#🍥|anime
1 messages · Page 85 of 1
Rip im getting bored of SD alr
Another edit!
Tried your prompt, very nice
Not rly mine I just stole borrowed it from someone else
take a break before returning 
Turning a hobby into a job is really good way to make sure you get burned out on the hobby.
on the flipside, if you truly love it, it can be the best job in the world
are people still donating to you for your bra model? 
jesus
ya, you should probably take a break, dont get too burned out like what nubby said
what if, and i mean what if, you just crazily make a few new versions, like v6 v7 and v8 in a week, then slowly release them
i mean, you put a lot of effort in a short time, and then rest
true
For me, when i get a little burned out on this, i just divert my attention to other stuff for awhile, and then my energy is back
im taking a short short break soon
at least that works for me
what else have u been playing with

Don't get too focused on perfection though
That's burnout any% right there
not really been playing with much, just watching stream, playing games, and catching up with other stuff
might i interest you in my kami-sama https://www.youtube.com/watch?v=xsc3Z53QTC0
21-06-23 Neuro-sama V3 Karaoke! 🎤✨ stream clip
Neuro-Sama channel: https://www.twitch.tv/vedal987
Original song: https://www.youtube.com/watch?v=ESx_hy1n7HA
but this is some rich people problems shit though lol
like im thankful this job is pressing a few buttons for training
and the rest of the time i can play games or slackoff or wtv
probably the model 
candy cigarette
rip right arm
Snow is such a comfy prompt to gen

oh man what's inside that
I'm too afraid to find out as well
🛌
Damn, looks really nice, did you use lora for this?
I think 3 LoRAs? 🤔
I kind of want to use this as profile pic
Probably not
most models can do guitars pretty decently without lora
man you always make such big images, how do you do it so well????
does it really make that much of a difference
like I tried it a tiny bit myself and it seemed like the same as the built-in script
I get more control with this extension + that's what I'm used to lol
My only issue with that is it seems like its best used in i2i, and meanwhile i prefer just genning in t2i
I didn't know they did galleries
i WANT to do more gens, but my airco can't get rid of the heat enough XD
i was out, so didn't have my airco on
nwo it's already pretty hot, and adding more heat into the room certainly doesn't help
If its a relatively small room it definitely doesnt help
it's not perse small, but it's an A+ certified isolated home
so any heat i'm building up, will NOT dissipate
Cries in summertime. I like it in winter when the GPU can double as a space heater.
so, i do have a small mobile airco, but for that small unit, it's an uphill battle
i mean, my house heating hasn't been active since februari, my gpu is enough to keep my house heated.
does anyone know a good model that uses SD 2.1?
interesting
well... not much at least
need direct venting of computer case to outside 😂
lol
I wish I was more proficient at HVAC
it's... not a totally stupid idea actually...
imagine reusing your computer heat to heat your water tank
it's extra free heat
idk how long a DVI cable you can run 😂
it'd be nice to have the noise somewhere else too
(i got a boiler -> got no water tank)
I have a boiler that heats a tank
thinking it'd be smarter to have an on-demand deal
they tend to be a bit more expensive
I mixed darkalfa with akkaimix 50/50
you made it more horny again?!
well, if it helps with it's general knowledge, it helps, right?
i mean, everyone is searching for 'the ultimate model'
I'm not sure
I have to compare the models more
a lot of the really cute models struggle with hands
I also need to try that hands lycon eventually
alfahentai 1.1 is REALLY good with hands
and i hate i couldn't fix it in darkalfa, but the general knowledge increase is just incredible, and while hands can be fixed with lora's, compositional knowledge can't
same img in darkalfa
darkalfa has the same issue alfahentai had, it's obsessed with stacking stuff 😄
don't EVER ask it too do doors lol
i mean, you aks for a hallway with doors... you WILL GET DOORS
lol I can't post the akkaimix version
too lewd?
imma need that in DM tho ❤️
for science
she's just showing her... goods 😄
bent forward, bent over too strong 😂
aight, darkalfa lesson. don't prompt showing her (x)
doesn't matter what. she's showing alright xD
goddammit @stable crane now you got me generating more heat into my room again!
DO NOT LOOK AT THE FEET ON THE RIGHT
@stable crane how do you like the assets for my android waifu?
ooo nice
darkalfa is so good fr
thx ❤️
BUT. BUUUUT... when it screws up... ITS EVEN BETTER. love the failed generations!
🍑
does your OG have a name @stable crane
yeah it's Lotta 😂
originally it was a random result from some other prompt
then I refined it
somehow was drawn to that style
On the whole art mascot thing, I can relate
I need to create something like that for me
just like @gloomy scroll's girl. i love seeing your brand characters ^^


