/vt/ - Virtual Youtubers

Request anchor

Smart anons of /vtai/, I am once in dire need of help to fix this error I have for latent couple extension. https://hastebin.com/share/qokimoduto.javascript
In the readme it says to apply the patch via the command but I don't know how to do so in auto1111. Maybe I need to put it into a specific folder or something? Thanks in advance.

:3

>>43244614
:3 :3

Wawa sure is pretty!

>>43244577
that's a cake not an anchor

I thought I left a LoRA on at first but no it just looks like that lol

>>43244577
Anya & Petra running for USA 2024 presidency office

>>43244767
She's got an anchor pattern on her jeans if you look closely!

>>43244820
As in together or opposing?

>>43244577
Holo of your choice with abnormal amount of arms, legs, eyes, heads or breasts.

>>43244904
anyer as president & petra as vp would be neat

Let's get those prompts {{{bigger}}}, /vtai/.

>>43245425
bigger succubot

Did anyone manage to generate nyotaimori?

https://litter.catbox.moe/h5hos6.png
https://litter.catbox.moe/i00pns.png
https://litter.catbox.moe/en7xg7.png

:3

>>43244820
Couple of tries but kinda failing :(

>>43246032
:3

>>43244577
>>43245425
Requesting a {{{Bigger}}} Smug Mio for science.

>>43244577
Thicc AkiBot shooting lasers at enemies.

>>43246126
:D

>>43244577
Is there an inverted nipples lora?

>>43246283
catbox please?

>>43246259
search on civitai

>>43246388
Thank you.

>>43246311
https://files.catbox.moe/b6vqmk.png

Lamy's new outfit is SEX

>>43244577
Cutest Nene you can make.

Thank you Kiara for convienently hiding your hands out of sight.
Is this pose kinda weird?

>>43246372
>>43244601
Nice. Feels weird seeing them the same size

>>43244577
Whorish Petra. Latex mini skirt, slingshot bikini top. Boots with high heels.

>>43246880
Here's my attempt at it.

Highres looks kinda iffy and Im not a fan but here you go.
https://files.catbox.moe/0j8jvx.png

Accidental giant wawa

>>43244577
80% of requests from last thread wasn't granted anyway.

>>43244762
true

>>43244962
i could only do it with cthuluesc style
https://litter.catbox.moe/fovytp.png

>>43247520
Thems the brakes. Most of the requests from last thread were pretty meh anyways.

>>43246737
thanks
>>43246775
love this one too

>>43247109
I just put what you said into prompt form.

>>43247527
Facts.

:3

>>43247939
;3

>>43247736
objective truth

What vae do you use?

>>43247587
That's pretty cool. More please?
>>43246372
>>43244601
Very nice. Is it even possible to make one of them smaller than the other, or does NAI insist on equalizing them?

>>43248052
I just started using this vae based off a recomendation from Marineposter last thread. I think it's pretty good. It's a bright but not as ridiculously bright as the wdv4 vae.

https://huggingface.co/stabilityai/sd-vae-ft-mse-original/blob/main/vae-ft-mse-840000-ema-pruned.ckpt

Shame about the arm...

:3

>>43248087

>>43248327
=3

polka-ina

too much junk in the trunk

You ever get cucked by xformers? Get a fairly nice result, rerun prompt and seed with hiresfix on, don't even get to the hires pass when 2 fingers have vanished. Multiple runs various fingers vanishing. So far no luck with i2i upscale either.

>>43248872

>>43220391
New miori? No way lol
You might find barely enough decent art for a set if you sift through twitter for 2 hours.
I made a shitty embed in pre-lora times but its barely better than grugging it.
https://files.catbox.moe/roj5qp.pt

>>43249054
no
the differences are tiny to nonexistent for me
there were differences back when it was new but there have been a lot of patches

>>43249143
Now merge her with gyaru form polka

https://files.catbox.moe/banz0h.png
https://files.catbox.moe/hviasq.png
https://files.catbox.moe/dg9pxy.png
https://files.catbox.moe/70t06r.png

>>43249354
holy sex

>>43246880

>>43244577
Those rare images with perfectly generated feet/soles, any chuuba.

>>43249041
no such thing. catbox?

>>43246193
a bigger grin?

>>43249629
Sure
https://files.catbox.moe/1mg6in.png

Last wawa for me. Goodnight, I'll see you guys tomorrow!

Goodnight from me too.

>>43249803
>>43249947
Goodnight, anons

>>43249728
>>43249803
G'night, anons. Gonna be turning in as well.

>>43244591
>go to your webUI folder
>right click -> open in terminal
>copy/paste this
git apply --ignore-whitespace extensions/stable-diffusion-webui-two-shot/cfg_denoised_callback-ea9bd9fc.patch
>press enter

What do we call the silver dress again?

>>43244577
Squ in luxurious wheels dress

>>43244577
more fat nipples like this >>43246931 plése

>>43223294
no regrets

>>43250973
interesting how you/the AI made Sui topless when she wasn't in the original image

>>43251850
There wasnt much to tell the ai that the shirt exists so it just gave her a detached collar and made the few creases into floating light lmao
It did give her a shirt few times but the mouth and hair was totally fucked up and I wasnt gonna spend 15 minutes on this

>>43245425
Aye aye captain

>>43245837
delicious, more?

