#🍥|anime
1 messages · Page 37 of 1
so basically your generation will be 1 - with easynegative, 2 - without easynegative
how does one do this
x type change to prompt S/R
then copy the trigger word for your easynegative embedding to the x values
and put easy negative on x values?
yup
anything else i need to change?
something like that
so sweater is your easynegative
then add ,
with a space after that
that's important
oki lemme do that real quick
ahhh, that's why there's 4 generation each batch
change it to 1?
if you change it to 1, it will be easier to compare yes
also i love my new model i just trained a custom style lora and merged it with a few models and now its this awesomeness

well, there's a lot of negative embedding
so u know how people have like
multiple comperassions?
how do u do that
cuz this is just one what if i wanna see different strengths of easynegative
use something like this
thank you!
@wheat falcon Stable Star?
im impressed with how good it is
one thing i really struggled with this prompt is the day time always being morning
took a lot of gens to get a night time one
but this is actually nice
now im gonna try with the animepunk lora i made
pretty useful i never used this tool until now
and then end up with 20000 images I have no use for that just sit on my drive taking space 😛
"oh wow this one is great, I'm going to come back to this one and make a cool wallpaper"
... 3 months passes....
Motion photo. Nice.
I tried those but was not very good at it. 😄
PhotoMirage I think I tried? something like that
its my fave thing to do with ai photos
those newfangled moving pictures... :3
I generally just end up going back to generating Tifas and trying to make widescreen wallpapers 😛
also realized the comperasion for my lora there wasnt fair cuz i merged my lora into this new mix
so its gonna look burnt on that ckpt
XD
fair!
i have been having no luck with outpaint tools
only with infinite zoom
Right now I'm working on a submission for the pow thing.
I never have luck with outpainting either
Thats nice, what did you use?
oh yeah the pixel art
I usually am just generating at lower res and doing a hires fix or img2img thing. Sometimes with inpainting or sd upscale. Sometimes I'll merge two images.
Awesome thanks^^
motion leap is mobile only right?
Her expression... 👌
oh ya thats a 10/10
my model can do a pretty good yandere
I wonder if my model can make that expression
never tried it
go for it!
What prompt you use to get the expression?
my model has the crazy expressions lora merged in as well as my own anime punk lora
so it has plenty of yandere reference to go off of
original character , 1girl, (crazy smile:1.2) , short purple hair , bangs , crazy eyes , hands on face , blood splatter on face , red and black color scheme, open mouth, (wide-eyed:1.2), glowing eyes
neg : (easynegative:1.2) , (Worst Quality, Low Quality:1.4), Poorly Made Bad 3D, Lousy Bad Realistic , bad hands , tangled fingers , extra fingers , sunlight , day time, combined eyelashes , bad hair , bad lines , bad knife, colorful , nose , missing fingers
I'll avoid the hands on face. I'm scared of what kind of abomination I might get xD
i always make sure to add realism models in cuz it makes hands better always
but theres the better hand lora u can merge into your own model
Got this... I'm disappointed 😔
THATS SO CUTE THO
crazy women ❤️
yiis its why i merged it in my own model
She need to wipe that ketchup off her face

Got this with my other merge.
It's an improvement but still not quite.
i wonder if the evilness will improve with my anime punk lora