rainbow hair, best hair
the model wants what the model wants
aint that a mood
impressed that we can do rainbow hair now
I gave the supermerger thingy a try because of that screenshot you sent yesterday
that's a pretty fucking quick generation there my man
I'm addicted 😔
I don't post right after generating, I make the set, save then post it :p
oh nice lol i havent tried the blockweigjts myself yet. got anything cool made witj it?
I WISH my hardware was that fast
ah, you had it saved up?
3060 go zoom
I tried merging my merge with dark sushi
oh yeah very cute @gloomy scroll
lowered the face part to the max because I hate how dark sushi does faces
I'm kind of happy with the results
Usually don't pay attention to discord or anything while generating and working on the art. Kinda zoning out while I'm on it. So finishing the whole set I'm on and then post in one batch is just easier for me
shit... i only got like 80000 images saved... i might need to cull... i might have an addiction
although, i generate less nowadays
we all do
lets see
I have only few saved images
what lora did you use to get this style? it reminds me of twitch emotes lol
oh shit, this is a god combination!
But then, I also don't really generate much quantity now and tend to keep working on one image until i'm happy enough with it, which can take a while
filthy casual!
Testing some old prompts on it rn
old prompts testing is awesome!
I have only bought Skyrim its own M.2
Man, I feel inadequate now, the stuff I actually felt is worth saving isn't even 1GB currently xD
ty, could i have the parameters for that?
I don't even save stuff
I just post it here and delete it afterwards
I only save the prompts/configs
you'll get there... 😄
ty
I save only the ones that I feel are good/interesting enough, which means the ones that went through refinement / manual editing and the whole stuff
so cute
I wish my model could make this 
convert to darkalfa 😄
I think by now I spend as much time in CSP as I do in txt2img and even more in img2img
Now that I made my private discord server I could probably remake all of these with same style and then upload some emotes
doesn't private discord need 2 boosts for emote usage?
no?
boosting only increases the amount of emoji a server can have
I believe non boosted server have like 50 emotes
😮
and 5 stickers
Yeah, private server can easily hold enough for personal use
just make more private servers!
^
but honestly, if you want your own model/mix to do stuff like that, make sure to do a add difference mix which knows the things you want to do, and stuff you want to remove
I'd actually have to make any emojis first, 50 is more than enough for me xD
did you prompt "mommy"? XD
The entire supermerger interface confuses the shit out of me
Nope
this is a fart... isn't it?
I can't unsee it now
looking at it again... i'm remineded of fern gully's Hexus
Damn, really like this art style
Looking through, turns out I lied, I do have some stuff that only had a tad of img2img stuff saved
that's img2img?!
txt2img -> img2img inpaint -> img2img sketch -> img2img multidiffusion upscale
what's the difference between ultimate sd upscale script and control net tile mode?
in img2img
Was impossible to make the ears and horns work without inpainting
controlnet tile can be used without img2img
ultimate sd upscale => "just a method to prevent OOM"
and it's pretty damn good to be fair
Do you happen to know how much vram you need to sdxl inference?
like is it even out yet
the ti was just 100$ more in my case
still, got a rtx3080, 10gb , collecting dust. takers?!
@wispy canopy Can you fill me in, been away for a long time, since new poe league came out and ff16
anything interesting?
I saw drag gan was out
poe league => again more standard league plus minimal league mechanic
yeah the mechanic itself was awful
has a few specific FOMO mods on the league mechanic, other than that, just standard league
(exploding totems broken)
I was hoping you would fill me in on AI happenings not poe 😦
tbf not a lot has happened ( I think)
all I saw is sdxl being showcased and a couple more updates to uh controlnet
and about AI happenings -> not a lot
Also tensorrt is pretty fast
as in, nearly nothing
@wispy canopy what's your fav model currently?
i think controlnet killed it
my own, darkalfa!
it has a very good grasp of general knowledge and styles
I'll check it out
i've tried some of my dumb prompts on other models, but they kinda fail... (either in compositional knowledge, or just not being able to put down what i want)
the only thing it doesn't respect is booru prompts.
What's the base?
uhh... the base would be alfamix 1.0
I stopped using masterpiece completely. I'm a changed man
I did and it's pretty amazing, has limited res options and tokens
this.
the res options and tokens
i REALLY hated losing the flexibility
Oh and biggest con is no lora support on the fly
have to compile every time with new lora
didn't you have to bake in lora's in compile?
this.
but i've been so much playing with LLM's
LLM's are just as much (if not even more) fun than SD
i'm honestly REALLY bummed stableLM only released a single model
i mean, for as long as it lasted, theirs was MILES ahead of the competition at the time
but LLM models is cutthroat wild west evolution territory
llm is giving me some really weird prompts
@wispy canopy I see your model really likes g cup sized breasts 👀
darkalfa IS an ex-hentai model

