/vt/ - Virtual Youtubers

Request Anchor.

head hort, going to bed. may good hands bless your prompts, goodnight.

わわ

>>41169615
We need more extremely thick shortstack Mel, with big tiddy goth gf aesthetic, and the yandere eye shading.

>>41169492
>>41169731
perfection !! thank you anon!!!! THANK YOU!!! goodnight!!!

>>41169554
Hime ballin'.

>>41169767
Could you try a black hair Kiara?

>>41169615
IRyS trying to fit in Ina's clothes

>>41170167
She's cute, do you want anything specific?

>>41170287
Nah, just wanted to see it in your style, thanks!

>>41170634
wawa negro kakkoii

>>41169615
Robot bunny girl Aki pole dancing

>>41169615
A graduated chuba granny sitting in a retirement home

>>41169615
More Luna playing sports, football also?
also pregnant

>>41171317
>symmetrical docking
Now I need to see

>>41171445
>outfit swap
>cuffed together
seeeex

>>41170767
curious... can you do negro mori? (lol)

>>41171317
Thanks for the new tag

>>41171700
catbox?

>>41171750
https://files.catbox.moe/h6muvj.png
>>41171713
Some work came up, maybe in a few hours

how do you make LoRAs? same way as embeds/hypernetworks?

im at work for the next couple of hours- but i was wondering if anyone has had any luck getting this sort of style with any hll models?

when i was testing yesterday i wasn't having much luck with the millipen tag.

https://files.catbox.moe/oz35u9.png

Fuck, symmetrical docking is the best

>>41169615
Tokino Sora with lots of armit and pubic hair in a bikini on a beach.

>>41171713
It really doesnt want to be fully black

>>41172047
+
https://files.catbox.moe/28h8ik.png

>>41172565
Mestiza mommy Sora...

>>41171257

>>41169615
more biker Luna pls
>>41162700
https://litter.catbox.moe/43f4qb.png

>>41171257

>>41172831
meant american football but that's just as good, mucho thanks
>>41173040
I need to replace this potatoe with something that doesn't start screaming at me while freezing whenever I try to generate, navy federal be damned

>>41173040
nta,
ARRRGGHHHH MY EYE!!!! NOOOOOO

can i request luna hime eating a lolipop? please non lewd, thank you in advance!

>>41172913
https://litter.catbox.moe/rzljmm.png
https://litter.catbox.moe/wiqqcx.png
https://litter.catbox.moe/aw7jqg.png
https://litter.catbox.moe/ml0i16.png

>>41173311
lollipops are awful to make. the swirl is always detached from the stick

>>41173987
ahh.. dint know that, but this is fine! thank you!

>>41173666
thank you kind anon

>>41173666
this shit is going to put that guy cutesexyrobutts out of a business with how similar it is to his style....

>>41174242
Lets hope it does
drawing [trending character but fat ass and backmuscles] isnt a job

>>41174242
yeah I'm using CuteSexyRobutt tag with two brackets, I really like his style + the background always looks amazing.
but in the end you can copy every artstyle as long as you have enough data

>>41174406
without people like him, we wouldnt be able to do any of this

>>41174432
So? He has served his purpose then

>>41174242
>cutesexyrobutts
I like that person's art but I hated their name. They specifically put the word "robutts" in the name yet they rarely draw robots, let alone robutts.

since we're on the topic, can we have Roboco in robutts style?

>>41174656
donkey
https://litter.catbox.moe/ylvo0a.png
https://litter.catbox.moe/e1sjat.png
https://litter.catbox.moe/uknkd3.png

>>41175025
hng
thank you anon

>>41169615
Moona in an almost clear white bikini taking a sunbath

pekoland

>>41174447
When holoJP 7th gen is released, how will we get their design into the models? artists are still necessary, but I think they will hardly get paid in the future

>>41175360

>>41175729
>how will we get their design into the models? artists are still necessary
For now sure

>>41175360

>>41143948
Morning my dudes

>>41172565
Thanks for sharing, i tweaked it a bit.

>>41169615
Gura fishing on the docks in overalls.

>>41177201
....why dint i get grossed out by this? why? WHATS HAPPENING TO ME!!!!

>>41177393
How could you ever get grossed out by daisempai. Only a nutter could.

but can it do the possum?

good morning /vtai/

>>41177962
ayyy morning!!!
need more 10ma, but as a loli baba with glasses if possible with theo in the pic!

>>41177706
I don't know what you mean

>>41178038
Dunno if it'll be possible to get Theo in without post-processing but I will definitely be doing more cute 10ma gens.

>>41178074
kanpai!!! look forward for more 10ma!

>>41169615
Tennis player IRyS

>>41178060
the possum machine

>>41177201
holy sex

>>41176727
Extremely nice fishnet details, did you prompt for it somehow or just gacha?

