#🍥|anime
1 messages · Page 77 of 1
darksushimixmix?
meina
:)))
(realistic:1.3), SLE, mksks style, best quality, masterpiece, detailed background, highly detailed, intricate, detailed face, long eyelashes, detailed eyes, (standing, cowboy_shot:1.3), (skinny, slim, thin:1.1), (medium_breasts:1.1), (white_background, simple_background:1.3),
(one girl, 1girl:1.1), (solo:1.3), (nekomata_okayu:1.2), light purple hair, purple eyes, choker, cat ears, cat tail, pale_skin, (skindentation:1.3),
black leotard, center opening, cleavage, cutout, stomach cutout, latex outfit, succubus outfit, sexy, bare shoulders, bare legs, straps,
(light_smile:1.1), blush, nose_blush, lora:GoodHands-vanilla:1 <lora
-1> lora:add_detail:-1 lora:Mossa:0.5
Thats the prompt
Also, lil soosi got new waifu? lil soosi cheatin? 
well, you can't have too many waifu right 
Cheatin! 

Hot tho :o

it's kinda scary with semi realisim

ahoy
I've been trying to get an outline image by following this guide, but it doesn't really work (at least for me)
idek what exactly is wrong, like, I'm just copying all I could but the results are so off.
could someone help me to make an outline image out of mine, please?
What's the image you're trying to get the lineart of? I've found that with images that have a ton of detail, it doesn't work as well
Using canny, even on a simpler image the lineart still needs manual editing
Oh ok, following the video tutorial I was able to get this cute grayscale image. Had to put "color" in the negative prompt
Removing all negative prompts except "color" gave me this
So yeah, might just be worth messing with the prompt and denoising to get the result you're looking for
https://civitai.com/models/5891?modelVersionId=91871 @merry tendon
does pets works on any model ?
I am using controlnet with rev model but when i try for example a human character and beside a pet like a wolf or dophin it doesn't work
I tried 4 models but sometimes it works and sometimes not like the chance is 5%
damn
that's pretty tough
I tried to use the canny model too, but didn't get any close to your results
actually it was just drawing another pictures, yet they had similar composition as the original input, but anyway they were too damn off

I love
anime screencap
well this worked semi-well i think
Did you use inpaint? That might be what's messing it up
I mean img2img
The main image should be a plain white background and the image in control net is what you want the lineart from, not the other way around. I almost did it the other way around by mistake lol
I'll steal that image and see what I can get it to do if that's ok with you
yes plesae
yeah, that's literally what I did
I don't think it's going to work on my end because there's no data in the png, but I'll try lol Nvm it's working fine, I just don't have a lot of memory so it's taking a while
Best I could get. I think the detail of the base image isn't suited for canny
That's AI
Why is it on paper then>
Oh, you printed it
Ok, who gave her all this ice cream
use a lorea or detailed eyes looking at viewer etc at the start of the prompot
I find it a bit sad tho that my lora gets absolutely no attention lol
which?
Yameroyandere
To be fair, i did leave a link to yours in the about version section.
Honestly i dont expect much attention to be given to mine in the first place with how unneeded it probably is these days
Your lora is not unneeded, yours can do constricted pupils much better. and thank you for the support❤️
What images did you use for the dataset btw?
U askin for the zip?
Nah, I mean where did you get it from, like gelbooru or smth like that
Gel, dan, various really, i was using the grabber program
Ok nice, thx
O, Pika 
Hey hey
Nice to see you around again 
It's good to see you, too ^_^
o.o
hello
o/
is there any guide for controlnet you guys reccomend?
i am using colab and i want to make certain poses but idk where to start
personally i just scrolled youtube and looked at the github
Damn those are really creative :O
Yin-Ting Tian Style? :O
github?
not that i know of
controlnet has a github
how does one achieve this, asking for a friend
https://civitai.com/models/82370/ruteko-style-monster-girl-style-lora im using this in conjunction with a giant monster girl wildcard prompt
A few days ago I came across an artist that drew really interesting monster girls and immediately wanted to train the style. The dataset is, ofc, a...
many thanks
yall ever have times where you regret your actions, specifically creating a public resource
@merry tendon
It looks a lot like their style
Also, naidong now on gelbooru apparently
Tis a better example :o
this is giving me ptsd flashbacks
turn off Restore faces
only meant for realistic faces
you can link colab to github? sorry i dont really know how to use github
No clue, I noticed yesterday :o
And true 
That's ridiculous
you will pilot capsule cosmonaut
what
idk i just woke up and saw ptsd and the image reminds of a Russian cosmonaut
nice controlnet
Bruh
studenten was the wrong day to take the bus....(Swedens graduation from gymnasiet)
What's good gamers I'm back, what AI image gen evolutions did I miss
Other than Control Net, I still have no idea how to set that up
I don't think you've really missed anything lol it's been quiet in here
Definitely look into getting controlnet set up though, it's fun
oof
@elfin geode
Remember to report too :o
already did
Yeee
I was wondering what chat I had open when I came back from kitchen 
XDDD
should i try to ping more mods? 
I'll do one :O
@finite herald :O
banned!
Good thing Fruit didn't have to see it xD
Yeee xD
I was not expecting to see Ace like that today 
Same XD

