left hand but otherwise pretty damn neat
Pretty cool concept
Honestly was inspired to try doing some space suit shots due to getting back into No Man's Sky after this most recent update (I say back into but I never put much time into it before this), but it's funny you assumed Among Us
I assume every spacesuit as among us
Why's that?
The starting look in NMS is too XD
yeah... it is kinda useless to debate like that, the iphone mentality isn't something to break free from easily in all honesty
not that im trying to shit on you mate, but your answer was just way too much of a stretch
ffs i tried to help you... what do you want from me?
if you came here with an answer before asking, then why waste my time or anyone elses?
and in a bratty spoiled way, i don't owe you a thing, nor anyone else here or anywhere else
Calm down brother keep it together
im not gonna force you to do a thing, do what you want yeah?, you're talking to people who most likely would help you for free and complain?...
if he isn't gonna get resistance his whole life - he is going to think that he is the lube
he's gonna have to hit a wall somewhere
I mean obvs they're trolling lol
How to deal with trolls > you don't
They come to the anime channel just to insult someone who surprise surprise has an anime pfp
Which isn't the crux of it but it's a dead giveaway
And it's an alt account
Oh- I think I came at a bad time-
Dw carry on
Nah ur gud
Here's a robot dog or something
Neat
/anime
Considering I don't know what Sabu is, probably not XD
Dammet 
Link?
https://civitai.com/user/LUUNA It here 
Also its not as lewd as it looks, I just had to clickbait 
@magic waveWhile it's lovely and I'm definitely going to play around with it, what made you think I was using it? XD
Well, kinda like the color palette and brush strokes :o
I use a big mix of LORAs, but the one that probably influences those aspects the most is this one
https://sevenstyles.com/p/artist-lora-for-stable-diffusion-1-5-8209429/
Tis a pic with V2
Ohhh, thats fair then :O
👀 looking through it you must've clickbaited pretty hard
Well, I want people to get it and enjoy it 
And it do work, my considering my first Sabu ones are the most downloaded LoRAs I got xD
Okay, time to benchmark the 4 Sabu ones
I've got a standard prompt and seed that I use to thumbnail my style LORAs
Ayyy 
I got Civit Helper and scanned my LoRAs so all the thumbnails got set automatically xD
Made using the loha version, the model i use is kinda kinda anime-ish so not very visible
I want my thumbnails to be on my model and prompt though so I can compare easily
Fair fair :o
Doesn't seem visible at all xD
Trying v6
Testing it without keyword and then with
I will say that style LORAs in my opinion should NEVER be trained with a keyword. Styles should apply when the LORA is applied, period. That way token space isn't taken up by a keyword
But that's just my opinion
@magic waveInterestingly, despite my rant about training style LORAs to need a keyword, it seems that yours actually don't. I'm getting very similar outputs with and without.
Doesn't matter how you dress her up, Bochi is still a hilarious trash gremlin
Correct
Well, yeah, keyword just helps a bit more .o
Nice clock bro
@magic waveMy usual LORA setup with your v6 added on top
Oh damn that looks pretty dope :o
that looks very nice
the colors are a bit washed out, are you using a vae?
?
I will upgrade it, because want to achieve specific style and then I will apply vae
testing right now
oh you're making a model?
Juicer
since everyone is posting art, i'll drop one too
Just one extra arm
Last one for today (maybe) I'm sleepy
that's what I say before making 20 more gens


I should sleep for real, got a tryout tomorrow
Gn
dman, love that style
(empty eyes, expressionless:1.5)
Any fine tuners here
uh does anyone know why it starts off normal and then turns into an abomination when i turn on high res fix?
could it be your denoising strength too high?
doesn't work with this model 
imma try a different one
i'm using anythingv5
I guess it's better for flatter/simpler models
those two keywords are danbooru tags, and anything models are trained on danbooru
so they're much more effective
thanks!
CamelliaMixLine
those are decent hands
gets stabbed and never sees her again


cute 
moody, me like

"A booger shaped like texas flying out a beautiful woman's nose, hitting a rhinoceros. Greg Rutkowski."

