Thread 8714630

Just take a shower edition

Previous Inspection >>8707841

>LOCAL UI
reForge: https://github.com/Panchovix/stable-diffusion-webui-reForge
Comfy: https://github.com/comfyanonymous/ComfyUI | https://comfyanonymous.github.io/ComfyUI_examples

>RESOURCES
4chanX Catbox/NAI prompt userscript: https://rentry.org/hdgcb
Tags: https://danbooru.donmai.us/wiki_pages/tag_groups | https://danbooru.donmai.us/related_tag | https://tagexplorer.github.io/#/
Inpaint: https://files.catbox.moe/fbzsxb.jpg (embed)
Upscalers: https://openmodeldb.info
Booru: https://aibooru.online
ControlNet (OLD): https://rentry.org/dummycontrolnet

>TRAINING
Guide (WIP): https://rentry.org/yahbgtr
Anon's scripts: https://mega.nz/folder/VxYFhAYb#FQZn8iz_SxWV3x1BBaJGbw
Trainers: https://github.com/derrian-distro/LoRA_Easy_Training_Scripts | https://github.com/bmaltais/kohya_ss | https://github.com/Nerogar/OneTrainer

>NOOBAI-XL
https://civitai.com/models/833294/noobai-xl-nai-xl
https://huggingface.co/Laxhar

>NOOBAI SHITMIXES
102d custom: https://civitai.com/models/1201815?modelVersionId=1491533
291h: https://civitai.com/models/1301670/291h

OP Template/Logo: https://rentry.org/hggop/edit | https://files.catbox.moe/om5a99.png (embed)

>>8714630 (OP)
XXX Edition

Previous Thread: XXX

>LOCAL UI
reForge: https://github.com/Panchovix/stable-diffusion-webui-reForge
Comfy: https://github.com/comfyanonymous/ComfyUI

>RESOURCES
FAQ: https://rentry.org/hggFAQ
Wiki: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki | https://comfyanonymous.github.io/ComfyUI_examples
Training: https://rentry.org/59xed3 | https://github.com/derrian-distro/LoRA_Easy_Training_Scripts | https://github.com/bmaltais/kohya_ss | https://github.com/Nerogar/OneTrainer
Tags: https://danbooru.donmai.us/wiki_pages/tag_groups | https://danbooru.donmai.us/related_tag | https://tagexplorer.github.io/#/
ControlNet (OLD): https://rentry.org/dummycontrolnet | https://huggingface.co/2vXpSwA7/iroiro-lora/tree/main
Inpaint: https://files.catbox.moe/fbzsxb.jpg (embed)
IOPaint (LamaCleaner): https://www.iopaint.com/install
Upscalers: https://openmodeldb.info
Booru: https://aibooru.online
4chanX Catbox/NAI prompt userscript: https://rentry.org/hdgcb
Legacy: https://rentry.org/hgglegacy

>TRAINING
Guide (WIP): https://rentry.org/yahbgtr
Anon's scripts: https://mega.nz/folder/VxYFhAYb#FQZn8iz_SxWV3x1BBaJGbw
Trainers: https://github.com/67372a/LoRA_Easy_Training_Scripts | https://github.com/kohya-ss/sd-scripts

>MODELS
https://civitai.com/models/833294/noobai-xl-nai-xl

OP Template/Logo: https://rentry.org/hggop/edit | https://files.catbox.moe/om5a99.png (embed)

Here is the lora. Hopefully this thread is more chill.
https://pixeldrain com/u/XB3nX5zV
>>8714634
Ok we'll use this one next time.

I really hoped we could merge back this time

>>8714638
thanks anon

>troll op
nah

>>8714630 (OP)
>(embed) (embed)
Everything else looks good to me, thanks for the bake

>>8714638
what lora?

>>8714652
Open it and find out :^)

>>8714647
Make a new thread.

Would you hire a clumsy elf maid that regularly trips and stumbles onto your dick?

So you're gonna keep making those garbage threads and pretend like nothing's happening? How are subhumans like you live with yourselves? Why not put a gun to your head and end it right now?

NON-TROLL THREAD
>>8714671
>>8714671
>>8714671

>>8714665
depends, is she on birth control

>>8714665
only if she has the correct number of fingers

>>8714677
No, but elves have low fertility or something
>>8714680
rebaked the lora a ton and hands aren't getting any better. zero lineart styles are a pain. obviously not gonna inpaint

>artist goes from kino to filtering ai slop
many such cases

>>8714694
eyeshadow always looks wack to me

>>8714694
>4 arms
Scary. I hope she has a spider body.

>>8713852
https://files.catbox.moe/n539e8.png

>>8714698
>>8714701
BOOOOOOOOOOOOOMBACLAAAAT

>>8714700
Nice.

Wonder if it's the anti-split schizo from the early threads or iaschizo

>>8714700
red rocket tip and anemic foreskin lol

Other people's gens are always more arousing than mine. I guess what they say about food made by others is true.

>>8714631
I never bake chars. I've had much more success with high rank loras when the goal is to make them work on their own on base noob. if you're on 102d or something pretty much anything is fine.

>me thinking someone is copying the same shitposts to both threads
>realize I had /hgg/ open twice with no /hdg/

>>8714700
ooops I catboxed the wrong gen
https://files.catbox.moe/gc7291.png

>>8714712
Food made by others is only better if you're a tasteless skillet so it makes total sense when it comes to your gens.

>>8714724
You can't put love in it when you cook yourself. unless your a narcissist or something

>>8714724
I'm sorry you're so unloved, anon. I will always appreciate the way my mom makes chili, even if I can make her recipe just like hers. Anon's gens are nice, even if the composition is simple enough for me to have done it myself.

>>8714727
I love your mother's chili too ;)

https://www.youtube.com/watch?v=fIBMepIBKhI
Did anyone try this? Can Aoi Yuuki tell me that she loves me now?

just wanna point out I hate dubfags

>>8714733
vaffanculo

>>8714733
No but you can have the Gabibbo tell you are his favorite besugo.

>>8714733
so that's why microsoft nuked vibevoice

>>8714733
voice transfer has been a thing for years
the worse part is the tts, voice transfer doesn't really vibe with me

>>8714700
Cutest, most overlooked Pokegirl

>>8714651
looks like he copied it from some previous op like a retard instead of using the updated template https://rentry.org/hggop/edit
>>8714634
>(embed) (embed)
>FAQ: https://rentry.org/hggFAQ
>Legacy: https://rentry.org/hgglegacy
>placeholder
kek
>https://huggingface.co/2vXpSwA7/iroiro-lora/tree/main
the fuck is this? is it any better than noob controlnets?

>>8714792
>https://huggingface.co/2vXpSwA7/iroiro-lora/tree/main
actually it's even included in here so whatever https://civitai.com/models/136070

>>8714792
it's anytest
I must've repeated this 50 times by now >>8711179

>>8714794
it's the old one for pony and animagine
the huggingface page also has newer versions specifically trained for illustrious, they work a little better

>>8714792
>(embed) (embed)
we have 600 characters spare so we can do this for a while
>placeholder
Op seems to be motivated so he can write something about shitmixes and the old lora rentry here
>actually it's even included in here so whatever
didn't realize this, anytest was fairly underground in the past.