>>43244577
Big booty Iron Mouse

>>43236526

Whe the tentacles have a mind of their own

https://litter.catbox.moe/szwb5v.png
https://litter.catbox.moe/hjm50f.png
https://litter.catbox.moe/5xs1fz.png

>>43252373
>two eyes
cursed. nice saggers though

>>43252534
hiding heterochromia is cringe

>>43250973
Ina will apply correction to this bratty idol

https://litter.catbox.moe/eo9pfn.png
https://litter.catbox.moe/lenru1.png

>>43252852
Please, can you correct Kobo. Onegai, Ina-sama.

>>43252426
Hot. More surprised/shocked Ina getting railed by her tentacles please?

>>43252426
this >>43254280
I love her shocked face, also tentacles going for wrong hole would be nice

>>43254392
>also tentacles going for wrong hole would be nice
Who knows maybe its the right hole for tentacles

>>43254280
>>43254392
tentacles are busy noe correcting the indo brat >>43253659.
will do it after.

>>43239521
>ask for sana
>get lactation for free
excellent stuff anon, thank you very much!
>>43241579
I'll be waiting anon

>>43254392
No such thing as wrong hole in tentacle porn

>>43244489
>coolest marine edition
honored...

>>43243066
>>43243196
guys, for once, before we start sharing hashes and mixes, can we agree to put HLL in position B like we're supposed to...
>>43249054
yea that's why I've been running with it off lately.

>Controlnet practice
Holy shit
How do I use the poser though? Saw a post that it has like a 3D bone that you can even multiply to make an army.

>>43253659
https://litter.catbox.moe/ta8hp1.png
https://litter.catbox.moe/6eoilo.png
https://litter.catbox.moe/xsj27m.png

>>43252852
one more Sussie
https://litter.catbox.moe/hgg06f.png

>>43255502
You can use blender for posing, or images as pose reference

pic: Mumei show her dominance with T-Posing

Owl laid an egg

>>43255666
Do you know of a dummies guide to blender?

>>43255666
Can I assign which model is for character A, which model is for character B (and maybe more), or is that still up to SD's grace?

>>43255794
I think Easy Pose can be an alternative for that, it's available on Steam and it's cheap. You can just screenshot the pose you made.

Eroge like Koikatsu or Honey Select can be an alternative too I think.

>>43255666
nice

>>43244577
Amelia ASS
Wearing whatever you want yoga pants, jeans, mini-skirt, thongs and so on

It's just need to be an Ass-focused prompt please

>>43244577
Matsuri dressed like this please!!

https://imgur.io/a/5lT2CNS

>>43255502
I have to figure out what an ahoy is so I can go eat it...
>>43255961
>Eroge like Koikatsu or Honey Select can be an alternative too I think.
They cannot. The Openpose model isn't meant for identifying poses from anime/drawings. It might work, but it isn't supposed to. It's for human poses, and if you feed it color coded Blender bones, you're just making its job easier lol
>>43255810
Probably with 2Shot, yeah.

>>43222625
Not exactly what I had in mind...

>>43256542

>>43256092
close enough

>>43256174
I thought the openpose one is supposed to work like img2img but it only tries to convert the pose from one image into a bone then reconstruct it as a new prompted composition

>>43256976
canny is more similar to img2img

Use pic rel to make this
>>43256167

>>43257051
Canny is more like tracing and recolour

>>43257193
Which is good as a LORA alternative to get a pose that hasn't been taught to AI?

Top : Canny
Bottom : Openpose

>>43257386
What settings do you use for openpose? it does jack shit for me

>>43257193
If you turn the weight down to sub .3 it becomes more open to drastic changes while staying somewhat faithful. Depends on the reference pic. Same reference as the gura, way different style. Not just traced.

>>43257419
Oh nice, i don't realise there is weight slider there lol

>>43257465
Ive been using .5-.7 for things I want similar, and .3 for things I want more variation from the source. Works well for me.

>>43257415
Is it showing you the pose in the generations? If no try a real life picture as reference.

>>43256976
They're all "like" img2img in that regard - I've actually only been using ControlNet with img2img prompts lol. But yeah, openpose will read the pose from your input human image and pose your output in the same pose.
>>43257051
>>43257193
canny is edge detection. the ones that aren't openpose, depth, and seg are kinda similar, with differing results. and the "edge detection" ones explicitly would probably be canny and scribble. hed/mlsd are, but not really

picrel is canny's output

>>43257415
I uh... I just set the weight to 1.2 that's all
I left the Annotator Resolution at 512, haven't played with it

>>43257581
I crank the sucker to max lol.
>>43257577
picrel is HED

>>43257568
>Is it showing you the pose in the generations? If no try a real life picture as reference.
Yeah it doesnt I just get a bunch of watercolor impressionists paintings and a black square at the end.
I use the lower size model some dude posted here and openpose as the preprocessor.

>>43257636
>a black square at the end
it's not able to detect a pose. you have to play with the sliders for your input image then or you're using anime/2D as input for a model meant for 3D poses

>>43257636
If the black image is blank its not detecting the pose, try a simple, clean, well lit image of a real person.

