#🍥|anime
1 messages · Page 191 of 1
on the door to the tavern :
get crushed.
for what it's worth, i really like the creativity possible in pony (cfg is just 5, tag used is 'limited palette')
i've been messing with that a little, haven't really gotten any good results so far though
yeah it's just kinda.. weird 😛
problem i've been having is that the prompt travel doesn't.. travel..
just trying a simple blink, but it's not closing the eyes and even the slight movement is on nearly the last frame (16) instead of frame 8
tried changing weights in the prompt, but it doesn't seem to help
As night falls, the faint moonlight falls like silver catkins, gently covering a peaceful mansion. The swimming pool outside the window reflects the clear night, with sparkling waves. There is a luxurious sofa inside the window. A handsome man in a straight suit is sitting on the sofa looking out the window and thinking deeply. The flowers and plants around the mansion are swaying slightly, and the night breeze is blowing.
this isn't a bot channel
thanks
hmm...
which motion model?
mm_sdxl_v10_beta
not really any point in using it if it doesnt work with my xl model
just curious if it'll work 😛
ooo apparently nubby put out a new kakigori... i'm behind the times
finally getting somewhere
yeah, it completely trashes the quality, could do more frames and make it look less choppy but it barely works to begin with
it's basically a "oh, that's cool" thing, but not actually functional for the most part
had to change samplers and drop the steps from 30 to 10 to make it even get anything at all
Why drop the steps
I don't understand how it works 
apparently some samplers send multiple steps to the unet, but animatediff is hijacking the unet for it's own steps, so it doubles the input
like, euler 10 steps is acually 10 steps sent to the unet, but dpm++ 3m sde karras is equivelent to 20+ steps
it might be an a1111 thing, not sure
just something i found after an hour digging around in closed bug reports on the animatediff github
since none of the useful information is in their instructions
I don't think 3m SDE or 2m SDE should do that. 2S a and regular SDE both do. Restart and Unipc both do some funky stuff that might mess with it.
to be fair i don't know if they actually do, but I couldn't get 3M SDE to output anything but basically a static image, and changing to euler with lower steps fixed it
The devil should always be a baddie.
based on the comment from the dev on that bug report, switching to euler OR lowering steps should solve the problem, but i had to do both 🤷♂️
I haven't tried the 3m samplers much because I've pretty consistently gotten worse results with them than I have on 2m Karras or 2m SDE Karras.
it seems most people are using it with 1.5 and using the motion loras though, the 'prompt travel' is what i was interested in and it barely works at all
for the most part 2M or 3M perform about the same for me, but with my model I liked the overall results better when i was testing them
realistically, I could use 2M just fine, at this point I don't remember what made me go with 3M over 2M
i think it had something to do with better performance with a greater range of CFG scale
but even then it's probably marginal
could be, I don't change my CFG scale much. I pretty much stick with 7.5 on almost all of my models and 7.0 for Pony.
i don't actually change it much at all either
If I want something that it struggles with in that range then I use Dynamic Thresholding at either 15 or 20 with mimic scale set to 7.5.
damn, civitai down yet again.
civit is up
Р
well, kinda
it's friday night, people are bored, horny, and likely drunk, so..

it's probably not that. it's midnight PST so the majority of the US is likely asleep, it's saturday morning for europe, and I don't think Civit is all that popular in asian countries where it's currently saturday afternoon.
could be a maintenance window for them or their host and running on limited resources. or just civit being civit.
,
That cat on the hoodie lmao
😄

Its one of these lmao
palworld pal
I was trying to remember, haven't played in a few days
There is another one that looks cuter like one the hoodie 
Sweepa = Best girl 


promt: street fighter
i wish i was that piece of furniture
ha, I hadn't noticed it was similar in all those.. interesting
ran this large batch last night... I definitely need to work on my ability to get better hands with kakigori 🙂
models or images?
Human girl with rabbit ears is a Knife queen
a woman sitting on top of a wooden table next to a window, computer graphics by Nagasawa Rosetsu, dribble, plasticien, anime, sensual, photo
that's nsfw, not posting it 😛
a woman in a blue dress posing for a picture, a photorealistic painting by Lü Ji, cgsociety, rococo, ultra detailed, anime, detailed
there's no bot here btw
"a man having muscular body and lion head"
and good day to the second hemisphere!😁
how are you?
everyone is kind of busy...
crazy man!
none of my icons are loading
in any folder
folders
images
I figured it out
I didn't have enough storage left on my drive
Well, it’s only Saturday there, but here it’s almost Sunday! 😁
Nay?
try use Cnet
It's just one from a discord bot
I'm lazy in bed
lazy man in bed! 😁 you puffy bear!
to be fair the StableDiffusion folder does also include my textgen stuff as well.