>>8714795
>anytest
It was always shit for me on pony. But anyway does that mean anon was lying when he said to use the old sdxl 1.0 CN for canny/poses?

>>8714798
I use anytest4 illu A at 0.3-0.6 strength with mostly koikatsu stuff as input and it works pretty well.

>>8714795
>it's the old one for pony and animagine
yeah, looks like the ones uploaded to civit are a bit crusty
the description links to the anytest huggingface page however and i personally never download from civit due to retarded ui (if hf is an option)
>>8714796
>so we can do this for a while
i don't see the point if you can just reply to the op and i'll update the template if you want to change something
>Op seems to be motivated so he can write something
don't mind this

>>8714804
>don't mind this
Then ask gpt for one

>>8714807
i'm not the one writing it, there was a skipped "i"
>gpt
why are you like this

>>8714798
It's meant to be used with regular images. Canny needs an image that's just lineart, or the output of a canny preprocessor that converts it into lineart. It only sees the lines, not depth information. So it's less able to recognize limb salad poses like 69, one person standing in front of another, etc. It also works poorly with images that don't have lineart, like photographs. But this "blindness" to depth can be helpful in other cases. And obviously it works better if all you have is lineart, like coloring manga, or turning your sketches into gens.

>>8714816
Oh okay I see, thank you.

did someone upload lilbro's sfw masterpieces to reddit or something kek
people are fucking hilarious with how possessive they are of aislop out of all things

>>8714828
huh?

>>8714838
i'll let you figure it out

I see the schizoposting has already begun. Couldn't even hold it in.

good morning ladies

>>8714841
Morning cunt, up early are we?

>>8714840
think you need to take your meds bro

>>8714696
That mean more handjobs!

>>8714848
Why are you even looking at his obvious shitposts to begin with? Just to seethe? He's been posting those mp4s in that style for a while now. Just keep scrolling.

>no bro you must be thoughtpoliced by me
kys

>>>/pol/514979679
what causes this

>>8714855
shitposter-kun, the crosslink has to have an immediately controversial image or text post
nobody's gonna click a /pol/ link just to click another link

>poltards discovering a meta from summer 2024
christ

>>8714855
the only AI song I fully listened to

it's the freetard poljeet schizosaar here to seethe about novelai's superior saas model

saars

>>8714856
I apologize, apparently the six of us here are all fomo idiots

is this confirmed? >>>/v/719991030

>>8714871
it's actually poopdickschizo

now 7 :^)

seeing lots of good novelai gens recently, starting to get the urge to resub

>>8714871
That's actually a good thread for aerithchads like myself :^)

Reminds me that I still haven't finished rebirth. kinda stopped playing in the mushroom village since the gameplay was ass

saars please stop raiding aur servers...

>>8714881
Gameplay is good though and fighting the summons is the proof of that. The major drawback has always been the changes to the story.

why it's so ugly, tho

>>8714886
I really don't like the repetitive open world stuff. the fights themselves are cool but retarded vantage point towers should have died out with ac3

for me it's huniepop (unironically)

but that's /aco/

cunnypop

that's anime

almost done with silksong I guess, new patreon update soon

>>8714889
For me it was that the character development is spread out over a bunch of side quests. Where the pacing was quick in remake, it ground to a halt in remake, even if there were some good moments.

>some random artist i baked based on 1 pic
>13 pics in dataset
>ends up being my favorite current style
i do be like that sometimes

>/hgg/ post highlights
>>8714314

Baking on doujins is so easy nowadays...

been out for a week
anything happened in that time?

:^)

>>8714733
i tried this some and it works really well in the 5% of gens that aren't random guttural noises or musical interludes or a random person singing

didnt there used to be a thread for finding doujins? or can I just ask here?

How can I make an AI generate hentai images based off of a certain artist's style?

>>8715020
sub to novelai 4.5
should be all you need

Which one is the best position out there and why it is full nelson

>>8714678
>still up almost 12 hours later
Thoughts?

it's nai

>>8715085
sensei's face is almost merging with her shoulder, weird pussy, weird fingers, background could have been better, nice bite marks, cute girl

moneysuckleschizo
cathag
omnipotentnaigodschizo
witnessmeschizo
birdschizo
Ricky DeNero
noodlekek
iprelayschizo
robloxdudeschizo

>>8715041
No, suspended congress.

pixelsmith and golden noise will save local

the chroma anime finetune will save local

I'm looking for good hires fix upscaler. I've been using remarci but I'm wondering if there's anything better.

which ones do you guys recommend?

I project and transfer all my suffering to this thread so that you all may suffer instead.
SUFFER!!!

>>8714330
What loras/styles are you using for this look again, anon? You shared it a while back but I guess I didn't save it or had an aneurysm deciphering the cumfy moon runes.

>>8715152
I'm using 2x-animesharp-v3
but I also feel like it doesn't matter, if your denoise is higher than like 0.1 it already erases any differences between upscalers.

>>8715163
On that topic, anyone have that comfy node to provide idiot-accessible metadata? I've seen a few catboxes that I couldn't import but that showed the prompts and sampler settings in a human-readable format.

>>8715167
nah it's pretty noticeable even at ~0.35. 0.5 and above yeah it doesn't matter at all

>>8715168
are you talking about stealth metadata?

>>8715184
I don't know, am I? I only saw the effect, not the tool producing it.

>>8715184
I think he meant a node that can present the comfy metadata in forge-compatible format that won't look like garbled mess anywhere outside of cumrag.

>>8715197
Yeah this is it. I believe chadmix anon was one of the few who went out of his way for it thobeit.

damn, i feel for the nai meme. I want my money back

>>8715200
GMS. GOOD LUCK ON THE REFUND OF THE RUPEES.

>>8715152
never use upscalers in hiresfix

>>8715210
As someone who understands what hiresfix does under the hood, I have to ask you to explain in more detail. Your statement doesn't make sense.

>>8715216
no :)

>>8714638
sorry, I downloaded and forgot to do anything with it

interesting how it mixes flat anime faces while the body and background are almost as if traced from a photograph.

>>8715235
another >>>/e/3000137
style looks like it would be perfect for easy img2img into photorealism

>>8715210
retard

What a gay thread

>>8715245
Maybe he's just bad at expression himself.

NEVER. USE. UPSCALERS. IN. HIRESFIX.

>>8715249
R. E. T. A. R. D.

>>8715210
>>8715249
Look, hiresfix is literally just an upscale step and a second AI pass. If you're not doing the upscale then it's not hiresfix anymore, it's just img2img at the same resolution. Do you mean to not use esrgan-based upscalers but only algorithmic ones? Which sounds questionable but at least isn't outright nonsense.

I hate AI btw.

>>8715255
N. E. V. E. R. U. S. E. U. P. S. C. A. L. E .R. S. I. N. H. I. R. E. S. F. I. X.

Imagine using upscalers with Hiresfix, couldn't be me

why would one not upscale in hiresfix

run a 1:1 and you'll get it

I guess it comes out more detailed right away and then just img2img details a bit more

its way fucking slower tho

you can't use hr.fix without an upscaler, retards. You'll have to increase either the size of the image or the latent for it to do anything.

just use nai 25 dollaridoos

>>8715282
its not really a hiresfix but you can just skip the upscaling

if no upscaling is so good then what rest to gen on , a good thing about the hiresfix upscale is that you can even gen something like 1216x1824 without much of a problem with the upscale while genning that from the beginning tends to get a fucked up image.