>>41167151
no you're not alone
>>41167712
>>41167869
>>41167939
wew lads
>>41170140
>we

>>41178074
make some angry/raging 10ma too pls

>>41178988
https://files.catbox.moe/mt7h01.png
I did use a fishnet tag "fishnet legwear", but it's not in the correct area so I guess a little bit of gacha too

>>41179183
Neat, fishnets are always tough for the AI, that was some good rng

>>41179479
May i have a catbox? This is a beautifl wawa.

>>41179555
https://files.catbox.moe/q04q3w.png

The credits haven't recovered...
Now what, do I go back to requesting?

>>41179893
Can I offer you this shirowawa in this trying time?

>>41179932
Thanks dude.

>>41169615
How accurate is twintails Lamy to the original? Could anyone make her, and also wearing a night gown?

>>41169615
You accidentally walk in on big tiddy Subaru changing clothes in the locker room.

>>41169615
Elira in lingerie about to kiss (You).

And now Firefox has crashed so now I can't even save what I didn't already.
Okay definitely have learned some lessons for next time if there is a next time.

>>41180341
Is she fat or pregnant?

>>41180457
I prompted her to be fat. (franchise) (character) pregnant (number) would supersede (franchise) (character) (lingerie/bloomers/fetish outfit [such as nun or nurse])

>>41180487
Put her on a snowy mountain with a blizzard and she is a Yuki onna

>>41180487
shiro wawa + oni horns = ayame?

>>41171257
As the anon who made the pregnant Pippa as well as bball Luna, I'm suddenly concerned lol

>>41169615
>all those credits burnt for only about 12 images to show for it
An extremely and intricately sad uuu.

>>41179046
You should try the 7th layer mixes, I think V2 G has a simplistic style by default, I might try it out myself.
Picrel is V2c
best quality, ouro kronii, cowboy shot, tachi-e,
with default negatives that I use for Holocream

>>41180564

>>41180956
ayy nice, thanks

>>41180341
firefox is buggy with the webUI, I only use edge

Would an AI understand being told to draw a Vtuber (i.e. Choco, Kaela, Kiara, Elira) painting her toenails?

Stupid sexy wawa

Hey shitstains, you dropped this garbage in our general so I'm bringing it back. Please make sure to keep your trash where it belongs, you have an entire board to simp for your twitch thots so keep it here.

>>41180341
cheers, \hololive\s\w+\s\w+\s\d+\ anon!
next time you use it, go back to my colab notebook link and re-copy it, I made a few changes, it will load some embeds now
>>41180917
I'll try it out. liking HoloElyV2 and Cilantro, lol
>>41181104
unfathomably based
>>41181308
abayo

>>41171257

>>41181281
>It feels like I'm wearing nothing at all, nothing at all, nothing at all...

>>41180693
IS THAT A FUCKING Tampa Bay Buccaneers jersey?!
kek

>>41181308

Toyota...the trimming...

>>41181505
that's just the best one it spit out for me, was trying to get it to be the same light blue as her stockings, but it gave me Buccaneers instead

>>41181586

>>41176155
Don't be retarded, you still need to go through multiple prompts just to get decent fingers and toes. I don't know what artist killed your family but we do still need artists in case anything happens to the AI technology.

>>41181476
Is Mori homer?

>>41169615
Biblically accurate Towa..

>>41180711
Jewgle wins again

>>41181586
for light blue stockings, maybe try a team that has a similarly colored jersey? or a similar color anywhere in their kit. Dallas Cowboys probably easy for the AI, LA Chargers but the AI will probably expect you to say San Diego, or Tennessee Titans might work best.
idk, Carolina Panthers, Detroit Lions, uh, Seattle Seahawks, NE Patriots, Philadelphia Eagles, Miami Dolphins, Denver Broncos, Indianapolis Colts. only teams with any blue or teal that I can think of.

>>41181370
How long does it take for colab credits to replenish?

>>41181744
It hurts.

>>41181808
I'm using paid colab, so IDK if it's different for me now, but I've had better luck if I don't let it run until they disconnect me. i.e. I close it myself by using "Disconnect and delete runtime" in the runtime menu.
Before I switched to paid though, if I got told I used too much, usually 24hr later it was good again.

>>41181863
Okay so that means around 5 pm EST

Rentry of mixes updated with the new hashes, 7thHolo2c and HoloAbyss7th1.1

I don't have a link for HoloAbyss7th1.1 and I haven't made the mix so share it or wait until I am free to do it and upload it.

>>41181762
it doesn't really want to take to the teams jersey's on my end sadly, it did spit out this at least which looks more Luna-esque. I'd have to train a separate LoRA on football uniforms to get it right I think,it seems to keep wanting to make soccer uniforms

>>41182109
kek she's cute.

>>41181762
>>41182109
actually, kinda scratch that, I refuse to believe it's getting this number from nowhere when I put in Indianapolis Colts

>>41182152
payton qt3.14

Make your prompt {{{bigger}}}

YOU, SHITHEAD, WHAT DO YOU THINK YOU ARE YOU DOING IN THE MIDDLE OF CLASS

>>41182813
I did but they're gone...