I deleted like 100GB of images and 6-700GB of models a couple of months ago too
just like you, I'm still a schoolgirl😂
Wow
fern?
yesh
still sleepy
you have the wrong idea, those are hers.. she's not the one in danger..

Damn that's clean

How did you do that? Is that in sdxl? I tried that pose Yesterday and failed. 
yes, it is a SDXL model
super simple prompt as well, except for those required tags
1girl, solo, white skin, pale skin, scales, dragon horns, dragon girl, holding heart,
Negative prompt: 3d, abdominal muscles
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 825314876, Size: 768x1024, Model hash: ac006fdd7e, Model: autismmixSDXL_autismmixConfetti, VAE hash: 63aeecb90f, VAE: sdxl_vae.safetensors, Denoising strength: 0.51, Clip skip: 2, ADetailer model: face_yolov8n.pt, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer mask blur: 4, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer version: 24.1.2, CFG Rescale phi: 0, Hires upscale: 1.5, Hires steps: 5, Hires upscaler: 4x_fatal_Anime_500000_G, Version: v1.7.0
all my generation should have pnginfo
should
ok, easyfluff and ponyxl have to be some kind of meme
123
A boy with a wolf cut is carrying a cute rabbit on his shoulder. Background: Pastel yellow style-s 750 --niji 5 - Variations (Strong) by @peak trail (fast)
nah slow

v11.2 (Note: you must download both the .safetensors file and the .yaml file with the same name. For A1111 you just need to place the two files in the same folder and they must share the same filename. For Comfy, I believe there is a node to choose the config file which will be the .yaml) https://huggingface.co/zatochu/EasyFluff/tree/main
hll is a line of lycos that are specifically meant to make anime work on EasyFluff. I use either hll6.3-fluff-a8 or hll6.3-fluff-a9. It also adds a large number of artists that you can prompt for different styles. https://huggingface.co/CluelessC/hll-test/tree/main/lyco
here is a list of the artists trained in a8 - https://files.catbox.moe/xecm6x.txt
here is a list of the artists trained in a9 - https://files.catbox.moe/c1jlaq.txt
He's really putting his heart and soul into this
my fav artists fr by nai diffusion by stable diffusion
What's the difference between a8 and a9?
a8 works somewhat decently on other models. and some of the artists that I've tested look a bit different than the same artist does on 19.
a9 doesn't work very well on other models that I've tested and it gained 68 more artists.
@native halo What do you think about dragon maids
hire or not
It's worth just trying both to see what you like better.
dragon girls are yesh
Probably not, she doesn't seem to be doing her job
need that pose but the camera needs to be placed (from behind,from below:2.5)
(from below:3.4)
Only see the feet

Hell, possibly.
What model?
D:
Forgot my negative lol
Is it any good?
forgot my neg
very good
Its like pony but more stable with styles and quality
yup, like what @nova remnant said, it's for lazy ponyXL users
the style doesnt change that much
cool
very consistent 👍

No grabby
Feet
A small pond surrounded by trees and houses in the background.

do you have a discord setup just for emotes and stickers? 😄
Yes
I've thought of doing that
Although I have added only 3 stickers
you about to catch these hands
lets go
Anyone made some good abstract or expressive stuff recently?
how?
why?
Who?
when?
dont blink!
ok.. it's A finger
well.. ok then
man, control net is super nice when it works, it's just unfortunate it doesn't work more often 
what model
For the lady!
I would say Soushiki


