#🍥|anime
1 messages · Page 70 of 1
right click any of the column labels at the top > select columns > check the box next to Dedicated GPU memory which is near the bottom of the list > click OK. Then you can click that column to sort by it and it will easily show you which processes are using vram and how much.
highly likely might be memory leak 
what can i do about it?
The simplest solution is to completely shut down your PC then turn it back on and launch SD again.
is that a permanent solution?
do you have the multidiffusion extension?
It's also possible that you're just using a larger model than you were before. Possibly an fp32 model or one that hasn't had EMA pruned.
nope
inside that extension, there's a button called
it works for me whenever i have memory leak with webui
ill try downloading it
but i haven't had memory leak since updated to the latest webui version
could you send the link to me? just so i dont download the wrong one
maybe
ty

What upscaler are you using
With which model can I get something similar? Something similar to those thick linens
Upscaler = 4x_fatal_Anime_5000000_G
Model = ExpMix_Line_V2
Lora used =
Yotsuba - Char
ligneClaireAnimeStyle - Style
1girl, orange hair, blue detailed eyes, green hair ribbon, volleyball uniform,
<lora:Yotsuba:0.7> nakano_yotsuba, <lora:ligneClaireAnimeStyle_v1:0.5>
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, censored, monochrome, greyscale, nsfw, nude, loli,
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 22, Seed: 1829849095, Size: 512x768, Model hash: 4d651c7638, Denoising strength: 0.53, Clip skip: 2, Version: v1.2.1, 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: 23.5.18, 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
Used embeddings: bad-artist [2a38], bad-image-v2-39000 [b03e]```
I put my yandere and tsundere chars in the same room. What could possibly go wrong?
can I use anime tags for training on sd 1.5?
I am trying to train a lora on myself but on an anime model, 100 steps vs 200 steps
bruh wtf is this
Your first mistake was training with images that were rotated. That's what we in the business call a....
the images aren't even rotated
probably problem with ui
or something
maybe bucketing problem idk
They definitely are. Phones do some weird shit with rotation of images sometimes. They'll save a preference of the orientation that you want to view the image, but not actually rotate them. Not sure of the best way to make sure they're actually rotated correctly. I would probably just open each of them in an image editor, make sure they're rotated correctly and save.
image editors show normal
Hi I need help, I have a short clip of a man sitting on a chair and I want him to move his hands a little. How can this be done? If any.
you move their hand
photoshop
I tried but he considered it sexual harassment
🤨
What is the best way to do this? This is a video I made with AI and all I want is for it to move the hands according to the speech or conversation
Hope it was nothing too serious. Good to have you back.
deforum + trial and error, i suppose
@supple raptor what service is that?
eh, got 5 teeth pulled, the Ds are good to me right meow
my brain literally just ignored this message until nubby replied to it
lol
nice to see you back and that you didn't die
Why is it still not possible with the help of AI to move a person's hands or certain movements according to speech in a relatively simple way
still waiting for my stable society role, big kek
The chat bot screenshots I posted? It's SillyTavern. All running locally using the Ooobabooga API.
it's a little too specific, you'd be better served doing it with animation software roughly, as a guide
then have AI make it a bit more fluent fluid
thanks!
there are plenty options, look for vtuber software
That's no fun, but could be much worse. Having braces through highschool makes me sympathize with anyone experiencing mouth pain though.
well i'm paid to be home and drinking milkshakes, could definetly be worst
The AI simply doesn't understand context in that way. It isn't the way it was designed. There's no way to simply tell it "move the hand up a little". You could use controlnet with the openpose model to create skeletons of each frame in your animation and manually adjust those skeletons, then generate images based on them.
this makes me wanna become a tongue twister
i need to get into that openpose shtuff, haven't had the time to look into it
only time i had was work and phantasy waifu online
man is soo outdated holy shit
could I have the model and prompt?
outdated but 15k in the bank~
It's pretty cool. Some of the other controlnet models are really neat too, but that one is probably the most helpful for animating.
I deleted exif of all the images
seems to be working better
👍
Nouse
Have you tried just looking better?
bruh