>>41182901
>SHUT THE FUCK UP BRENDAN!!!!
https://www.youtube.com/watch?v=ekqxCOhVSto
is what i heard lol.

>>41182901
I can hear her yelling at me for backseating NieR Automata. I love it.

>>41183026
STOP WITH THE FUCKING DONATIONS I AM TRYING TO TEACH

I think that's still my favorite 10ma stream of all time

>>41169615
Vincent Von Gawr (Gura)

So yeah, Gura, but Van Gogh style

>>41182901
>>41183465
That's an embedding/LoRA, right? Can you share?

Bratty dark elf..

>>41183616
It's a LoRA
https://huggingface.co/sylveriate/lora-characters/tree/main

>>41180118
AI still has some issues with transparent materials

>>41182813
AH! 7 layers of {{{bigger}}}!!

good afternoon /vtai/ :)

>>41184037
Beautiful. May I have a catbox please.
I'm a sora fan and i love it when she's all wet.

>>41184371
https://files.catbox.moe/6oyf5d.png

>>41184037
request teen sora in jk uniform. non lewd

>>41136339
Let me know if you've got this, since it's been a while.
https://litter.catbox.moe/mahsm3.png

rurudo is pretty cute

>>41183902
Oh man, I thought that was Abigail Dovely, not Abigail from the hentai of some gacha based on a hentai light novel.

>>41178135

>>41166721
quick test, eh...

>>41184548

>>41171257
>>41180693
>>41182109
>>41182152
last sports Luna, was attempting mixing in a couple other LoRAs with this one

>>41181476

>>41184660
A Sister Abigail LoRA, huh? Not entirely out of the question.

>>41179646
>8thCream
What's this?

>>41185921
weighed sum merge of 70% HoloCreamsicle and 30% 7thHolo2C

>>41186012
NTA
Is WS @ .3 A:Cream B:7th holo?

>>41169615
Anon, generate some more IRyS paintings please. You are an absolute chad

>>41186088
I thiiiink I did WS @ 0.7 A: 7th B: Cream but I'm not 100% sure

>>41186184
ok, will check hash anyways, thanks

>>41185050
wow, really impressed the arm and leg didn't fuse together for this pose.
I guess it must help that it's a front facing angle maybe.

>>41179046
>>we
Your mum

>>41181606
>Don't be retarded, you still need to go through multiple prompts
What part of "for now" don't you understand you illiterate faggot?

>>41169615
Hags laying in bed and getting massaged on her breasts, thighs or asses

>>41186519
Quit your bitching and post anime girls

>>41186671
Can you add a cameltoe?

>>41181966
do you have the hash for Abyss7th1.1? tried to mix HoloAbyss7th1.1 but got hash 5da439ba20 when loading. trying to figure out which step went wrong

>>41186671

>>41187152
Give me a sec I am mixing as well, actually, I got a different hash as well, and is different from yours

>>41187152
aeebd61508

>>41187152
>>41187237
nta, but from a previous thread, another anon has linked someone from /h/ where they talked about how they changed how they label models with hash, so you may get different ones compared to what is listed

>>41186671
sorry my bad

>>41187215
yeah that's better, thanks

>>41187152
Wait I fucked up kek this >>41187326
is the hash of 7th layer v1.1

Abyss7th1.1 gave me the hash c6dea6594c

>>41187353
But we should get the same hash if we are using the same models and steps right? is not randomized numbers

>>41187237
i had [1b2e5611]ca for Abyss7th1.1 and [5da439ba20] for HoloAbyss7th1.1
not sure if it makes a difference here, but i also have medvram set

>>41187491
correction: [1b2e5611ca] for Abyss7th1.1

Since AI I've been noticing that a lot of actual art has fucked up hands (though not as much as the AI's) or very clearly uses poses that hide the hands.

>>41187491
I have medram too, but never affected the hashes before.

Are you mixing as safetensors and Don't right?

>>41187672
i'm a little retarded. did not have it on Don't

Is there any Rushia/Nazuna stuff? I'm not even asking for some to be made, just wondering if there's some pre-existing stuff.

>>41187728
Don't worry, I think the original might have used A, B or C, I might try that as well

>>41187758
Why not check the Mega?

>>41187763
I checked last time and it doesn't affect the hash. It just generated a yaml file.

>>41169615
Pregnant Kiwawa baking.

>>41186137

>>41183690
Go on keep it coming.

>>41188400
Black hair Wawa hits different

>>41188210
she is speaking the language of the gods...