you just need to feed it the stuff of dreams!
nvm, god bless llm
I like this one
honestly, best LLM character is @wind girder's stablecat short
i've done a LOT of attempt of making image generator characters, but his stablecat short is honestly the best one i've used
if you need the char yaml, gimme howl
It gave me this earlier, but it was the lora that was fucking everything up
bonus hand and fingers
I'll try with controlnet now
ohh better
@wispy canopy did you add any specific characters to the model?
none
Yeah I've noticed that it can't do megumin very well, like barely
usually all the models can do megumin
Basically, the LESS characters it can do, the better
having general knowledge is better
characters can be done by lora
yeah agreed, was wondering what you were going for
general knowledge can ASSIST lora, AND improve image quality at the same time
so, i just tried mixing so i'd kill of all the unneeded stuff, trying to add to general knowledge
i do have ideas for training, but i'm not sure if i could pull it off
not to mention, having to manually tag all the bullshit when SDXL is around the corner
general knowledge/reasoning in (SD or LLM) model is just stronger than ANY aspect the model MIGHT have
if i'd go train model, it'd be about general knowledge, so it KNOWS how stuff relates
i'd love it AND hate it at the same time -> right now my SD prompting is relying on glitching the prompting
well, when it gets 'conflicting prompting elements' which do not hinder each other
The way I glitched mine is by giving it high weighting on tokens (between 1.3-1.7) and just giving a LOT of tokens
then it would start creating interesting stuff unrelated to the prompt
that's making sure it glitches! interesting stuff happens
yeah
although darkalfa hates high CFG/high weights, so usually 1.2 is enough
but using my verser script is best option tbh, i've had the most creative stuff happen with that one
@wispy canopy I gave chatgpt a prompt to generated a long list of tokens to switch every step
it generates something like this for the related prompt:
(Cute|Scene|Girls|Hug|Heartwarming|Adorable|Affectionate|Joy|Smiles|Happiness|Precious|Tender|Soft colors|Gentle lighting|Serene atmosphere|Sweetness|Innocence|Connection|Bond|Friendship|Love|Human connections|Warmth).
in A1111 it's nothing, so you'd ither have to do {2-5$$and then the stuff you mentioned} or [the stuff you mentioned]
I think the | just alternates the token with each step no?
not with ()
alternating tokens is []
random token from list (using dynamic prompting extension) is {}
oh yeah those are put inside of it, () is just of emphasis
prompt?
?
I meant the actual prompt.. sorry for that
girl, rainbow hair bleding into universe, cute, abstract surrealism, ink watercolor art, universe background, ethereal ghost, clouds, flawless goddess
this one?
yes thank you!
do note my spelling error, blending was spelled wrong
ye, I fixed it
working towards your ideal prompt output feels satisfying
chat died 😦
F to pay respects to chat
true and real
F