???

Jesus fucking Christ, you utter retards. No upscaler in hiresfix doesn't mean the image doesn't get resized, it just does it normally instead of using a meltie upscaler that gets denoised into irrelevancy anyway.

>>8715291
nigga there is no normally, you always need some basic algorithm which is lanczos most of the time

>>8715292
Waow, you're so smart! Now tell that to the person who called the option "None".

so what you're actually saying is to not use an upscale model

i mean yeah it takes some major skill to grasp that from "never use upscalers"
truly

>>8715291
If the image was upscaled then there was an upscaler used to do it, otherwise how would it happen "normally". If you do not specify an upscaler it uses Lanczos, if you do specify it uses your upscaler of choice then makes up the difference with Lanczos. But it's still an upscaler.

You throw insults around but you can't express your thoughts clearly, and you don't know what's actually happening under the hood.

>>8715295
eh, I already guessed that much here >>8715255
he just kept being schizo instead of confirming cause he didn't know himself

retard

anon wants to say stuff but is bad at expressing himself

>>8715302
just line in my japanese animes
except it's not a cute girl this time

AnimeVideo 6B is where it's at

>>8715304
more like 6BBC's you fucking cuck

https://civitai.com/models/1935380/ring-light-reflection?modelVersionId=2190457
finally

>>8715309
I like how all the sample images are off-model gachasluts, he knows his audience.

>>8715311
I have developed a seething hatred for that blonde in the bunnyslut suit just because of AI
I don't even know the anime/game

>>8715311
Ok but does it actually work?

Someone in the previous thread posted this, and I just wanted to say thanks. This is high-quality anime porn. Love me big titty goth girls, simple as.

jej?

>>8715348
Would be better if it were /ss/ however

>>8715216
He uses latent to upscale. Also make sure you change that setting for img2img in reforge, anons. Set your upscaler manually there so it doesn't use latent in img2img.

>>8715354
Img2img doesn't even use latent by default, just lanczos.

>>8715179
Nice pose. Pov 69 is a real treat.

Still haven't tried the model anon recommended but I'm dropping by to ask for more Hilichurl related gens.

>>8714009
Basedo

>>8715327
Even in styles where it shouldn't. But then adding skin reflections was never a problem, more the other way around.

>>8715326
Blue Archive. She's one of the few adults in an otherwise loli game, so gets a lot more attention that she would in say Azur Lane or similar.

>>8715371
>adult
she's 17, you sick fuck

>>8715371
I just hate that she has the bean mouth

>>8715376
>he doesn't know

>>8715152
nomos8k or 2x-AnimeSharpV4 but it doesn't matter that much if you are going to inpaint after that

https://civitai.com/models/1572905/chromalinemix-or-noobai-v-pred?modelVersionId=2182143
bluvoll has now arrived where zoot was a year ago

>>8715373
Ah, my bad for assuming a maid and a secret agent would have old enough for either of her jobs.

>>8715380
I don't like any of the preview images

how can i check if my pixels are not underage and legal?

>>8715402
they insist upon themselves

>>8715402
Are their breasts AT LEAST (huge_breasts:1.4)? Good, now check if there is a single piece of official art of them wearing a scho*l uniform or any other kind of outfit associated with pedo CSAM. If there is none (and if your gen doesn' contain any of that), you can now check their official age. If they are at least older than 38, you are clear and your pixels are now fully legal and wholesome.

Just post them and let's see :^)

>>8715407
>If they are at least older than 38, you are clear and your pixels are now fully legal and wholesome.
That also depends. If you're in 18-25 range then yeah I guess it's fine. If you're in your late twenties or thirties pursuing women that young is kinda creepy though. Technically legal, yeah (for now) but you really should just stop being a creep and go for the 50+ age range.

>>8715407
But oppai loli exists. Will the cops still hate me?

>>8715393
what causes this

>>8715411
Oppai loli is the most pedophilic one because you don't even have the balls to just jerk off to a childs and try to pretend it's a regular woman

>>8715412
I don't know but I agree, comfytroons should just fuck off to jeetvitai.

how is that THE most you fucking disingenuous imbecile?

>>8715415
Quick everyone, we have to get to the truthnvke shelter before the blast arrives!

uh oh

my brain make up the story for money

what caused that?

why is he copying my bait

it's not yours

chat, I still have a fucking lot to do to 100% silksong

why are you transgender

>>8714630 (OP)
What's the box for this one?

>>>/v/720041763
why anime looks always so pedophilic?

you're brown

>>8715456
the japs are tortured by their meaningless adult lives so they have a severe case of autoephebophilia to cope with that

There are too many TRVKEs being dropped ITT, we'll end in a TRVKlear winter like this.

>>8715402
ask gpt

its crazy how good novelai 4.5 is
and its so cheap to sub

Buy $25 of nvidia stock instead and profit every time a dumb naigger hits generate

>>8715422
As if we didn't get enough of this with naruhodo. The market wants what it wants.

>141/58
ugh why is hdg beating us, sisters?

>>8715410
If you are in 18-25 range, you shouldn't be having sex anyway. Having sex before stable career and mature brain is degenerate. When you are 40 is the right time to start.

>>8714630 (OP)
The lora posted earlier.

>>8715483
I can smell your circumsized penis from here

>>8715455
Bro literally posted a lora for that one couple posts below. How come you aren't reading the thread?!? It contains so much useful info.

>>8715483
Good point. That only applies to men, of course, women are free to do whatever they want as long as they are not having kids before 35.

>>8715455
See >>8715485 The catbox extension loves to fuck me over sometimes.

Bro who posted bakushin last thread and gave .toml your's actually kinda worked I just need to probably double the epochs. Thanks

im jewish
im circumcised
im subbed to nai

>>8715483
>>8715489
tel aviv or cali?

NovelAI is insanely good, and they just keep getting bigger. I remember reading posts from localjeetykekschizosaarmeltygroids about how NovelAI would be dead soon, but here they are over a year later still thriving.

4.5 images were literally so bad they killed /hdg/

>>8715495
This. You can mix characters in ways local could only dream of, plus the 16 channel vae allows for way finer detail. I was a long-time local user, but after seeing local continue to decline I felt like I had no choice but to buy a sub. Here I am now as a happy Opus subscriber generating things I never imagined were possible. Thank you, NovelAI!

>>8715496
Now tell me, how petite is too petite?

>>8715498
and /hgg/ is next. it's time for /elg/ the ecchi lora general

>>8715501
>do they look like a child?

>>8715495
Free trials are always the beginning of the end. Take it from a long time mmo player

>thread gets flooded with blatant kike propaganda
>immediate naishit flood follows
Their incompetence and disregard is almost comical. Eat lead, kike.

>>8715495
>>8715499
based! at 25 dollars theyre practically giving away a SOTA model

it's soo easy to make localbrown seethe. why does he hate superior saas models so much?