>>41187852
I did it just to confirm, you right gave me the same hash, guess I am gonna test if removing medram changes anything

>>41188741
Nephilim language

>>41187852
Can I ask you to remix it following the recipe in the rentry? I think that is the ultimate test

>>41169615

Coomer: holoEN (or VTuber of your choice desu) sweating in a tracksuit slightly too small for them

Not coomer: Kronii smoking or drinking on a balcony in front of a city nightscape, bonus points for neon aesthetic

>>41189037

Just tested, for the record medvram or xformers do not affect hashes of mixes

>>41187353
>>41187465
>before
webui displayed...idk, some kind of hash
>now
webui calculates a SHA256 hash, and then I think truncates it and only shows you part of it. you can see the full SHA256 hash in the console as well
you two tards should just ask each other if you're on the latest git pull or not, kek
I haven't updated, so I've only seen the new hashes on colab
>>41187672
>>41188858
VRAM switches don't affect this shit at all. Mixing models is purely I/O. lol, don't waste your time/space doing this.
>>41187852
if you know how to calculate file hashes outside of webui, check and make sure the files are identical? if you've got 7-Zip installed it likely added a "CRC" option to your context menu when you right click a file, but there are better tools if you haven't
>>41187758
Plenty of Rushi, nobody really prompts her that often. Nazuna is harder but doable.

>>41189499
see above, lmao. you can mix a model without CUDA installed at all, from the command line if you were so inclined. the HoloElyV2 mix in the MEGA, and some of the steps leading up to it? no GPU in the system that spat those files out.

Looking at all of these images i can already imagine what we'll have in future.

>>41189619
>>41189653
I have trust issues I need to see it with my eyes.

Yes, I am updated-ish, at least until the hash changes, new hashes are 10 digits, old ones are 8.

>>41189427
Lovely, thanks!

>>41189838
Another

>>41187758
You can generate Rushia with base nai, and Nazuna-lite with hll, everyone just forgot about her.
Her recent twitch streams had ~2k viewers. Even Pippa gets 3k these days..

>>41189499
Hash algorithm got changed recently, if you updated in last 3 days your hashes will be different
https://github.com/AUTOMATIC1111/stable-diffusion-webui/commit/a95f1353089bdeaccd7c266b40cdd79efedfe632

>>41189930
nta. but i hope to see a smoking one. sorry to bother you anon.

>>41169615
Barista Miori Celesta lactating

>>41190027
Doesnt seem to like cigarettes too much, got this

>>41190457
its alright anon, its close enough! thank you!

>>41190457
>>41189037 (me)

It looks like an Asian smoking thingy, I like it

https://files.catbox.moe/es5yye.jpg any tips to get a similar image for this? i dont know what prompt to use for the mouth to look like this, and the rope under the boob

>>41185050
>>41188210
God damn! The camel toe on this one!

Holy shit! First time using creamsicle and it's fucking insane how good it is!

>>41190968
The mouth I have no idea, there's a bondage LoRa with ballgags and other stuff.
The ropes would be shibari.

>>41190968
"open mouth" + "gag" seems to work about 25% of the time (on holo nuts) with the other 75% being AI body horror
rope bondage seems to work perfectly fine

>>41191759
thanks! but how do you make it so the mouth is opened much wider like in the pic i linked?

>>41169615
IRyS with collar around her neck, that would be great.

>>41190508
This has me fucked up, I give up

>>41191939
RNG? you could also try the bdsm model

something's wrong...

>>41192441
Can I get a catbox for one of your Tenmas or the base appearance tags you're using?

>>41192489
https://files.catbox.moe/axcfxv.png

>>41192071

>>41185897
NTA but I would be eternally grateful.

>>41192434
https://files.catbox.moe/maal39.png
https://files.catbox.moe/wqm2ld.png
im getting there

>>41192546
めっちゃめっちゃいい

>>41192546
Thank you. I would like to see some more bdsm themed pics with IRyS if i don't bother you ;)

>>41192538
Thank you. I'll see if I can generate some good lolibabas tomorrow.

>>41192434
bdsm model?

>>41192843
Yessir

>>41169615
Iroha in a sexy luxurious dress

>>41169615
fat brats

>>41169615
are lewd requests allowed? Nerd Fauna on her knees, mouth open full of saliva, sexy expression

>>41187152
>>41187237
Make sure you are on the last update, I got the expected hash
I will upload it into a pixeldrain and update the rentry anyways, but it will be until a few hours tho.

what's her lore /vt/

>>41192843
One more

>>41194533
you've created a weapon to surpass metal gear, the loudest, angriest chuba in existence. run, run far far away, with OSHA approved hearing protection.

Have any LORAs of the holo girls been made? I'd like a Fubuki one

>>41169615
a very cute little Fauna stealing your food. If possible give her kind of a smug grin and have her looking at (You)

>>41194970
We only have Kanata

Moona

Is Catbox still working? Trying to upload but I keep getting a 404 error and have no idea why

>>41195028
Moona's massive moonas...

>>41195128
Just uploaded, works on my machine
https://files.catbox.moe/2j8apa.png

The fuck is this?