I can't make V pose with my NAI style 
hmmmm 4070 ti super is 980 dollars "return from customer" with used up campaign codes for games but they've tested it through sharpener stresses and there's nothing wrong with it. Maybe the dude overestimated how big the upgrade would be
But at least I have this
4070 super all new no ti is 750 dolla
ok
i hate auto correct
it always fucks up the most menial of words and sentences
it even fights me on the word fuck

Soushiki with these loras for style: https://huggingface.co/Aotsuyu/LoRA/blob/main/artstyles/[style] zunusamav3.safetensors
https://huggingface.co/Aotsuyu/LoRA/blob/main/diffLora/8dim/[diffLora] darkstroke.safetensors
and this lora for middle finger pose:
https://civitai.com/models/7016/middle-finger-lora
these middle finger loras seem much better than the old one I was using https://civitai.com/models/190750?
rude lora
model pls 3:
copilot pro
How to use this?
like this

Oh... Well a surprise to no one I guess.
got my fair share of good images made by that too but lets not disrespect sd here lol
and I like to tinker with SD. Yes, I don’t know how to draw beautifully. and even more so with paints. but I love illustrations and I love the style of Anime and Manga. So why not generate illustrations? even if what I like comes out on the 30th or 50th try. and after 100500 corrections. 😁
am not picky, i'll use whatever can output the style of image I want the best at the time, sometimes it's sd, sometimes designer takes the lead, still waiting on the WDXL update... one day maybe or a future SD update (never midjourney though. they don't accept paypal kekw)
wat?
I wanted to make a picture, and the system called me here.
bot- here in offline
Are all offline?
think so. everything people show here was created on their local computers
local or something like dreamstudio/designer/myriad of other things
if not local it usually costs money though
Yes, it’s possible and some services on the network
well
why not try using it locally? it's very interesting and exciting!
I'll try hahaha
I almost never use sd for this stuff because i did before, and i used sd so much i noticed the little things and stuff. Like the ways it draws eyes, coloring... and so I began to hate the art sd creates.
still fire for other things tho just not... this
sounds like a skill related issue
Hey guys, I'm looking for an anime model that supports things like colored sclera (black sclera) and synbol-shaoed pupils (heart shaped, etc)
TKS
Exactly like that
any model can do that,u just need a lora and the adetailer eye model,then u put the eye lora on the adetailer prompt
I was told there were models that could do black sclera at least, but aside from a model called pony diffusion, I haven't found another model that could do that. Thanks for info on the symbol shaped pupils though
im actually basically a mixed of that model, and it's pretty good if you can run sdxl based model
if this is too much, let me know
there are loras that can do that
Furry models are far better at black sclera from what I've seen. Pony for XL and EasyFluff for 1.5 would be my first choices. The LoRAs I've tried for black sclera haven't been very good.
if u only add them to prompt it wont give good results u have to use them with the adetailer eye model to get better results
depends, if it's sd1.5 model then probably, but sdxl based model should be able to handle that better
but it will take longer so its probably about the same time
I think for now I would not be able, since I have a 6600xt for a GPU until March that my 24gb GPU arrives, but I'll try
nah its standard issue
u could also try to just use an airbrush to paint the eyes takes me like 10 secs on CSP
Yep, what model/Lora are you using?
EasyFluff with the hll lycos to de-furrify it would probably be your best choice.
with controlnet + my drawin tablet i can create anything i want with SD if u have those two and know how to use controlnet which is easy u can pretty much create anything you want
1girl, solo, white hair, double bun, black sclera, heart-shaped pupils, yandere, hanfu, looking at viewer, head tilt,
Negative prompt: (worst quality, low quality:1.2), 3d, censored, abdominal muscle
compression artifacts, jpeg artifacts, jpg artifacts, blurry, noisy, grainy, low resolution, low quality, worst quality, bad quality, badly drawn, incoherent, error, fault, bad antomy, censored, blotchy, beginner, flawed, crude, muscle, sketch, wip, render, cgi, photo
Steps: 35, Sampler: Euler a, CFG scale: 7, Seed: 1052218664, Size: 768x1024, Model hash: ac006fdd7e, Model: autismmixSDXL_autismmixConfetti, VAE hash: 63aeecb90f, VAE: sdxl_vae.safetensors, Denoising strength: 0.51, Clip skip: 2, ADetailer model: face_yolov8n.pt, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer mask blur: 4, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer version: 24.1.2, CFG Rescale phi: 0, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Version: v1.7.0```
Thanks a lot, I'll have to take this day off to learn to use adetailer
Love to get back into sd so if u can prove me wrong thatd be great
i would like to as well but is there any benefit for me
ok then
came here looking for good stuff, find the good stuff, asks for model... and its Microsoft's children dammit😭
takes a while to create something magnificent with a drawin tablet + sd like for example something with like 12 hands doing ninja signs
i have to say i wouldnt waste my time creating that just to post it here for free
i type words, press button, and "art" comes out
we used to post lots of styles here when the chat was alive so like 2 months ago try to scroll up maybe u can still find them
No way discord will load those images for me lol
FINALLY. SUCH AMAZING ARTWORK WHAT MODEL IS THIS??
did you link it to the gif or am i insane?

