#🍥|anime
1 messages · Page 187 of 1
you need to use hires fix at like upscale x2, then even when genning at 512x768, you can get good generations


damn what is this model/lora? i really like the style
it should have metadata
1girl, white hair, bob cut, short hair, purple eyes, sitting, sofa, window, indoors, maid,
<lora:aoi_12:0.2> <lora:tafIsAllYouNeed_taf2022B4:0.2>
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, text,
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 22, Seed: 66833500, Size: 512x768, Model hash: 51e4b0d5e1, Model: SoushikiV1.0, VAE hash: 0468eb8aab, VAE: kl-f8-anime2.ckpt, 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: 24.1.1, CFG Rescale phi: 0, Hires upscale: 2, Hires steps: 10, Hires upscaler: 4x-AnimeSharp, Lora hashes: "aoi_12: 1aae14b8ed57, tafIsAllYouNeed_taf2022B4: faa401305beb", Dynamic thresholding enabled: True, Mimic scale: 7, Separate Feature Channels: True, Scaling Startpoint: MEAN, Variability Measure: AD, Interpolate Phi: 1, Threshold percentile: 96, Mimic mode: Half Cosine Up, Mimic scale minimum: 4, CFG mode: Half Cosine Up, CFG scale minimum: 4, TI hashes: "bad-artist: 2d356134903e, bad-image-v2-39000: 5b9281d7c6ad, negative_hand-neg: 73b524a2da12", Version: v1.7.0```
tysm! could i have the link for the loras? i cant find them in civitai
https://civitai.com/models/8717/aoi-artstyle-lora-semi-realistic this is the aoi lora
i was using the test build, but it's the same
https://civitai.com/models/10088/taf-is-all-you-need this is the other lora, might have nsfw artwork
tysm again, really nice style

can someone tell me why my image have this weird waterpaint texture?
lora:Yelan_Hard:1, ((masterpiece)), best quality, 4K, HDR, Realism, VFX, Volumetric Lighting, Volumetric Light, Volumetric, Natural Lighting, high details, detailed and intricate, intricate details, high intricate details, absurd amount of details, absurdres, high resolution, Medium Shot, Single Shot, Shallow Focus, Low Angle Shot, fine art, soft colors, black outline style, elegant, very detailed Cel-Shaded With shaded detail's, soft shadows, in frame, solo, woman, short hairs, bob cut, diagonal bangs, bangs covering left eye, blue hair, multicolored hair, green eyes, blue earrings, breast, large breast, black kimono, infront of a traditional japanese mansion,
is it due to your lora? If not, then try using a really simple prompt
then see if problem persist
definitely not lora, though i'll try your advice and reduce some prompts
mornin'
the only recent changing i remember doing is changing hires. fix upscaler to swinir 4x, could it have caused it?
possible, do a plot xy
man, i hate that. it take soooo much time
what's wrong?
you can do plot xy with just two settings, your before upscaler, and your new upscaler
hmm, i might first wait for the next batch to complete, it's already 40% done
There are many different safe tensor models in various folders.
Huuuge
I tried it and there are still things it makes in the background. But I think big blank spaces are filled by SD1.5, it doesn't like empty patches so it fills them (in this case with a person and flowers). But the output is better than the other ones so far. Cleaner
But I'll keep experimenting
i changed the upscaler and removed the prompts that might have been affecting the style, it's still doing that. Now even i'm thinking maybe it's the lora.
have you tried the lora with the model it was made for?
doing 1 without any lora just to check
or maybe something simple like 'watercolor' in the negatives
interesting results
i feel like i need to write a story to go with it 😄
damn, looks like a manga panel
that's what i prompted
I like the color renders better though
the light colors and shadows are so good
hm. but the b/w looks more mysterious
easy change
nah man, it's definitely something in my settings or prompt. This one is without any lora.
since you said you changed upscaler, try changing that
or maybe not? in this one on the clothes looks water-colored
that's a lot of prompt...
A work in progress of the second page of my manga test
a better question is why does your prompt have stuff for realism and photography terms, all that extra garbage isn't helping you any
you can tell because your prompt says black kimono and it's coming out blue
ah, i though i deleted the black part
4K, HDR, Realism, VFX, Volumetric Lighting, Volumetric Light, Volumetric, Natural Lighting, high details, detailed and intricate, intricate details, high intricate details, absurd amount of details, absurdres, high resolution, all of that is pointless
even if you wanted to keep some of the terms, there's no need for 3 'volumetric lighting', it doesnt change things
i mean, technically it changes the render, but it isn't meaningful
i see
did you find out it was the lora?
I think it's the model
what's your default prompts? I mostly have those and I copy-paste them on every prompts
this is an sdxl model, and i turned off the vae/face restore, but otherwise it's the same data/steps/etc
I don't have default prompts, I type out what I want and then refine from there
It's different for every model, and sometimes it's different on the same model but different goals
oh, i'll have to try that approach
just as an example, these are simply anime girl with fox ears and anime girl with white fox ears and anime girl with black fox ears
basically the same prompt, but notice how major details change, like the clothing style and the backgrounds
result with my default upscaler, before i changed to check if it was causing the issue.
simple things like a color of hair/ears can change how the rest of the prompt behaves
Definitely prompts then sigh
anyways, thanks for the help, it's already around 11 pm here. Good night.
was trying to make a fancy long winter coat, but it makes it so shes ONLY wearing the coat 😄
can't wait for summer.. ❤️
mods online 