>>43257621

>>43257723
I had to use HED to get it to actually see the background. canny threshold sliders are a pain in the ass

>>43256076
How about cowkini?

https://litter.catbox.moe/s96oog.png
https://litter.catbox.moe/768gp8.png

>>43257684
>>43257691
Oh so open pose doesnt work for illustrations at all? whoops

>>43257882
It can if you have a decent image, but its not meant for anime/illustrations.

>>43257882
it CAN, but it's really not supposed to. this is one of those cases where RTFM helps
https://github.com/lllyasviel/ControlNet

Now the Mecha Lulu division can roll out their biggest guns.

Some quick AOM3 comparisons, using hll3.1 lora:
https://litter.catbox.moe/zke433.jpg
https://litter.catbox.moe/ry0hvh.png
https://litter.catbox.moe/zcqkbt.jpg

Use hll loras from here for quick testing, no need to merge models:
https://mega.nz/folder/rBVGjIpQ#WCl9DL65mMRyaBmyIPA9WQ

Also, i just started training HLL4. Looks like --noise_offset (https://www.crosslabs.org//blog/diffusion-with-offset-noise)) actually works, darker images are possible. Still ~30 hours of training left.
https://litter.catbox.moe/mfuhdd.jpg

>>43257770
it's good too, thanks Anon

>>43259439
>Also, i just started training HLL4. Looks like --noise_offset (https://www.crosslabs.org//blog/diffusion-with-offset-noise)) actually works, darker images are possible. Still ~30 hours of training left.
Neat, might wait on merging with AOM3 yet.
Any other changes to the training? Larger dataset? More indies?

>>43259439
>Also, i just started training HLL4. Looks like --noise_offset (https://www.crosslabs.org//blog/diffusion-with-offset-noise)) actually works, darker images are possible. Still ~30 hours of training left.
>https://litter.catbox.moe/mfuhdd.jpg

>>43259303

>>43255644
Can we hope for Ina to correct more brats?

any advice on a mix that has a flatter artstyle that would work better with my asanagi lora?

>>43239452
she needs (You)

>>43259563
Mostly technical changes -
random cropping to partially solve cropped-off heads and feet problem,
noise offset for better color range,
booru + WD tagger(swinv2 model) for better tags.
speech bubbles, multiple views, low quality nfsw filtered out.

For dataset - all vtubers, filtered for low quality (half of indies got filtered out, kek) and duplicates (118k images) + 60k extra images (mostly nsfw). nsfw part is slightly larger than in hll3.1 and more diverse.
Tag counts: https://litter.catbox.moe/e0qhju.txt

I like putting collar on a girl's neck

>>43260113
>flatter artstyle
the same time anyone asks this question. one of the 7th mixes. v3_C is the flattest in my experience. thickest lines too.

>>43260849

>>43260825
>+ 60k extra images (mostly nsfw). nsfw part is slightly larger than in hll3.1 and more diverse.
doesn't whatever NSFW training is already in the models we mix HLL with do that work? I thought that was why we got 3.1 after 3 - NSFW back out

>>43260938
plopping a titty out isnt even hard the entire purpose of this is to get accurate results for as many chubas as possible no?

>>43259439
Please sir will futa be included this time sir thank you sir have nice day sir.

>>43260890
I always forget I have a 7th model in my list... appreciate the retard alert, anon.

>>43261059
"accurate" is not how I would describe most NSFW art I've seen of most chubas. kek. if I were selecting from my own danbooru or pixiv bookmarks for even remotely on model art, MOST of the nsfw would go flying out the window

>>43261166

good mornings vtai. any tips on what tags to add or remove to improve my subaru breeding session? https://files.catbox.moe/1e2uhv.png

>>43261272
The obvious choice is (arthropod limbs)

>>43259439
i like the AOM3 A2

>>43260811
WTF anon, you left after you got her pregnant?

>>43261584

>>43261271
I think Kronii has a skin disease...

So hard to get it right...

>>43261949
Cute.

>>43244577
Botan onee-san lounging on the couch
secretly in heat

>>43261272
Listen to this anon >>43261347

>>43259625
>>43259303
Catbox please

>>43262090
I've made enough scary flesh beings I'm scared to see it mixed with bugs https://files.catbox.moe/juzn9u.png

>>43260925
What about an collar with a leash?

>>43262044
I don't know how much of a secret it is but

>>43260938
"virtual youtuber" tag had only 9k 'nsfw' images left after filtering. So if you don't add more, it will be like old Anything v3 - can't even do a simple sex scene properly.
Not having extra images will also cause drift on some concepts - for example: if there's Ina and her tentacles in dataset, but not enough other tentacles or tentacle sex - all tentacles will look like Ina's. hll1 had this problem.

>3.1 after 3 - NSFW back out
Old 3 had 78k vtuber and 140k nsfw images and nsfw images were not filtered properly.
3.1 had something like 80k vtuber and 30k nsfw
4 have 118k vtuber and 53k nsfw. Ratio looks fine to me.

Personally, i mostly prompt nsfw stuff and use AOM mixes, so having extra nsfw content in hll simplifies things. Because even AOM2_hard cannot do some types of nsfw content, and using a separate lora for every little thing is annoying.