Thats the joke fam
that's what i was wondering, sorry my joke factor isn't that advance 
here something really good
Oh my god!! i came twice looking at the image!
hot gentleman ♥️

I dont mind it personally
Welp, until I get a new GPU I'll just have to make do with LoRAs and inpainting, thanks everyone for your help
if any of my gen is too much, please let me know 
I think you guys mostly know whats good and what isnt, it tends to be the newer users testing the waters with some stuff imo
Your gens look awesome, that model you use is superb, sadly it's only sdxl
I'm looking for similar SD 1.5 models but can't find anyone
hmmm, i guess that's the kinda the difference when comparing sdxl with sd1.5. SDXL knows the context a lot better than sd1.5
Yep, I'm looking for model with the same style but nothing so far, I was using anything v5 but man, it's really limited compared to others
direct me some images with the style you are looking for, and I'll see if i know any sd1.5 model that can do those
cook your own chibi
grabs tail
no need to grab her by the tail. she is good. and sweet... and kind...
that all! nothing more....

I feel like slippers are about to fly at me again
look like result on-line service SD
awwww! I love hiking!
me too
I actually have a picture very similar to this that i took of my niece on a hike
after a short hike, the destination is reached
setup for the night 😄
morning in hiking
still sleepy, 5 more minutes
wait I postd those already oops
I love your pictures so much
Does anyone know what style prompt would best describe this? I was thinking ‘digital painting,’ but I was curious if there was a better alternative.

Thanks, I'll give them a shot.
ffs, not a single black sclera lora I tried works, goddammit. Maybe they are not getting applied, who knows.

Whyyyyyy
Lmao
I tried like 4 models and not a single one is doing what I want
Pony or autismmix can do them pretty well, but they are sdxl
I think I'll have to bite the bullet and use sdxl, I'll have to be patient, that's all
I've heard sdxl can work on 8GB VRAM
RX 6600XT, a Tesla P40 24GB is on it's way, but it will arrive on March
So I should probably just stop trying to achieve black sclera on peasant sd 1.5 models lmao, or follow the guide a user sent me on easyfluff and anime lycos
Linux: check
Comfy: not checked
So, maybe i'll move into inpainting and an inpainting lora I found...
or stop trying to achieve black sclera for now, ffs, I could have gotten a 2070 but I went for maximum penguin compatibility
@supple raptor sent me a guide. Another model works but looks horrible lmao
better wait two months
my demonic gens can wait until then
Sounds like prompt or parameter skill issue. EasyFluff looks good. Here are a bunch of examples. Other than the 3rd image (because I drew the ocarina) they all have metadata so you can get some ideas for prompting.
Might also just need to find an artist or combination of them from the hll lora that you like the look of. There are hundreds after all.
What model do you use?
I can answer that for him. He is using ponyXL
As I feared 😦
I went to play game
and forgot what life is and discord