>>8715496
we should have it in the OP

>>8715520
In the FAQ ironically
>placeholder

>>8715496
>are they trying to seem like a child
what does this even mean
do they need to be chainsmoking with a margarita in hand to demonstrably prove "look this is totally an adult!" or something?

>>8715523
Probably referring to hags cosplaying in kindergarten uniforms or something.

>>8715523
ikr? kek
>40 year old woman puts on her high school uniform to relive her glory days
>not allowed
But of course in good faith I'd say it means hypnosis, age regression, unironic retards, dementia/alzheimers, and things of that nature that make an adult act like a child/baby is not allowed since they can't consent mentally.

pedo uncs sure love to yap all day

>>8715525
cosplaying is a very adult pastime though.
kids don't have the income to do that shit.

but I mean, "sexual situations" clearly isn't valid here, even though it should be. If someone wants to see "noooo that's a child!!!!" they can see it in fucking anything they want. Scantily clad girl with C cups? She's clearly 17 you fucking sicko pervert. D cups and a youthful face? SHE IS NOT 18 YOU PEDO! Chest completely obscured with wide hips and a fat ass? That's clearly a loli and you're trying to skirt the rules you sick fuck.
May as well say that if the girl doesn't have visible wrinkles it's not allowed.

>>>/v/720051665
it's happening

>>8715528
>cosplaying is a very adult pastime though.
Nah they actually don't even care if it's an actual adult or not. Even a D cup 30 year old bitch cosplaying as a school girl is now unacceptable for redditors because it's pedo-coded or something so it's forbidden too.

>>8715495
does nai only do /e/ or is /h/ a 35$ option?

>do they look like a child
>40 year old woman
based skim readers

>>8715495
thanks for the nice mix!

you got baited into discussing pedo again

>>8715538
>implying it isn't pedos favorite past time to discuss how persecuted they are

>controlnet
AHAHAHAHAHAHAHAHAHAHAHAHAHAHA

>videos
Jittery rotoscoped slop with endless yapping
>loras
Literally a shitty cope for not being able to finetune models properly on underpowered local hardware
>controlnet
lmao, lol
>unlimited upscales
A cope for the obsolete 4ch VAE melty slop local has to deal with

>controlnet
worst shit that happened to local unironically

didn't kurumuz PROMISE controlnets for v3
and then just never delivered because they were hackfrauds and couldn't even manage that much?
it's bizarre how hard they got carried by their mistake in SMEA implementation.

>>8715551
For me it's Krake modules
>textfags third year of waiting for having something to spend anals on

it's fucking sunday what the fuck is wrong with you retards, go outside

>>8715557
lilbro writing from the corner at the party

>>8715557
we are. novelai is superior saas which allows us to prompt from our phones.

>>8715561
kek

haushauhasuahaus even

>can prompt from anywhere
>lightning speed
>superior character recognition
>better styles
>16 channels
>proper segmentation without regional coper
>unlimited generations
I am seeing no reason to cancel my sub anytime soon.

it can't do cripplefus, it's shit

it is literally a regional prompter though, you even get to pick the regions

Reading this shithole of a general is like watching a trainwreck. Holy shit you people are actually miserable.

>>8715577
Fact Check: FALSE
Kurumuz confirmed that NovelAI's character prompting technique is NOT regional prompter, but instead an advanced feature built into their model's architecture. Here is the precise quote:
>regional prompts is just an inference hack, our model already natively supports character placing
This confirms that regional prompter is nothing more than a cope inference hack.

>>8715579
>he says while being in the 22nd edition of a general
jej

Post ai bitches sucking oversized cocks instead of yapping COME ON

>oversized cocks
/ydg/ is that way

>>8715584
Then normally sized or tiny just post PORNNNNN

>>8715580
and my dick natively supports placing it in your ass

>porn
/israel/ is that way

>>8715499
Wait, this is actually crazy good?? How have I not heard of NovelAI before? Checking it out now and the results are INSANE. Looks like you just won over a new subscriber

anyone remember how to switch ez scripts from lanczos to nn or something? I lost my install.

>>8715582
Wow, looking through my gens it's actually been a while since I did one of those

>>8715579
This thread was made to get away from the console/subject war shitters and focus on genning in the first place but they can't stick to themselves

>>8715641
Thread splitting is a market for lemons.

>can prompt from anywhere
>lightning speed
>superior character recognition
>better styles
>16 channels
>proper segmentation without regional coper
>unlimited generations
I am seeing no reason to cancel my sub anytime soon.

>>8715641
>subject war
what is this?

deja vu

>can prompt from anywhere
>lightning speed
>superior character recognition
>better styles
>16 channels
>proper segmentation without regional coper
>unlimited generations
I am seeing no reason to cancel my sub anytime soon.

Most of us moved on to the undercover /e/ thread thoughbeit

Hey mukuruz, how long until NAI Video gen???

>applies jitter filter to AOMslop
Here's your video gen bwo

We went from 2024 localcope calling nai v3 blurry
to 2025 saascope calling vidgen jittery

>>8715687
Seeing people seriously praise awful garbage that current videoslop is made me really empathize with artfags who were seething over seeing people drooling over ugly broken aomslop. Guess back then I was more passionate about the tech and wasn't seeing the faults as much as now, but man the videos people are posting just really fucking suck.

>>8715669
I was only in the /e/ thread because wanschizo was spurging out the hardest there.
now that he seems to be taken care of for the moment it's just the same old boring slop again with no real entertainment value.

>>8715696
On one hand this general did aom for two years
On the other after two years of aom people should know better

>>8715697
not /edg/, i'm talking about the secret thread. We obviously baked it with a fake op

Too bad local is so shit he needs to spy on the superior SAAS general.

>>8715699
I just wonder if they actually think this shit looks good or just like "ah yeah it's new amazing tech, I mean it's better than it used to be right???". Second attitude would make sense I guess but plenty of anons actually unironically seem to like those

>>8715709
anons say this and then go back to watching the newest queen bee release

vidgen is sick, actually, it just can't do anime at all

>>8715709
How quickly you became those you hate.

>>8714698
is she real

I ain't your sister, faggot

>>8715582
>>8715745
i need to do more non portrait resolutions

I shouldn't even need to prompt, this technology sucks.
The algorithm should just figure out my tastes and feed me whatever is right up my alley on a daily basis

god dyamn

>>8715756
Ok but somebody should finally make a service like Criticker but for anime tits

>>8715754
This but landscape specifically. I'm not on my ultrawide so I can't make good use of it but man it's the shit.

jesus christ bitch got stretched into the 4th dimension

>>8715764
I need to make more landscapes but upright
Trying to do non melted on side sex is an exercise in patience at best

>>8715754
I just leave on the random resolution thing in reforge it's been the biggest game changer since wildcards.

>>8715774
Where's that again?

>reforge is unarchived again
My favorite schizophrenic maintainer :3
>>8715774
oh yeah right inbuilt extensions i removed them
i hope you can set a bunch of them up so it only chooses from a pool
sounds real based if so, there's so much variance that gets fucked by prompting only portraits

>>8715775
Literally near the bottom of the list for me not sure if you have to turn it on or not in settings. Was in one of the later updates before his initial hiatus and

>you can
>try it
>first thing i get is cowgirl but in landscape and it looks ten times more kino than portrait
It only just began...