last one, gotta go, gn
make the smallest resolution at least 512
and try for 25-30 steps
SD is trained on images of 512 pixels, and tends to work very poor when it goes below
also not the greatest above, but, we have lots of trickery to fix that
tbh, it looks pretty good without hires fix
the low resolution is hard to fix tho
i was getting good results before, weird
did you turn off hires fix?
@gritty patrol it could be that you have two tags that bite each other, i have had that too. those are hard to find tho
perhaps something in your loras?
damn these ketchup producres, they shoudl make the bottles better
ayo guys, is there and Satellite Z30-B-14w keyboard replacements under 40 euros?
I do (lots of blood on face, lots of blood on clothes, blood)
I get this 💀
?
are the replacements any good?
depends on what you get, check the reviews
all depends on the seller
but the site is pretty good, bought tons of stuff and parts from there
only downside is the shipping, takes 26 light years 💀
what model is this
anything v5
thonking
🤓 light years is not a time measure

are you using hires fix?
im off👋
if not, then, at low res, it's probably blurry
you might wanna tweak your hires fix settings to better fit your pc
i said it cannot handle
had to remove sword, my model doesn't handle that very well...
medvram always fails with hires fix for me, even if i put on x1
when I was on the rx580 a couple of months ago x2 upscaling took 6-8mins on a 512x512 image 
@gritty patrol you can try the vram estimator extension
that might give you an indication what you can use
there's also multidiffusion, I think it uses less vram
it said 896
it has downsides
i have it installed but i don't use it because it normally gives me duplicates
it's not great when using it for txt2img
mmm, I never tried it
it's great for img2img tho
yes, it's the best
1.5x hires fix
used tiled vae
pog
lol, i really don't know why it wants to do that thing with the legs, but i'll take it ^^
👈
cutoff is really cool tho, so much more image coherency 🙂
he became edgy
fear me.
there's something wrong with this image but i can't point it out
there's a LOT of things wrong with that picture xD
like what???
the simulation is breaking
shadows, random white pillars
this image is so beautiful, i wish her pinky heals up soon ❤️
sssshhh
she's just one of those blessed people who can bend their pinkies like that
how did you guys get that stable society role? i'm lazy to see FAQ
the TRUE HORROR of long hair!
do well in the PoW
what's that
Picture of the Week
ah okay
aight, ran out of memoery (DAMN YOU FRAGMENTTATION)
but that's a signal i probably should go to bed
byebye all ❤️
guess who just got multi control net working for the first time finally
your mom
Congratulations on your future endeavours.
Omg suwako
Be right back in about 3 hours
jesus, that looks amazing
guys, i need tips for generating better hands
.
what does she use the third wing on her back for
best i could find online, haven't tested yet
looks promising
it does work dayum
this is absolutely perfect (except for the extreme exposure on her face)
but holy crap, how did you get this?
(highres, promotional art:1.2), ((black hair)), solo, tohsaka rin, medium breasts, two side up, blue eyes, black pleated skirt, red sweater, thighhighs,
Negative prompt: (lowres:1.5), (worst quality, low quality:1.5), (realistic, 3d, furry:1.2), (bokeh, bloom:1.5), (blurry, blurry background, blurry foreground, depth of field, motion blur, film grain:1.5), (text, signature, watermark, card (medium), title, meme:1.5), (lips, teeth:1.2), (holding, holding gun, holding weapon, dual wielding:1.5), (monochrome, greyscale:1.2), (bad hands, multiple hands, fused fingers:1.4), (very long fingers:1.4), (crooked fingers:1.4), (bent fingers:1.4), (thin fingers:1.4), (short fingers:1.4), (disembodied limb:1.4), (disembodied:1.4), (fused:1.4), (6 fingers:1.4), (4 fingers:1.4), (3 fingers:1.4), (1 finger:1.4), (fingerless:1.4), (mutated:1.4), (duplicate:1.4). (multiple views:1.4), (cropped:1.4), (3 arms:1.4), (3 legs:1.4), (amputee:1.4), (long legs, long arms:1.4), (bad anatomy:1.4), (blurry hands and fingers:1.4), (monster:1.4), (thick thighs, wide hips, cameltoe, skindentation:1.2), (badhandv4:1.0), (EasyNegative:1.0),
Steps: 40, Sampler: Euler a, CFG scale: 7, Seed: 1707692594, Size: 1200x600, Model hash: 6e430eb514, Clip skip: 2, ENSD: 31344
what would "two side up" mean?
it's like twin tails but you have back long hair also
not really. last two are neg embeddings though
negative: badhandv4, EasyNegative, (holding:1.4)
vs
negative: (lowres:1.5), (worst quality, low quality:1.5), (realistic, 3d, furry:1.2), (bokeh, bloom:1.5), (blurry, blurry background, blurry foreground, depth of field, motion blur, film grain:1.5), (text, signature, watermark, card (medium), title, meme:1.5), (lips, teeth:1.2), (holding, holding gun, holding weapon, dual wielding:1.5), (monochrome, greyscale:1.2), (bad hands, multiple hands, fused fingers:1.4), (very long fingers:1.4), (crooked fingers:1.4), (bent fingers:1.4), (thin fingers:1.4), (short fingers:1.4), (disembodied limb:1.4), (disembodied:1.4), (fused:1.4), (6 fingers:1.4), (4 fingers:1.4), (3 fingers:1.4), (1 finger:1.4), (fingerless:1.4), (mutated:1.4), (duplicate:1.4). (multiple views:1.4), (cropped:1.4), (3 arms:1.4), (3 legs:1.4), (amputee:1.4), (long legs, long arms:1.4), (bad anatomy:1.4), (blurry hands and fingers:1.4), (monster:1.4), (thick thighs, wide hips, cameltoe, skindentation:1.2), (badhandv4:1.0), (EasyNegative:1.0)
same seeds
Hands 
Good morning
I feel like you could do better.
I gave up shorty after, was getting too cursed for my liking
The lightsaber was unintentional
can you give me prompt + model?
does anyone how to do eye direction stuff? like looking at a specific object etc
Pretty sure there's a controlnet for facial expressions but i don't know much about it
Yeah Veggie's new Anihex model is 🔥 from the little bit of testing I've done on it.
It's on HF?