Game instead of genning waifu? 
I will go sleep instead of genning waifus

Sleep instead of genning? 
What if you dream about genning images
The zest look
ur thoughts guys
my first thought was that the last episode this season was garbage
whose dad is this
oh yea still havent watched JJK waitin for the s2 bluray 😔
you probably shouldn't post peoples drawings as AI
even if it's your drawing
can I post ai drawings of ai?
But you cn turn it into AI art 😄
you could, but adding a fake background and putting your IG handle on it isn't cool to begin with
or what would you call it, hmm up-arting with ai
post taylor swift instead
as long as you drew it, who cares
that's mine bruh
so you're mrc-k?
i mean, that's fine if it's true, but a drawing from ~3 years ago doesn't belong in a AI channel
dont screenshot my nft
use ai to make reference art then draw it and then ai it
u talking to me bruh
bruh
bruh
bruh
u talking all the fun captain varghoss
NOW it's AI art
🤷♂️ i have a thing against stealing peoples work
but its mine
don't think that word means what you think it means
fixed it
ok, but like i said, even if it is yours, why would you post it here
uhhh really?
everything on this server is ai
you can post normal art here if u want,i have posted a few from random artists here as well
i stole this one
naah
I'll just go offline for another year
my honest reaction
real anime char 🗿

😄
i like closeups
there switched to my anime merge
lol, added 'taylor swift' to the end of my prompt
the number of time's i've had to regen/crop/inpaint something because my model is too horny...
well that wasn't the problem, kinda... it had a painting next to her that looked nude so.. no
yeah that happens to
i was making rave images and it kept making random nudity in the backgrounds 😄

in the background lmao
I've never seen that happen
i don't usually make images with crowds in the background because they look so bad, so i almost missed it
@native halo oh no, theres fire, help the priestess scape 
How do you fix the faces
we need to spray her with water to protect her from the flames 
https://www.reddit.com/r/StableDiffusion/comments/1ajxus6/stablediffusionwebuiforge_for_low_vram_machines/ anybody gonna test this?

interesting
Forge backend removes all WebUI's codes related to resource management and reworked everything. All previous CMD flags like medvram, lowvram, medvram-sdxl, precision full, no half, no half vae, attention_xxx, upcast unet, ... are all REMOVED. Adding these flags will not cause error but they will not do anything now. We highly encourage Forge users to remove all cmd flags and let Forge to decide how to load models.
seems to have some interesting performance increases, not just vram usage, if it's accurate though, massive speed boosts for 6-8gb card users
im tempted, since im a 8gb vram user
but im lazy to test, so kinda want someone to tell me if it works or not first

I don't really need the speed boost with my 4090, but it supposedly speeds up other things like lora loading and vae processing
also it's the same person that makes controlnet, so that's built in
apparently they also make fooocus
i didn't even realise that
tbh, i didn't read all of it so 
but my confidence level of this goes up a lot more now that you mentioned that
i recognized the name on the repo link, illyasviel 😄
i was thinking the name was familiar lmao
Thanks to Unet Patcher, many new things are possible now and supported in Forge, including SVD, Z123, masked Ip-adapter, masked controlnet, photomaker, etc.
also, built-in lcm and turbo samplers
i wonder if it works for low ram users like me,i still stuck with 16gb 😔
mine just finished installing
low ram and low vram? 
lol
let me know your result 
is 16gb ram considered very little now?
i now have a 3060 12gb but have a lot of programs open that cant close so sdxl gets laggy and animatediff 1.5 also lags when i have loo many programs running
well, right off the bat, i'm getting about ~7it/s instead of 6it/s on a basic sdxl gen, seems to be on par for the 1-6% increase stated for a 24gb card
vram usage seems to be WAY more stable though, idling at 8GB, only spiking to 9 during gen
a1111 for me usually sits at 9.5-10gb and spikes to ~12 during a basic gen
are you using same model folder?
yeah
it comes with a model, yeah
ahh okay
but, i use command flags in the user.bat for my own folders
--ckpt-dir "D:\AI\Models\checkpoints" --hypernetwork-dir "D:\AI\Models\hypernetworks" --embeddings-dir "D:\AI\Models\embeddings" --lora-dir "D:\AI\Models\loras" --vae-dir "D:\AI\Models\vae" --realesrgan-models-path "D:\AI\Models\upscalers" --esrgan-models-path "D:\AI\Models\upscalers"
makes it easier to have multiple installs of UIs since that models folder is nearly 580GB
no more 1060 steven
loading loras is way faster
used to have a ~10s delay after adding a new lora to the prompt before it actually did anything
now it's maybe a 2s delay
its still there in a box