Aw thanku 
definitely milkers 
I'll accept cats
HMMMMMM
here cat
Here's green cat
smol smushy face
NOM
Is this kind of cat ok?
female @wind girder :o
True XD
Remember to water your chibis
O das adorable :O
lemme show you guys my mochicats
i couldnt find the watering can emote lol 🚿
There isn't one? I just looked, and I thought there was one :O
Wao das weird, big time mandela effect :O
Fair, I mean that kinda works XD
Luckily you missed it, it wasn't that bad, i think, but it was pretty cursed xD
Typically no! lol
We do from what I know have an influx of apps, but they are being reviewed!
Ohhh, thats fair then! I mean I sent mine like... I think about a month or two ago? Maybe? So I assume it just got rejected already :o
Oooooo~ i'll see just in case

Yeee, and alrighty, though I applied for Community Guide since I heard thats what you had to do first 
Though I think I checked off both boxes just in case 
I dunno why the leather strap, but I love tis style :o
Can you share prompt?
1girl, teal hair, bob cut, sitting in flower pot, cute, adorable, uwu, water drop,
Negative prompt: (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (bad-artist:0.6), bad-image-v2-39000, negative_hand-neg
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3216954448, Size: 512x768, Model hash: 931aaff934, Model: Newly Downloaded_expmixLine_v3, Denoising strength: 0.53, Clip skip: 2, Hires upscale: 2, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Version: v1.3.1
Used embeddings: bad-artist [2a38], bad-image-v2-39000 [b03e], negative_hand-neg [b740]```
Thanks sir
that's a lil sus i guess
In what way? 
Cute :)
Well what are the current best models then
There's not really a best model, everyone has their own preferences
Dreamshaper and meinamix continue to stay at the top on civit though so imo that says something
Or use some obscure model from huggingface and end up finding some hidden gems lol
That too lol
I think I'm the only person who doesn't like auora 😂 I can never make anything good with it, my prompts just don't mesh with it well
The endgame would be merging stuff yourself though lol
Yep, I only use my own merges at this point, unless I'm playing around with something new to see if I want to add it in to a mix lol
Woah cute
Hey anime folkes!
Your feedback has always been incredibly helpful in the past so I encourage any and all of you to consider filling out our server survey to help shape what the future of this server looks like!
https://docs.google.com/forms/d/1J96YgbX69fP9h3QXYaSFx1jYmJVkBBp47VQ_ucjZusU
Thanks!
Welcome to the second Stable Foundation Community Survey!
The first section on this survey will focus on the different community events available on the server with the second section focused on discoverability and how you use the server itself!
The survey itself will take an estimated 2 minutes to complete. Please be as thorough with your ans...
The survey will be open for a while yet, but nothing wrong with getting your thoughts down if you are willing!
Plus it's another chance for @acoustic agate to ask about the Miku role 