I don't look anything like this alien looking mf
projectai be like
Did someone pinged me?
Model is a private one. The prompt I stole from @supple raptor and fused together with an older one that I have saved
yo it kinda worked, do you know a tutorial for this extension?
masterpiece, high quality, high definition, intricate details, illustration, beautiful detailed eyes, beautiful detailed hair, Glowing neon, 1girl, highly detailed face, (mystical bioluminescent flowers:1.1), mid shot, full body, 1girl, mist, serene, magical, long hair, white dress, barefoot, sitting, sitting on a branch, holding a crystal, surrounded by glowing butterflies, a gentle smile, water ripples in the background, an excited expression, joy, (dark background:1.1), (darkness:1.1), a beautiful vortex of energy swirling around her, vibrant colors, finding beauty in the darkness, superb, 8k wallpaper, extremely detailed, intricate
Jesus, the difference a model make is fucking brutal
guys what adetailer model to use for faces?
fawk. my previous work is all on a borked drive ><
any idea for recovery ? seems the MBR is AWOL
what's the prompt for this? I like the red hair vs shiny black. I have struggled to get good shiny black.
It very much starts to feel like whatever can be prompted in terms of quality has been. The next step should be in terms of control, moving from macrocontrol to microcontrol over the smallest nuance
may not work for your model cause this is basically 4 loras stacked on top of eachother (this should have pnginfo btw)
hmm yeah I guess describing the materials used is the way to go... forgot about that. Brain fart. 😄
guys, amazing lora I found
this works too good on anime
for some reason
without vs with
show some more
how fix?
no
All the adetailer model is doing is finding the face, hands, or person so as long as it's consistently finding the thing it's supposed to, you really can't go wrong. I always just run face_yolo8s at default settings and it works consistently for me. I think the the media pipe models are only for realistic faces so they will likely struggle with anime images.
feels like i took acid again
Did you bring enough to share with the class? 
always 
I love the AI attempts to make kanji
Some of them look very convincing to me as someone who knows how to read exactly 0 japanese characters.
I imagine it makes as little or even less sense than the AI's attempts at writing in english though.
or weapons
Occasionally it can make some really cool weapons
Wonder if she's generating AI image or is that a CCTV live stream? 
looks like CCTV to me
That's an idea. Put the waifus to work making more waifus. Our output will grow exponentially.
yeahh, probably why she looked anxious.

if anyone updated locon extension
could you try replicating older images that used locon
cause it's not doing it for me
anyways I go slep now gn
I am pretty sure I also updated the webui itself
I am on dev branch though
why you using locon extension anyways? we moved on to the lycoris extension a while ago which handles locons as well as the other alternative lora types and allows specifying separate unet, te, and dyn strengths.
You'll have to move your locons to a new folder, but it's worth it.
didn't have to
but it's just doing the same thing the locon extension did tho
can't replicate older image
Yeah I wouldn't expect it to help with that specifically since you weren't using it before. Locon extension also hasn't changed in any way that should matter for generation since Apr 7th so I suspect the difference lies somewhere in something that A1111 changed.
probably
I can replicate images where I used additional networks extension
just not ones where I used the thing in the prompt
If he changed something about how loras are loaded then that makes sense because the locon and (I think) the lycoris extensions both hijack the built in lora implementation, while additional networks works completely separately.
me when https://github.com/KohakuBlueleaf/a1111-sd-webui-locon this extension is not for additional networks
I'm still pretty convinced that it is.
They have to go in .\models\LyCORIS and then to use them in the prompt as <lyco:NameOfLyco:1>. You'll also have a new tab in the UI to select them.
lora:LoconLoraOffsetNoise_locon0501:1 and lyco:LoconLoraOffsetNoise_locon0501:1 give me same output (I put it in both folders)
run civit helper pls
yea xD
Almost none of my Loras come from Civ. Most of them are from HF or Mega.
make some damn previews 
Ain't no body got time for that. I got random waifus to make and chat bots to make fight each other. 
any idea?
Bold of you to assume I've ever used most of these without 2-3 other LoRAs at the same time 

time is an illusion

what if he deletes his msg
moment
lmaooooo
lawddd
Oh no... whoever made that gif left a single frame from the scene change in it. 
Lol, which one got you? The snake or the tentacle face?