>>43262175
here you go :)
https://files.catbox.moe/lf3pr3.png
https://files.catbox.moe/csetzt.png

>>43262233
Hard to get leash right

>>43262398
nahone. makes sense. thanks for clarifying king

>>43262398
>for example: if there's Ina and her tentacles in dataset, but not enough other tentacles or tentacle sex - all tentacles will look like Ina's. hll1 had this problem.
also brb, gonna use controlnet and 2shot and hll1 to make ina fuck all the things with her sentient squishies
no I'm not. but I'd love to see someone do it

>>43244577
Zaion LanZa waiting for (You) on a bench on the park while wearing a cute dress.

>>43262198
Aw, come on, arthropod limbs are hot
Flesh monstrosities can be, too

>>43263141
YES! more THICC Ame please

>>43263699

>>43262552
Thanks for trying anyway

>>43263711
AME ASS
LET'S GOOOOOOOO

>>43263756
Heres one with a fucked hand I dont want to inpaint

>>43263808

>>43263823
The hand isn't that bad desu
I've seen worse

>>43263868

>>43263993

>>43244577
Robot nurse Polka

>>43264195

>>43264686

>>43264768

>>43264877

>>43254280
>>43254392
>>43254428
>>43252426
https://litter.catbox.moe/gsn7ho.png
https://litter.catbox.moe/c6i6ff.png
https://litter.catbox.moe/hkunh7.png
https://litter.catbox.moe/8xemyx.png
https://litter.catbox.moe/98ha2n.png
https://litter.catbox.moe/b3y831.png
https://litter.catbox.moe/xyrlth.png
https://litter.catbox.moe/sjkzlu.png
https://litter.catbox.moe/1kh731.png
https://litter.catbox.moe/md5tt6.png
https://litter.catbox.moe/0y18j9.png
https://litter.catbox.moe/cxd7k7.png
https://litter.catbox.moe/86eh19.png
https://litter.catbox.moe/adxjxh.png
https://litter.catbox.moe/0pdv5j.png
https://litter.catbox.moe/qxuhp4.png

>>43260076
Who?

How do I stop making Towa bare her shoulder? Black listing "Bare shoulders" doesnt help

>>43265338
Can I ask you to prompt Ina smirking while tentacle fucking Amelia?

>>43265338
Amazing stuff, anon.
>eldritch text
Kek
>Who?
Shion, La+ or Anya

>>43252373
Oh nice, thanks. I really should get around to setting up controlnet.

>>43264981

>>43250880
https://files.catbox.moe/u3s7g1.png

>>43266274

>>43266318

>>43265338
>https://litter.catbox.moe/hkunh7.png
Amazing, i really loved the last panel

>>43266227

>>43250184
I appreciate you trying to help but I'm on google colab, not local anon. I don't have any console other than the one I can't input anything into that starts the SD in the first place.

>>43266543

>>43260811
Lower part of her dress isn't supposed to wrinkle like that

>>43265540
already did that a few days ago

new
https://litter.catbox.moe/pyryyd.png
https://litter.catbox.moe/al385j.png

old
https://litter.catbox.moe/nwoniq.png
https://litter.catbox.moe/q2eb64.png

>>43266607
> I'm on google colab
How do you strengthen tags on colab? {} or ()

>>43266616

>>43244577
Gura as the little cheeky grinning restaurant cleaning lady. A cap on her head if you can

>>43242112
just woke up
https://litter.catbox.moe/a45gz8.png
https://litter.catbox.moe/nfye8k.png
https://litter.catbox.moe/8is7yu.png
https://litter.catbox.moe/rls1t7.png
https://litter.catbox.moe/0ekrby.png

>>43266616

>>43267003 (me)
clicking through all these, looks like i needed to do the car ones as an inpaint desu.

>>43244489
>Colab for Complete Retards: https://colab.research.google.com/drive/1STL60qfoY-iSlhRb9zFETRLTqhNbznRf
I've added an experimental "feature", and by feature, I mean "PR'd someone else's fork and merged it directly into stable-diffusion-webui"

It looks like this out of the box: https://files.catbox.moe/dt7dar.mp4

You do have to tick the "Alternate UI" box on the Colab for Retards to use it.

If you're not a GPUlet like myself, this is the relevant PR: https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/7519 and the main reason why it's not implemented - he added a fucking ko-fi link to the footer
and this is their fork: https://github.com/anapnoe/stable-diffusion-webui
Not sure what the git commands are to "pull" a specific PR locally for testing. But if you want to just git clone it "separately" to play around with it, MAKE SURE YOU ARE NOT IN THE PARENT DIRECTORY OF YOUR STABLE-DIFFUSION-WEBUI FOLDER and then "git clone https://github.com/anapnoe/stable-diffusion-webui"
or you can "git clone --branch anapnoe --single-branch https://github.com/vt-idiot/stable-diffusion-webui.git" to grab it without the "gib moni pls" button, kek

I'm only playing with it because supposedly it allows you to resize UI elements, and its existence was pointed out to me after I tried to change the img2img tab UI to look like the beeg Iroha I posted the other day https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/7886
I have to touch grass for a bit, but when I have a free hour or two later tonight, I will play with it and see if it's worth it or not.

>>43266975
I used both, but I think only () works. Why?
I always thought {} is used for characters to focus, where () is used for everything else to strengthen the tag.