pingu role wen
Whenever you create and hand it out 
Can't decide which vae I like better... anythingv3.0 on the left and clearvae on the right, same prompt and seed. 
Kinda leaning toward anything just because the pastel/slightly desaturated look is more of what I was going for
I switched to ClearVAE as I found it to have better detail clarity, like in the eyes.
Ah yeah the eyes are definitely more detailed with clearvae, I've noticed that too
I think anything gives me better hands, but it could just be my imagination/rng lol
what does "Are there any events that are not currently available that you'd like to see in the future?" mean?
not currently available?
Like a server-wide celebration for Miku Day
For sure, happy to help
Bless
lmao looks like you got the idea
have a link?
anyone have any new models they're digging?
man, what's up with this server icon though
The 200 is gone 
are u using a Casca lora
yes, checked it with PNG Info
Which model are you using?
guys do u know a good model to make animal characters ? i tried dream shaper and rev but 90% of the times they make a human character instead of animal character
Animals specifcly, none come close to my mind. But I had same problem as you said, models lacks understanding of animals.
Didn't realized this also happen to realistic models too, thought it only on NAI Clips.
yes most of the time it generate a human instead of animals
was trying for days but couldn't figure it out 
I even used some Lora models still the same
I see, I’ve also been noticing good hands. Is it a model that does this or Control Net
Mix of model and adetailer for me personally
If I get a gen that I love but doesn't have good hands or eyes, it's an easy fix
Also prompt helps I've noticed, and negative embeddings
You got any specifics I can look into?
Omg lmaodmdmsn
Also is adetailer an extension?
nice
I see
Yeah it's an extension! https://github.com/Bing-su/adetailer And honestly, the best advice I can give you is basically just to browse huggingface or civitai and download whatever models you think look cool and just play around with them. Check images people have posted with .png info and borrow prompts too, that's how I started
oh boy..


👍
hmm that got interesting
Woah
I dig the colors on that though
Speaking from experience, models that does Block Merging (MBW) with other models tend to gives better hands generation, since NAI models struggle with good hands and you can test out this theory for experiment.
https://rentry.org/Merge_Block_Weight_-china-_v1_Beta#34-repair-hands-provided-by-white-sheet-er
ControlNet also improves hands and so does other methods. Others are generally inpaint and cherrypick.
I'll have to look into what that is
As the rentry stated;
(Note, poor hand drawing is a fundamental problem with AI, and this is only a solution to alleviate the issue, not completely cure it.)
In general most people try to find models with good hands, and usually ended up with realistic models MBW with Anime model.
This method was very similary to very old days when we used F222 to merge with anime models for anatomy improvement.
learning to use lora's and hyper networks and inversions 😄
Same! I've never even heard of block merging before. I'm still a noob at all this stuff lol
the eyes mess up on some of these
Popular model that does this would be AbyssOrangeMix2
https://huggingface.co/WarriorMama777/OrangeMixs
They use only Anythingv3+BasilMix for their merge.
WarriorMama does very well with their documentation 😄
I'm using elegantentropy
Oooh ok! I don't think I've tried that one before. What I did with my models was just do checkpoint merges and then run a test prompt and if I liked it, I kept merging different models until I was satisfied with the result lol
I try to use different one's. merges for me work at first then well go screwy
If it does well then I don't see anything wrong; In general MBW is very painful process and tedious.
shrugs
That's one of the things I love so much about all this though
There is autombw, but I prefer manually for customization.
Everyone does things in slightly different ways and it's fun to compare and learn
yep
Not sure if this count as self promoting, but I have cookbook for everyone who are eager to learn as well.
https://www.figma.com/file/aESyZAxHxBJjE63gog5ExZ/KENSHI?node-id=0%3A1&t=2ULQWeLUSIWhk1aE-0
https://www.figma.com/file/3nat1kAMIxJm2lFwW56EC4/Aurora?node-id=0%3A1&t=Qb4N2OHue67ROg5W-1
Ooh cool!
I promote for you don't worry 
Kenshi was made half-year ago all checkpoint merging so there is not MBW but have theory and clear concept I had for the model.
Aurora best model 
I wiah i could update mine but im still taking care of kittens and too much going on
That's fair then :o