No offset vs with offset
I think I need to train at like 0.5 offset, or 1 offset XD
No vs with again, I feel like there is barely a difference
try making a dedicated noise offset lora trained on extremely dark images
a friend, that i waifu-ized
playing around with depth and 3d maps
woahhhh
I should try that
tbh i havent really got around and do that yet
pretty good, sure, what's up?
its been forever since ive trained a lora are these settings fine?
sorry for the super late reply, but I'm not sure if there's a tutorial available for multidiffusion
Guten morgen
Must be a Windows Desktop... 
I need to ask you for genrate images for me
I can't use sd
I have errors
Cool lipstick
How can i do this
Is like putting embeddings and negative prompts without putting them?
click the floppy disc to save your current prompt, click the dropdown to select your saved prompt
it's that simple
Ok thanks
They are taken into account. You could "materialize" them in both prompt textareas with another button. They are actually described - you could see description of all headings like i.e. what is denoising above slider or here, with hover over it
Which upscaler is better, extras, sd upscale or ultimate sd upscale?
Ok thanks
do you have controlnet?
most people use multidiffusion or tile controlnet
i prefer tile controlnet
I think they both pretty much use the same method don't they?
controlnet has better tile accuracy
but same method yes
Would it be just this? I saw in a video that apart from this you also use it together with ultimate sd upscale.
What is clip skip?
yes you use ultimate sd upscale with it
clip skip changes the way stable diffusion reads the prompts your typing
and do upscale as a last step
Should I use it? What difference does it make?
what he sent lmfao
it changes your results it could either be better or worse but you should try it
you could also go to your settings , user interface, and scroll down to quick settings list and type in CLIP_stop_at_last_layers
then it will show here
depends on models, reason people use it is because models such as NAI and some newer models train the models at clip skip 2 or even 3 because they noticed it helped (NAI did)
aida your english is very good lol
Should I use it with toonyou and aom3?
aom3 based on nai so yes
about toonyou I never heard of it
toonyou recommends clip skip 2 in their page so yes
also I don't like the detail
preference
Which model do you recommend that has a similar style?
similar style?
idk
Doesn't toonyou have lora also or do I remember wrong?
No
How do you use cfg scale 22?
using dynamic thresholding
Thanks
extracts it perhaps
I don't understand what is going on
problem with vae?
I will try in automatic
adetailer?
it's active 
adetailer op
it's good with the face module, but not the hand
that's why hands in pockets 
@lyric zealot
I don't want to wait 2 weeks but I'm too lazy to read 

I have tried these vaes and leave it on none but it still comes out wrong
ahhhh, actually i haven't even watch that show yet. But kinda know about the story
try no half vae arg
oh it's delayed?
amazing show
mushoku tensei s2 gonna be better though
I already have it
try removing it
Something else comes next week probably recap
yup, it was the only anime that beat FMA on mal for awhile
but it's back to fma at the top
obv
fma fans going to all other anime and downvoting 
hope i don't offend any fma fans here 
To beat fma both animes would drop under 9 after the war
it's never gonna happen
fma fans are too hardcore 

Yeah my model doesn't know Gura time for Lora
Can you run stable diffusion on an iPad Pro with the M2 chip?
shark girl 
there's that Draw Things app on iPhone, might be on iPad as well
Yeah I could probably just describe Gura

Gawr Gura INTERNET YAMERO 【hololive MMD】
全ての制作者様に感謝いたします。
お借りしたもの
◆Music:INTERNET YAMERO
◆Model:げのげ様
◆Motion:楕円
◆Effects : Rui様, あや様, what the様
#mmd #gawrgura #がうるぐら #shorts
dafug is this anime lmao
For next one I'll add lora activation tokens
wait you almost did it without lora tho
ctd be like
That's with lora and Gawr_gura token
gawr gura, tail, shark tail, blue hoodie, shark print, shark hair ornament, this is what I meant
success
Classic Mood Experience The best masterpieces ever recorded in the music history.
Join our Youtube: https://goo.gl/8AOGaN
Join our Facebook: http://goo.gl/5oL723
Ornette Coleman - The Shape of Jazz to Come (1959)
00:00 Lonely Woman (1959)
04:56 Eventually (1959)
09:14 Peace (1959)
18:11 Focus On Sanity (1959)
24:58 Congeniality (Nomad) (1959)
3...
this good
jazz
i prefer weeb songs like https://www.youtube.com/watch?v=51GIxXFKbzk
https://linkco.re/cSbbuymU
https://whysoserious.jp/needy
歌唱:KOTOKO
作曲:Aiobahn
作詞:にゃるら
キャラクターイラスト:こかむも、 お久しぶり
ピクセルアート:ねんない
背景:ccnaov
MV:TOHRU MiTSUHASHi、前田地生
Vocal Direction:まろん (IOSYS)
Special Thanks:Megumi Torii (Xemono)
Produce:さいとーだいち(WSS playground)
Aiobahn
https://aiobahn.net
https://instagram.com/aiobahn
https://twitter.com/Aiobahn
カラオケ...

i prefer trash music!
close enough
widdle hands

Anyone wanted catfish?
That is kinda vibes 
Well
Big kinda
XD
smile post big bitty thick girl right now.
I like the hardbass
The only acceptable type of catfish
then you might like internet overdose as well