Is there any easy way to compare a bunch of different loras? Like to compare the finished result with a bunch of the partway outputs from training. It doesn't look like it's in the X/Y plot options unless it's called something else there

>>41194490
on the latest update. did some digging and found that my 7th anime 1.1 model is the problem. getting a different sha256 than what huggingface lists and terrible models out of it. seems to just be an issue on my machine

>>41195294
gotta update your plugin an webui bruh

>>41195294
They added addnetwork models to x/y plots recently. You should be able to figure it out I think.

she's coming

>>41195260
You got pickled

>>41195728
So now what?

>>41195788
I'm joking I don't know, it says you need to reinstall something

>>41195861
I'm doing everything exactly how I did yesterday.

>>41169615
Zaion LanZa cowgirl seggs with her hands tied up and raised.

>>41192441
Is there something such as hagification?

Updated the additional networks extension and the additional networks drop down is gone. Is this a bug or did the functionality change?

>>41195245
Freckle Mori...

Also, weird
It went to shit a day or two ago and I have no idea why, it worked perfectly fine before but now it's just constant 404's

>>41196109
the embed doesn't make her hair cones correctly

>>41197757
After segs is ok

>>41169615
Ayame or any of the hags sitting on a gaming chair and surrounded by a gamer setup.

Does anyone know why my images look kinda blurry?
Catbox link in case you want to check if I messed up somewhere in there...
https://litter.catbox.moe/q4ilht.png

>>41197036
>>41187215
>>41186671
>>41188400

god blackhair wawa is something i didnt know i need very very nice

>>41198438
"blur" accounts for too little of your negative prompt
stop using meme negatives

>>41198438
Stop using (((multiple brackets))). webui uses either (single) brackets and (weights:1.2) instead.

>>41198642
Webui can use multiple brackets.

>>41198608
I thought the schizo negative prompt was good... I'm just getting back into this and playing with the new stuff
>>41198642
I can try this

>>41198438
Try reducing the scale as well 10 or 8 are good enough.

Daily wife

>>41200190
Nice RNG getting braid/drill (?) style hair.

>>41198608
>>41198850
This is the best one I've had so far but others still come out blurry

>>41200344
Try this negatives
(worst quality, low quality:1.3), signature, watermark, username, artist name, text, censored, cropped, exposed nipples, text, speech bubble, letterboxed, depth of field, blurry, blurry background,

>>41200335
Think it's biased because of the mimonel or the afrobull lora, didn't really prompt for it

>>41200503
https://litter.catbox.moe/nnorv6.png
I feel like that's better

>>41183902
How do you use LoRA? I've been so outdated with this

>>41201959
https://rentry.org/lora_train

>>41195260
>>41196015
then you wouldn't have ended up with PyTorch CUDA 11.7 and torchvision using CUDA 11.6. I'm not even sure how you got them both that high up, doesn't it normally use CUDA 11.3?
>torch_command = os.environ.get('TORCH_COMMAND', "pip install torch==1.12.1+cu113 torchvision==0.13.1+cu113 --extra-index-url https://download.pytorch.org/whl/cu113"))
is usually somewhere in launch.py
go look at launch.py on your system, and then look at actual launch.py https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/master/launch.py#L182
also >>>/g/sdg okayuposter
>>41198069
>Ayame
>hags
KEK

>>41202095
Thanks

>>41195260
>>41196015
oh, and I guess check "webui-user.sh" and make sure the TORCH_COMMAND line there is either still commented with a leading #, or wasn't changed to something mismatched like that. also, it's checking in your local libs, does it not usually launch on *nix in a venv like it does on Windows?

>>41202155
Is there a way to completely start over from scratch?

>>41198069
Some fucked up hand, horns and a feet hand
https://files.catbox.moe/cf6esn.png
https://files.catbox.moe/0rm0xy.png
https://files.catbox.moe/5u29ln.png
https://files.catbox.moe/win2sd.png
https://files.catbox.moe/vr1r2t.png
https://files.catbox.moe/4rajuu.png

>>41202449
Why would you start over from scratch for something so simple???
>open bash. do not sudo/su, I am guessing that usr/local/lib shit is like %APPDATA% site packages for windoze python
>pip uninstall torch torchvision
>pip install torch==1.12.1+cu113 torchvision==0.13.1+cu113 --extra-index-url https://download.pytorch.org/whl/cu113

>>41202362
The lines are there in the code where I hope they're supposed to be.
I don't know what the second part of this means.

>>41202584
No what I mean is uninstall all the stuff you downloaded and reinstall it. That seems like it'd be way easier.