That's my mindset as well, I love learning 
Man yall are making me want to do some Sakura blossom images but I need to go to bed lmao
hmmm not exactly what I had in mind...
ahh there we go thats better
well ok then
ok fixed...
sorta
and then it went wrong....
Ah, so the days when I stopped doing AI shit for a while, are the old days now
Except it's only been like five months 💀
So how exactly does adetailer work
i try it , that add ons just make your image not blur and bring more detail in face, body etc
What do the negative prompts and whatnot mean and do
And the first and second model things
There’s like some prompting stuff under the adetailer tab
i never use it need more research
i just enabel it and pick what part you want to be good
I see
adetailer does an auto inpainting, if you leave it blank it will use your full original prompt, or if you specify it, it will use what you specify for the inpainting
I see
It can be bad or it can be good
O thats really cute :o
My better guns lora really works 
gn 🙂
Weird gun but kinda cool
anyone still remember this game?
Hm, it seems adetailer makes very slight, yet noticably good changes
It seems to have very noticable changes when the object is far away.
ikd who this is, be his images look super sick(made with my lora) https://civitai.com/reviews/140510
5 star review
Well, isn't ADetailer for inpainting faces?
O they do, das Clementine from Overlord :O
Yes
ima need that prompt, settings and social security number please
Try using it with my gun to head lora?
I haven't :o
Try it, see if the gun turns out even better
V4 SUMMARY In the attempt to revisit this and make it better, V4 of this concept has now attained better functionality when generating handguns. It...
Ayyy thanks
Seems better :o
How many tries did it take
Kinda
That was the first gen xD
This the 3rd gen
4th gen :o
Gun gunned to hard in this one, 5th gen
6th gen
Ye seems definite that its better @compact warren
Awesome. U gonna post it? Ill link it in the suggested resources section of mine
I think I posted it already
This LoCon is meant to improve guns, it is not the definite cure to bad guns, but it will help to make better guns. Guns that were in the dataset a...
Aight ill do it when i have the chance
food
is that a comfy node for AD?
Adidas Vodka
that's like its main purpose
what is the thing that makes the faces like that for you?
1girl, solo, pink hair, twin braids, yellow detailed eyes, full body, jacket, hands in pockets, wading, reflection, lake, forest, (night sky, dark, meteor shower:1.1),
Negative prompt: (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (bad-artist:0.6), bad-image-v2-39000, negative_hand-neg
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 22, Seed: 656077053, Size: 1024x512, Model hash: 931aaff934, Model: Newly Downloaded_expmixLine_v3, Denoising strength: 0.53, Clip skip: 2, ADetailer model: face_yolov8n.pt, ADetailer conf: 30, ADetailer dilate/erode: 32, ADetailer mask blur: 4, ADetailer denoising strength: 0.4, ADetailer inpaint full: True, ADetailer inpaint padding: 0, ADetailer model 2nd: hand_yolov8n.pt, ADetailer conf 2nd: 30, ADetailer dilate/erode 2nd: 32, ADetailer mask blur 2nd: 4, ADetailer denoising strength 2nd: 0.4, ADetailer inpaint full 2nd: True, ADetailer inpaint padding 2nd: 0, ADetailer version: 23.5.19, Hires upscale: 2, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Dynamic thresholding enabled: True, Mimic scale: 7, Threshold percentile: 96, Mimic mode: Half Cosine Up, Mimic scale minimum: 4, CFG mode: Half Cosine Up, CFG scale minimum: 4, Version: v1.3.1
Template: (detailed background, best quality, masterpiece:1.2), cinematic lighting, highly detailed, absurdres, hyper realistic,
1girl, solo, pink hair, twin braids, yellow detailed eyes, full body, jacket, hands in pockets, wading, reflection, lake, forest, (night sky, dark, meteor shower:1.1),
Negative Template: (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (bad-artist:0.6), bad-image-v2-39000, negative_hand-neg```
sorry i was on dnd mode and didn't realised your reply 
Pretty, I like the kind of painting texture
i really like it as well
it's the same prompt as above 
What you mean like that?
this style
that is like more realistic and stuff
and also consistent
add some realism tags
Yes
It is a specific style
Persona 
Style prompt, and then 2 style LoRAs
That reminded me, persona 3 remake 
faceage LoRA at -1 to guarantee mature faces, add detail LoRA at -1, and then Mossa style LoRA at 0.5
Is it? :O
And is very consistent
And I really like et 
So I think I have got the gist just looking around a bit, but auto detailer just auto selects and inpaints the face right?
Yeah
Yes you can read this more on Github
Actually I think the embedding already explain what the extension does 😄
what does these mean? excluding same prompt
What you think of this?
good generation, you might want to change your vae tho

well, for layman terms, it's basically for better colors
go to settings in your webui
https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/vae/kl-f8-anime2.ckpt Get this, put it in VAE in your models folder, and select it from the settings
yup that ^
Oh, y'all use the website
My phone can't handle it I ran it once and I had to restart it
This is the stable diffusion group right or did I go into a different one again?