Any model with this style?
Try setting LoRA to 0.6 weight
And remove a lot of these details things from your prompt
Maybe remove cel-shaded even, it doesn't seem like the model understands what it is
guys, do you listen to mili? they released a fantastic cover of an older piece
hello all
who?
[ MARCH 20, 2024 ] Mili CONCERT in TOKYO
REGULAR ADVANCE TICKETS: https://l-tike.com/mili/ (Japanese site only)
ONLINE LIVE STREAM TICKETS: https://www.zan-live.com/live/detail/10389 (English or Japanese)
[ links only work after premier ]
Album streaming available J...
this group
wow
eheheh
I use img2img+canny SDXL controlnet but the problem is the eyes look at the viewer instead of the direction the original image does.
I am loin these pics
I've only listened to a few of their songs, but those ones are bangers
milli vanilli?
no
this piece is insane https://www.youtube.com/watch?v=tQ77IdoGgFA&list=RDGMEMCMFH2exzjBeE_zAHHJOdxg&index=4
Provided to YouTube by MAGES.inc
嘆きノ森 · Ayane
ARCHIVE LOVERS
℗ MAGES.Inc.
Released on: 2013-11-20
Auto-generated by YouTube.
mark wahlberg as batman?
chris evans as batman
using wildcards I got : anime Drawn style, ((gary oldman))as Bakumon, posing in world war 1 with gradient, HD-DVD, pixel art render, art by Toei Animations,
I love wild cards now
dont be ridiculous .. thre is nothing obscenely revealing
but it does appear to be so yeah that can get ya in trouble but you do you bud. I tried
your imagination getting the best of you
If you've seen Goblin Slayer, Mili did the op for season 1.
can i get the prompts you use for those diagonal bangs and hair covering one eyes?
oh yes still waitin for the blu ray release of S2 to watch from beginin 
Here's a gem of a meme template from early in S2
thanks torq
lol, those blood stains on the floor doesn't look ominous at all
that was a random thing was suppose to blood red part of clothing or something shrugs
stephanie szostak standing in biopunk area, wearing cowboy outfit that is Blood Red, hair is fishtail crown braid that is Blush Pink, art by Norman Foster,
hmm should add a random anime artist too
what's with that 4gb vram thing, i think you're the 3rd or 4th person with that in name.
ah i see
swimsuit 2piece
a girl in orange spandex, posing, glamor and beauty, fantasy theme backdrop
behold my masterpiece
mine looks sexier
dam, that looks like abhorrent love child between sexy squidward meme and ugly bastard tags from offbrand anime.
when i'm using two textual inversion like easynegative and badhands, do i need to put comma in between them?
its probably not required, but it makes less work for the tokenizer
usually yes, but you can also add space in between
tested with just spaces too and they work fine