I find it weird because I merged it with king crimson
idk that model
ooooh oki
hmm idk
try to increase the offset on the yandere prompts
This other mege also can't
this is better tho!
This merge is more focused on the cuter side
I feel like the second was the closest I could get
Haven't tried the crazy lora yet
This lora is perfection.
idk what i pressed help ><
Im surprised its still useful these days lol
probably the image looking icon under generate
my prompts are rght to left
that lora is the only lora that actually works on all models that i've tried
Yes, months ago
its really good!!! i hope its okay that i made a merge with it
i heard yandere
generated an entire concept work there wth haha
https://civitai.com/models/52346?modelVersionId=56800 my animepunk model is out\
guys I wanna make a model that only outputs this image. what should I name it (gonna make whole fake huggingface page to prank people)
I decided on anything v7
All in one model
unless you have better ideas
I think maybe uploading literal garbage to HF just to prank people might not be as funny as you think. Personally, I appreciate that they offer their hosting free of charge and wouldn't want to encourage them not to do that.
call it luckyroll
people will think its a gacha
but really its a GOTCHA
anything v7 mworks
is it not?
Consider for a moment, HF models basically only spread through word of mouth. If it isn't a good model, nobody is going to tell other people to download it. Maybe you get a couple dozen people to download it at the cost of a few GB of storage on HF's servers. Was that really worth the electricity it took to train it and the the storage space it took up on HF's server? It would be far funnier to train something that's actually good, but include the rickroll on a commonly used tag within it, something like masterpiece
Also, now that you've talked about it both here and toohoo server, where do you plan to try to spread it? A prank becomes infinitely less funny when you tell the targets that you're going to prank them.
nah, it's meant for more standard ai art users
like normal civitai users and such
didn't really think where I could spread it though
for 1 model, no but if everyone starts doing it, then it becomes a problem that they have to address.
oh
I see
good point
I really wanna do it though. is there any way I could do it without uploading to hf?
can't civitai, it's against tos to use images not representative of the model or something
like google drive?
You could throw it on a g-drive, but then you just look like the sketchy dude in a long trench coat peddling .dangertensors files in a back alley.
yeah
I mean I might be able to convince some with it being safetensors
but still
what about github?
github not non-profit right?
or is it?
Not sure if their non profit or not, but it's the same situation as HF IMO. They're offering free hosting to us and intentionally abusing that isn't very nice or responsible.
I care less for github than huggingface
though yeah
I mean, g-drive basically the same no?
pants
tights
thighhighs
skirt
Not really because you're using your own limited quota of storage space that google provides you in exchange for harvesting your soul data
nah not pants
thighhighs
idk man
sun ray
what
let me show you
the zaddiness of my model is 
You could try adding robes to the prompt. it may help to encourage her legs to be covered more, though if you have any mention of thighs in the prompt you'll be fighting against that.
no i put that in negative
i will try robes
Sea Empress
found out my model is pretty good at multiple subkects
i think its adorable the one giant hat that fits 2
out of surgery since 8 hours, 6 hour surgery. Attending nurse said everything went according to plan. Doctors wont stop by until tomorrow but that pretty much means it went well caues otherwise they dont wait with the bad news. Theyre not that cruel....Thank fuck
I can stop running from the giant boulder now, I found the gap
were you trying to get sun beams to cover there?
robes didnt work
now I just have to learn to walk well enough that theyll send me home...ooh and poop
yes they want you to poop
you don't get to go home until you poop
"Waiting to void"
in my case its cause they just took a substantial part of my small intestine
With any luck, you'll have a "hat" in the bathroom to use.
so the GI tract goes into recovery mode and wont shit until its well enough
hat? what fucking hat?
You don't know about the toilet hat?
how do i get the stable defusion ai?
Hi, guys i'm noob about IA, i need some help about many things, can someone help me?
Probably, just post the questions and if someone can they will.
I'd recommend checking out a YouTube video about the various "repos" you can install for Stable Diffusion, along with the hardware requirements.
Then you can pick what works for you.
ok so about the textual invesion for negative prompt with one are always used, i use this: bad-artist bad-hands-5 bad-image-v2-39000 bad_prompt_version2 bad_quality, but where can i see the version of these?
Embedding names often contain version number
Yea its a temporary Team role i got 🙂
can I also get temporary team role?
@finite herald can I also get Miku role?
when miku role back?
Sadly its not back
I am asking when it will be back (it will be)
When we succeed bribing Fruit
ok so, i can find the version of hugging face just typing the name of TI and then if they are old i replace it. Right?
Yeah but if you searching new versions for those negative embeddings you shared I think all of those are newest
Ofc there is some other newer negative embeddings that wasn't on that list of yours
Tried something
And I was wondering why civitai search didn't work 
I was about to make the perfect prank today on huggingface, but nubby came and used facts and logic to convince me against it because it is not the most ethical to do that. now I am bummed.
that is why you should bring back the miku role.
The prank model idea ? xD
I already have the model man
I just know it will be unethical because bwm decided to ruin it for me
What does the model do ?
I will show you, just give me a prompt
1girl, armour, copper, choker, ribbon, blonde hair, long hair, straight hair, samurai sword, reflection, dynamic light, engraving
here you go
(should have trained for more epochs)
I also made a model for this
Ive seen it on civitai it looked really sketchy xD
it got taken down lol
I uploaded it
at the start I forgot it was ckpt so everyone called it a virus (even though civitai literally checks for pickles and stuff automatically)
then I uploaded safetensors
but people said preview image is clickbait
even though it wasn't
and that's the reason it was taken down
I will (apply to) become a community guide if you change the role color
ur too young 
nooooooooooooooo
fuck
ok then
I'll wait
18+?
Bro is desperate
only like 10 months left