Will always stay alive in my heart
i want this done with sdxl ree
Still in the testing/fiddling stage before refinement but I think on a good way for the first go on something with a wider aspect ratio
step away for 30 minutes and chat is already fucking dead
i wanted to start using stable diffusion again, but it just said "nah, but you can for sure have this error!"
please 😢 i need help
Is there a way to generate with transparent backgrounds? I wana try making some sprites
I usually use "white background" in prompt and then remove it by hand or use the remove background feature under the extras tab
Is that the "PBRemTools" I see in the extensions list?
yea, that's an extension and not vanilla, I'll download the couple that I saw.
PBRemTools is the better one. The other one is just ok.
Open up a Cmd
Type: Pip cache purge
Hit enter.
Then delete the venv folder.
Restart the webui-user.bat
nice, i need to reinstall the fkin packages AGAIN
everytime i start that shitty thing up something goes kaboom, and i need to reinstall the packages
nice
not on my line tho, dont have the motivation anymore
Does anyone know which model this is?
I gave up relying on the environment that a1111 creates. I started using conda to manage the python environment, and when something breaks because of some extensions, it's way easier to fix, and the packages are cached
Today, it was adetailer that downgraded torch and I wasn't able to go beyond 768x768 anymore
conda is as shit as the env, may be better but HOLY FUCK! i am currently using it to install some fkin packages for sillytavern extras and its NOT FUN
The advantage is that the breakage is limited to that specific environment
its SHIT to control, one fukin ssh error out of nowhere?! then my env cannot be found WTF MAN
All settings and masking options lead to errors, I think I'll just continue doing it manually.
I think I've had my venv break in A1111 like 1 time in all the time I've used WebUI that wasn't my own fault or some broken extension. Vlad broke something almost every other time I launched it on the other hand.
yeah, vlad breaks EVERY time
that's why i kinda stopped usign it
sure, it has new(er) tech, but having it break like every other update, just not worth it
well i use vlad
i am not going to switch AGAIN, i cant do this anymore, its jsut getting on my nerves so bad until to the point that i wont use stable diffusion anymore(what happend in the last months, thats why i havent been uploading at all here)
as anyone of us
that's why i'm sticking wiht A1111 for now, vlad is just TOO unstable
With this, imma go to sleep, gotta work tomorrow too 😄
i habe vlad and auto1111 symlinked for models so i doesnt clog up space lpl
HAHAHAHA go ahead if you want