New thread when?

>>8715774
The problem with this shit is that some positions are tied to resolution so it's a recipe for disaster.

>doesn't seem to work with prompt sr
It's over...

>>8715774
I should figure out random/alternating routing on comfy to switch between landscape and portrait with appropriate (random) controlnet input but I feel like the workflow is going to become mega autism

>the other random resolution thing is a script so you can't use it with prompt sr either
SO SO OVER IT'S NEVER BEEN SO OVER

I just need more cunny

>>8715814
I can't post any of my cunny gens

>>8715826
It's over
never recovering from this

Sorry to hear about your 1boy

>/hdg/- 168/ 70
>/hgg/- 311/ 33
lol

I'm fine without the bibici tifa, thanks

what caused this? >>>/v/720069635

>>8715857
don't you have a better thing to do on sunday night

>>8715857
hank hill ass

what caused this?

>>8715526
Stupid thing is fictional characters have no agency. They can't consent or not consent. They aren't real.

>try to gpt a fix for it
>just doesn't work
oh well it was a nice thought

don't care

(he cares)

>>8715865
no, i'm out of ideas what to slop

kek
>>8715857
>>8715845
>>8715530
>>8715520
>>8715510
>>8715456
>>8714848
>>8714702
>>8714701

it was the cause

test

what is lil sis testing

>>8715814
https://files.catbox.moe/fp8ay1.png
https://files.catbox.moe/2u02eg.png
https://files.catbox.moe/h3mkwk.png
https://files.catbox.moe/78tq2r.png

I am disgust

tasteless normiecunny

touristcunny

>>8715745
>>8715754
BOX PLEASE?!

no :^)

>>8715942
get on your knees, open your mouth and beg :3

>>8715890
I'm challenging you to prompt slop your mom. No, seriously.

hm I didn't expect the style to be transferable with 100% of it being 1girl, samegirl
good setup

don't care

>1:20h of caring
cute :3

>>8715945
i'm begging uwu...

>>8715956
well ok :3 https://files.catbox.moe/tbsvhy.png

>>8715814
wolf no cunny da https://files.catbox.moe/a7tlj6.png

amerikanu wolf https://files.catbox.moe/uixc89.png

>>8715958
thank you... u.u

>>8715963
Absolutely would smash

>>8715963
she's been to them diddy parties

>>8715972
>Oh hell no, not again

>>8715951
Could you gen the other dokis in the same style, please?

>smutting with sexbot
>suddenly starts to write on chinese mid sentence
B-BROS?!

>>8715963
excessive pubic hair chad

>>8716001
deepseekGODs

>>8716001
Don't prefill you dumb faget

NEW THREAD WHENDORU?????

what causes this

>>8715960
>No (You)s.
>>8715963
>3 (You)s.
Told you americans are disgusting!

it's the samefag

what killed the /hgg/

The /hdg/ fags that dragged their feet over around #10. Been freefalling since. /hgg2/ soon.

>>8716171
What if they're simply primitive men and like the Venus of Milo aesthetic?

>>8716288
primitive men would follow social conventions without question, same as they did back then

>>8716290
Too early for me to get what kind of argument you're trynna have but have a fatty

I have to deliver myself out of the womb

baking

what are you baking step bro

>>8716306
3K with olde adamW
300-500 with all that newfangled tech, came/wrex fp8 gradient accumulation

i'm gonna came on you

i cum on your grave

>>8716310
A-anon, I…

>>8715972
Liru 3 waiting room. Man never gonna deliver and we'll never get an animation as good

>>8716324
I already didn't like 2 desu

I will probably end silksong today unless the game throw at me 5 new areas like the last time

>>8716354
Is the bug /d/ or /trash/?

>>8716356
/d/ I guess but I don't see how "people" can like the design in a erotic way, it's way too goofy

silksong is trash for zoomertards

fpbp

>>8716358
it's not too far from Raita

bump

>>8716361
zoomertard website doe?

I need pancho to give me auto-anti-aliasing for my gens

>>8716398
bro your 1x_ArtClarity_strong?

>>8716408
What a terribly awkward position

>>8716427
AI isn't flexible so that shit comes out stiff a lot it's a shame.

>>8716425
NTA, does this actually work?

>>8716433
yes it does, quite well imo
>https://files.catbox.moe/pagk0w.png (before)
>https://files.catbox.moe/temx5q.png (after)

>https://files.catbox.moe/o13mur.png (before)
>https://files.catbox.moe/572fua.png (after)

>>8716437
guess it wont fix botched clits but it'll make eyes a bit better at least, thanks

>>8716438
would it not be easier to just inpaint

>>8716427
she's being raped

>>8716440
not easy on my hands and honestly not worth it if I'm not getting paid for it, but I probably should figure out a batch detailer workflow sometime since it'd all use the same prompt anyways

>>8716443
>not easy on my hands and honestly not worth it if I'm not getting paid for it
bro... lol

>>8716437
am i blind or did that just like... darken it a bit?

>>8716455
yes, it slightly affect the colours too

>>8716455
you're blind unfortunately

>>8716455
the change is so minor i dont know why even bother
guess thats local for you lol

>>8716455
i put them side by side and it barely differs at all

Noticing a ton of high-quality NovelAI gens recently. With all the new activity, I'm surprised they haven't raised the sub cost. It's really just that good.

>>8716498
ugh, I know you are shitposting but I like that erofu, it's cute

Is sex on NAI a secret sub? I've ever only seen /e/ from it...

>>8716498
It really is just so good, it's hard to resist! I keep resubbing every month as they keep adding more amazing features!

Nai isn't that ugly btw, it's falseflagging localjeets

>>8716437
I don't like any tools that make image sharper like that. The lines that gain "extra definition" always look fake as fuck, more AI-generated than AI sloppa itself.

>>8716509
interesting take, agree at some degree, it doesn't look that good with some styles but with very minimalist ones I do think it looks great

>>8716503
you can just type whatever for tags and nai gives you an amazing image
its THAT good

what happened to illustrious? What is the last public version

>>8716515
Because it's the world's first vibe-based model rather than a tag-based one. It understands what you want even if your prompt is outside the bounds of booru tagging. It's one of the most significant steps towards AGI we've seen yet.

>>8716519
>what happened to illustrious?
everyone moved from it

>What is the last public version
2.0 which is honestly not great

>>8716519
they tried to scam everyone with 3.5

wai > nai

wakaranai

^ true

Where am I supposed to post tentacle gens?
>/ddg/ is for 1girl, futa, huge penis
>the non-futa /d/ thread is dead and for fetishes beyond my comprehension
>/trash/ is for diaperfur
>/b/ is for 2.99d pedoshit

>>8716567
Just catbox it here, retard. Noodleanon and I are using 4chanx anyway and there's no one else here

I'm here too tho

>>8716567
Yeah, I am disappointed in how the non-futa /d/ thread turned out. I was hoping it would attract all the people who couldn't post tentacles, breast expansion, stomach bulge and other less extreme fetishes not allowed elsewhere for whatever reason.

Instead it's been mostly amputees thorought.