4 LoRAs, all at 1 weight, 1 is hand LoCon, 1 is AsteroidIll style, 1 is Rain World scenery style (which is basically abstract art for the model), and then the offset Locon
why does it still look good
i've got some business
you should see what people be doing in projectai. imagine what you are doing but with also mbw inside prompt
then 10 extensions active at the same time, all doing minor stuff
<lora:chapu128dim-epoch-000009:1:1,0,0,0,0,1,0,0,1,0,1,0,0,0,0,0,0> <lora:mmix_v7:1:1,1,1,1,1,0,1,1,0,1,0,1,1,1,1,1,1> be like
Sounds like I should check that out 
I tried to do something myself but then I realized that I had no idea what I actually wanted to achieve so I gave up
Shes packin sum heat
they have a guide there
or something like that
Oh yeah I never actually checked that thread
why did I not get pinged when you sent that message?
maybe because it's like that?
m-me? 
well, yes. I'm using the darkKuKiv1 lora for dark lighting
but it's specifically made for holokuki model which im using right now, so might not work well with other model
I'll definitely try it out
Been looking for something like that lol
There we go, something dark for once lol
Actually tried this locon the first time, but yeah it doesn't quite work with anime stuff
ahhhh, then im glad you found this one
huggingface with its goldmines


What mean the number near the textual inversion?
she's gonna read a bedtime story
I would wanna listen to it 
omg the sky is sooo good
thx, yea thought so too!
really detailed love it
what model do you used?
used an own merge
ohhh i see
ohh poor thing





welcome to the community btw 


i´m still confused
yea welcome 😄
is this channel for any kind of anime or just Dream? Style
i dont know whats dreamstyle
okay good haha
i just started before some days using SD, its just amazing and i got so many good results already

yup, if your image is blocked by clyde then it's just that
even if there's no nsfw in it, it's still against the rule to bypass clyde
you can refer to here #🍥|anime message
ITS AMAZING
well, try passing it though img2img
and maybe something will change enough for clyde to accept?

then i guess you will have to post it somewhere else
sadge
there's probably a lot of skins right?
Blur it
what do you mean exaclty?
blur what?
like a lot of skins shown on the pic?
definitely wasn't ready to get jumpscared lol
ohh yeah its a kind of her kda outfit
then some other hehe
been trying to get a good image for pow for like... too long
Horror
uhhhh love it
did you drawed my sould? hehe

Easy negative is the only way I found to do a dark one easy
what do you call in neg promps to get those results?
I used an add on to select the hair on the original image (segmentation), then inpaint it. But you can see the edges of where the hair was. What do I do?
It's an embedding
A negative embedding called easy negative
U need to download it
It makes everything better no joke
Even if you are using a scuffed model it fixes everything
Original Hugging Face Repository Simply uploaded by me, all credit goes to https://huggingface.co/gsdf . Counterfeit-V2.5 (which has 2.0 as well) o...
Easynegative hasn't helped with darkness for me. The main thing people recommend is using a model or Lora with Offset Noise
This is what i usually see people recommend https://civitai.com/models/13941/epinoiseoffset
I do recommend using easynegative though, I use it in nearly everything
i just have the problem, if i load in Loras with additional network i have to many sometimes and they destroy my pictures
has anyone an idea how to fix that?
like using 3 loras in same time is around maximum
sadge but thank you for the information. I can just try to find a good balance right? Hmmm is it a better way then to create own lora´s if i wanna multiple kinds of lora expressions in one?
I mean, you could try that, but it does get complex
WELL
Also if you're in the WD 1.4 server
They know how to do some LORA block merging stuff
major horror
But I don't know which is worse, if the block merging for LoRAs or training your own
I use this as well! This is also wonderful
What webui u using ?
can you send me the link in private maybe? I will create my own one i already read about. If i´m going in love with something i sink in haha AI got me
SD Master
Ah ... I recommend a1111
isnt it A1111?
I mean, what does your WebUI look like?
i think its A1111
1s
Sure :o
i hope your folder isnt called stable-diffusion-webui-master
it is XD
rip xD
Looks like A1111
O, tru
Omg all these tabs
uhhm i working with the batches
?
Same tho 😭
thats just 1 browser XD
what would you recommend to fix?
and what do you mean with updates, which updates exactly?
updates of the webui, they come by git. your stuck on this version until you either reinstall SD or rename the folder and git clone the repo in it again
dw i know what you mean, i can handle it 😄
but thank you for the reminder
i will do updates tbh just manually
manuall by downloading it as zip again and override the other stuff ?
not the best way for sure
I mean, its easier to do git pull, takes like 5 seconds 
tbh i create a whole new directory for future updates
it is but i´m thinking like functionally, its kinda grown over years of writing code and working with unity that i´m doing this way
How much space do you even have, or how much bandwith, or how much disorganization