>>43267046
the car is a little derpy but this is a thicc migo

>>43267142
ohhhhh ina has to shit

>>43266975
>>43267156
webui does not use {}. that's novelai shit. you use (((houshou marine, pirate hat))) or (like, this:1.2)
or (like this:0.9) to match [novel ai's, like this]

>>43267156
Just started using colab. Got mixed results and didnt know what was working

Resisting so police Irys will beat me

>>43244577
Gura playing minigolf.

>>43267170

Denoising strength, what should it be set to? Before they changed how hires works I was at 0.7 just because and eventually dropped it to 0.55 because the latent option was fucking up when it was too high but now I'm using a non latent upscalers and have no clue where the denoising should be or even what it does to begin with.

>>43267783
I run denoising on 0.5, I think it gives me the best results for my prompts.

>>43267783
I use 0.5 for general purpose, higher for control net, lower for minor fixes.

>>43267783
.6

>>43266848
Yeah? so? this is the AI thread not anatomy 101 gtfo

>>43266376
I love it, thank you

>>43256834
>>43268317
These are really hot, thank you!

>>43267783
For me, it depends on how much you like the original prompt. On the non-latent scales, I vary between as low as 0.25 to 0.6 if I feel the prompt needs more details in general. 0.65 and above tend to deviate too much from my experience. On latent scale, never go below 0.55 and never above 0.65.

>>43268444

>>43244577
Requesting the other Towa supposedly an anon it was trying doing her some time ago.

>>43268945
What?

>>43267142
>image at the top of the screen on one side
Nice. I hate having to scroll down.

>>43269033
https://twitter.com/armadillo_daiji/status/1621492589152325633 This one.

>>43268929

>>43269033
But has Vtuber really been far even as decided to use what they want to do look more like?

>>43267783
TL;DR - Depends on a lot of things really. I don't use any one setting. They're different tools, with different strengths and weaknesses.

Latent - can add detail! Can add intricacy! Also seemingly completely destroys FINE detail in the first gen like fishnet fabric or other clothing textures or even cool things like "this panel on my robot waifu's leg had stripes and they're gone now" because it gets denoised during the upscale.

ESRGAN scalers - seemingly less destructive to source /detail/ but also less additive detail in the sense that it's less apt to give you "something new entirely" the way latent can and does at anything ~0.4 and higher. They do still add detail, after all, you're using the selected ESRGAN network to upscale in IMAGE space, and then denoising that with the original image in LATENT space ??? (I think. shit might as well be fucking rocket science). Think of the ESRGAN scalers -while you use them in webui- as "waifu2x. but in 2023. and with a generative image AI somehow calling some of the shots"
someone PLEASE correct me if I'm wrong on that one, kek.

I'm also lumping them all together. When you get up to ~0.6 strength, the different latent scalers differ from one another too...

She >>43267188 was latent bicubic antialiased - "softer" added details desired because of the style of the image at 0.6. If I know my image is being prompted sans hands - I'll go as high with 0.7 on any latent if I want "latent magic" in there, but only up to 1.5x resize. Latent itself at 2x upscaling becomes a balancing act/gacha thing at denoise >~0.5-0.6, especially if you have xformers on.

Most of the time I'm using one of the ESRGAN upscalers like lollypop - at 2x resize 0.5. That one doesn't really start to get until you get closer to 0.7. I also prefer lollypop to Anime6B at 2x and greater, but will stick to Anime6B when it's below 2x and I don't want latent fucking things up and just want {{{bigger}}}.

If you want a huge obnoxious comparison image, unfortunately with all of them at 0.7 denoising strength, lmk:
Denoising strength: 0.7, Clip skip: 2, ENSD: 31337, Hires upscale: 1.5, Hires upscaler: Latent, Script: X/Y/Z plot, X Type: Hires upscaler, X Values: "Latent, Latent (antialiased), Latent (bicubic), Latent (bicubic antialiased), Latent (nearest), Latent (nearest-exact), None, Lanczos, Nearest, 4x-AnimeSharp, 4x-UltraMix_Balanced, 4x-UltraMix_Smooth, lollypop, R-ESRGAN 4x+, R-ESRGAN 4x+ Anime6B, LDSR, ScuNET GAN, ScuNET PSNR, SwinIR 4x",
only problem is it is very hard to look at. I need to write a very simple gallery browser for individual xyz images...

And actually, looking at that wikipedia comparison of different image upscaling algorithms, I think it's neat that we can do "latent bicubic" or "latent nearest exact" but I want to see some freakshow shit like "latent xbr".

Played with xbr itself via an AviSynth script a few years ago while trying to upscale some SD - standard definition, 480p, not stable diffusion - footage and was kind of fascinated with it. I'll still use "SuperResXBR" to 4x stuff like old home videos and then Spline36 resize it back down to 720p. Looks significantly cleaner than just blowing it up to 720p with a "normal" upscaler, or letting whatever the playback device is, usually some family member's iPhone or something, handle it.

>>43269373
Bro what the fuck are you trying to say

>>43269345

Guy who has been asking for the fun prompt challenge stuff here. Sorry for not being here for a few threads. Will be next thread if it's not too late in the day.

>>43244577
Requesting Miko in some daisy dukes. Backside view preferred.