Hey guys, just doing my occasional check in lol

So much pretty stuff
ok, what i have to do to get arrested from her?^^ nice pic
I really like this one, the glowing butterfly thing against the black is pretty.
Thanks! That one is an old prompt that I had GPT-3 write for me. I've just been throwing random prompts at Veggie's new model to see what works and what doesn't.
Space prompt?
detailed background, turn off your mind relax and float down stream, sakura
For example
Hold on
offset noise went hard on this one.
With weights
like this one
She forgor 💀
"Miss, you seem to be missing your... skin."
do u prompt with "demon slayer style" look similar to that?
I use ligne claire anime lora then prompt with ligne claire, flat color, high contrast, chromatic aberration, colorful, vibrant colors, at the start
it seems the modell knows daemon slayer ^^
Dear god putting horror in my prompt was a mistake
Ok. I have to say I love multidiffusion more than I ever thought I would.
ᵉⁿʰᵃⁿᶜᵉ enhance ENHANCE
I don't even know what that is.
"Tiled Diffusion" (on MultiDiffusion method) and "Tiled VAE"
Works wonders with generated upscaling, is very forgiving with the prompt (most times), and in my opinion is much better than traditional SD Upscaling.
That and if you want to, you can have different prompts for different image regions.
Thanks, I'll have to check it out when I can.
try this with diffrent animals XED 🙂
Too lazy to inpaint
And adding prompts would probably break it
ahh^^ yes to hard to prompt
Heh. Who said girls didn't play MW?
been there done that
Guess this prompt
Multicolor, monochrome
"I don't want to set the world on fire."
Pumped up kicks 💀

These goofy prompts gives surprisingly good results
Well
And many thanks were given to the model trainers. How much more these models can do when intelligent thought is placed into the prompt creation. It's so exciting living through the birth of such technology.
Its like a random prompt
I use counterfeit 2.5 checkpoint, bocchi the rock locon, and controlnet random photograph of girl looking at window with hed guidance 0.5 and strenght 0.6. Prompt is "1girl, head rest, pink hair, looking at window, building at distance"
So what's offset noise do exactly?
Offsets the noise 
All SD models that are trained with normal methods generate images that have an average brightness of around 50% which means both very bright and very dark images are impossible to make. Offset noise widens the range of brightness that the model is capable of producing.
Ahh... I gotcha. Thanks for the easy to understand explanation. Sounds quite useful--I look forward to experimenting with it.
In the... "Un"stable diffiusion server...?
just write it in chat we dont judge you here^^
did u specify the accesoirs the nadd weigt and a better description maybe
les say u add jewelry and it dont appears so make (jewelry:1.5)
yes then give necklace more weight
specify (daimond necklace:1.5) or something else maybe?
so after readding anime as a channel, we instantly get new people?
Careful about discussing NSFW prompts here. This server does not allow any NSFW which includes both images and text. You haven't said anything that would get you in trouble, just wanted to give you a friendly heads up since you seem to be new here. Welcome BTW!
ah tagging 150+ images if dark fantasy, gonna try training this lora with a different approach. More epochs etc, less training passes
then I will be afk for x-rays
Time for exu rayu
What's that
X ray
X-ray?
CT PET
Yeah figured
Final one before surgery
do u make x-ray pics from weird stuff explored under x-ray? like this guy with a ww2 grenade in his a** a while ago (in real XD)
Hopefully it doesn't tell me I'm beyond saving but they are taking their time. Their sweet fucking time
Uh no
Not bad
Guess the reference picture
Booom
Same seed, different upscale denoise and cfg's 
welp i guess she literally is on the bed
half of her body that is
She is the bed
she's stuck
Would lie on it 10/10
Stepbed im stuck