I mean, thats
Fair I suppose
Do you symlink your folders at least?
i have on this pc 10tb of storage and dl rate is around 40mb
I mean, thats a lot, but still really inefficient for storage space since each venv is like 5-6 GBs, thats a download you have to make and install everytime
hahaha well i have this
i dont care xD i just buy me another ssd then
I mean, it would be easier to symlink
Since for example if you got a new install in a new folder
It wouldn't detect LoRAs from the old folder
Or models
Or settings
Or extensions
Well, actually extensions you can't even symlink
Oh
tbh i will create different folders for different projects in future
I mean, if thats your plan
I really recommend you just do it with git
Cause
10tb is a lot
But
I have 400 GBs of LoRAs/Models
10tb wont last if you get to that point
XD
i got you
hmm i know thata problem, my still not finished own game takes alone 800gb of space XD
After all, you'll even have to optimize your game too
So everything needs to be optimized to be efficient in the end
yeah i know, the game itself has less gb, the project folder has like 200gb and the rendered game currently around 30gb thats okay
but i know what you mean
Yeee
but

i need money for GPU
Maybe you could save money through data efficiency 
not server pff i can use my 2 pc aswell but i need a new main pc at all so yeah
uhhhm

a NAS could help too 😄
I mean, it is a good deal, could take you like, 30 minutes maybe, and you'd be doing a pretty good thing for the future
ALSO
i have one XD
@wheat falcon What are "Stable Star"s? 
You have a Lil Nas™️? 
tbh, i´m really bad in this, or i´m just to lazy idk
I have one for 4 plates yee
but not running everytime yet
its more a off storage currently
wait you mean smth else