>>41202449
>>41202584
alternatively, I guess you could
>pip install torch==1.12.1+cu116 torchvision==0.13.1+cu116 --extra-index-url https://download.pytorch.org/whl/cu116
if you were in fact using a higher CUDA version for some reason. did you use pip to upgrade everything or something??? IIRC the issue is that some webui shit absolutely requires exactly torch==1.12.1, and torchvision==0.13.1, and CUDA 11.7 is only available for torch>=1.13 and torchvision>=0.14 or something like that
>>41202926
then use windoze, kek. big boy pants for a big boy OS! those packages didn't update themselves automatically, unless linux does some shit to update python packages for you, but I don't actually use the thing
>open bash. do not sudo/su
>pip uninstall torch torchvision
>pip install torch==1.12.1+cu113 torchvision==0.13.1+cu113 --extra-index-url https://download.pytorch.org/whl/cu113
>>41203046
idk. "git pull" would fix anything he changed in the launch scripts

>>41194195
Sexy expression to me at least idk I'm autistic

>>41203094
I don't know what a git pull is.

>>41190084
best i can do is inpainting https://files.catbox.moe/1jrsln.png

>>41194974
>What are you gonna do now, m-fer?

>>41203370
catbox please?

>>41203094
I keep seeing this. I don't know how to make it go away.
I'm pretty sure I'm too stupid to do this and there's no way you can physically get onto my laptop and fix it.

goomba

>>41203876
Sick. Catbox?

>>41203980
ogey

>>41204094
https://files.catbox.moe/cnzd02.png
forgot the link

I give up.

>>41193385

>>41203807
Sure thing but it's using a dumb mix, you'll likely have similar (better) results with Creamsicle
https://files.catbox.moe/fjd0u0.png

https://files.catbox.moe/vz32u6.png kronsters

>>41192071

>>41203857
>laptop
wat. how did you get webui/torch on there in the first place? and yes, you kinda stupid. what happened when you did this???
>pip uninstall torch torchvision
???
and how are you trying to launch webui?
post a pastebin of the entire console window

>>41169615
Marine anal fisting

>>41187857

>>41206985
My lovely wife.

>>41203675
This spoiled brat...she's just too cute to punish
thank you!

>>41205613
oh. guess I can't help you any more then...
>>41203857
wait. wait wait wait.
>light mode
you bumbling...say you're using the colab man. for fuck's sake...

>>41207571
...yes?

>>41207794
I'm supposed to read your fucking mind and know that?! I'm launching the Colab now and seeing the same errors you did. Google updated some software on their side. Give me...10-15 minutes, I'll re-write it to uninstall their shit and downgrade to the versions that webui expects. If it's not possible, I'll post again and let you know.

Studying with the nerd

>>41207857
So I wasn't doing anything wrong?

If i'm out of pixel 20gb space, are there other ways to get a holo model? Was hoping to test out the latest holo abyss.

>>41208188
No, it wasn't you. It looks like the xformers wheel somehow changed its fucking requirements on its own, and I was using a static link to the exact damn version that works, so it makes no sense. If I can't figure it out I'll just make it launch without xformers.

>>41186594
>It's just a massage, bro

I think I know exactly what you want, but I don't know how to get it, I might try another day

>>41208188
You definitely weren't doing anything wrong, and you're not alone either. https://github.com/brian6091/xformers-wheels/issues/4

>>41208242
Doesn't just slows your download speed? The other way is to mix it yourself

>>41208242
use a vpn

>>41169615
chuubas getting skinshipped if that's even remotely possible

>>41208188
~10 minutes

>>41208968
Middle aged women and intimate relationships with them.

>>41208437
catbox?

>>41208864
What does that even mean?

>>41209051
https://files.catbox.moe/b7zqjb.png

Hi guys have a healthy Okayun. Mogumogu~

Hah, gotcha, I have your attention! It is I, the other LORA Rentry guy here.

Good news everyone! I found a fancy trick in the ninja scrolls, turns out you can basically make a LORA from the difference between a model and its training.
How is this relevant to /vtai/? Easy peasy, you can make a LORA out of HLL/HLL2 and use it with any other model without merging. It's also a much smaller file size. This can be a way to also improve traditional mixes by serving as a testing ground of sorts (so you can gauge how much strength you'd need to merge and keep details).

The magickal incantation is something like
"python networks/extract_lora_from_models.py --model_org ../stable-diffusion-models/Stable-diffusion/animefinal-full-pruned.ckpt --model_tuned ../stable-diffusion-models/Stable-diffusion/hll2-e12.safetensors --save_to hll2-e12.LORA.safetensors --dim 192"
Resulting file is a 432MB .safetensors file. I plugged the resulting LORA, generated with AbyssOrangeMix2 et voilá, works very close to the mix.
If someone tells me a decent file storage that can handle files that big and doesn't require me to make an account I'll upload it to save you the hassle.

I'll be putting more exact instructions in the rentry in the next ten minutes: https://rentry.org/59xed3

I can feel the disk space savings already.

angry meido
i don't know why i tried pointing, hands don't work that well with this model
>>41209490
I really should check into /vtai/ more, I'm stealing this rentry for /h/

>>41183514

>>41209490
hey other LORA Rentry dude.
I have a couple questions.

Is the default "best" model to train lora on right now nai?