Always , and then space after
Otherwise it looks ugly!
All you need is a space. The comma is just an extra token. Since TIs work outside of CLIP there's no benefit to adding a comma between them. That being said, I still do it because it makes the prompt easier to read.
do you use stable difusion?
Just a little. I've only generated about 2 million images 
nice
nad what do you do with these imagens?
Mostly share them on discord. Some become character art for TTRPG characters. I did a project for someone that needed a bunch of Touhou character images for their website. Sometimes I use them to practice drawing by picking a gen with problems and fixing it by hand.
its a game
you should have someplace to put the art
I put some of the ones I like on aibooru
a ok
Tummy
any one know how to save presets so i don't have to adjust all parameters again and again whenever i reopen sd?
after reopen SD just try press this is
this is load from file your last parameters
never knew that existed, thanks man
ammm.. i am a girl 😁
Real anime girl 
but still, is there a way to save presets? like if i use many differnt models and everyone of them prefer different parameter, i would like to make presets for them
Yes I think it's called styles
This button @winter fable
hm will look into it then
i see 😅 lol, thanks
should have ask this here from the get go, i asked this on sd chat, still didn't got a clear answer
guys always happy to help here. and I really like it!
I help if I happen to see someone asking for help but I'm not reading this chat that actively currently and also when Nubby isn't typing already
As soon as i open sd, it eats up 2 gb of vram. is it only me or it's suppose to do that?
It loads your model to vram
i understand
If you have problems with running out of vram you can install Multi diffusion extension and enable Tiled VAE to save some vram
https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111
megacool extention it true
It's not slower if your gen OOMs
ooms?
Out Of Memory
out of memory
the slow one it's me lol
yeah true, but it's good to know that's it's better to use it only in such situation
of course it not a 5 sec. 😁
slow is when a request to change just one line is still pending...
using comfy for example, i only use tiled when upscaling, by setting
😂
i know that feeling -.-'
I use it when gen would make my GPU hit 100% use since it would just slow gen and cause videos to stutter and in this case it wouldn't rly be any slower
pretty sure you are thinking about sdxl now XD
No with 1070 8GB I can do 1080p images with highres fix but highres steps take long time and pretty much caps my gpu
i suppose upscaling x2 may be a problem even for 8gvram cards
GPU hit 100% ?! hmm.. my GPU hit < 10% but VRAM
when upscaling
Oh it just dropped when I took screenshot
in task manager this is SD server
but for me, Cyrillic is an alien language
Why GPU is just GPU but CPU isnt
slangs are a weird thing
anyway, the models are loaded on gpu, but the gen looks on cpu
because the gpu is not staffed full
so, the generating work i think is running on the cpu
unless the gen is so light that it's doing it's magic in that 28%
I would gladly like to use at least 80% load when generating if this affects the generation time
didn't generate stuff in a while and got those 3 cute renders xD
if it's loading on cpu, unles you load more models, i don't think that's a thing
but that will only change the details, don't speed up the generation
when upscaling, tiled is a must, for low spec card
bad... bad...
for example, i can't gen images bigger than 688
what you mean?
in SDXL?
that's why I was surprised
so, i configured the upscale to set every tile at 688, but that's pushing my pc to the limit
without leaking
ah you mean upscale process?
well, i use a weird high res fix on comfy to upscale
with comfy, you can do very weird stuff
but when it works, it works lol
In a1111 Tiled diffusion should be able to do same thing I believe
damn I forget that you work at ComfyUI ☺️
yeah, you can set the tile
Streaming available SEPT 5!: https://linkco.re/1UbnMv9z
Smartphone "Sony Xperia 5 V" Preinstalled Track
スマートフォン「Sony Xperia 5 V」プリインストール楽曲
Mili on Patreon: https://www.patreon.com/project_mili
Mili Merch & Goods: https://projectmili.store/
Follow Mili
Spotify: https://open.spotify.com/artist/0K05TDnN7xPwIHDOwD2YYs
Apple Music: https://itu...
to inspire your gens
it's like that for the last 5 min 😑
sorry but only this can inspire me https://www.youtube.com/watch?v=c8I4iu7Qvyg
Provided to YouTube by Aniplex Inc.
Bios · Hiroyuki Sawano · Mika Kobayashi
GUILTY CROWN ORIGINAL SOUNDTRACK
℗ 2012 Aniplex Inc.
Released on: 2012-01-25
Associated Performer: Hiroyuki Sawano feat. Mika Kobayashi
Lyricist: Rie
Auto-generated by YouTube.
Send image from terminal
the vae is clogged, or it's crushed
what does that mean?
or just press "Enter" in terminal window
I just wanted to see what step it's doing
ohh i like that, so cyber punk