Fair I suppose
if you wanna help me and give some hints, i´m open for everything haha
a secret special agent role i got 😄 (sd team intern role)
I mean, I don't know the most, but symlink would probably be pretty useful, at least if you want to keep making more and more folders with "repeated" things
But otherwise like, would be good to just stick to one
Ohhh, can you stay as a perma intern? 
i´m not gonna do that because i´m losing the overview
Cause I mean, Stable Staff is cool
But you got a lil star next to your name
So it is cooler
the community guide or mod roles are not temporary so i guess yea 😄
I mean, then like, when you gonna get prompted into stable staff, intentionally slip on like a puddle and drop some important papers on said puddle, so you don't and you can keep the cool star, and just do that until you get bored of the lil star 
Big brain strats 
Cool lil star > Anything else
XD
but the stable star role is temporary 😮
Oh, I thought it wasn't
Thats sadge then 
Maybe you should start a like, staff poll to give important roles different colored stars 
That would be actually pretty cool tho
At least for mods and guides, since they look a bit like booster and member roles 
yea i liked the old mods role because of the blue sd logo it got
Ohhh, I didn't know about those, those would be pretty neat tho, would help differentiate, and help newcomers go like "OH! Hes got a fancy icon! He must be important and know whats up"
yea exactly, i will bring it up maybe we can get them back with different colors
I have a question,
if i mention that in prompts:
lora:morganaClassicLora_morganav1:0.8
and the screen, isnt it a doubled argument and what is he doing?
Yeee, I mean, I think differently colored stars would be pretty good, stars stand out, and could be colored to match like the color of the roles easily
What do you mean? 
looks right, first you select the lora type (in this case lora) and then the lora itself, then the weight
one time i set the weight in the additional netowkork and another time i set it up in the promps. Whats better or is thios effecting
but this cant be right?
Since I can open up the selection, get a look at the preview of what they do, and then apply them
Its in the extension's LoRA folder right?
uhhm no
i mean almost
its in here
F:\AI\stable-diffusion-webui-master\extensions\sd-webui-additional-networks\models\lora
you have to choose one lora option, you cant have boath
Well, the extension has a different LoRA folder to the built-in LoRA folder
there is the extension or the built in lora module
ahhhhhhh okay
True too
so if i use additional netowrk then i can ignore this: lora:morganaClassicLora_morganav1:0.8
You should like, just not use it
Since its not doing anything
Though if you're not using the Unet and text encoder weights of the extension
You should probably stick to the built in
Since its easier to use due to the selection gallery that lets you see preview images for the LoRAs
ohh damn then i just missed the mainfunction of it?
I mean, I'm not really sure how to mess with the unet and text enc weights, like I don't know how they affect the usage
So I just use the built in
yea put the loras in models/lora folder in then use the third button under the generate button to get to them
a new test fireworks1
new test fireworks1
ohhhhhh
uii thats kindda cool, i wanna start with that stuff upcoming weekend
You tap on the lil button here and all the LoRAs in \stable-diffusion-webui\models\Lora will show up
ohh thats way better 😮
And when you click on the thumbnails they get automatically added, or automatically removed if you already added them
whats the point on additional network then?
- you can set preview pics so you know what each one does
locon and lohas maybe
I mean, I guess the unet / text enc wights
whats that
other lora types
oh yea thats needed for sure
That just hijacks the built in LoRA system
ahhhhhhhhhh the one where like lora but not working with lora
So additional networks is kinda unnecessary now
not supported by built in lora
Hard Drives are OLD school
ASUS Hyper M.2 x16 Card V2 PCIe 3.0
And if you wanna get complex its better to just use the block weighted merge that the fancy bois at the WD discord do
But thats for real complex stuffs
The main reason that it was made is because the WebUI didn't natively support LoRAs at all when the extension was released. Additionally it allows specifying a separate weight for text encoder (basically how it interprets your prompt) and UNET (basically the part the dictates how the image will look)
Oh damn das a lotta stuff
Wait, why would you want to separate those two
i think linus techtips has the same one
yeah he did review this expansion slot, its really good, you can get 4xM2 2TB for like $150 if its on sale, no sale its like $200
wow a pci with m2 thats a fantastic idea
YOU GIVE ME NEW OPTION FOR UPGRADING HAHAHA
@magic wave will hate me more xD
XD
I use it for some loras that have especially bad, text encoders. The LoRAs that were extracted from the offset noise LoRA is the main one that I like it for. TE set to 0 with UNET set to anywhere from 1-2 looks better on those with an anime model.
thats awesome, but idk maybe i would buy a nas before
I see, thats fair then 
I've never messed with settings too much
Well, a bit
i gotcha i remember now why i started to use it XD bc of that, thanks for reminder
that will help solving my lora overloading problem
But atm my settings are like, 1e-4 lr, None Unet, and None Text Encoder
I tried uhhh, 5e-4, 5e-4, 1e-4, as recommended by distro, but it seemed to train too hard
Wait I just realize, 5e-4, does that mean its like... 0.0005?
exactly
It's just scientific notation.
I was like "ahah, funni number
"
Didn't pass my mind for a second that it was math xD
Same, It was the Moment i realise i Have to Go to bed My Brain has a Stack overflow xD
Fair XD
Thanks for all your help guys. Have a Good night.