There's a website and then there's the mobile app thing that I have
In the app there aren't any settings
That's how I made that
Oh I see
What's your negative prompt btw
Mine is low resolution, text, error, cropped, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, out of frame, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck, username, watermark, signature, mutated eyes, bad eye color, bad head covering, horrible art style, worst eyes, grayscale, bad color, bad clothes, horrible height, bad background, inaccurate, horrible pose, inaccurate background, deviation from prompt
Mine atm are
negative_hand-neg, illustration, 3d, sepia, painting, cartoons, sketch, (worst quality:2), (low quality:2), (normal quality:2), low res, bad anatomy, bad hands, normal quality, ((monochrome)), ((grayscale:1.2)), disfigured, kitsch, ugly, oversaturated, grain, low-res, Deformed, blurry, bad anatomy, disfigured, poorly drawn face, mutation, mutated, extra limb, ugly, poorly drawn hands, missing limb, blurry, floating limbs, disconnected limbs, malformed hands, blur, out of focus, long neck, long body, ugly, disgusting, poorly drawn, childish, mutilated, mangled, old, surreal, portrait, big pussy, ugly pussy, deformed pussy, extra legs
Though I use one embedding at the start and it doesn't sound like you can use embeddings
Your prompt is kinda sus at the end
I got it from someone else cause it does something weird to the lighting, so I just use that xD
tf is kitsch
Probably a trigger token
Googled it
Kitsch (/kɪtʃ/ KITCH; loanword from German) is a term applied to art and design that is perceived as naïve imitation, overly eccentric, gratuitous or of banal taste
Oh damn, fancy word :O
I wonder if it have any effects 🤔
@Komodo#0184
Never played Cyberpunk 2077 I guess. 
It's amazing how many things I've learned through gaming across the decades.
guys why lora models make the image in a bad shape ? trying to make a character riding an animal
Try lowering the weights maybe
3060 should be here teusday and can start training loras 
do you mean the weight from controlnet ?
the lora, i mean
I tried 0.3 but it doesn't add the lora and above 0.6 breaks the image
It is a poorly trained lora then, there are actually quite a bit of them
Overcooked momento
Its almost all of them are bad for me , even if the images are good , maybe i am using models that is not compatible with the lora model ?
what if you use it at around 0.4 - 0.5
sometimes, some lora is about finding the perfect weight
will try to do that
Todd Howard says zoom in for 16x the detail.
with 0.2 it works well but 80% of the time it doesn't apply the lora to the image 🤣
i think it's more of, it's applying at really low weight, that it doesnt really change the image much. You can try seeing if there's any activation token for that lora, and use the lora at weight 0.2, but the activation token at higher weight
trying to use this one
@magic wave i linked ur lora to mine
yup so the activation token is "riding on a"
will try to change model and see if it works well with it , maybe its my model , i am trying to make it anime way and i am using an anime models , maybe this is the issue
maybe
one good way of testing is to literally use the model that the author used
omg another cryo
I am here since a year , i remember you 😄
so first you wanna do png instead of jpeg because jpeg is the work of the devil
in my opinion jpg takes low space and its the same
yes i was here like a year ago but i stopped generating for like 6 months and came back ^^
looks like shit with all these artifacts and how it deals with gradients
will try png then , i tested it like a year ago and it was the same
It looks blurry af when you upload to discord
yea the image was generated on 640x640 and it has issues with the model
the lora working good with that rpg model but sadly its not anime version 
to be fair, generating an image at a measly 640x640 is going to make any image look terrible
yea i wish i could do more but my 3070 cannot handle more than 832x832 image with lowVram option
I can max do 1200x1200 with my 1060 6b
oh is there a settings u do to make that happen ?
sometimes it doesn't let me even 832x832
No idea, My webapp definitely does run more efficient than it did in the past.
I did do a torch update a couple months ago to try and fix my constant NaNs errors.
well more of a reintall than an update
I want to update torch but it always give me this warning :
Beware that this will cause a lot of large files to be downloaded, as well as
there are reports of issues with training tab on the latest version.
I will reinstall torch then maybe this is the issue , i could do 1024x1024 a year ago now its hard for me to do 832
I am installing the app again and will try to see if it fixed the issue
If you want a temporary workaround, you can upscale 3x-4x in the extras tab and inpaint sections you want at your max resolution
It's how I made the 3k by 3k image I posted a bit ago
it looks beautiful , did you make an 832 image and upscale it with highres by 4x ?
It's a 1K image upscaled by 3x I don't have a card that can handle a 3k highres fix, so I use the dedicated upscaler in extras tab
oh i see , will try that trying to reinstall the webapp and trying again ^^
can i ask something about mov2mov for stable diffusion
This is what the initial 1k image looked like when upscaled to 3k, big difference even without counting my photoshop effects
yea it look so stunning and the quality is too good ^^
i hope after update it can handle more pixels while generating
niji
I done, isn't the best tho
guys why does my 3070 doesn't even handle 832x832 😦
I even reinstalled auto1111 still the same
do any of you know how can i stop other views or people (like a close up of the character or smth) from appearing? i just want the full body and the background
anyone is 8GB GPU can run more than 832x832 with controlnet ?
you might want to try using multidiffusion
it's an extension
is it similar to controlnet ?
no. What that extension does is that it can allow you to generate higher res stuff that your gpu can't
so with that extension, you might be able to generate more than 832x832 with controlnet
oh this is cool , i will test it out 🙂
if you are using it, this is my settings on a 8gb vram gpu
can you generate something like 1024x1024 with controlnet used ?
not sure, never really tried that
just curious, why are you trying to generate such high res stuff straight away?
because it gives a better images , i always get bad images with less than 640
wanted to try 1024
are you using high res fix with that?
highres fix doesn't work even if i am using 512 images , dunno why
wanted to see if 1024 gives better images or not because i saw some beautiful images online on reddit that people are using 1024