And how to determine the optimal number of image repeats when training a certain number of images, assuming an epoch of 10.

So for example, if I have 100 images within an image folder, how many image repeats is the best? What if I have only 25 images or 50 images?

Thanks in advance!

>>41209490
Well shit, color me intrigued
Wormhole .app you can upload without an account, the link will only be good for 24 hours but I or someone else can just throw it up on a permanent mega

>>41209490
Just put it on pixeldrain we could re-upload if ever gets lost

>>41181708
She was a m*ale

>>41208188
go back to my original colab link if you copied the notebook into your drive. rename the old copy to "broken" or something. xformers is fucked on colab for now and I can't be arsed to figure it out
https://colab.research.google.com/drive/1STL60qfoY-iSlhRb9zFETRLTqhNbznRf?usp=sharing
bonus: it loads 50 embeddings up by default now, start typing "<" in the proompt box to see the list of embeddings only, otherwise it will also autocomplete yer tags.

>>41207995
cute!!!

anyone know if there's any way to get Ina's outfit? the embeddings i've found for her so far don't seem to help any

>>41210489
>embeddings
>ina
wat

>>41210540
...marine why are you so deep fried on good_shit with the SD 1.5 VAE, lmao

>>41210489
I just use a Holo trained model with no character embeds or Loras, works great for me.

>>41210355
Thanks for trying anyway. Have one of the few files I could salvage.

>>41210805
>trying
trying what? it still works. your gens will just be 10-20% slower.

>>41209927
Hey there.
Yes, NAI seems to be the only really valid one for anime. Maybe it's related to the presence of Anything v3 in some models, has a bit of a reputation for being a bad checkpoint to train from. When I tried a LORA that worked fine with NAI from nutmegmix it was a fried out disaster. You can probably train from models without Av3.

The image repeats are tricky. Seems more repeats of images are better than more epochs, at least it did improve results noticeably. Try doubling the images (using the number at the start of the folder names, like "1_Thing" make it "2_Thing") and run 5 epochs instead. I just did that test with an older data set and got a better result in terms of accuracy.
This is specially true if you use the "aug" options (--color_aug and --flip_aug) which randomly alter colors and flip the images (so you don't have to) at runtime, at the cost of not caching latents (a minor speed down on Linux, a bit of a harsher penalty on Windows, but still not massive).

If you have a very small amount of images you might have to get creative, specially if you aim for flexibility.
In the rentry I describe a method called "Dreambooth style" which uses regularization images, it seems to give substantially more flexible results while still learning details fine, but mostly forfeiting the style of your character's images. This can allow you to get more standard bodies out of chibified characters or adjust clothes or things better, at the cost of having to do a few more training steps (2-4 seemed to put it at the same level of accuracy of regular LORA training).
If the character you want to train has limited clothes or lots of details it's a good idea to draw (if capable) or edit (you can afford some sloppiness) images to have samples with and without that detail (and matching tags), so you can remove or weight that detail from the prompt. Keep tags clean, bad tags contaminate the result (avoid false positives, redundant stuff unless pair of a combo ("pantyhose + black legwear" for example, and other interrogator goofs).
Training time will be short for such a character, so make a couple tries. Use the first tries to potentially generate more images (be picky!) to boost your data set. If they look fine and they are accurate they'll work. Gacha for perfect hands if you have to.

I've specifically trained very obscure characters, so if that guidance is not enough, let me know, I'm pretty much specializing on reviving obscure shit no one cares about even for money, so I'm already no stranger to working with a lack of good source images.

>>41210102
Oh yeah I've downloaded from pixeldrain often, it should be AOK, I'll go check it out.

>>41211504
>HANINC I OM

>>41202507
>https://files.catbox.moe/5u29ln.png
>Those twintails
MUH

Nice images anon, thanks

>>41169554
Requesting Princess Luna exposing her baby royal pits, please

>>41209490
>>41210102


pixeldrain com/u/RjE3psRf

Alright, that's the HLL2 difference as a LORA!
Be mindful of the standard loss of precision when using with models that aren't NAI, but at least it can be adjusted on the fly which might save a generation.

I'll look into increasing the network dimensions to compensate for any other possible losses of precision. Hard to guess the right value.

>>41169615
Requesting Princess Luna exposing her baby royal pits, please

>>41211875
Downloading, thanks

>>41210886
Never mind, I'm retarded.

>>41180295

>>41211058
Wow. Thanks for the in-depth and well put together response!

Now that you've brought up tagging, there is also another question that popped up in my mind and is confused about. On another entry related to training embeddings, the anon said that it is best to remove tags that are an "essential" part of the character so that the AI can learn about said "removed" tags/features without having to actively prompt for it. Is this still the case for LoRA? Or is it flipped where you want all the tags that describes the image, pruning redundant tags of course, and remove only those that doesn't describe said image?