time to see if the TRT extension works on this
this is reaally working !?!!
no idea yet, some of us are testing it now
@vital raptor have got good result tho

it's functional, looks basically the same as a1111 but with a few extras built-in, still uses the same extensions (mostly) as well
i'm not going to be the best judge for most of the performance increases since I don't have a smaller card
i also noticed this torch: 2.1.2+cu121
As far as I understand, this is a certain version similar in appearance to the A1111, but much faster for cards with a small amount of memory?
also just realized the slight performance increase i do have is without xformers 😮
they basically took a1111 and upgraded it
probably because the a1111 dev is slow and randomly goes missing for months at a time
i thought they said to removed all of them?
so basically it might be working with xformers already, you just didn't have to put it in?
if I install it, it won't break my a1111?
as long as it's not in the same folder
I have a new 'forge' folder next to the a1111 folder, it's fine
oh shit
i saw that as well
so you mean with xformers is gonna be even faster/better?
maybe, maybe not
i think its using the sdp attention, which is about the same on my card as xformers
like maybe 0.5s difference
do extensions also work in it?
yeah
notes on extensions
controlnet is built-in, so that's not needed
apparently animatediff is currently maybe not working
but others should be fine
It sounds very cool if on my 1660 6Vram I get at least a 50% increase in speed with all the same functionality!
yeah, it's a game changer for a lot of people if it's really that good
faster funya
really?
I haven't tried
this gen took about 40 seconds for me
and if im not wrong, it was taking about 1 minute before
the 16 cards really would love some optimizations
adetaler not work in it?
should work
but not sure
i can't find it in extension tab

well install from URL it is
yeah i dunno why it's listed like that, it's called ADetailer literally everywhere else

also, if anyone was using path flags in their user.bat like i do, the controlnet one doesn't work
50 seconds for this with adetailer
but, there's a new setting in the settings tab for a controlnet model directory
it's faster definitely
omg, it auto-loaded my TensorRT profile
and how did you specify the paths to the folders of models and other content for this WebUI?
looks like maybe TRT isn't working, it's loading the profile by itself but it's not going any faster
Hi
👋

hi hi
I only knew about it because people linked me the extension
before talking to people about it in this channel for example I only called it after detailer lmao
Then I started going around #📝|prompting-help for fun and I was undecided on wether to call it adetailer or after detailler for a while cx
im getting to 38 seconds for sdxl generations

SDXL is being a pain in the ass for me
it's always a pain to use it, but it's finally hitting for me
I've been trying to replicate this style for the past half hour now
How the fuck does my old as shit model not keep throwing shitty flat shaded LED lighting all over the place
But animage does 💀
have you been generating with higher batch size?
no
cos if im not wrong, i can now go batch size 2, and it's the same speed as my old batch size 1 speed
i don't generally do batches
same i don't
but kinda interesting that batch size is not a lot more worth it to do now
batch size 2, took about 1 minutes 11.6 seconds
previously it would probably take about 2 minutes?
i can now run batch size 4 dafug
Guys i need help
hmm, it said something about using less vram from hires fix, so i guess i can test that
Why my faces are bad
show generation
skill issue
None of them were good so I didn't save any
try using extention !After detaler
my face is also bad,do what i did wear a mask
Okay....
I tried the inpaint method but it didn't work
Lmaoooo
well, firstly, if we don't see the generation, we can only assume what the problem is. It might be your prompt, your negative prompt, your model, or just the subject is far so you need to use adetailer
or your resolution is so low it can't do anything else
I use very small prompt
2 min?
3 minutes 9.8 seconds
with Hres?
Negative prompts I keep things like more than two arms etc
ahegao mask 
really try show something

yup, 35 steps euler a, hires step 10, upscale by 1.5x
@pallid moss go gen 1girl, portrait, maid outfit, high quality, best quality
and send here 
if you're using 1.5
WOW! i am install it and run it!
😄
omg i am forgot setup my A111 folders....
probably
you'd have to hardlink them though i think, soft links might break some extensions
hmmm
hard link?
Wait
a softlink is basically a folder telling windows "the stuff is over there, not here", and hardlink is more like "here and there are the same place"
hmmm i reading and use SymLink
Don't beat me up if the character looks like mashed potatoes and her figure looks like some kind of boss out of elden ring
。
i must setup this is too in set VENV_DIR=%A1111_HOME%/venv in webui-user.bat??
no
👍
my command line argument i must copied from old bat file??
I'll beat her up if she looks like a boss! 
I want loot
Meanwhile me lookin at my estimated 20mins time
what is your gpu
unless you need something special, you shouldnt need any arguments except the ones for paths
Nvidia GeForce GTX 1650