what's your settings for hires fix?
if you want to get better result, maybe generating at 768x768 then hires fix at 2 times
since most model is trained on 512x512, generating stuff at 2x that res might duplicate subjects like 1girl prompt might gives 2 girls
its the default settings
will try this and tell you if they are working ^^
good luck 
not sure still didn't let me do 832x832 images even with multidiffusion enabled
did you set the image height to 832x832? the highresfix will double the size so you would do something like 448x448 for the initial image size instead
yes i did set size to 832x832
i remember a year ago i was doing 1024x1024 but now i cannot even 832x832
is there a setting i can do to make it work without highres , i was trying to make images with controlnet with 832x832 but it always goes to 80% then give cuda error
will try that and see
I never quite grasped Tiled Diffusion
so where is your vae?
hmm thought I had that set
lol i cannot even do 256x256 with highres enabled to 2x scale
changed vae to an anime one
huh why does it go all black, I don't have the censor turne don. wtf?
better
guys anyone with 3070 GPU can do 512x512 with highres open 2x ?
doesn't wok at all , dunno why
there's definitely something wrong
are you running other stuff that uses a lot of gpu power
no i am not , just using this app alone
hmm have you tried the --medvram -xformers in the start config?
in the webui file ?
yes--> set COMMANDLINE_ARGS= --medvram --xformers
this is mine set COMMANDLINE_ARGS= --deepdanbooru --precision full --no-half --autolaunch
go into task manager > details, right click one of the label, and select columns > tick Dedicated GPU memory > OK, and sort by that
Do you use restore faces?
see if any other programs are using a lot of gpu memory
yes
You should try without it, because it usually fucks with the eyes :o
ok trying now
Adetailer with controlnet is usually better
Yeah, for anything that isn't 100% realistic I think its worse most of the time
wow
yes
Yeee :o
lineart
its basically automatic dude
some apps uses 7k and some uses 100K but python is using 7M K
haha added luisapkawaii hypernetwork...
does that mean python uses all of my GPU power ?
when it's not genning?
yes i am not generating anything and its using 7M K
then you probably have memory leak there
press the free gpu button in multidiffusion
basically take the cropped pic that adetailer made and pass it to controlnet and then generate it
Kinda messy, maybe less weight :O
hmm yeah good point
Am confused
just try it and you will get it
But how tho
I closed auto1111 and python went to 0 but when i open the webapp it turns to 7M K even when i am not generating
there should be a controlnet tab in adetailer
made it random seed and hmm no...
Yeah its still pretty messy :o
wait a min, 7gb vram?
or 7mb?