Have a good nite 
oh is it good?
im too lazy to download that
Also seems like that file is just broken 
no idea i used it on huggingface's space
Oh, fair then
very wonky but its neat lol
https://www.youtube.com/watch?v=w41-MUfxIWo if you guys are interested in installing it locally
let me know if you try it tho

as with anything fun its prob cuda reliant thus barring me lol
5, 3, 3, 11 GBs of download 



This one is better XD https://github.com/JonathanFly/bark
He showed that one off in the video
But also
Yall are fools
THIS
is the best BARK
Nah, this is the best bark

nah gotta go with some aurora bark
That is some spooky bark tho, it has a hole 
naked tree
Tru
partially revealing bark tree 

o shit loks like
we were playing with bark earlier?
was busy with controlnet stuff
sounds a bit worse than vits imo
or tortoise
depends on the settings probably
nice noise offset
I'm not sure who's LoRA I used. It came from a bit repository of links on gitgud.
oh. i see
I did this one https://civitai.com/models/23972/moth-girls-lycoris
Downloading now so I can try it
thank 
Seems quite a bit less overfitted than the one that I was using. I've gotta go eat now, but I'll try tweaking the prompt to make it work even better later.
I gotta go sleep now
got like 2 hours of sleep
very sadf
that's what happens when you try to root your phone
at night
anyways
GN
good luck! 🙏
tthere's some prompt guide stuff posted on the model page if you didn't see
It can give some interesting results. My current avatar was made with it, and I generated this as the cover for a model mix I did https://imagecache.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/04ac4f29-2b56-4b66-ec87-8433010da500/width=1920/00003-3263854852-2.jpeg
am i too late?
oh hi linus
"i sent u my techtips plz response"
Congrats on graduating beyond leafdom.
It makes me happy to see how many leaves we've had lately.
🌿 🍃 🥬 🍁 🍂
also would you like a free tech tip?
i feel so humbled and honored to be no longer a leaf
now i can be my true, moth self 😌
anyways free tech tips?
What is even happening here?
Linus Tech Tips and idk
it's jus the tip(s)
I see
Anyways... back to the moth waifus
pretty 
moth.. interesting
We need some different stuff
Do a moth lady flying over a busy city street while eating a hotdog.
In June.
😳
Just tired of so much samey stuff. 😛
Well, here's some of the stuff I did today. Two of them are men, maybe that's different enough for you? 😛
how about big res moff girl?
ha
that's sorta different
at least not just another portrait
with a character holding a pin or with the hands on a ledge 😛
ha
that's sorta different
at least not just another portrait```
Sorry Omnia. Looks like what you said as a prompt isn't all that different after all...
haha
Do a moth lady flying over a busy city street while eating a hotdog. In June. Sorry Omnia. Looks like what you said as a prompt isn't all that different after all...
g'night
happy boy eating ice cream with her mom realistic images
Hmm. Not working for me. I'm sure you got this Nubby
There is no image generation bot here. Pretty much all of us generate locally on our own hardware.
gigabot seems to not work more often then it does work for me.
Doesn't like us I guess. Darn Clyde spreading ill words of us, lol.
@obtuse yew
Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝|tech-support for any problems you encounter while installing it locally!
Tq

i just realised might dai name is pretty literal
That would be my face too
I think this is the most active channel in this server right?
anyone good with latent couple? whenever i add and syntax is goes wonky af lol
Are you capitalizing AND? If yes then my next guess would be the area prompts conflicting with the alpha (base) prompt too much.
could you dumb that down abit for me pls

but yes both AND are capped
OH you mean the AND prompts are conflicting right?
Yeah, conflicting with the alpha prompt. Usually I've had that problem when I try to have too much detail in the alpha prompt.
mmm okay just messing around with the content of both applicable prompt sections and their weights seems to be what i need to do
I used this extension instead of latent couple specifically (they work similarly) and I got some decent results from it https://github.com/hako-mikan/sd-webui-regional-prompter
For that one specifically I did a common prompt with all the boilerplate and placement stuff, then one ADDCOL with the first character, then a bit with second character ending in "BREAK"
it really wanted me to end it in "BREAK" or it uh, broke the extension
so keep that in mind if you get errors 😅
kek this is considered nsfw by clydes standards but not cleavage or anythingis included, just two faces
smh
but
the upscaled went trhough?

latent couple and tile diffusion can either do hit or misses i find
deff a hiit in this case
How is this action or pose called
Pocky game
Could've sworn there was a lora for it on civitai

What's todays
?


the pyramid at the end of the world
hello there 👋
yandere peach
now i know my model can make japanese curry
yum
that's looking mighty tasty
i agree
yeah, people can do amazing things sometimes ^^
i wish i knew the inpainting model they used or alterntively how to make a good inpainting model
i've installed pytorch 2.0 my man i reduced by 50% the time of text2img
reddit is really a good place
using opt-sdp?
What gpu
15 sec with highres fix/ UltraSHARP x4 (resize: from 640x448 to 1280x896)
RTX 2060Ti
Hmm
I wonder if the change will be as significant on a 3070
went from 4it/s to 7.15 it/s
so for me its actually a huge gap i would say
Oh so i just drop it in like a normal extension install?
nope