yo that worked
Sometimes it happens to me that the server seems to be on pause. and if make the terminal window active and press Enter there, the server will continue to work
and I don’t know why this happens, really
Only times this happens to me is when I click on terminal
oh
Did you enable Tiled VAE?
weird stuff
unless the tile is too big, he don't need it
I use only Tiled VAE and it helps
i was testing the time difference b/w default, only tile diffusion and both tile diffusion and tiled vae. So i'm enabling them one by one
but if i'm not wrong, the tiled is not starting because the end result is smaller thant the tile set
This is what I use when doing 1080p or larger images so Twitch doesn't struggle
Nope just watching streams pretty much all the time
ohhh
use it on upscale step?
Haven't rly bothered to learn how it actually works but yeah it helps a lot with upscaling
i mean you use it after HiRes.Fix?
I think it applies to all steps it deems tile is large enough
And I'm on a1111 if you thinking about comfy
yep
me too
those. Is this always active during generation?
Here's pretty much same settings I use for large gens
Yes
when the resolution is too small, tile will get ignored anyway
It doesn't rly matter if you always enable it since it only checks if it's needed
my favorite song of the moment
https://www.youtube.com/watch?v=Eoe5IZHbarQ&list=RDGMEMCMFH2exzjBeE_zAHHJOdxg&index=27
・Download & Streaming
https://orcd.co/yoasobi_monster
The Opening theme of Animation series "BEASTARS" 2nd season
Original Ver.「怪物」(Kaibutsu)
https://www.youtube.com/watch?v=dy90tA3TT1c
Original Novel(English Ver.)「I Press My Own Ear to My Chest」(Written by Paru Itagaki / Translated by Emily Balistrieri/Arranged by TranNet KK)
https://www.yoas...
that's what's truly inspires me
nice Anime! i love it
Damn scroll double clicking
never saw the anime, but the opening is a banger
Is DPM++ 2S a Karras rly twice slower than DPM++ 2M Karras
50 min?!
Normally that took 20min but I had different sampler accidently
I was wondering that generation felt like was taking a while and I wasn't wrong
and you use Controlnet also?
Yes I had hand refiner on
what kind of beast is this?!
ADetailer hand detection + cnet depth
hand adetailers is a thing i still can't implement on comfy
You don't have hand detection model for Comfy?
yep, but stuff is still rough in such sense
only a face adetailer with a weird noise with the form of eyes
i suppose it's stuff that help with face detailing
and if you use with hands, things gets messed up a little
model CN is model 1.5GB?
i have this is
that there is this other method, with a hand depth control net, but honestly works only with big realistic hands
rough stuff
the comfy model only works for Nvidia backends :cccc
i have nvidia and honestly doesn't work so well
I tried the hand mesh graphmorpher and the like that is supposed to use the control depth but it says backend not implemented :c although it does look like it kinda thinks and gets the hands, it should be amazing if you have nvidia
oh I thought it was better there, or at least workable. I have the depth inpaint hand model but my adetailer never detects it, even if its in installation folder and without any kind of symlink
it worked a little better with a 2048x2048 full body image
maybe with a close up, things are better
but hands were good enough from the start
oooooof that size would kill my gpu haha, but have you trtied the hand refiner models that come in with the ultralytics detectors? or do they not work
when you get distorted, or missng fingers, detections and correction still fails
some anime models produce three fingered hands even with TI
so, what's the point?
I guess then we would have to paint the hands in? or the fingers missing?
yeah, hands fixing is still a necessity
just to get the form right and than the model doing his magic
would you say it is possible to do img to img in comfy with some kind of mask for the hands?
SD is trained more on booba than hands 😄
I have seen that for clothing if you specify regions with colors in img to img it picks it up, maybe do hands like that?
tbh every single model in sd1.5 offers that lmao. Or at least some guy will push the model to creating that ^^
your right.... MegaBooba!
damn, also does anyone notice that some models struggle with controlnet hands, like either too long or just outright not following it?
use a better model then,a recent one that gets updated often
hands are also a very small part of the image, so it doesn't focus on them much
that's where adetailer and other solutions try to help, but they're not great
oh yeah the thing is I wanted this
but the hand came out weird, so I tried a little bit with controlnet but nah, any ideas on what I could do to fix em?
I mean inpainting works I guess, but I have heard using inpaint and controlne tat the same time
create 100 imgs until it comes out good or draw a good hand on top then img2img
would be faster to play with a 3d model of an hand
how would that work?
using the 3d editor tab or
that would take longer pretty sure i can paste a good hand from a random pic in like 20 secs
fix the pose and persective, make a screenshot, pass it on depth controlent and than hook it on hands
with and without hand adetailer, it doesn't even try to 'fix' the hands, it's just details, inpainting is still needed to get a non-deformed hand
that image is a little embarassing
on 67 generation of inpaint i fix her right hand
It would need like >0.7 denoise to work but I'm too lazy to actually try to get it work properly
personally I just use inpainting and generate a bunch until it comes out right, it's faster than anything else i've tried
WOAH 67? and how would you do that, like just prompting for a hand or with controlnet hand?
i just draw new hands, way faster <.<
just prompt
so like in photoshop you just draw over them?
ahhhhh I see I will try this for the image I sent, I loved the composition so I dont want to throw it away
Just take image of your hand in that position and then make that into depth map and inpaint with it
you artist!
how? If the hand is bad how would that work?
when i was younger i was very deep in drawing
do you knbow any kind of video tutorial I can follow for that ?
now i'm more focused in music
basically just photoshop a better hand on it and then img2img to blend them
it's not complicated
yep
no just sculpt a hand on zbrush im sure that one is faster i swear (real)
Nope but you can get depth map from cnet you would just need to clean extra
or model a hand from the blender cube