Eurobeat and j-rock is my life blood
Have cowgirl
I mean, just some part of it is to for me xD
Fair, not the biggest fan, but some godo :o
understandable 
Also listened to the cyberpunk 2077 soundtrack before?
Uhh, some songs :o
I really like Never Fade Away


Yeah refused really did a good job on those track
I need emote ideas!
Although I prefer archangel
Yeee 
Fair :o
can i introduce you guys to psytrance? https://www.youtube.com/watch?v=R_uS0aT0bG8
✨ Follow Diversity's Playlists on Spotify:
➥ https://divr.link/SpotifyPlaylists
Stream/Download:
➥ https://monstercat.ffm.to/guitarmass
🎧 Genre: #Psytrance
Infected Mushroom
• https://facebook.com/infectedmushroom
• https://twitter.com/infected
• https://instagram.com/infectedmushroom
• https://soundcloud.com/infectedmushroom
Monstercat
• ht...
Ohhh I know that one, I like et :O

THOUGH
1girl, elf ears, choker, tactical clothes, holding handgun, Glock 18, white hair, blue eyes
I do feel like the start is wasted
fzjkhngnkl jznhiznhjklzsnghijkwzngkzwnhiuzwnwh
Because taht would have been some sick buildup for a metal drop
It is, but I wish it had a metal drop xD
Huh
https://youtu.be/4eUs1Uj_nKA Kinda like this
Title - 春色小径 ~ Rest In Peace, Saith The Lord.
Circle - Demetori
Album - Determinism&DestruKction
Game - 東方花映塚 ~ Phantasmagoria of Flower View
Original - 春色小径 ~ Colorful Path
This took a little while; I had a lot of work, and then had to get out on friday/saturday because RL =/.
a lot of psytrance don't have big drops
Old image I made
I see
Clip from Bocchi the Rock! Ep7
This is gold
Well, yeh, kinda is in the name I'd say, but I wish it was like a different genre of song and it had a juicy drop
xD
well i get you 
https://www.youtube.com/watch?v=zG0gJsCaVPM then you might dislike this remix as well
🎧 Support on all platforms: https://monster.cat/hero-remix
▼ Follow Monstercat
https://monster.cat/follow
▼ Follow Pegboard Nerds
Facebook: http://facebook.com/PegboardNerds
Twitter: http://twitter.com/PegboardNerds
Instagram: https://instagram.com/pegboardnerds
SoundCloud: http://soundcloud.com/pegboardnerds
YouTube: http://youtube.com/Pegbo...
it keeps building up, but no drop

i guess if you really wanted the drop, listen to original 
i like both
for different reasons
Yeee :o
||https://youtu.be/h2dJ-JUzhVs|| Best song tho
All rights belong to Bad Dragon and Pepper Coyote. Im just uploading this one song.
Check out Pepper Coyote here:
https://peppercoyote.bandcamp.com
https://twitter.com/peppercoyote
https://www.patreon.com/peppercoyote
Check out Bad Dragon here: https://bad-dragon.com/
Lyrics:
My neighbors ask me why I'm limping down the way
And who that fe...
This is what I confused it with

Discord blurs embeds too now?
Thats dope as hell :O
I thought it would just remove it like when you use <>
Didn't it do that before if you spoilered link?
It didn't, pretty sure :o
https://www.youtube.com/watch?v=U6n2NcJ7rLc anyway it's friday so
Riton x Nightcrawlers Ft. Mufasa x Hypeman (Dopamine Re-Edit) 'Friday' is out now. Listen here: https://lnk.to/FridayRNMHyd
Follow Riton
https://www.instagram.com/ritontime/
https://www.facebook.com/ritontime
https://twitter.com/Ritontime
https://open.spotify.com/artist/7i9j813KFoSBMldGqlh2Z1?si=iumrX7YQSFOdzexkfPN6pA
Follow Nightcrawlers
htt...
my youtube be like:
MY TRIP TO TOKYO JAPAN, IT'S LIKE BEING ON A DIFFERENT PLANET, LOL.

Where does one post their art now that pixiv and fantia banned monetization for AI art?
this is what I call mall music
IG, Arstation, Deviantart... Twitter?
Patreon? Various sites I guess
mall music? 
music they put at malls
no where i guess
maybe artstation
but people hate ai art so much on artstation so
get ready to be bashed