\
Trying some sprites/paperdolls out.
venv is still like 6GB for each
if you have 100+ gb of models like i do its nice anways lol
I wouldn't know anything about that https://cdn.discordapp.com/emojis/997241451555590164.gif?size=48&name=ChenEyes&quality=lossless
Digital hoarding.
i hope you symlinked
When my LoRAs take up basically the same space as your models https://cdn.discordapp.com/emojis/956601970733490197.gif?size=48&name=EikiScare&quality=lossless
behold!
I have one chance today 🙏
Upscale + ControlNet
I do a little hoarding
Lol, yes I symlinked for both Vlad and the UX fork.
Edit: Oh yeah, comfy is symlinked too.
good lad
What are your upscale settings and how much VRAM do you have? You can probably make some change without effecting quality to get that below your VRAM limit so it isn't swapping and speed it up a ton.
I want to change the image, but it's not possible with plain img2img because of the size
And inpainting the image is giving problematic results
after some more work and refining I'm pretty happy with the result
@drowsy dove No, I'm talking about changing things like latent tile size and batch size or tiled vae sizes. You'll still get the changes you want. The problem you're having now is that it is swapping to system ram because you don't have enough VRAM for your settings (maybe even pagefile based on how slow it's going).
I interrupted the process. Gladly I did, I would lose 2 hours
The problem is using ControlNet. Without ControlNet, the same img2img takes 10 minutes instead
And 768x768 seems to be the biggest tile usable (I was using 512x512)
Yeah, controlnet takes extra vram. So I take it you're using SD Upscale or Ultimate SD Upscale currently?
Yes, I'm using Ultimate SD Upscale
Does the ultimate upscale provide any benefit that the default one doesn't? I tried it and I didn't notice anything.
I'm running a few tests to check vram usage for you before I recommend specific settings.
Try using Multi-Diffusion/Tiled Diffusion instead of Ultimate SD Upscale. With these settings it used only 5117.592 MB of VRAM for me. Confirm that these work well for you, if you still have VRAM headroom you can experiment with increasing Latent tile Batch size 1 step at a time to find the limit for your system.
The main benefit of Ultimate SD Upscale over SD upscale is that it can be set to go over the seams between the tiles to further reduce the chance of visible seams. It allows slightly higher denoising because of this. I've almost completely switched to Multi-Diffusion upscaling instead at this point. I get better results with virtually no chance of visible seams.
Also gonna throw my hat in the ring for Multidiffusion upscale, never used anything else after trying it out
The ControlNet option must be that one? I ask because I would like to use another ControlNet option (so I would have to add another unit)
Other ones should work fine too, I just prefer the tile model for upscaling.
It seems to be working, but this Tiled VAE completely messes up the terminal output
Yeah, terminal outputs get a bit messy with tiled VAE enabled, but it's worth it for the vram savings. I've upscaled to 16K x 8K with it.
Tiled vae is the way
It works! And took just 11 minutes, instead of 120
Ok, it seems I misunderstood which of the morbillion "upscaling" related options Ultimate SD was.
It's an extension that works similar to the SD Upscale script that's included with WebUI, but it adds more settings.
I tried it out, multi diffusion came out a bit wonky, but I need to play with the settings
not that I know what any of the settings even mean.
I love that the spear turned into a person on the third image lmao
I'm going to try and 4x multidiffusion while I go walk a trail, it will probably be finished by the time I get back.
Ignore all of the noise inversion stuff, it will usually just make your images worse. Method is just two different methods for the tiling to be done. I like Mixture of Diffusers for upscaling, but they're both pretty similar. Latent tile width/height/overlap are just the size of tiles and how much they overlap (This isn't important to know but all of these meauring latents, not pixels which means you would multiply them by 8 to get the pixels size equivalents.) Usually between 96-128 for the size dimensions will be safe, and I like 32 for overlap. Latent tile batch size is how many tiles will be processed simultaneously. Upscaler is just the upscaler you're going to use before starting the img2img process and Scale factor is how much larger you want the final image to be.
Where do I get additional upscalers?
that was the multidiffusion one, my Denoising strength of 0.4 was probably too high.
Be sure to check the architecture of the ones you download so you can place them in the appropriate folders inside the ./stable-diffusion-webui/models folder ESRGAN and Real-ESRGAN are not the same thing.
Yeah, I usually start getting some artifacts above 0.38 denoising. It can vary based on the sampler though. DPM++ 2S a and DPM++ SDE both need lower denoising. I use 0.26 denoising for those two.
While I'm away you should see what what I've made by upscaling through inpainting. https://catbox.moe/c/9q1x8v
These are quite nice! I've seem a lot of them in here already, but they're all worth looking at again
What can I do to make the hand holding the ball OK? It's so close of being good 
tried with controlnet canny?
Not yet, just inpaint
i suggest you controlNet, is very useful
also the badhandv4 embedding can be helpful
I've had better luck with this negative embed for hands https://civitai.com/models/56519?modelVersionId=60938
Other than the days I installed the original Stable Diffusion, the basujindal fork and automatic1111, I think I never did so many changes on my install as today
I think the hand is good now
seems a bit smudgey
what are you using?
whyyyyyyyyy
here's your problem
The biggest issue I see is the images are 1920x1088 pixels but have 512x288 detail. Whatever upscaling method you are using doesn't redraw anything.
i should try a different workshop then
At least the upside of braving the Civitai horrors is that I've found a new oil painting LoRA to try out.
I still can't tell the difference between Shoeonhead and Boxxy
Hmm... I *think * that's Shoe? That's my first impression anyway.
It's shoe, a boxxy gif would be 144p
I'm not sure that's what people mean when they ask for twintails.
Chen has two tails
Who's that?

Ahh. With all of my anime knowledge, I know nothing of Touhou except that it has a planet-sized following.
I wonder If I should start making my main OC, or continue making muscle girls.
sadly only 2/4 of my OC's are muscle girls.
you have a oc?
4 of them actually! all exist in the same story I have in my head.
honestly one of my all time favorite generations
I just thought of something. What am I supposed to do with all my stuff when an anime-trained SDXL model comes out? 
you need more diskspace 😄
sʜᴜᴅᴅᴜᴘ
ɪ'ᴍ ᴀʟʀᴇᴀᴅʏ ᴜsɪɴɢ ᴀʟʟ ᴏғ ᴍʏ ᴍᴏᴛʜᴇʀʙᴏᴀʀᴅ ᴘᴏʀᴛs
Also, I really love this style.
that's pressing my random button 😄
would've wanted to say, lot of skill involved... but well, the skill involved was creating my random prompt like ... waaaay back lol
I doubt you'll need to worry about that any time soon.
We'll see. I hope I'm wrong.
@untold glacier I just keep bashing the generate button, until interesting stuff happens 😄
trying to flesh out outpainting n get better with it
Yeah. Emad has overpromised before (see: the **entirety **of 2.x), but this looks pretty good thus far. We'll see indeed.
LoRAs in your wildcards?