O she stuck in her married
that's kinda scary welp
She is going places
Autocorrect strikes
Tonight I'm making a Lora with nothing but blood effects, gushes sprays etc. For any color
Any color?
Yeah not everything bleeds red
Oh nice
sounds interesting, good luck

Also gonna see if i can get the alien blood working
Won't be easy cause it has to learn "gravity" for lack of s better term

Classic autocorrect
Yep
Here have a chtulu
I need to get a file that's stored and ready with words we use
Cute vulva face Cthulhu
Dear god
It's getting worse
I don't see them having vulvas
I wanna get home so I can finish dark Gemini 3.0 with the example images and all
Then will i guess upload it
For all to enjoy
Well all who like dark stuff, grim gritty. Gothic of all sort, horror, sci-fi, Lovecraft, event horizon
What model has it has? 0
alfamix
i don't think i can work without cutoff anymore 😮
the added image consistency is just amazing
How does cut off work?
Ah yes my favorite singer
it's an extension, you can mark work that the model should watch out for -> for example when you say red ribbon, and then mark "red", it tries not to apply the red to to other elements of the image
in my case, i have gold accents, blue wings
so i mark gold and blue
but i've also already used it for other purposes
Guess I'll apply it when home, make sure my character has white hair. Dead pigmentation etc
And the psycho character has nmn
this is the normal image, and the image with dynamic thresholding and cutoff
for example, now i can make sure the ruined is applied to the city only
among other things
and the gold is less applied, as it should only apply to accents only
How does that work? That you only count ruined to the city and not the rest?
prompt: fantasy cyberarmored girl, visor, biomechanical, humanoid, refined, halved, red helm, white painted body, gold accents, evil, horns, messy ruined city, confident stride, angel crystal energy wings, blue wings, pristine, [anime lineart:0.3]
cutoff: red, white, gold , crystal energy, blue, messy ruined
as far as i learned, the words you put in the cutoff prompt, it tries to apply them only with the tags they are grouped with
i read that as "hey, i can try to make stuff apply more local, instead of global interactions"
Yeah but cut off, does it isolate the tags from the rest of the tokens?
I see
still not failproof, but enhances, surely better than nothing
green scarf, red shirt, blue pants, blonde
how is everyone doing ? 🙂
changed blonde to yellow -> shoudl be clear 🙂
hey floydian, how have you been?
well, i'm just the usual i suppose, i'm alive
just the same old, i suppose?
am good, recently back from my work trip in Korea :3
oh yeah. i'm old. i'm turning 40 in a couple of weeks 😮
damn, i'm jelly
i bet you wouldn't be jelly of that 60hr/week tho !
if it does payout the proper amount.
did a 60 hour 3 day stunt once. never again
not the best pay, but the benefits and my situation makes it worth still 🙂
6x10h
1 day off
still, sounds like hell
no time for much, but i did get to visit the city and buy gifts for family (and some for myself too !)
my next assignment is in the US, and i should have more time on this one, hopefully
"if you die early, i'll resurrect you and kill you again!"
is the operation planned @cyan burrow ?
It's yeah but very late and trying to get it moved to earlier
yeah, all the days you can get it done earlier is better
need to get back into the game, anything i should be aware of with new SD updates ?
@terse jacinth i think the last update is ok again, it was broken for a bit. you might want to install last version of A1111 in new fresh folder
you can install pytorch 2.0 and use --opt-sdp-attention instead of xformers, about as fast, but it's deterministic
xformers isn't relevant anymore ? D:
although, if i did read correctly, pytorch 2 SHOULD be faster than xformers when properly programmed
especially on A100's... heh. i need to get that other unit serviced so i can get my clutches on our A100's again
lucky
i saw my job is running RTX quadros, wish i could play with those
Oh lol, my keyboard had protection from offensive words
tho they power it with cheapo bronze PSU, that felt shameful
Lol
THE KEYBOARD HAS NSFW FILTER?!
and as for updates to A1111? i recommend the dynamic threshhold and cutoff extensions
they are great for increasing image fidelity/consistency
noice
i'm so rusty now i'm hoping that the update process goes smoothly, i have not touched anything since january
It works for him 8 to say Bambi killed his mother for YouTube likes
only thing i played with was Bing AI and it was kinda neat, it wrote me poems about political tensions between North Korean and the USA
oh yeah. gpt4 happened of course, and now the AI wars have truely started
guess my old versio still works 😄
What is the diff between cutoff and negative prompt?
yikes....
negative prompt is basically saying to avoid generating it, cutoff is limiting certain prompts/traits to certain object/subject
interesting, thanks