True

send me the silent meta data extension link again
https://www.youtube.com/watch?v=1TewCPi92ro this was the beginning
THANK YOU FOR WATCHING! PLEASE LIKE, COMMENT & SUBSCRIBE!
LIKE, COMMENT & SUBSCRIBE!
Official Page for Cousinskeether
Download New Mufasa Ring Tone Now!
https://itunes.apple.com/us/album/hoo-my-god/1480763807?app=itunes
Social Media
IG: https://www.instagram.com/cousinskeether/
Facebook : https://www.facebook.com/CousinSkeether/
https://twit...
I don't think that works anymore since discord supports exif again
Yeah
really?
Thats what I was thinkin
I thought that will still works with copying and pasting
待望のTVアニメ第2期!『無職転生 Ⅱ ~異世界行ったら本気だす~』
2023年7月2日より毎週日曜放送決定!
<TV放送>
TOKYO MX 毎週日曜24:00 / サンテレビ 毎週日曜25:00
KBS京都 毎週日曜24:45 / BS11 毎週日曜24:00
AT-X 7月3日より毎週月曜22:00(※毎週水曜10:00、毎週金曜16:00リピート放送あり)
<配信>
ABEMA・dアニメストアにて地上波同時・最速配信開始
その他動画配信サービス順次配信
【スタッフ】
原作:理不尽な孫の手(MFブックス/KADOKAWA刊)
キャラクター原案:シロタカ 原作企画:フロンティアワークス
監督:平野宏樹 シリーズ構成:大野敏哉 キャラクターデザイン:嶋田真恵・齊藤佳子 美術監督:三...
I believe Ashen dropped support for Stealth pnginfo
New season? :o
1girl, pink hair, twin braids, yellow detailed eyes, jacket, hands in pockets, nightclub, neon lights, dark, dance club,
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, censored, monochrome, greyscale, nsfw, nude, loli,
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 22, Seed: 4187648823, Size: 512x768, Model hash: e4fe0bcfc8, Denoising strength: 0.53, Clip skip: 2, Version: v1.2.1, 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: 23.5.18, 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: 4ËÚ"A IDATxœÌýyÐmÉqˆý2«Î½÷[ß÷ö_ï•$¸ ÁIPÔÃÑŒF¢ö±#ä˜ ÿá /š Ç8lÇ8Â8ì```
ignore the weird characters
this is like the third or fourth of fifth trailer already lol
i use edit with notepad++ 
Like he did for SAG, that hoe
they announced s2 a long time ago now

Oh damn, they been cooking for way too long xD
I just remembered this https://www.youtube.com/watch?v=BrmeOWbyOXw
captions obviously in the style of @ZHL242
Download transcriptions at https://georgecolliermusic.com and check out the Discord!: https://discord.gg/3qxNDA3NCy
Original video: https://www.youtube.com/watch?v=krLYZmPRtnc
Performed by: Leo P and the Metropole Orkest
Piece: Moanin' - Charles Mingus
Transcribed by Tom Pickles
MEMBERSHIP: https:...
literally her
ufo attacking? 
https://youtu.be/Z83mEbMJ_zk Also for all of you not watching Tengoku Daimakyou / Heavenly Delusion, go watch it, you missing out 
Provided to YouTube by avex trax
innocent arrogance · BiSH
innocent arrogance
℗ AVEX ENTERTAINMENT INC.
Released on: 2023-04-02
Composer: 松隈ケンタ
Lyricist: 松隈ケンタ× JxSxK
Auto-generated by YouTube.
Now I go back to milky posting

I will bite your toes while you sleep 
bite my toes tho
Fine, I will
the songs jammin in my ears rn as we speak
Good, cause its gud c:
Try listening to this
Taken from the album War Music out now https://refused.lnk.to/WarMusic
Economy of death dictates the division of east and west
Cold war cultists design an arms race for safe space
The working class is betrayed while capitalists do the bidding
Your occupation becomes that you die for a living
You’re so fucked
You’re so fucked
Economy of deat...
Oddly chill video for the music XD
I like it tho
And this too
Music from the game Ultrakill. Developed by Hakita and published by New Blood Interactive.
https://store.steampowered.com/app/1229490/ULTRAKILL/
Follow the game:
Youtube - https://www.youtube.com/user/Hakitarkus/
Twitter - https://twitter.com/ULTRAKILLGame
Discord - http://discord.gg/wbJZqfZ
Artwork - https://www.artstation.com/jericho_rus
Mot...
doom eternal but liquid dnb jungle with a touch of breakcore be like
This is basically most of ultrakill soundtrack
what is this https://www.youtube.com/watch?v=NvM28WMdRLQ
この町では変化は<悪>
岡田麿里監督最新作
MAPPA初のオリジナル劇場アニメーション作品
2023年9月15日(金)劇場公開
恋する衝動が世界を壊すー。
映画公式HP:https://maboroshi.movie/
公式Twitter:https://twitter.com/maboroshi_2023
公式Instagram:https://www.instagram.com/maboroshi_2023
■あらすじ
突然起こった製鉄所の爆発事故により全ての出口を失い、時まで止まってしまった町で暮らす中学三年生の正宗。いつか元に戻れるように、住人たちは変化を禁じられ鬱屈した日々を過ごす中、謎めいた同級生の睦実に導かれ、製鉄所の第五高炉へと足を踏み入れる。そこにいたのは...
I know that, I love Ultrakill
xD
ultrakill very good
Best fps game
Although I play a lot of ready or not compared to ultrakill
I didn't fullscreen the other one, but I did fullscreen this one
MAPPA
THE GOATS
GOD DAMN
Why are they so good
breakcore
Agreed xD
I want to train a Lora on this holy shit
Bad aim? 
is very good
Maybe
okay why stable diffusion just generate a rat .