I have a 3050 ti 4gig 😄
And where do I get the lineart model?
You should send screenshot :o
here is a screenshot when i am not generating anything
added moxxi lora
i guess its using 5GB vram while not generating images
that might be okay

that is pretty light dude
I mean
Thats like uhhh
About a hour of download for me
My download speed is abysmal
xD
I have 2 GPU in the same PC 1660 and 3070 , should i disable the 1660 and try again ?
i dont think automatic1111 allows using both GPU (I might be wrong on this)
oh so you dont have the dank ass asian broadband speed?
so basically it's using only 1 of them
yes it doesn't , i will disable my lower GPU and try again
oh haha I have the waifusim turned oh DOH
good luck
I don't :O
you don't what 
thats fine... just wait for 1 hour
cursed
O
Bad aim
XD
I'm going a lil schizoid

I wanna watch stuff tho
cursed
then play a game or watch in 480p
I usually watch in 480p 
Also my download speed is uhhh
600~ kpbs :o
And my upload is something like 150 kbps I think xD
ick
how in the fuck you survive like that?
I dunno what that is :o
nm
Well, for most of my life I had 300 kbps download, this year I was able to get 600~ xD
you got a ethernet cable right?
Yeah :o
you connected that to your desktop right?
Yeah :o
are there a lot of people using the network?
Just me atm xD
turned on hi-res fix
What kind of bra is this guys?
I'm telling ya, internet is just shitty here xD
Seems like just a crop top :o
you live in australia?
Worse, Argentina
💀
that explains it
Yeah xD
tbh at that point try and swap to starlink
I mean, hopefully this month I'll have optic fiber here, so that'll be neat, really speed stuff up
Damn that looks sick :o
same issue , i guess maybe torch or auto1111 added features that takes more VRAM , i remember a year ago i was doing 960x960 easily without any issues i even could 1152x1152
now i cannot do 832x832 😄
im just resuing my prompt, over and over again 
O wait whats your prompt?
are you still unable to generate at 512x512 with hires fix?
1girl, solo, pink hair, twin braids, yellow detailed eyes, full body, reflection, jacket, hands in pockets, lamppost, street,
Negative prompt: (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (bad-artist:0.6), bad-image-v2-39000, negative_hand-neg
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 22, Seed: 2248291441, Size: 1024x512, Model hash: 931aaff934, Model: Newly Downloaded_expmixLine_v3, Denoising strength: 0.53, Clip skip: 2, ADetailer model: face_yolov8n.pt, ADetailer conf: 30, ADetailer dilate/erode: 32, ADetailer mask blur: 4, ADetailer denoising strength: 0.4, ADetailer inpaint full: True, ADetailer inpaint padding: 0, ADetailer version: 23.5.19, Hires upscale: 2, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Dynamic thresholding enabled: True, Mimic scale: 7, Threshold percentile: 96, Mimic mode: Half Cosine Up, Mimic scale minimum: 4, CFG mode: Half Cosine Up, CFG scale minimum: 4, Version: v1.3.1
Used embeddings: bad-artist [2a38], bad-image-v2-39000 [b03e], negative_hand-neg [b740]```
this looks like one of the cs go map
O, it does :O
cs italy
this part
yea i cannot , tried everything still no luck
Ty
do you have xformers?
hmm think i'll remove that cyberpunk it's doing to much
no i don't , the only one i have is deepborro
as in in your run.bat file, do you run with --xformers agrument?
I do and it helped thats why I suggested it
this is the argument i use
set COMMANDLINE_ARGS= --deepdanbooru --precision full --no-half --autolaunch
add --xformers and try again
hi-res pass.. wow
hmm think i'm overloading it
changes some settings... and....
are you using highres with Latent ? its working and when it goes to 99% i get a CUDA error still
you mean my upscaler?
im using 4x fatal anime upscaler
yes , i will try that upscaler and see
hmm boy
im wondering are you using vae?
yes
Probably a low step
A couple of those seeds look pretty good, mind if I spice those images up in photoshop?
meaning?
I think they looked diff caused I changed some settings in style pile

😄
lighting and post process effects
What you mean by style pile 
gah....
it's an extension
allows for things like conceptually, image types, direction, mood, colors, some preset style influences, art movements and artests and other thing
Oh, so its like just a prompt helper?
yep
Fair then :o
Added my Asteroid Ill LoRA for better background
jumps off the edge
What you mean changes to?
Oh, I mean, looked up Elegant Entropy, and it looks pretty good on its own 
-1>