\stable-diffusion-webui\scripts <===== run the commands in there
I see i see
type activates then your cd should be (venv) and then run first command, and second command when the first is done
If there’s any improvement on my end ig ill come back and report on it as a 3070 user
@vagrant falcon i just ended up getting a large dependency error on the first command
ignore every error
do the second command anyways and launch your webui
this command right?
yeah
I made this in 3d without any SD, like the webtoons. Then, used extracted lineart from a plug-in and fed it through controlnet, and added img2img to keep the color
It turned into this
that actually looks super nice, gj! 
I think webtoon creators can save a lot of time by incorporating it into their workflow. You can tell the rushed webtoons spam a lot of 3d assets without paintovers, but SD can do it automatically
they actually can feed their training with own work to gain a lot of time
and paint over if there is mistakes
The character in the image is 3d too, it uses a plug-in to make it get line-art and flat colors. And it can change expressions because it has blend-shapes and joints
The day of 1chapter per day is arriving
You can see in the background, the straight lines on the door. A telltale sign of 3d assets. I think a little bit of img2img+denoise can smooth out those problems. I am running a highres fix on this workflow but it takes a bit long, let's see what comes out
I was thinking this also. I think artists will be squeezed even more rather than being given more break. But who knows? Maybe it will give them more time to work on the story aspects
The workflow is through ComfyUI. It looks complicated but I basically copied off the example workflows on the official page. This way, once the workflow is set up, it just needs a few inputs,. Having a few of them should save extra time
Did too high res and couldn't do the highres fix. So I changed the render res on the 3d but it made the lineart so faint and no idea why. I guess that's the problem with longer pipeline workflow; more moving parts means more chance of problems... But, this is an encouraging start.
It didn't work because she's not a fish
do you have any tutorial or similar for it? This would really help me with my game to get some assets i really need 😮
I don't have a tutorial, but I can help out however you need. You mean for Stable diffusion, or the 3D aspect of it, or both?
the 3d aspect. i´m a bit familiar with SD already and start working and trying extensions. But this is awesome i wanna do it too 😮
It takes a while to learn 3d, and even then there are still problems I have a lot of trouble with, so I have to use work-arounds (for example, I couldn't connect the head of another character onto a body, so I had to move them individually). I never actually went through random tutorials, but knowing what I wanted to learn, I learned what I needed. So, what kind of things would you need to do, for example?
i´m working since some years with unity, so i know a bit about that topic at all. I actually wanna turn a picture into a 3D Asset.
Oh, I've never done that before. Just the opposite, turning 3d into 2d. I've seen a few videos that show turning 2d into 3d, but I think they are just experimental
ahhhh i see then i read it in a wrong way 😄
There are some things that can automatically generate textures on 3d assets, maybe that can help
ohhhh do you know the name?
Dream textures for Blender
There is also an add on for blender that can generate renders with SD within blender. And a few controlnet rigs that work within blender too.
i love making ai art on laptop
6k rpm

idk how it is on laptops but you might be able to undervolt the GPU to get lower temps @ same performance
thank you i will learn a bit the nex t weeks i guess
AI is a tool to help in the workflow of an artist. I think learning any skill in addition to AI puts one far ahead of others. Ie a 2d guy can fix hands and stuff, 3D can do a lot, etc.
true, but it also open me ways to create something i wasnt able before and improve it then
might be because of my tabs because it wasnt like this before
or maybe the webui update it became more laggy around that
I don't think it's tabs, I have an embarrassing amount of tabs open and it doesn't change anything really
how much vram you got
tabs in browser, especially in chroma using a lot of RAM but almost no GPU capacity, except you watching videos or similar
anyone else having this problem with civitai?
all the models i saved by liking changed to random ones
keeps changing everytime i open it
I don't understand how that site took off as the favorite. Seems to run on band-aids and bubblegum.
man i saved some good ones there
should have saved somewhere else
I just do this 
If only we stopped procrastinating and just downloaded everything
