Anyway
ぼっち・ざ・ろっく!-SPECIAL STUDIO LIVE-
「青春コンプレックス」
Vo:長谷川育美
Guitar:音羽-otoha-
Guitar:三井律郎
Drums:比田井修
Bass:高間有一
SSLはこのライブのために集まったスペシャルなメンバー構成で、
スタジオ収録したライブ映像をお届けする企画です。
=======
TVアニメ「ぼっち・ざ・ろっく!」
各配信プラットフォームにて配信中!
🎸あらすじ
「ぼっちちゃん」こと後藤ひとりは、ギターを愛する孤独な少女。
家で一人寂しく弾くだけの毎日だったが、ひょんなことから
伊地知虹夏が率いる「結束バンド」に加入することに。
人...
guys she sent me full booty butt nsfw.
XD
what do you expect 
Mini milk booti
I did? 
not yet try again
Its really good :O
Fair XD
Housewife mini milk
Beef? Cannibalism?
Hmm try no milk
Milk good tho :O
Then maybe try making femboys
Mayb :o
how do i separate this into dates? where is the option
some sus folder you got there 
do you claim innocence for yourself? 😄
I definitely don't have a nsfw folder
Thanks

a SPECIFIC nsfw folder. it's all mixed, good way to talk your way out of it! 😄
tbf, i only have 678 items inside 
i have many more... i don't delete xd
well, maybe because i don't generate a lot of nsfw stuff

same here. although i did make alfahentai, now did i?
well that didnt work 💀
Maybe increase the unet lr a bit?
will do
0.0001 is a bit low
And do try playing around with different schedulers
looking awesome! same model as yesterday?
yos
ay rip
if it comes public at any moment would love to try it
the results are amazing
New OC?
not really, just kinda like black hair with neat hairstyle 
foreheads ❤️
🍞 Reminds me of this.
where did you place the loras? last time that I added something only VAE and model existed
models > Lora
thanks!
no way thats stable diff 😮
its my own lora yes
is this the local stable dif?
yes it is
bruh

looks like sd to me
what was the prompt ?
you know that almost no body uses the vanilla sd models right?
I didnt know there were different ones
i don't think you get what i was replying to 
I didn't read
just joined now

let's see you read prolegomena to any future metaphysics
I think you kant
chat gpt, please read and summarize prolegomena to any future metaphysics for me

shit
got outplayed
now rewrite the summary in uwu speech

that reminds me to download sillytarven
💯
which llm are you using tho? 
I like TheBloke_gpt4-x-vicuna-13B-GPTQ but you can use whichever model you want since it uses the Ooba API. Pretty sure it supports Kobold API too if you prefer that. Or you can always use ChatGPT API if you don't mind spending money.
I think you should be able to run the 7b quantized models, Like pygmalion-7b-4bit-128g-cuda or TheBloke_WizardLM-7B-uncensored-GPTQ. I'm not sure if there are better ones out there now, but the WizardLM-7B model was pretty damn good when I tried it.
yup yup, i think i've tried the wizard llm before
Juxtapositions and parallels are drawn between different clips to deliver the original narrative of the video in a transformative manner that falls under Fair Use and provides unique value to my viewers. Juxtaposition is an act or instance of placing two elements close together or side by side. This is often done in order to compare/contrast the...
Oh, hewwo there! I'd be dewighted to give you a cute widdle summary of "Pwogwemena to Any Future Metaphysics" in a sweet and uwu way! Here it goes:
In Immanuel Kant's adowable book "Pwogwemena to Any Future Metaphysics," he expwows the fundamentaw questions of our understanding and knowledge. Kant is like a smawty-pants cuddly bear who wants to figuwe out how we can know things for sure and what stuff we can be suwe of. He's all about finding out how our minds work and what makes our widdle brains tick!
Kant's goal is to answer questions like, "How can we know anything at all?" and "What's the deal with the stuff we can't sense?" He wants to uncover the wules that govern our thinking and reasoning. He talks about concepts, intuition, and the cute widdle limits of our understanding.