>>8716567
Yeah that's super stupid, as the other anon said, if you really want to share them use catbox

>anon forgot about me
bruh...

boomeranon and rin/emifag are the same anon tho

>>8716573
>>8716578
thanks noodleanon, but you don't have to pretend. I don't feel lonely

b-broski?! don't skip on those meds please, you were doing fine so far

>>8716584
I stopped filling my prescription so I could keep subbing to nai though

>>8716570
think we said six people yesterday?
the funniest part is even with so few, loli still gets reported and deleted within minutes

>>8716308
Rex/Came is superior due to the time, otherwise they're both the same. No, your activation tags are not needed unless you're baking on the artist tag. Even then, OP proves that still isn't necessary.

>>8716628
trigger tags are weird. some styles are totally fine without, while others just refuse to stick and simply adding an activation tag lets me cut the training at 20% of the steps I would otherwise have to do

>>8716632
Eh, black box.
>lets me cut the training at 20% of the steps I would otherwise have to do
I always bake 30 epochs these days and manually look at each epoch or save every 20 steps and make a bunch of grids before I pick. I used to hate testing but these days it guarantees that whatever errors are in the lora are a result of AI, not me.

>>8716632
>>8716634
Still just perpetually not figuring out how to do charas at all, like it just feels like it's never baking no matter what have to boomer prompt features to get the character to show up and it's driving me up the wall at this point

>>8716637
Post your dataset

>>8716637
t a g d r o p o u t
a
g

r
o
p
o
u
t

>>8716639
Turn it off or on?

>>8716639
Gimme the numbers senpai

>>8716640
>>8716641
use your head
if you don't use it, you only teach the model to generate the character along with 100500 tags that go along with the character name
you should drop every tag except artist/style tag and a character tag
aggressive dropout + slow bake will give you best results
>how to do it in easy scripts/kohya
dunno, you can try using ||| or keep n tags

>>8716644
Could have just said shuffle captions and walked away, 'nonnie.

>>8716645
that wouldn't solve the poor character tag binding issue retard

>>8716637
I thought boomer prompting meant natural language. Are you talking about having to prompt out all her individual features every time?

Normally that's be solved by pruning tags, but I suppose >>8716639 might work better. The model learns faster when it knows what it's looking at, and then just occasionally remove them to show they're not necessary. Though both make the lora less versatile, if say her hairstyle or outfit gets baked it becomes difficult to change.

>>8716644
It does make the lora way less versatile, in case you wanted to change her hair color, hairstyle, clothing, make her cosplay, etc.

>>8716646
>you should drop every tag except artist/style tag and a character tag
Are you implying shuffle captions + keep tokens = 1 isn't this? Why are you mad?

>>8716644
Meant that caption dropout numbers in optional args. I'm aware it's just never been that big of an issue before on any model til illustrious and it just doesn't play. I've removed all the erroneous tags for features as well on the dataset and scrubbed it down once and it's the same result.

>>8716648
Dropping tags and shuffling them is different

>>8716649
Why haven't you posted your dataset so we can all test these baseless theories and be done? All anyone ever wants to do here is bicker.

>>8716648
>Are you implying shuffle captions + keep tokens = 1 isn't this?
no it isn't, shuffling randomizes the order of all existing tags when you want to drop them instead
>>8716647
>It does make the lora way less versatile
unless you drop all other tags with 100% probability, it doesn't really make it less versatile, that's how noob was trained in the first place
>>8716649
>caption dropout
iirc in kohya it's a different thing at it's related to cfg
don't use it for loras, it drops the captions entirely which breaks the model output distribution

which one arguing is noodleschizo and which one is chadmixtard?

>>8716654
Cause I don't think it's the issue I've run chara datasets this same way on all the previous models and it's been fine until illustrious.

>dropping tags
at no point has this ever been a good idea for loras/small scale FFT.
using any form of dropout on small scale will just fuck up your training. The entire point of it is to try and extend the time you can train on a dataset before it fries out but that just isn't really an issue unless you're doing a full finetune with a massive dataset.

>>8716653
Your artists/lora, sirs?

>>8716661
>https://arxiv.org/pdf/2409.19946
>Finally, we adopted Multi Level Dropout by dividing the dropout into 4 stages, ranging from minimal, critical tokens to full tags. This allows the model to adapt to varying levels of caption detail. By 30% chance, we utilize max(30% * total tokens, 10) tags, 20% chance, max (40% * total tokens, 15), 10% chance, min(6, total tokens), 4% chance, min(total tokens, 4) tokens.
this was never intended to "extend the time" of a bake, it was meant to solve the poor prompting issue

>>8716656
>it doesn't really make it less versatile, that's how noob was trained in the first place
And you can try putting a random gachaslut in something different than she wears normally, see how it goes. Unless she has dozens of pics in different outfits you're gonna get leaks all over from her default. >>8716610

Though it's definitely better than pruning those tags entirely.

>>8716662
>(goki buri,akai sashimi:0.7), hamunezuko \(nezukonezu32\), (nanjou asuka:0.8)
now go and make more gens of Sakura's butthole being used for the nedful saar

>>8716667
>Unless she has dozens of pics in different outfits you're gonna get leaks all over from her default
this is what will happen if your tag dropout is way too aggressive and you didn't bake it slowly
either way you can lower the probability of dropping the tags that leak

>>8716666
>this was never intended to "extend the time" of a bake
It is what it does, though.
>it was meant to solve the poor prompting issue
It can only really do this properly if you use a very specific logic/scheme to the dropout, which you do not have the option or ability to do without writing your own training code.
Using a blind % based dropout across the full caption pool is just going to fuck things up.

>>8716672
More likely it's gradual. Even the first dropped caption "should" lean towards it in theory, just the more you do it the worse it gets. There's always some baseline level too, from outfit pieces that do not have accurate tags to describe them.

Damn you guys, I didn't need any more stuff to experiment with.

>>8716619
box please?

>>8716673
>It is what it does, though.
this is what any dataset augmentation does and it's a side effect mostly
>It can only really do this properly if you use a very specific logic/scheme to the dropout
yes, which is why easy scripts and kohya scripts are fucking ass, the most advanced feature in there is keep_tokens_separator https://github.com/kohya-ss/sd-scripts/blob/main/docs/config_README-en.md and it's not simple or easy to hook your own dataset processing logic
>>8716677
>from outfit pieces that do not have accurate tags to describe them
i thought the most common solution to that is creating outfit tags yourself

>>8716681
>i thought the most common solution to that is creating outfit tags yourself
It is, if you have enough pics with and without.

>>8716659
I don't bake at all silly willy, I'll rather larp as an artist than a MLchud

>>8716680
https://files.catbox.moe/6mijc6.png
I painted the breast veins manually before upscaling.

>>8716685
why not pull an oekakischizo and do both

>>8716684
i don't think you need a whole lot of them, if the effect is too weak it probably means that your dataset isn't balanced

>>8716690
Because then you'd suck at both, just like oekakischizo.

>>8716671
>(goki buri,akai sashimi:0.7)
Based. Use them for one of my mixes with haguhagu.

>>8716690
is he in the thread with us right now?