brain is gonna blow up thats too many prompts for me to account for😭
bruh do 90% of those negative prompts even do anything?
If you haven't already seen it. I wrote a quick guide for outpainting that's in the pins in this channel.
yes 🙂

that's the v1.1
so don't get your hopes up

i just took a look i might give it a try tmrw ive been using openoutpaint for mine
Then it's even more versatile than I've tried. I need to start being more inventive with my prompting.
HANDS... HANDS
that's what i kinda did try mixing the model for 🙂
Yeah, the artists are probably having a big effect. I need to experiment more with those again in anime models.
ooo this is cool
I'm still impressed by the black backgrounds now that we have noise offset.
My controlnet method allows the model to get better context from the image and also allows using hires fix with it. You can also add a second controlnet using Reference only to help match styles and image contents even better.
they do, that's the influence of the generic models
yea the issue ive mostly had with openoutpaint is the long render times and it struggling to get the context of the scene
ok lets see how long it take for me to upscale 4x using ultimate SD
assuming it even lets me
yall mad creative im stuck doing anime portraits💀
I made a health potion
Nothing wrong with those. We're mad from already have made so many of those, lol.
Why does it look like what she's sitting on may burst at any moment. Is she really that heavy? 
im still relatively new so it hasnt bothered me too much yet but ive tried stepping out of portraits n doing more full body n stuff alike but the end results just end up being awful😭
They look terrible for everyone, you just have to use inpainting/upscaling/ and some negative embeddings
Half of the battle is just knowing your model. How to use language in such a way that works best with what it does.
let it all be awful ^^ just have fun while doing it, and let us enjoy the stuff that comes out too
The other half is just experience. Knowing how to word things, what the settings should be at, what extensions to use, etc. It'll come with time.
For example (masterpiece:1.3) is just going to add a fancy painting in the background
thats so true i finally created a good prompt structure for cetusmix n ive been spamming it ever since😭
Unless you're using a booru-language based model like Aurora, heh. As for something like DarkAlfa, sure it'll pretty much not help in any great way.
what "universal" negative embeddings should I have? the only ones i use atm are easynegative and badhands
that's a good start 🙂
Yep. Good start. I can also recommend Auroranegative as an alternative sometimes to easynegative. That and I've seen some good things about deepnegative a month or so ago. I use Auroranegative all the time though.
I also stopped using badhandsv4 a while ago, now I use this #🍥|anime message
yea badhands helps sometimes but its still a hit or miss from my time using it
True. To be fair, these negative hand embeds don't really do what most people think that they do. It's very difficult to gen hands to begin with, so the embeds can't help too much in that regard.
The best thing that helps hands and other small detail cohesion is a good model.
Uh oh. Nubby is typing. I may have made an error. 
They just improve your odds of getting good hands. What I don't like about badhandsv4 is how much it effects the style of the image. I don't want a negative embedding that's meant to help with hands to also influence the art style.
😮💨
eh.. scuffed images, bad hands... if it happens, it happens 🙂
Man, I hate how blurry discord makes that. The original is much clearer.
Lol
Is your artist wildcard custom?
actually, it is 😮
is that an actual bubble-butt?
scuffed face, too lazy to turn on adetailer
Thanks for the new reaction face.
Explains why I had a terrible experience with Aurora/kenshi, I was prompting with it as if it was Dreamshaper.
yeah, some models respond really bad or good on prompting styles
Yep, Kenshi does pretty well with a decent amount of natural language mixed in with booru prompts, Aurora does best with very little natural language.
Yeah, Eface really hates masterpiece, lol.
ᴛʜᴇʀᴇ ᴀʀᴇ ᴍᴜʟᴛɪᴘʟᴇ ᴘᴀɢᴇs ᴏғ ᴛʜᴀᴛ ʙᴛᴡ