>>41209075
>skinship (uncountable)
>(particularly Japan, South Korea) Bonding through physical (touch, skin-to-skin) contact; particularly between family members, relatives and loved ones. quotations ▼

Metro Fauna, idk

>>41212192
Works great.

>>41212192
>>41212484
lol. also, I think I might've figured out how to force it to properly install xformers. definitely not going to rely on that facebook employee keeping the .whl file on their github forever, so I've uploaded it elsewhere while I test it. if it launches with xformers successfully after this, I'll edit the notebook again

>>41212546
Wait I got a really stupid idea.

>>41164860
dozo
https://files.catbox.moe/xhyf27.png

Good evening, ladies.

Bread?

>>41212786
It took a couple tries but it worked. Faun Lennon.

>>41210645
the model i'm using only wants to default to nude instead of an outfit, there a good model to use that handles lewds while retaining that level of accuracy?

>>41211913
here you go
>>41213094
kek. anyways, yeah, xformers is working with my little work around. I'll re-edit it once more...

New thread doko?

>>41212940
Thanks babe.

>>41213191
Ive been using HoloCreamsicle for lewd and non lewd. If I dont prompt for clothing it mostly defaults to their outfits.

>>41213077
I can bake. Catbox?

>>41213423
Oh I did just get in so suggest me an OP because right now I'm defaulting to Flare booty unless you guys have other ideas.

>>41213423
Sure thing
https://files.catbox.moe/cnjajq.png

>>41213191
You can use the safe version of Creamsicle if you find the constant lewds of base Creamsicle annoying.

>>41212786
>>41213094
next-next time you use it, copy once again from that colab link. it launches properly with xformers like it used to.

>>41198069
made fun of you for implying ayame was a hag earlier, kek, here you go too

new bread
>>41213977
>>41213977
>>41213977

>>41212191
Eyy thank you

>>41211913
When I first tried this, it gave me a nude Luna, even though I had nsfw in my negative prompt. When I tried it again but with pink dress in the positive, it crashed my tab as soon as it finished. I think this prompt is cursed.

>>41181708

>>41212258
Depends.
Generally speaking, adding the character tag will not be a bad thing as long as it's an unique tag.
Without tags you are basically dumping all the training into tags like "1girl" or "park" or "checkered background". So yeah you'll be able to get the character without doing anything fancy but if the tags aren't clean that's gonna do into random shit and have unexpected effects.
Adding the tag is extra "glue" for the concept, which can be weighted to use it as a little boost.
So yeah it's slightly better with (specially if your character is rare and already not in the current models) but if your training set is good and varied enough it won't matter much. But not all characters have that luxury so it's good to keep it in mind.

BUT

If the character DOES exist in NAI or your base model in any capacity, even one unable to make a good image, you DO want to tag. It'll benefit from whatever training is there (or guide it in the direction you want).

As for specifics, the whole thing works with pixel nonsense and the tags to guide its attention. It's gonna attempt to find the things specified in tags in the pixel soup.
Mandatory minimum work for quality is removing everything that doesn't belong to the image, specially if it was captioned automatically (deepdanbooru, etc), because that thing makes a lot of goofs, inserts unrelated character names which will make a mess internally, and so and so. You'd be spreading the training into random nonsense and losing on "general purposeness" for it or directly contaminating if it's out of hand.

Then depends on how perfectionist you get. I am training more LORAs for flexibility so I can change costumes and backgrounds easily, so there's a bit of tedium involving teaching about the AI about the different layers of clothing (and without any clothing if possible) and accessories (and without any accessories) and have tags reflect the difference.
As said before, it's best to make test runs and YOLO it and see what specific details need fixing. For example I trained a LORA on Shauna from Pokemon and the thing absolutely insists on putting bows on every piece of upper clothing, so to fix it I'll need to edit or draw a picture of her without the decorative bows on her shirt (and no bow tag. Then make sure every picture with the bows does have the tag) for it to get a clue and stop doing that. Did it for other characters, it fixed the issues 90% of the time (can do it more times to probably eliminate the issue or keep it to 99% good)
It also has a compulsion to keep alternate costumes "in the same color theme" but I actually like that, so not gonna bother to fix it too hard, but you can do it the same way.

Note you can train more than one concept in a single LORA (use separate folders and tag accordingly) if you want to reinforce a specific object, background or concept to go with the character. In my experiments the multiple concepts won't overlap unless prompted for (and it doesn't step into other tags the character is training). Specifically I trained one of Windy from Panel de Pon with a bunch of glitter and sparkles (because that game is girly as fuck, was fitting) and I can turn them on and off at will from the prompt. Helps giving that fairy energy.
It was the "Polka eyes" theory, I finally got to implement it now that it doesn't take weeks to train, lol.

I'm terribly verbose and I'm sorry if I digress too much, but hope this helps. It gave me "is this fucking magic!?!?!" results from absolutely lacking image sets. (That's also why I post about findings in /vtai/. I'm sure someone has an unpopular oshi and this might help)