>>8716690
I ((tried)) to bake some loras before, all of them fucking dogwater, sure I could always ask for help here but there is still many things to learn and test about that to the point it feels like a chore (to me)
Not to mention the (not really that long if I am being honest) bake times and god knows how many retries, I'll rather doodle around to make my gens more interesting

...I will take that as a yes

>>8716687
You didn't like the way they looked with the tag?

i just fucking wish someone puts up a systemd-like dataset preprocessor, it's a fucking genius idea
you write a small independent logic pieces which you then can reuse for different datasets and it it's all sorted out through dependencies

>>8716692
he seems to be doing okay though.
>>8716699
in the walls, etc.

>>8716704
They are pretty inconsistent and tend to come out too defined, I prefer drawing them on my own in most cases.

Seems like NovelAI's superiority has caused localkeks to go crazy

https://files.catbox.moe/3gd83h.mp4
I should probably take out bouncing breasts and lower the motion slider when it comes to absolute flatties lmao
The tech's fun tho

>>8716687
thank you

>>8716733
Even really small girls will have some bounce to them. The only way you're not going to see any kind of jiggle is if they literally have nothing.

do you need to prepend char: or character: on noob?

>>8716773
nope, just the character tag is enough

>>8716774
huh, weird, some artists work better with artist:

>>8716776
correct, artist, because it was trained that way but with characters is not necessary

>>8716778
alright, thank you

is this the thread?

>>8716733
reminds me of this one

>>8716531
nice style, I really like the shading and the hair

>>8716437
>>8716509
it almost makes it look like vector art

>>8716567
they go in the non-futa thread. that's what it's for. just because the thread isn't ALREADY full of your preferred content doesn't mean you can't post them there.

>>8717017
No it's clearly the right place, it's just that thread is very slow. Chicken and egg though, it will not speed up unless people learn to post there.

I dump my stomach bulge there when it's over the line, and cumflation. But mostly just when I find a new artist mix, it's been a while.

>>8715362
box?

Chat, I thought I was at the end of it but the game really did throw at me 8 new areas and a bunch of new side quests, fucking hell

>>8717155
what

>>8717160
silksong

>>8717160
use nai

>>8717161
you need to play more games if you didn't expect that

>>8716996
Style came out a little strong on base so I had to slop it on 102d

>>8717155
if you 110% the game you unlock the secret hornet sex scene

>>8717195
hornet_(rance) > hornet (adhd zoomer game)

>>8717196
I get that hating popular stuff is what /v/ does best, but is it really more "adhd zoomer" than the first game from 8+ years ago? And also, aren't you in the wrong thread.

just baked a lora using maximum possible effective batch size with grad acc for shits and giggles (total 15 optimization steps) and it turned out surprisingly well
now gonna train a baseline to compare against

>>8717215
i just started baking for the first time in like 6 months

too many video games

>>8717215
the one trained with batch size 1 and 1320 optimization steps is melting, lmao
now this is time to do bake a lora using just 1 step

>>8717257
AIIIIIEEE ANON NO YOU'LL INVERT THE TIMELINE COVELANCY

>>8717258
i unironically think that maybe the optimal batch size for stable diffusion is from 250 to 1000
if there's 1000 noise levels, then batch size 250 will at least consider 1/4 of all of all timesteps at all times

>>8717275
the original SD models all used extremely high batch sized but i'm not sure that simulating it with GA will actually help

>>8717277
>but i'm not sure that simulating it with GA will actually help
that should work, that's how they train using many gpus in the first place

lora baking with high batch size kills style. just averages out all the subtle details that might not be perfectly represented in 100% of the dataset

>>8717155
hm. i should gen more hornet futa having gay anal sex with anon.

hmmm a novelai sub to own local sissies... why didnt i think of that

your age?

>>8717298
26
I occasionally gen kuro and asuka

>>8715163
sorry was busy with work, its relies a lot on comfyui custom nodes so i don't think it'll translate 1:1

https://files.catbox.moe/m7pn2k.png

>>8717313
box before text?

>>8717257
>1 step
doesn't look like it learned much of anything about styles lmao
>>8717280
it kinda do be like that it seems, although i didn't aim to learn the style here
anyway, regarding the effective batch sizes:
lise_1 - 88 - learned features the best of all 3, almost didn't learn any style
lise_2 - 1 - burnt, leaks pictures from the dataset, all melty
lise_3 - 1320 - learned the features a little but didn't learn styles at all

So can WAI do flat anime coloring or is it doomed to always gen shiny 3dish

>>8717321
that's a really old gen that I never post before so I lost the very base gen, I also used as contronet reference that gohan meme
>https://files.catbox.moe/kxw4yr.png

>>8717312
careful bwo. her age is showing.

>>8717331
boxo please good man

>>8717330
I will try r3mix cause I just need to get off of custom I need something new.

>>8717351
I unironically just merge r3 with 102d flavors

>>8717356
Can you just merge vpreds like the quickie model merge and it doesn't just make it flaming shit?

>>8717360
yeah if they're both vpred, simple weighted sum merging with whatever works for me

>>8717351
r3mix is nice but way, way too bright even on dark settings so be aware of that

>>8717363
I liked the brightness plant milk oat is one of my favorites and I've been trying to chase that look with vpred or dodge the shininess with that model and I can't do either.

>>8717337
Saul Goodman: https://files.catbox.moe/n08imk.png

>>8717379
Is it literally just Zankuro giving the solid flat coloring style or is 291H good at that kind of thing?

Does anyone have comfyui workflows they'd recommend for Wan2.2?

>>8717429
the schizo workflow that comes with easywan is fine, use the python 3.10 installer if the basic doesn't work
make your own prompts down the line, the presets are esl shit that might not do exactly what you want

>>8717390
it's 291h

anyone have a decent asanagi lora?

yes

it's up on da' cord that came in a dream.

up in your ass

Why spam?

>>8715814
I have literally thousands of saved cunny I could post, let me know what kind you want. Most are slop though.

Same, but I just keep it to myself.

Seeing my gens saved by other anons is a nice feel.

>>8717759
i've saved over 5k gens from anons since the start

>>8717762
Kek I have around 7k and that's without 1.5 gens as those aren't sorted in any way.

>>8717764
i try to be a bit picky

I have saved exactly 0 gens from other anons, they are that bad.

>>8717765
Yeah lately I am too.
A lot of the time I saved them to maybe go back and check some artist tag even if the gen isn't amazing, but I never actually do.

>>8717770
i started keeping a notepad of artists and now there are hundreds, it's nice

I probably should be a little more picky with what I save.

>>8717777
the numbers agree

>>8717692
So it was all a dream?

>>8717777
I'm doing a pruning again and finding old shit I really like and wonder why I swapped to vpred, then try out the model again and gradually remember

>>8717852
This nigga needs a liver transplant ASAP

>>8717777
My lora folder is smaller than that. Do you really need to save all the slop that gets posted here?

shill blogpost warning
after 3 years of a1111/forge i finally tried out comfyui lads
it's not bad, but it took a day to figure out how to do img2img, a highresfix equivalent, BREAKs, rescaleCFG (and other placebos), controlnet, and fix it all up into toggleable groups so i can do pretty much everything in one workflow
one annoying thing seems to be regional prompting, i saw how to do it but it's very autistic. ironic thing is the reason i tried switching was because regional prompter was being shit in my reforge installation...

it's safe to say that nai lost HARD

we already know

>>8717860
Infinite image browsing is the only reason I will never leave auto1111. Until cumrag implements he can get fucked.