this is step i am undestand...
I get to the point of getting depth map of the image :c
that's what i was thinking about
You are Monster! 😁
or just buy a wooden model of an hand, hook it and shoot it on a white background lol
that works too
or just hire a real model and be a photographer, simple!
too much work
nah, it's easy, but models are a pain to work with
i was a wooden hands at 25€
or using DesignDoll prorgamm... yes!?
something like this but would need correct depth map first and then inpaint with cnet enabled
don't have that on comfy
what is that?
as i said, hand fixing on comfy is still rough
Depth library for a1111
okay I see, so it adds that depth mask and do you send the image as is to img2img?
hey this is now working in A1111?! Really?
Extension
or do you need to convert everything into a depth map and then send the image in?
this is what it actually gives you
the version I tried didn't work
Oh so what do I do with that?
(I am a complete noob in sd srry)
one of these days maybe i'll play with controlnet stuff
I'm not rly sure how inpainting with cnet works but I guess you just put that in cnet and mask hand area for inpaint
oh and how do you use it then?
I want to replicate your workflow if that is okay ^^
Don't have such thing for this
I don't bother fixing hands I just know this should be one way to do it
ooooooh I see I see, well I will try it out later tonight if I get the chance ^^ thanks for info though!
But there should be some guides how to use Controlnet with inpaint
Also kind of unrelated, (does anyone know how to check the ARIA2 queue)
but eventually I'll have to use this in Cnet?
I think I may have added at least 30 models into a queue with 2.5 MB/s speed and I want to check the queue to see just how long it would take
Yeah you just but that kind of image to cnet and then in t2i or when inpainting cnet will use that for hand
couldn't resist, sorry...
https://www.youtube.com/watch?v=Ci_zad39Uhw
Vocal:しぐれうい(9さい)
Words:まろん(IOSYS)様
https://twitter.com/maron47
Music & Arrange & MIX:D.watt(IOSYS)様
https://twitter.com/wattchan
Illust & Animation:ががめ 様
https://twitter.com/utsugame
Illust:しぐれうい
Movie:鬼灯わらべ 様
Narration:犬山たまき 様
Recording Engineer:丸岡詩織(studio MSR) 様
Mastering Engineer:片田浩史(Sony Music Studios Tokyo) 様
2022.5.25 RELEASE
1st...
bruh, that thing keep haunting me everywhere
the animation in that video is so good
just like Chika back in the day
especially this one
https://www.youtube.com/watch?v=srjLwgtcxYo
Making this to address shortcomings of previous Short of Shigure UI's 9mm Go Bang edit that features Missouri.
But I ain't stopping there!
Now featuring:
• THE FULL SONG with the appropriate UI's dance sync
• Yamato, for having the bigger 460mm caliber guns
• Cinematic aspect ratio
• Turret movement
• Better quality environment and render
• So...
lol
smooth animation 👍
gura posted one of her on twitter recently
i've seen that animation reused so much
I'll just add this as reminder
Discord shows twitter embeds again
doesn't look like it embeds the gif though
good to know
Chika Dance also great!
https://www.youtube.com/watch?v=eqjFmsZGBSc
Kaguya Sama Love is War ED / ENDING Theme 2 "Chikatto Chika Chika♡ " by Chika Fujiwara
TV Anime "Kaguya-sama wa Kokurasetai: Tensai-tachi no Renai Zunousen" Special Ending
Lyrics by @NeobraneEffect
⎚ Anime : かぐや様は告らせたい~天才たちの恋愛頭脳戦~ / Love is War Season 1 (S1/1st Season)
✎ Artist / Singer : Chika Fujiwara (Konomi Kohara
♫ Song Title : Chikatt...
not that i use twitter much, just something to scroll through cute anime stuff while i'm sitting at work for an hour with nothing to do every morning
i have this one trick model and i can't chose if it's flawed, or just very specialized...
i stopped using it when they changed it so you can't even access twitter without logging in
i should just delete it, becoming boring pretty fast
maigh characters lora can help
i login just so i can bookmark stuff for generating later
i just follow a ton of anime related artist accounts and block anything else that shows up on my feed
I just browse art and some game company news
https://github.com/wywywywy/sd-webui-depth-lib
https://github.com/jexom/sd-webui-depth-lib.git
what is the difference?
Depth map library for use with the Control Net extension for Automatic1111/stable-diffusion-webui - GitHub - wywywywy/sd-webui-depth-lib: Depth map library for use with the Control Net extension fo...