boy lives in the past 
only recently, since i was in asia for a few months 
i'll be in your TZ next week thou
thanks, it was lovely in Seoul 
its raining here a little bit
I mean, at least in Korea, there's seasons. 
Why is that so xD for me its 5pm
But why 30.01
that's the way to tell EPOCH time
12 am for you is 05pm for me @lyric zealot
i mean, that's the date when i updated "about me"
time = distance to X for absolute time
Ahh okay
since now it's 7pm for me, it's 12am for you now?
nope it now for me 13.00
Yea ik i just wondered of the date
A1111 finally worked it out, yay
1pm
its a bit complicated to convert from asia TZ to american because of the summer daylight saving time ><
@lyric zealot
I'm gonna be honest, still have no idea what's daylight saving time is
really good idea
it's to adjust time to get more daylight in the days due to seasonal changes
it's a bit dumb but it has a few good arguments (safety and body mechanism)
you are like 6 hours behind?
why is there a need to do that, is the real question but oh well
sure am 
Nobody knows and EU trys to revert it xD
not if you're in China, then you gotta be annoyed by a singular timezone
the argument was more suntime for body functions, and safety concerns for children walking to school in the dark, i suppose
and other such things
actually, i can respect that
you from the UK i think
It was largely made for northern, northeastern northwestern etc wireless
Nope
what
f
ah the wireless region, love that place
God damn auto is poison
Trust me, it ain't that great. Unless you're a reindeer
gib me that 5G wireless country bro
it's the land of no wire, it's probably high tech af
pushing remote working too far : remote wireless living
I swear VR working is gonna be a reality soon
We have 5g in bigger cities. We shall have it everywhere soon enough
Maybe Singapur?
my country is so smol, there's 5g everywhere
Seoul had 5G but only for korean citizen, i was stuck in LTE
Then the waifu will spread
poor work visa pleb noises
So rare seeing people spell Singapore that way. 
isn't LTE hella fast too
it's fast enough for my uses
considering they only recently retired 3G, and 5G isdn't widespread yet. probably a while 
does someone know if i can get a higher res while upscaling with tiled vae?

I bribed my way through
Ye
probably? and also using "extra" tab with upscaling works as well


uwu
same... i've been waiting ><
I had my 1 shot to get it with the valentines POW, but I was late, now everyone is too good 
try the weekend pow. 
it's shorter period
Imma go have dinner