>>43267783
>Denoising strength, what should it be set to?
Between 0.5-0.7 usually.
>Before they changed how hires works I was at 0.7 just because and eventually dropped it to 0.55 because the latent option was fucking up when it was too high but now I'm using a non latent upscalers and have no clue where the denoising should be or even what it does to begin with.
>>43269399 sums it up good. Its complicated
>If you want a huge obnoxious comparison image,lmk
One step ahead of you brother https://litter.catbox.moe/qts38v.png

Thanks for all the big booty Kiara lads. I've been meaning to increase my Kiara folder for awhile and this is the best way to do that.

Need help? Trying to make Ina's but most of the images come out with only one of her head flaps? Tried in painting and assuming I'm doing that right can't seem to get the other to pop in and not sure if I need to prompt it somehow or if there's other ways to fix

>>43269757
You only gotta worry about upscaler when your shit looks weird

>>43269593

>>43269944

>>43269757
>>If you want a huge obnoxious comparison image,lmk
>One step ahead of you brother
I think you may have misunderstood "huge" and "obnoxious" me brudda.

>>43270100
Yeah im sure you can properly compare images that are 15 screen lengths apart nice one

>>43270310
>>43269399
>only problem is it is very hard to look at. I need to write a very simple gallery browser for individual xyz images...

I do actually need to find a way to do this easily, since I still have the individual images. Some kind of web page that will show you buttons you can click to switch between them.

>>43269849
Prompting waving is like cheating though.
Any other complicated hand pose like peace sign, holding, grabbing or whatever the AI decides to do with it unprompted sometimes. I dont really have a word to describe it but its like its grabbing a hairstrand or something just fucks with it alltogether and latent upscalers + highresfix just fucks your shit up.

>>43270100
>174MB
kek
Is this every upscaler?
Why though. Just generalise it between Latent and non-latent really.
Comparing non-latent ones between eachother is like playing spot the difference.
Much easier to see what a latent vs non-latent does.

>>43270086
Having fun generating Shion, anon?

>>43270086
time to tinker with resolutions and aspect ratios again

>>43270534
always :) that brat makes me feel things
>>43270542

I feel like I fap to these threads more than any other porn platform nowadays.

>>43270611
I understand. I've been on a crazy streak generating Pekora myself. What model are you using? Or is it a lora mixed with a model? I havent seen this style before.

>>43270659
Install the AI yourself, and let your wildest dreams come true.

>>43270527
I wanted to know all the differences, kek. It's every included upscaler + 4x-AnimeSharp, 4x-UltraMix_Balanced, 4x-UltraMix_Smooth, lollypop

looks like this might work...
https://github.com/loetcodes/multiple-image-comparison-slider
but I'd have to manually fill it out

>>43270687
I'm using using the v3_C variant of 7th, here's the catbox if you wanna see my prompt (schizo negatives warning): https://files.catbox.moe/mpno0u.png
>>43270611

>>43270659
I'm genuinely scared I might develop priapism from edging whilst generating.

>>43270687
also, love that Peko! Hope to see more :)
>>43270783

>>43269465
>>43269373 (me)

I'm just shitposting, I have no idea what the initial cryptic message was about

>>43270783
Gotcha, thank you! Also no need to warn me because I use those schizo tags with every single of my gens as well.

There exists another version of negative tags.

"(worst quality, low quality:1.4), (depth of field, blurry:1.4), (greyscale, monochrome:1.2), (bad_prompt_version2:0.8), (censored, censorship:1.3), (text, speech bubble), error, bad anatomy, (bad hands), sketch, red eyes, multiple views,"

But I think they're more fit for more lewd stuff, I experimented with them today and the negative tags I posted above are well suited for lewd more than tame stuff. Otherwise you might get a bit too "burned" results. Also if you're using a model like yourself, the tags such as "dof" and "blurry" are redundant. It's simply for models using AOM such as creamsicle(pic related) or grapefruit(the newer creamsicle)

>>43270898
Be my guest, I've generated a LOT for the past 3-5 days.

>>43271149
ah, that's the negative prompt I was using before, but I think bad_prompt_version2:0.8 is an embed? So I was getting a lot of weird gen errors using it.
Keep sharing those Pekos! :)

>>43262365
yesyesyes.mp3
Can you do it with her loose shirt outfit?

>>43271250
>bad_prompt_version2:0.8
Honestly, I have no clue. I figured some anons included it into the colab itself as a script perhaps? So it wouldn't be a wall of schizotags. It works though, try it yourself. (Or maybe that's just a placebo due to worst quality, low quality:1.4)

>So I was getting a lot of weird gen errors using it.
Odd, I didn't have any issues with it myself.

>Keep sharing those Pekos! :)
Glad you like them.

just some mother nature

https://files.catbox.moe/1beyhk.png
https://files.catbox.moe/fydc6o.png
https://files.catbox.moe/a22i7f.png
https://files.catbox.moe/2t4fzl.png

>>43271634
will have to give it another shot then :)
Appreciate the info!

>>43271837
You're welcome. I wish I could get the latent couple script working on auto1111, still don't know how to apply the damn patch.