i use it for news, it's way faster than journals and tv
yes some guy called alex jones is pretty fast with them news
a fast search for the origin and that's it
The 1st one is forked version and has newer commit
first in the official list of extensions in A1111


Not sure if you know, kind lazy to read the answer but use the PNG info tab, your images are basically your settings saved, I have a little side folder with them and pick an image I like the parameters for to gen again
You didn't want to read my answer
style presets are good for some things, fore others I use this extension for settings https://github.com/Zyin055/Config-Presets
The main question in a keg is beer or honey?
hand lotion


No reading only cute elves from you

Do you have one looking down
what's the name of the extension that shows sd vae beside sd checkpoint?
It's a settings option
go to settings > User interface > Quicksettings list, then add Sd_vae
Settings> user interface

More 
Had a weird dream
And there was a pink lizard @lyric zealot
Very common that I dream with stuff that I was seeing/doing before sleep
Lmao
got it, thanks
Not pink bunny?
bunnies


yes

i believe there's metadata 
i have no idea what i did, but i think if i copy paste from browser, there's metadata as well

it works for other server, those with prompt inspecting bot
day 112349865 of waiting for this server to implement that bot

Do you have that shadow prompt or what ever it was called extension?
I don't remember its name 

Yes
Yeah bots can read it but I don't have it anymore installed
Guess I need to install it again

I just checked that there wasn't iTxt info in the image with notepad++
I think Sealth pnginfo encodes it to actual image
So even if all other data is stripped you can still read prompt if you have the extension
If you have the extension enabled
I just check this 1st normally
Cuute
so charming lol
when using detail lora, a lot of the time fingers tend to get messed up
and im only using a weight of 0.3
0.25 seems to work better

this one is the best.