don't think about my gen times, it never makes anyone happy
interesting that the interrupt button takes two click now
except me 😄

hmmm .. not sure special it or not COMMANDLINE_ARGS=--medvram --xformers --no-half --always-batch-cond-uncond --theme dark
good for people that do more than one batch count i guess
Bro
you can keep the --theme dark part, nothing else is needed
With that xformers command line i was getting some error
Like something called Nan, try using ---nohalf command
xformers will greatly improve your performance if you can get it working
Can anyone explain to me how refiner Hires fix works?
Sis 😉
well to start, those are 2 different things
refiner or hires fix? it's two different things
Yeha ik but what do they do
for refiner, it's basically like changing model mid generation
hires fix basically takes the image generated, upscales to a higher resolution and then re-generates a few frames
to clean up details
i say mid, but you can define at which step, example you are doing a 40 steps generation, and you put refiner at 0.5, then your model will change to your refiner model at step 20
Okay so i was right but does it help in any way

depends on what you meant by help. Example you can use model 1 to generate something, then use model 2 to change the style to that of model 2. One benefit of this is, if model 1 is good with hands, or humans, and model 2 is bad at that, but good with style
Oh thanks
maybe try a more standard model like couterfiet or something
@lyric zealot what is the syntax in specifying the path to the shared installation folder a1111? now i am have this is set A1111_HOME=e:/3D_Projects/stable-diffusion-webui_v_1 but cant see my models in WebUI...
that one looks like everyone is either using loras or massive negatives to make it work
you're asking 2 different things
you don't have to share the install directory to share models
im not using that thingy, im just using commandline_args, might be easier
actually now that i look at it, it's the same shit
Perhaps I haven't formulated it correctly.
just they made it multi line
hi]
now i have thit is ```@REM Uncomment following code to reference an existing A1111 checkout.
set A1111_HOME=e:/3D_Projects/stable-diffusion-webui_v_1
@REM
@rem set VENV_DIR=%A1111_HOME%/venv
set COMMANDLINE_ARGS=%COMMANDLINE_ARGS%^
--ckpt-dir %A1111_HOME%/models/Stable-diffusion^
--hypernetwork-dir %A1111_HOME%/models/hypernetworks^
--embeddings-dir %A1111_HOME%/models/embeddings^
--lora-dir %A1111_HOME%/models/Lora
call webui.bat
and I don’t see my models that I used in the old version A1111
Only no half
hmmm
I'm not sure we're talking about the same thing
try "e:/3D_Projects/stable-diffusion-webui_v_1"
try
We are not.. XD
i cant run it 😭
is it not running, or just not loading your models still?
no no it runnig but it cant see my model folders
take another screenshot of the bottom of the cmd to see what it says about model
looks like it cant load that model for whatever reason
yep any models
you need to put the quotation mark
like --ckpt-dir 'C:\Users\User\OneDrive\Desktop\Stable_Diffusion_Models'
' or "
my last vertion bat file look like this ```@echo off
set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --theme dark
@REM Uncomment following code to reference an existing A1111 checkout.
set A1111_HOME='e:\3D_Projects\stable-diffusion-webui_v_1\webui'
@REM
@rem set VENV_DIR=%A1111_HOME%/venv
set COMMANDLINE_ARGS=%COMMANDLINE_ARGS%^
--ckpt-dir %A1111_HOME%/models/Stable-diffusion^
--hypernetwork-dir %A1111_HOME%/models/hypernetworks^
--embeddings-dir %A1111_HOME%/models/embeddings^
--lora-dir %A1111_HOME%/models/Lora
call webui.bat
does that not work?
this one says
e:\3D_Projects\stable-diffusion-webui_v_1\webui
but the other one you posted says
e:\3D_Projects\stable-diffusion-webui_v_1