>>8717898
you can use it as standalone

>>8717901
>standalone
mmmnyo?

>>8717902
that's literally how I use it, without launching reforge, separate install

>>8717898
what's that?

ip relay to and from here from haagen daz come bolster hq defensce now

>>8717904
https://github.com/zanllp/sd-webui-infinite-image-browsing

new tagger, anyone try it yet?
https://huggingface.co/pixai-labs/pixai-tagger-v0.9

>>8717941
>Finetuned from: SmilingWolf/wd-eva02-large-tagger-v3 (encoder frozen)
>new

>>8717947
>finetuned from the best current tagger
how is this bad

>>8717857
This is my own slop, anons didn't post anywhere close to 29k of images in these threads.

>>8717926
neat but i'd never use it

>>8718008
>I'd never try to find old pics via their metadata/tags
Why?

>>8718011
i can just scroll back to when i genned it in explorer, i have a good memory so this is not a problem

>>8717941
i might be a retard but where the fuck do i point it to a folder to actually use it? they say to use handler.py to use it locally, cool

>>8718012
>i remember the exact dates of every pic I've ever genned since my memory is good
Why aren't you doing anything productive with your super power?

>>8718017
i'm designing games in my free time and if the rest of my family gets off their lazy asses we'll actually release our first one

>>8718018
>you call this level design, mom? bitch have you even finished symphony of the night?!!

>>8718014
you can host the demo locally, why do you need handler.py?

>>8718026
idk man i'm just trying to test it out but i am not a programmer

roblox dude cut my hands off to hurt me

Was it ever figured out what artist styles were actually mixed in to make nyalia?

>>8718085
Shitty ones.

>>8718085
Yes.

>>8718089
>>8718091
And so what was it?

I always used the -ai_generated tag but I recently saw it's gotten much better. I've been using novelai to gen my favorite hentai images with new characters and better art styles and it's pretty stunning. Way better than when I briefly used local stable diffusion like a year or two ago. Might get addicted to this.

novelai is really really good

geeg. nobody tell them that today is not opposite day

oh I see that novelai is a shitposting point here
Which ones are the contenders then? I have to assume the best models are all propietary so they're all online services.

>>8718115
ask in the nai thread
>>8714686

>>8718117
Do any local/open source models compete with online stuff? I would prefer to do local seeing as how I have a 5090 and don't wanna drop cash on online credits constantly.

Very normal and genuine posts today I see

novelai is genuinely amazing

Just wanted to let you guys know that the kike bot farm is spamming bot reports on my posts as "off-topic" meanwhile their incessant shill spam is supposedly on-topic. Yeah.
Hey nigger, maybe you should look into the bot farm that has been terrorizing this board ever since the NAI leak? No? It's so fucking obvious you'd know if you cared about your useless job even a little bit.

if youre anti-novelai youre pro tifa blacked spam
only thing that makes sense

>>8718139
Based whore enjoyer

chat, is this real? >>>/v/720318842

in the end, ruining this general because i got mad about one nai shitposter was worth it

d e e p s k i n
e
e
p

s
k
i
n

my condoms are sheepskin

>train artist lora
>doesn't look like the artist that much but looks nice
what the FUCK

>>8718180
That's how most decent inbuilt styles work too.

actually i think i might know why

>>8718180
>>8718185
this property of sdxl is what makes it easy to distinguish ai gens

>>8718191
i mean, doesn't matter the model, they all have their own quirks, like flux has that greasy shine

>>8718192
>like flux has that greasy shine
it's all rlhf slop, they were aligned to do that
bytedance managed to ungrease it and bake in fotm ai slop styles instead https://huggingface.co/bytedance-research/USO
style mixing is kind of impressive

>>8718201
it doesn't fix the excessive depth of field blur so it's still shit

>>8718205
i didn't mean to say it's not shit, it's not even anime
all fotm ai slop styles look like that, but hey at least they've managed to combine several slop styles in one model

>>8718209
impressive (?)

>>8718210
they are consistent, don't fluctuate and don't spontaneously mix together depending on the prompt or the current moon phase unlike anything sdxl
imagine if you had wai, autism and animagine all in one model, they all have different slop profile

>>8717360
>>8717362
Some anon talking about merging a while back made me look into it and on a whim I merged 291(the super sekret mix) with silly greek letter 1.5 and haven't looked back since. I did 2 merges with one as the dominant model in either one. One was mid. The other was goat. No longer remember what the weights were.

>>8718245
Fuck it you know what? I'll waste a few hours doing this. I'm watching some movies to get inspiration for my textgen story anyway.

>>8718245
291h? I din't get the sekret mix I kept missing it when it was here.

The sekret mix is 24r2+28m? because that one is not that great either lel

the truth is snake oil will never cure your illness

my snake oil is 25$ a month~~

honestly a fair price for the best model ever made

ah sorry, wrong thread for shitposting

>>8718282
Nah. 291h is kinda close but the anon added some lora to 291 for stabilization or whatever iirc. 291 without something to stabilize is pretty schizo but the artist fidelity is as close as you are going to get slopping.

>>8718299
Is it really that much better? I never saw much improvement over 102d. Do you use quality tags?

>>8718301
>Is it really that much better?
291h? If so, I flipped between it and 102d. There were sorting artist mixes of mine that worked better on one or the other.
>Do you use quality tags?
Just masterpiece and newest. I meme'd myself into believing I get less minor errors when I use masterpiece than when I don't. Otherwise, I've never seen a noticeable influence on artist style to NOT use it.

>>8718307
>There were CERTAIN
is what I meant, SIRS.

>>8718307
>There were sorting artist mixes of mine that worked better on one or the other.
Yeah this was my experience too. Guess I'll just stick to it.

>50/50/50 mix of 3 models
>Can't tell if it's better or worse

>>8718307
>>8718309
I've set up a wildcard for just different quality prompts and just let it roll cause I've been schizo pilled that badly at this point otherwise yeah just masterpiece, best quality is the default.

>>8718314
>50/50/50 mix of 3 models
>Can't tell if it's better or worse
Do dark-low lighting settings and grayscale test gens

>>8718314
>50/50/50 mix of 3 models
This us a very bad way of doing things. Take the time to think about what you're doing anon. You have to have a purpose in mind when doing it. Find a prompt that showcases what you like about Model A, then Model B then work your way up methodically swapping A and B position every 0.05. Better to do it with to models then at the third and repeat once you've created a good merge.

how do I gen the 1girl such that she looks like she weighs 60 pounds

how many kg are those

>>8718392
Do you mean like she's just tiny or anorexic?

>>8718397
ideally both
make it 25kg in fact

imagine if the shitposter was accepted as a janny

>>8718481
seems like the new one killed Tifa spam at least

>>8718483
wdym? just checked /hdg/ and it's still there

>>8718484
oh yeah, only the later ones
which I guess means he's using multiple IPs

>>8718139
Almost good. Extremely broad shoulders though, makes her manly, can't get erect to that...

>>8718464
(skinny:1.5)

I haven't genned in a long while, what's the meta nowadays? I tried 102dcustom and it didn't seem to work for me. Have some blacked 2hus.