>>43271803
ナイス

>>43271984
you run it in GIT Bash in the webui folder :)

Actually managed to make a model that spits out a half-decent Roboco if you squint hard enough. Might have to pass this one through the succu-prompt...

https://files.catbox.moe/r1g6s4.png
https://files.catbox.moe/kxigbq.png
https://files.catbox.moe/a8uvs7.png
https://files.catbox.moe/bht53f.png
https://files.catbox.moe/qkzgu4.png

>>43272114

>>43272114
Uhh, I'm sorry. I'm quite stupid when it comes to this. Do I need to put the script into one of these folders or what? https://files.catbox.moe/e515o6.png The scripts folder perhaps?

>>43272450
oh you're a Colab bro!
Sorry, I thought you were using the local install of webui. Unfortunately I'm not versed in Colab stuff :(

>>43272578
Yeah. My gpu is 970, so I'm far from being to able to enjoy local gens, hahah. Thanks for trying to help me, regardless.

>>43272647
no worries, hope you get it sorted soon! :)
And I feel that. I was on my 1060 until the holidays, just got my 3090 all set up and it has changed my life with stablediffusion lol. Making loras and all that.

>>43263868
>BTLSIN
God damn do I love Bulin Waion.
Catbox?

>>43272741
Yeah must be nice to be able to switch models and loras on the fly. I am stuck with multiple google accounts if I want to switch up models + there's also colab limits, so yeah. Once I upgrade my pc I'll join you and everyone else on the local throne.
Also I remember few months back an anon in these threads who also got a new GPU and got into AI and constantly posted Okayu in a swimsuit, you remind me of him, but Shion.

>>43259439
I am not getting anywhere near these results with aom3a2. It's absolutely fucked for me.
https://litter.catbox.moe/4dsjxj.png
Litterbox same prompt but removed the lap sitting lora in case that's what was fucking it up, and it did seem to help but not much.
hiresfix doesn't seem to help much either.
https://litter.catbox.moe/gl4n6e.png

>>43272908
here you go :)
https://files.catbox.moe/j1g7iq.png
>>43272927
haha I love when that happens.
and I really do look forward to seeing your content when you can join us on local install!
I tend to do batches of stuff at a time, like the asanagi lora and certain characters/poses. I also sometimes do plugsuits, once called myself plugsuit anon, even. lol

>>43273244

>>43273244
Thanks! :)

Fuck feet and fuck fishnets

>>43270310
I will make it work with a comparison viewer.
>>43244577
It's Friday, post Coco.

>>43272450
I'm not sure where this starts, but which Colab are you using?

>>43273244
>seeing your content
I mean. I can do most of the stuff already loras and so on, and I'm very thankful for it, AI is truly a miracle.
And I'm certain I can get latent couple working as well, just need to figure out how. I refuse to believe that there isn't a way to make it so on colab.

>I tend to do batches of stuff at a time
I know I can do batches of different tags as well, I didn't figure out how though. There's a script included in auto1111 but I haven't fiddled with it at all. I just do batches of regular images and tags.

>I also sometimes do plugsuits
I mean, we all have our niches and uses for AI, don't we? Something that lacks in actual art market we can make real. You want Shion in a mecha? You can make that or Coco in a racing queen suit? Easy. And so on, I generated a lot of slingshot swimsuit Pekoras today for example.

>>43274095
I use lastben's fast_stable_diffusion_AUTOMATIC1111.ipynb mainly and recently poked around with voldy's colab for retards but I didn't like how it butchers output quality sometimes.

>>43274217
>voldy's colab for retards
It's not Voldy's, it's mine desu. I kept the original filename.
>butchers output quality
Nothing about it is intrinsically different. Uncheck the xformers option.

>>43272380
She looks like half BIONICLE and I dig it.

>>43274217
>>43274276
I dropped this. Those two extensions are just part of the start-up workflow if you choose to use them, patches and all.

>>43274217
plugsuits are life <3

Had to brave literal hurricane force winds to get here, but now it's proompting time!
Good evening /vtai/!

>>43272380
(((more technic)))

>>43274356
Damn is that an easily accessible colab option for that posing thing? Was gonna start fiddling with getting it up and running, but I guess this might save me a lot of trouble lmao.

Goodmorning, /vtai/

>>43270659
Believe me, once I got this thing set up, I've fapped myself dry from hundreds of generations of different chuubas. So dry, that I got tired of fapping all the time and rarely generate lewd shit anymore. I'd rather just do wholesome shit now lol

>>43274559
Last two OPs were made using it, yes. It makes at least installing it dummyproof. if you're already using webui on your own hardware, installing it locally is dummyproof too - check the previous thread and watch the saga of luluposter unfold as he installed it. if he can do it...

I'm not sure if 2Shot still works or not if the required patch(es) changed but I rarely use it, and nobody's reported it as broken.

>>43274276
Oh, thank you very much for setting up the colab in the first place. I like how you included a bunch of models to play around with.

>Nothing about it is intrinsically different. Uncheck the xformers option.
Strange, I actually tried that and didn't notice any difference. It still looked "a bit" off - the artifacts around the eyes like here https://files.catbox.moe/cmdz2d.png But I will try again next time I boot your colab up.

>>43274356
Thanks! That saves me and a lot of other anons the trouble. But since you're a tech savvy, can you point me out how do I get the latent couple working on auto1111 colab? Where do I put the patch to make it work? Or do I need to edit code somewhere in one of the js files?