Using Animagine XL v3.0
a girl in a backyard garden, flowers, plants, lamppost, evening.
Using bluepencil 3.10
guys, there are news, some dudes leaked viruses to corrupt models
copyright policy is one of the biggest bs around today
nothing new
teachers should copyright their teachings so students dont build career upon those teachings
you can't expect any good from a society based on competition
This one right?
Also you can remove nightshade with: VAE Encode -> VAE Decode
Apparently this is the super complex anti nightshade workflow
too much stuff for me
such sofisticated technology is too much sofisticated for my limited brain
uahaha
ok, added to the image loader
It was challenging, but I did it
!
ty for the information XD
We are: 
This is not a virus. It also doesn't work.
good to know
following the whole thing has been pretty hilarious tbh
Doesn't work
You mean it actually doesn't make better loras!?
no, it does, but not in the way the developers of the virus intended
I have no idea how that works
it sort of does that sometimes, but even that is not consistent. all it does consistently is make the original art look worse to humans

in my eyes, it may be an improvement
considering my workflow pretty much works in the same way <.<
giving the ai another chance of thinking out of the box, is without a doubt an improvement
is trying to answer to my stupidity without offending me XD
Huh? Their intention is to add invisible noise to an image that will trick blip captioning into thinking the image contains something that it doesn't, for example a cat when in reality the image shows a dog. In turn, if that part worked, we would end up with models that you ask for cat and you get dog, or ask for 1girl and get eldritch abomination, or whatever other tokens they decided to poison. Also, the noise that is added is very much visible to humans.
yep, you ask for a power ranger with a gun and the model answer with a power puff girl with an eiffel tower
what do you want more?
i must add it to my workflow, better than a wild card
It just doesn't work though. The poisoned data still captions correctly and trains correctly.
unless it mess with the color of my gens, i can't forgive such offense
they must improve it!
oh well, i will continue with wildcards
it's pretty much the same thing
jokes apart... it's pretty sad how there are people out there ready to ruin what simply works
I can't blame them for the attempts to protect artists, like Glaze and Mist (I think that's what the other one is called). Nightshade is a different story as that's a direct attack and if it were to work I could potentially see it eventually resulting in a court case. The whole intention of Nightshade is to intentionally upload poisoned images in a place that they know they will be scraped. The other ones are at least designed to dissuade the scraping in the first place by disclosing that the image has whatever countermeasure applied.
it's like the cure for the aids.
we had it at hand, the world knew and could prepare his welcome, but they chose to destroy it, instead of changing the way they make money
I have no clue
If AI dies so be it, if it doesn't, so be it
make Booba while it's here
just put the poisoned img on img2img with like 0.10 denoising and your "poison" is gone
About me?
Or nubby message
no, their mentality, to poison what is right
but i can relate to have fun till it's here
and move on when it's gone
People have too many problems that don't really affect them and want to talk about on the internet
i feel it's more a form of disillusion.
some, want just to hate something, but others are looking for a resemblance of virtue
it literally cannot die at this point
but i suppose i'm out of topic
Well I guess it can if we have total societal collapse.
eheheh
and there's no computers left
only realistic scenario where there's no AI left.
government crackdown would never work.
@native halo 
Not super easy to get 2 good characters through text to image only
imo people online whinging is mostly about a combination of insecurity and envy.
reminds me of Howl's Moving Castle
they were bitchin to japanese artists even before ai
its just how twitter/reddit works
I do not know what that is
Yes those discussion spaces are driven by insecurity and envy for the most part.
Prompt is just 2girls and luck
an anime
they are poisoned by negative yin energy.
insecurity may be bull's eye, but envy?
the two very much go together.
she can hug me anytime 
charming
she needs a hug too 
nice Rena 🙂
2girls 1luck
looks like some starwars guns