it just causes big boobs, well... more than i already have in images lol
yea, I think I'll skip both ultimate SD and Multidiffusion for now and just continue doing normal inpainting.
terrible results
Use slightly lower denoising and change the prompt to only quality and style tags. No character or background description for the Multidiffusion/SD Upscale prompt.
I'll try, but these will take probably 20 min each time I want to test a variable.
6gb vram and all that
Yeah, I getcha.
I could spend that time making muscle girls instead.
I love Castfbblanmmn! Such a great movie by Bellelenamalmmin.
I think it's actually spelled Shyamalan, though. 
The settings I used here used less than 6gb even with a controlnet enabled. Depending on how much is being consumed by the rest of your system is using, you could reduce latent batch size to 1 to save a bit more VRAM at the cost of some speed. #🍥|anime message
Still has five fingers, so I guess it is technically a win
Your yesterday prompt is working great with revAnimated. Finally trying out new stuff now that I've stopped myself from downloading more.
Meinamix
img2img is cheating
Controlnet is cheating. i2i is a good start. 😛
my openpose control type doesn't even work anymore lol
canny still works though
Yeah, it's been a couple months since I've had Controlnet installed and working. I need to get back on that eventually. Canny is fantastic.
She's sleeping
Gnite 
Night
Nighty night
Does anyone here potentially know what lora or prompting was used to get arms (and legs) like this? Sadly no way to post the legs as well without it getting nsfw. Image has no metadata and as it was shared along multiple corners I also have no idea who the creator is.
Maybe a model merge of a normal model with a furry model
can anyone share their workspace i need one fo upscaling images
For some reason this gives me some vibe of Saber.
how can i improve?
Improve what
Just use hires fix, or a different model
Wouldn't hurt to try tweaking your prompt too, throw in some negative embeds to help with the multiple arms and wonky eyes
The adetailer extension would help clear up the faces a lot too
aDetailer
okay
i think i should use a different template for the work space
i think i need another template
^^
Im still scared of your pfp
It's perfect
same prompt,different ENSD 😄
wtf
O_O oh wait .. it reloaded the checkpoint to SD 2.1 instead of the NineKey i had set :/
INK INVASION!
Ink break
Looks like you might be using an untrained VAE. Are you using Automatic1111's WebUI?
Just gotta update y'all that vsync got banned from this server
???
Yes Automatic1111, with Aurora as you told me. What's a vae and how do i train it?
Wdym by untrained?
Like vannila stable diffusion vae?
Do you know why?
No need to train it yourself. The VAE is the part of the model that turns numbers into an image. It influences textures and colors mostly. You can grab one of these two VAEs. https://huggingface.co/nubby/kl-f8-anime2-blessed/tree/main
Put the .ckpt file in ./stable-diffusion-webui/models/VAE. Then go to Settings tab > User Interface > Quicksettings list and add sd_vae to the list. Apply and reload the UI. This should give you a dropdown list at the top of your WebUI that is labeled SD VAE. You may have to click the blue refresh button to the right of that dropdown before you'll see the VAE I had you download in the list, select that and be amazed at the improvement of the quality and colors of your images. The difference between the two VAEs I linked is the contrast level, 08 is lower. 09 is what I personally prefer.
He reached out to me so I would update y'all
So I asked what happened
He said he is not a pedo and was just trying to debate logic
Yeah. I think it might actually be one of the approx VAEs that Auto uses for cheaper live previews that gets loaded on some of the models with pruned VAEs.
Makes sense

RIP
damn it's sooo much better, thank you!
Sometimes, it's just not worth it. Thankfully though, unbeknownst to us. Those of us who are Final Fantasy 14 players, had an augmentation installed in us. An AI that has spread a synthetic cybernetic webbing across our brain and body. So surely' we'll be safe
Anybody else want some chocolate milk?
hmmm does the mp3 file work for everyone?
works for me
ok not what I was going for but.. and I think she killed something/someone?
hmm interesting
Dude, you're driving me insane, lol.
huh?
Hmm
Hmmmmmmmmmmmmmmm
Yeah, thanks!
Why your pfps same character?
consistency is the exact opposite of random lol.
so i'm consistantly randome?
Been my profile for years, I figured I'd keep it the same char. That and I'm a musician so it's a good fit.
Why this character specifically
so... me driving you insane, good or bad? or .. i'm confused
Kagamine Len, vocaloid.
Oh

ok, it's that wierd hyper network I have in there.. removes
almost did it again
caught myself. bad habit
took me a min to catch on what ya meant. sorry
oh wow
ok back to anime 😄
I would watch this anime
...
looks like your model has no clue who Lucy is. Still it's a neat gen.
this is prob why Unhinged Drawing:1.1) of (Screen print:1.05),(Surprising:0.85) crazy style, faded poster paper, Manga panels, chat bubbles, Detailed background, background: fashion stage in japan, cartoonish line art, full body portrait, Anglo girl, standing, medium view, frilly goth dress, power puff, one color, drawn, outlined. sketched, colored-graphite-chalk, poster, art by makoto shinkai, showing victory with fingers,(by Artist Yoji Shinkawa:0.7),(Glitch Art:0.2),(Technicolor:0.9)
Oh, I assumed you ran my prompt
oof that face