You need to stop mocking god
Ask how is synthetic apriori possible according to kant
Mock who? They dead already lol
People have just yet to realize it
@trim current pls
Any of you use the adetailer (after detailer) extension?
wondering if it is any good
Yep
yeah you like it?
Yeah, it's really nice. It's just automatic inpainting at full resolution so you can get high quality faces even when the subject is farther away. They also have models to identify hands and full bodies, but I haven't tried those much.
Nice, maybe I'll give it a whirl.
Messing with a few different negative prompt combos. Which of these is best do ya think?
1 and 5 are probably the best. 2-4 have weird legs.
wydm, your bike doesnt have a foot for stopper like on tile 4?
ha 🙂
the first one i cant immediately see anything weird
Interesting. I'm gonna run another set.
ive been taking it easy messing with my deck, did pow atleast
does anyone know what the best anime model currently is?
The shoe is a little bit weird on 1, but it's not terrible. The bigest problem with that one is the back tire, which is off axis and the part above and below the exhaust doesn't line up. 5 has the same problem with the back tire.
yeah maybe a motorcycle wasn't the best thing to use for this. 🙂
We're at a point that there isn't really a "best" any more. There are too many very good models that all accel at different things and many of them are very similar to each other.
People tend to be mixing them with Loras a lot now, too.
Pretty much pick a popular anime model on Civitai and it will probably be very competitive with all of the other anime models.
A lot of the models just look the same to me. 😐
atleast tell me best way to fix hands
mine, totally not biased or anything
Which one is yours, @trim current ?
This embedding is pretty good for that. https://civitai.com/models/56519/negativehand-negative-embedding
This is my third merge. With my new SD base I can now comfortably merge with my weight recipe. PNG INFO DIDNT STICK Please use DPM Karras / UniPC /...
is a good model
Yep, I think it's his newest embed. It still won't fix all of them, but my success rate definitely went up since I started using it.
nice, gonna check it out
its clean af
yeeeeeee
Embedding, wow.. there's a throwback
facebomb helped in it
Lizzie, what's your license?
if you use the GGML type files, you can do LLM on cpu
"Please use DPM Karras / UniPC / 20ish steps /"
yikes, i don't use those like.. ever 😄
i mean
I do like Lizzie's model, some others that I use are Aurora and Dark_Sushi
I hear more and more about UniPC... I should look into that.
Here's the negatives again with another set.
yeah kinda interesting
Its fast and gives a plentiful amount of detail
My default is 20 steps
I especially like UniPC for artstyles that have more simulated brush strokes.
I always default to 23 steps. It's like a lucky number I've been using for a long time now. 🙂
Unfortunately not too many extensions support unipc
Like cfg threshhold and even the normal inpaint function
Mines 28
nice 🙂
the inpaint thing is because UniPC can't do img2img at all. It's configured by default to use DDIM for any img2img requests, such as Hires Fix. You can modify one of the files to change that to a different sampler though.
It's in .\modules\processing.py Line 950 as of whichever commit I'm currently on. Just change the img2img_sampler_name to the exact name of whichever sampler you want it to use. I like it with DPM++ 2M Karras, but I haven't changed it back to that since the last pull. It looks like we'll be getting the option to specify a different sampler and prompt/negative prompt for Hires fix in the next Auto1111 update so I probably won't bother.
face_yolov8s is the most accurate model for anime faces btw. the mediapipe models are only meant for realistic faces.
cool
@trim current's model is nice