That is a good point :o
getting to that new SD-WebUI thanks to @wispy canopy 
just need to transfer all my previous content to the new install, reee
all wildcards, all models, all embeddings ><
and re-do all my styling and custom defaults
why didn't i think of that....
i guess being away for 90 days ruined my brain, feels like i'm new all over again
I will make sure Kizuna AI is in my next model. The gothic version of her
I fucking hate my autocorrect. For some reason it has forgotten both English and Swedish
Ooof
I've just turned it all off except for anticipated words during swipe
it works better then
I can barely type on a cellphone is weird. too small to grip
just had mala and i'm dying welp
trying to load the latest file in directory, but since its being copied
or are CKPTs now loading anymore ?!
why wouldn't they?
not sure, vanilla installs now come with safetensor
but i guess it's just a "copying files over and it's picking up the latest which is still being copied" issue
nobody wants to use them though . when you get more stability security etc from safetensor
those are my months old files, safetensor was just out of the oven when i left, not quite a standard yet.
i have to recap 90days of updates, bear with me ><
if you have questions just ask, I'll try my best to help with stuff i know 
We need that wallpaper 
i remember that one, upscaled that one on the A100's
i think it took around 30gb vram to fit it all at once. now we have multidiffusion and ultimate sd upscale XD
shinobi style?^^ nice
What model is this? very similar to the one i usually use
Also good evening
I literally just off computer, but it’s holokukiv2
With Lora
Oh so it's just the ligne claire doing most of the work then
Yup yup
love me some artistic lineart ❤️
i feel i might be doing something illegal here...
(no controlnet or latent coupling was used)
she's sending signals.
they tried. not quite a heart tho. at least they get to use their hands after this attempt. the others... were less fortunate
she kaboomed the shit out of whatever that was 😄
Made alot of "kaboom" art as a side product of making my PoW
i was just thinking that... looks like for PoW?
Doesn't really represent the theme so i scratched it
The one i submitted however
Absolute masterpiece
Feel free to screenshot
OH
Floydian, you should check out Nerf's Aoi Ogata LoRA, its really nice
i will for sure
need to get back on this training too
Dat one
oh @terse jacinth, if you didn't have them already, the offset locon(new type of lora thing) is a nice thing to have too
loras are native
locon isn't native
but it seems locon works better than lora
could you send your settings for this? i never get good results like these, i wanna see if I'm doing something wrong
it allows you to darken your images considerably
I mean, as far as I know LoCon works better for styles :o
Is what I was told :o
ALSO
well, offset would be a style then, right? XD
honestly i prefer img2img with completely black image, gives me better results and is more configurable
what's wrong with your results? 🙂
ooh, that's a good idea too
kinda burned and overall bad quality
Also Holokuki noise offset is best noise offset 
imma try that one too soon then
let me reply to a result i got before
one sec
Yeee, I think its the on their huggingface, civit page for the model got the link for their huggingface tho
mefistic, try this extension with these settings: https://github.com/mcmonkeyprojects/sd-dynamic-thresholding
i feel it raises image quality by leagues
Wait wot isn't it backwards?
it's backwards 🙂
but as long as we both have been here, have you ever seen me do anything remotely resembling sane?
what exactly does it do?
i have 10 images for a character but I can't make a lora
in the way i set it up: during generation of the image, it keeps raising the cfg, this enhances image fidelity
dunno how XD and basically, my fun comes from prompting 🙂
then why is your minimum and maximum the same? it's literally not gonna change that way

Ah, then I could teach you 
XD
On vs Off. i think it's clear which one has more details
can you make a lora for me?
https://www.youtube.com/watch?v=9MT1n97ITaE
this is what someone recommended to me for learning how to make loras, it seems quite straight forward and ive tried making 1 thats im still testing. should give it a bash 😄
I think Holokuki's noise offset is stronger :o
Maybe 
got a link ? i wanna try it out 😄
it's just the dynamic CFG -> the lora is at 0.4 strenght 😛
i woulnd't have an idea what to train XD
hmm what about your cyborg person from earlier ? just as a test subject to understand the process then go from there
i can just recreate it in anything i want -> that's the power of generic models 🙂
These were made with Holokuki noise offset
Lemme grab it
this face style is cursed asf
in the native A1111 implementaiton it's part of the prompt...
oof
like, tf did you do to okayu
But the one I got is 36 MB
The scary one? xD
the okayu looks ugly
Oh
it's a weird point between realistic and anime
Seems pretty regular to me

It's style man dw about it
I think thats called semi-realistic
Have a doomslayer
Also, spooky ahead
wtf you put makima's face in okayu?
Wot?
Its just a expression 
i wanna see people try this
better than offsets imo
my images have got all messed up now, like its missing something but i haven't changed any settings, trying a reboot just incase 
Never did offset or that, might try one of those
restart application... why reboot?
yeah i mean reboot application lol
anyone wanna try?
Weird, though I also had something like that happen
Where I tried PNG info of old pics I generated but they weren't the same
just follow these schematics. perfect waifu everytime
some internals were changed over time
And drastically different considering it was all the same settings, same seed, and same everything
Damn 
screenshot a black solid color and put it on >0.8 denoise img2img
haven't tried, but I'm pretty sure you can tint the black to get better results
Also tis is the Holokuki noise offset, I don't know if its the same as the hugginface one, but this file isn't up on the Civit page anymore @proper trellis 
perfect. waifus. everytime. Generic non-anime model.
lets see what happens 😄 same prompt, using black background
first attempt. hmmm...
Shes got a "hmmm..." face xD
0.75 XD
try 0.9
0.95 😮
it's purely the prompt my dude 🙂
ah
girl schematics, [blueprint:0.2], [anime lineart:0.2], labels