>>43274095
Here's your Coco tax.

>>43262044
She may or may not be as inconspicuous as she thinks she is
https://files.catbox.moe/xrmhd4.png

>>43271634
>bad_prompt_version2
found the file on huggingface.
this prompt was done with the negatives including that embedding, seems to work well now!

>>43274711
Neat, guess I'll give it a shot and see how it goes, excited to try this stuff out, appreciate the work you're doing to make this more accessible to us less mentally available anons.

>>43274635
Wholesome Mioposters together strong.

>>43274922
>Wholesome Mioposters together strong.
I'll drink to that

I like how it made heterochromia apply to her glasses too

>>43274450
I made an android embed way back in the pre-LoRA days, might use that and do a gothpunk android series after this.

>>43274635
Rule #1 of dealing: don't coom to your own supply!

>>43272380

{Bigger}
https://files.catbox.moe/bjbyui.png
https://files.catbox.moe/h82rov.png
https://files.catbox.moe/rxb2bc.png
https://files.catbox.moe/idb1yx.png
https://files.catbox.moe/r3vy15.png

{{{Bigger}}}
https://files.catbox.moe/yor2ib.png
https://files.catbox.moe/lqzkoy.png

>>43274861
God I love that look of "I should behave but I need to breed badly"

>>43274635
Morning anon, hope you slept well.

>I'd rather just do wholesome shit now lol
It's nice to keep the balance between the two.

>>43274922
I have some wholesome Mios too.

>>43274371
I'm more of a fan of bodysuits myself. But I guess they're pretty much the same thing?Just more mecha, I suppose.

>>43274919
Oh great! So it's a script or extension, lora or what is it?

>>43275142
>>43274999
:3

>>43275084
>Rule #1 of dealing: don't coom to your own supply!
Don't worry, I learned that the... hard way.

>>43275142
>It's nice to keep the balance between the two.
True, but it'll take a lot of wholesomeness to bring me back to the middle lol

Fellas how do I tag the neck area above the tits?
Even if I tag out all the detached collar nonsense it loves to fill that space with random lace.

>>43275142
its a textual inversion embedding, added to the 'extensions' folder of webui. basically a single tag that counts as multiple tags, and the benefit of being able to modify the weight.

>>43275296

>>43275325

>>43275505

>>43275381

>ShareX a GIF of all these Choco'seses filenames changing in realtime
>Use Windows 3.1 File Manager for the lulz
>ShareX cucks me with this
aw

>>43274790
Again, nothing about the one for retards is intrinsically different as far as the generated images go, other than xformers being enabled by default. Also, if you want to be sure, launch with it disabled so that it doesn't get installed. I think launch.py tries to look for it whether or not you ask for it with --xformers.
One other thing, these two options are also enabled by default, because I'd been using them:
>Enable quantization in K samplers for sharper and cleaner results. This may change existing seeds. Requires restart to apply.
>Make K-diffusion samplers produce same images in a batch as when making a single image
And one thing that I can't change on my end - there's a pending PR github to fix it though - SDE samplers (all of them) are non-deterministic when batched vs. when generating single images, I think they've always been like that. But that would also apply on thelastben's unless he's merging that PR in at runtime. Extra annoying because a week-ish ago, webui stopped saving batch information to the PNG info...
>>43274790
>can you point me out how do I get the latent couple working on auto1111 colab? Where do I put the patch to make it work? Or do I need to edit code somewhere in one of the js files?
Auto1111 usual means the actual github repo for webui itself, no? You mean TheLastBen's "fast-stable-diffusion" notebook from the OP? Sorry m8, I haven't used his notebook since December. Last I looked at it, it symlin...sorry, "mount --bind"ed a bunch of stuff to Google Drive, which is obscenely slow. And also means that a bunch of files that webui uses "persist" somewhere on your Google Drive when you're done. One of the steps for installing 2Shot patches webui itself in more than one place to make it work - so depending on what gets saved and what gets downloaded fresh when you re-run it, you'd be "stuck" with it installed.
If his code cells are split up in a similar way - i.e. "cloning the webui repo happens separately from launching it" you can try copying the cell that I have for installing it in right before you launch. The only thing I kept from his notebook was the little block of code to allow switching to localtunnel, and the on the fly edits to blocks.py to make text green and print the localtunnel URL in the console. The rest is surprisingly all me at this point...

>>43274922
Cheers

>>43275296
Nice Zaion, what model is this?

>>43275282
:D

>>43275325
Gotcha! Thanks for the info, I'll download it asap and try it out myself.

>>43263167
>jeez there you are...what took you so long?

Im gonna become the strongest zaion poster if you guys keep leaving me to do all the requests

>>43275686
>Nice Zaion, what model is this?
I'm currently testing out my own custom mixes, but it's mostly from HoloCreamsicle_new.

>>43275686
happy to help! hope it serves you well :)

>>43265860
https://litter.catbox.moe/51xnfs.png
https://litter.catbox.moe/4qxjbv.png
https://litter.catbox.moe/ndezw7.png
https://litter.catbox.moe/znotiz.png
https://litter.catbox.moe/6kh291.png

>>43275829
happy to see Shion getting some correction.