and it's correct!!!!
lmao
yeah needs more flesh
it's going to be great when we can generate Rena ripping out intestines from Keiichi in 1shot
biting them
makes me wonder if dalle could do it
guys lol
AI in 10 years gonna be nuts
gore and extreme violence is banned by discord, i don't think that the last Rena is even permitted here lol
Probably not gonna take 10 years at this rate for you to think that.
unless they lobotomize the teach to make money
I can only hope
I feel like nvidia has had amazing tech even 4 years ago
and it just never saw light of day
some of their youtube vids from that time period on shifting through styles of art are still amazing (must not be realtime)
I'd share the vid if I could find it 😢
oh, that looks promising too
I wish my model understood tails better
i like big feathers 
More hugs
@native halo you're now her servant 
licks
lick what 
the ice cream im eating 
what prompts or textual inversions you guys use for hands?
arms behind back 😂
(please give good hand rng god:999)
but in all seriousness
there's the goodhands lora
sometimes work
there's also the badhand-neg to put at negative, it's an embedding
lol, that's one way to do it
imma show you how it will fail 
(massive giant gargantuan colossal epic rtx 4k HD hand : 2.5)
4 fingers jumpscare
Cute
three hands
😖
big hands
give that girl an arm
very grainy
Yup, probably due to the character Lora
I tried to use ram instead of vram to store a llm and i got a bsod, lmao.
memory management error go brrr
oom?
the colors are very warm
wow
so pretty
Miku, Miku .....
https://www.youtube.com/watch?v=AufydOsiD6M
HATSUNE MIKU EXPO 2019 / MIKU EXPO 5th Anniversary Theme Song Official Music Video
Hatsune Miku’s world concert tour series “HATSUNE MIKU EXPO” is celebrating its 5th anniversary! In commemoration, we are proud to present the music video for the 5th anniversary theme song, “Lucky☆Orb feat. Hatsune Miku” by emon(Tes.). The video is a special co...
what a surreal effect
Also got this out, which I kinda love
conceptually! 👍
An ethereal vision of a wireframe angel with wings of light in a dark void 
wow
this is a 1.5 model? O.O
Yup
first version was better on my option
2 different models, this one reflect better the prompted age
and fixing the hand is too tiring
not worth it, but yes, i agree
damn, trying to install an online game, but god, 100gb for a single game
what's the deal with them?
Laughs in 400GB skyrim install
what? O.O

full modded?
1900 mods
is possible to stably load 1900 mods?
totally insane
But im also running a 13700k and a 4090
i would need a new hd for such a weight
Its on my nvme 
by the way. Don’t you want to try replacing the model for hands in Adetailer?
i have no hand adetailer
comfy is lacking a decent hand fixer
or at least i couldn't make one work decently
sad
yep
Laughs in terabytes
lol
maniacs!🤯
Keeping the downloads is madness
That's not rly problem for me since that's 4TB hdd although I should clean some older mod versions at some point
My only problem is that my Skyrim m.2 is getting full
I only got a 2 TB NVME and 2 500 GB SSD
Downloaded it, played for a few hours and forgot about it
The endless cycle
I wish starfield would have filled the void
I mod it until I brick my modpack and have to start again just to lose motivation
What model?
Fun times 
getting +2k mods to work is though
I know 
I have been there before, the modpack was my lazy option
Dont have the time to spend weeks on making my skyrim again
i know that feeling
I took ready made modpack but there is still many mods I want on top of it and some changes and the challenge is that that pack is near esp limit with all of its own manual patches
you guys have nvme? here i'm running game and sd on dramless ssd lol
Check out my other models: https://civitai.com/user/aji1/models ···················································································...
laugh till you can, meanwhile i burn this model in oblivion with passion
mt head hurts looking at the hand
can't even tell if it's left or right
right hand coming out of left sleeve? I think
@vil will make this a cute and dangerous elf!
eheheh
UWU wakaiiii!
nice
yep. like it
breath taking, what model is this?
https://civitai.com/models/182723?modelVersionId=205068
With 2 loras, should have meta data tho
Be sure to read the NOTICE in THIS POST before downloading the model! 23.11.29 Kakarot 2.8D V2.5 Update patches is below 'Show More' < cli...
@wooden coral
I had more mods to enable than I remembered
Luckly about 700 are armor mods that I can just merge
Any model with this style?
looks like an 80s style
crap! I am surprised by the work of Controlnet. Having a clear picture, his models manage to ruin everything completely. 😭
this model is promising lol
what happened?
t2iadapter models work very bad.. or my hands are crooked...
ah Thanks! I was actually talking about my hands😂
uahaha
i wasted 2 or 3 days with such models, meaninglessly
the results where truly mightmare fuels
But what do you guys say about “Suzume no Tojimari”, is it worth watching?
its makoto shinkai so you already know half the plot , boy meets girl across the time/universe and they fall in love
well, the artistic compartment looks promising
yes its an eyegasm
the musics and dubs too
"your name" for a new picture and a new look?
its good if u like romance
new musics too
then.... lets go!
when i think i need to clean my collection of loras