Adetailer is also really nice to blending facial stuff, blend eye colors, invoke a more consistent heterochromia
Gosh that gorgeous
I should boot into linux at some point tonight and some pretties
Some pretties? :o
Like nubby made
This voting system in pow/cow is so flawed. You could ask your friends on another server to join and vote for you and you win
or ask friends on here
It's demotivating cause then it's no longer about the art
@cyan burrow did you see these ones I posted?
I did now, what about them?
Just thought they would be something you'd enjoy.
They're good as far as a prompt, but for me I have to feel a sense of dread or eeriness looking at monsters or elder gods
What's ADetailer? Is there any guides, please?
It automatically detects faces, hands, or full bodies depending on which model you choose then inpaints the rectangle around whatever was detected at full resolution to increase the level of detail. Optionally, you can enter a prompt just for the inpainting, but if all you want is increased detail that usually isn't needed and you can just leave the ADetailer prompts blank and it will use the prompt you used for generation.
Yeeeeep lol
It's why I hardly ever put any effort in
Wow thank you! Are there several ADetailer existing or just this one? https://github.com/Bing-su/adetailer
Unfortunately, there isn't really a good way around the voting problem. Could make it staff votes only, but that takes away from community engagement and would likely lead to accusations of favoritism. Anonymous submissions wouldn't fix the problem of someone potentially asking all of their friends to vote for them.
I mean sure, a separate group and name color can make you stand out as more of an "authority" but in the end it doesn't really matter
That's the one that I use. I'm not sure if there are other forks or implementations of it.
@supple raptor have everyone register in some way to vote and their voting ID is then tallied against their votes and if they keep voting for the same person or persons, with no deviation at all. Then it could be suspected that they are favoritism votes
We got trexdal admitting to me that they had their friends voted for them for it.
If people take that route..did they really earn it? What if those who couldve won naturally? Its a shit pot lol
Thanks!!
Yeah, I could ask my FF14 FC to jump on and vote, though that might stand out if I get 100 votes
Hahahahaa
Get on WoW
"Okay to do this weeks raid yall NEED to vote for me in here server"
but I wouldn't feel like I won then and I admit in some situations I will cheat, but not when it comes down to art
Dps only tanks and healers freetaxi
or like proving myself
alr so i made my noise-offset lora
how are noise offsets trained? On night and day images?
same image night and day, dawn and dusk?
or well dawn/dusk
What if I just happened to like the submissions from the same person for multiple weeks in a row? That's a real thing that I've done and I don't look at names until after I've finished voting. Even worse, what if someone starts doing it on their alt maliciously to try to get someone disqualified?
well mostly on dark images that the dataset maker find the most pleasing tbh
I said they would be suspected or like observed. To see if they're at all a part of the server or only joined to vote
if they only joined to vote, my suspicion would be high
I also feel there's a strategic advantage to not post or reveal your submission until last minute
to not give your competition ideas
if there's someone that the staff like, it's definitely me. I won the first cow as staff picked
I'm kidding btw
It would still be easy to abuse so it would mostly just add additional work for the staff. At that point, I wouldn't be surprised if they just decided the whole thing isn't worth running at all. At the end of the day, it's supposed to just be a fun, friendly, competition. They already offer a prize with a real monetary value in a contest that definitely doesn't drive very much new traffic to the server.
but when it's so easy to cheat, it's demotivating
I totally get it. I quit participating weeks ago due to similar feelings.
I do need to learn blender though so I can make characters that SD can then render up
I'm more motivated about outside and personal success now than I am an easy to cheat contest
progression!
I'm just gonna put "halloween costume" in my prompt and see what it gives me 😂
I'll give it a try too
Ishtar ❤️
halloween with high brackets tends to make interesting backgrounds i find
Happy Halowee
a green robe with a white undershirt
got one with her panties showing for some reason
Maybe it's a costume of some character she's familiar with or something, idk lol
probably will get bonked by clyde
hallolee
How does one make profit from generating AI pics? I just want a little extra income since this is my hobby anyway.
great way to get turned off by a hobby is turning it into a job.
not really
CUTE
only real way is to take request for really niche stuff. be that training stuff or image making
masterpiece, high quality, high definition, intricate details, intricate details, 8k wallpaper, beautiful detailed eyes, beautiful detailed hair, 1girl, Halloween themed, (asymmetrical bangs:1.1), pumpkin hairpin, mid shot, dethatched sleeves, standing, smug, black dress, dress, Halloween costume, stockings, night, bats, pumpkins,
This was the only decent halloween gen I got lol
That's what I get for not specifying a specific costume I guess
Don’t underestimate how lazy people are. If people will commission others to make specifc loras and other lycoris models, and then theres damn sure a lower rung that wants that and then some
Sick
I saw this prompt on civitai ''CyberPunkAI ( | neon } soldier , detailed background '' . This '' | '', is it good for something or was it just a mistake when typing the prompt?
No Idea. I remember using it for prompt mixing back in my Novel AI days
the abnatomy on this generation is particularly accurate
apary from the fact ofc but the stomach muscles and fat reserves are all in roughly the right place despite her lower half being cut off
this one is also good
wtfffffff
?
It's not even that good tho
I mean I like it
I mean, yeah
it's pretty
but I just got orange flowers from the "halloween costume" prompt xD
think it might be the trans in me talking with that one chief
That's why I said I am disappointed
also got this one today, and it uhhhh makes me feel a certain way
how does SD manage to make someone so pretty
goddamn bruh
Now that you mentioned it










