#🍥|anime
1 messages · Page 71 of 1
1girl, ugly, ugly girl 💀
ugly girls aren't real
also wtf happened to the first image
Also, have ugly girl
that's not even a human lmao
XD
Ugly maybe? Its also kinda of a eldritch monstrosity tho so that adds hotness points
frog girl
💀
howd you gen me???
Ish so pretty 
Exactly, you is 
🐸
evenin'
"|" is used in a couple of functions, but that isn't a valid one.
[word|otherword] alternates between the words ever step in the generation, this one is built in. The prompt blending script, which is a separately downloadable script, uses {word|otherword} which averages the two together. Dynamic Prompts is an extension which also uses {word|otherword} to randomly choose from the two options for each batch in the generation.
evenin'
gettin' weird in here
creating comic panels is not particularly enjoyable
Started boarding out a comic to see what a workflow would be like.
Lots of elements, still pretty tough. 🙂
discord sorta cropped those images funny 😛
luv the horror style

epik
hmmm rtx 4060 ti with 16gb vram arriving in July at recommended price of 500 usd
maybe I should sell my 3070 ti before then while I can still get 400-500 dollars for it. Settle with being on a shitty 1060 for awhile and then buy the 4060 ti
which should be at least as powerful as 3070 ti
but double the vram
Hello, the anime pictures you put on disk are very very nice and cute.
4060 ti 8GB reviews are already out and it's basically exactly the same as the 3060 ti in gaming. The 16GB variant is getting the same GPU and will be limited by the same 128 bit memory bus so the performance will be the same except for scenarios where 8GB wasn't enough. I haven't seen comparisons in ML tasks so it might be worth looking for that to see if it would at least be better for AI than it is for gaming.
wait....a new generation and its the exact same for gaming?
the 4000s series is quite the disappointment
Yep. It's embarrassing.
4090 is impressive if the price wasn't so absurd, other than that it's all been one joke after another.
It does a little better in games that use DLSS 3.0 and/or RTX, but that's still not that many games.
get a 4070 instead
why? that's only 12gb and a 2nd hand 3090 is like 800 euro while 4070 is 700-800
I'd rather buy a 3090 then
what do you need 24gb for bruh
yeah I thought it was a unique twist on anime that noone is exploring
thank you very much
🤍
nope
no
192*
not 128
Everything I've seen says 128. I just double checked to make sure I wasn't misremembering.
unstoppable

XD
my favorite hairstyle

to luv ru darkness
u keep sending the same over and over
Yes, because it good 
Though I just decided I wanna do mini milk now
Its not even kink related tho, it just looks cool :o
sure
sure
mommy milker kink
Is it? XD
For cow girls or for big boobs? xD
both
boffum
She has small boobs tho 
sorry wont happen again
like yo~
what
what
🔫
I just showed my 7 year old nephew some of the images people posted here
too many women
he was like "where are the men"
so I told him people like girls more
though there were a few cool images of guys
waifu diffusion?
i am currently training a: completely retrained, 500pic + dataset(300 pics on 1.0 version), new settings, WAY more flexible version of my yamero yandere lora
nice fresh installed sd cos was being whacky, got torch 2 and 5.4 rocm without needing to build
omg model name for the second? I’m dying to generate male characters
the second one oh my god I’m dying to know the model and prompting 😭😭
Its absolutely gorgeous
wanna test it out?
you first
i did the thing... i put the new 65b model on company A100 XD
Hey!
i'm sorry but this is Nijijourney.. so not realy a model for stable diffusion
Okay I'm gonna get nijijourney
i can realy recommend it it's awesome
I can see from the server!!
Do you recommend any stable diffusion models tho?
I want a mixture of both~
Anything you recommend for male prompts? Too many waifus.. I really wanna generate vocaloid men lolol
midjourney
Well, if you like the look of Niji. Here's one pretty nice one https://civitai.com/models/47067/pikas-new-generation
i'm not realy generating males
sorry
Yes I really like niji!! I've seen that model but I'm still new to ai art so I can't really understand how to get masculine men like niji does
I'll just go for niji as of rn until future models drop
niji is just insane for those
WOOOAHH
is niji like stable or is it like a website/bot ?
What were the prompts?
it's a bot on discord
It's a paid service.
It looks absolutely amazing I just bought it rn
uh i have no idea i would need to take a look
just enter male and i think it will give you what you want 😄
It says the dms will be public ? Fr?
Okay!
just make your own server invite the bot
Omg I probably need to do that
I'm gonna get sick of generating the same anime boy eventually
what i also love to do is niji to sd
llooks like this when done
niji
after i'm done to sd
with edits
😄
OOOOO
magic 😄
photoshop + inpaint
Ooh okok
@frozen olive srry one last question, for which style did you choose for the vampires? I see people using "style expressive" or "niji 5" and I'm still confused..!!
ok ping me i'm excited to see
will do!!
nah it's for emotions
So it generated like 4 images, how do you view one by itself?
They're all so cool I'm so excited
started yesterday and damn, love stable
Oh so I just crop it?
noo just click U1 - U4
hamburger goddess
HAMBURGER GODDESS. WORSHIIP THE BURGER
damn, guanaco giving waay to good prompts 😮
there's more buns than just the hambuger here...
"Hey guys we finally arrived on earth, will take another pic in a few minutes when this strange forest is gone" -- Alien
I wonder how long before stablestudio surpasses auto
I think when it allows for extensions, embeddings, lora, lycoris, control net, hypernetworks etc, it will become much better unless Auto or Vlad step up their game
vladmandic keeps updating hard though
yeah too hard
he just adds shit
and often what he adds makes previous things conflict or not work and suddenly your entire approach to it is altered
or your prompts no longer work
i've had vladmandic break down on me yeah
when it comes to tech, he's very much on the game though
Yeah but I can just look at Vlads and pick things out that I like, find them and add them to auto
until comfy have a massive speed boost over auto1111 i will consider it
X/Y/Z plot will create 64 images on 1 32x2 grid. (Total steps to process: 1280)
time to test 64 models
X/Y/Z plot will create 150 images on 1 30x5 grid. (Total steps to process: 3000) there that's better. Using multiple styles now too, easiest way to do several prompts
different results from different models, yet weird how they retain a very similar style
same can't be said for the other prompt
this will be very informative on what models I want to merge, I gotta look to each strength and weakness
maybe add a few more styles (prompts)
This prompt really does showcase the different concepts they have
heyo
I can't even paste the grids
130mb each
what was that plugin that like... fixes adjectives
e.g. "red hat and a brown sweater"
would avoid the issue of both items being red
X/Y/Z plot will create 300 images on 1 30x10 grid. (Total steps to process: 6000)
uh cuttoff? or something?
evenin'
Anyone have an issue where controlnet still seems active even after u turn it off?
haven't seen that
making comics with this is really pretty tough
might be easier if I had a fast openpose workflow
man we need someone to go and clean up that strawberry jam
Sometimes SD's failures can be spectacular
Ok I'm pretty proud of this one
The WD tagger got a new MOAT model. https://huggingface.co/spaces/SmilingWolf/wd-v1-4-tags
just discovering ADetailer :O
No more needing to highres just to get a good face on wide shots
crazy how good it works with such a low res
ADetailer is pretty sweet
it's a lifesaver, I don't always want to upscale the image but sometimes you have to if you want the face to come out
how about upscaling the image, can ADetailer also handle it?
it can with mixed results, it upscales just the face and can create a weird box around the details behind the head
adetailer is literally automating the processes of using inpaint at full resolution.
lower denoising is the solution to that problem.
oh i see, does it fix bad hands too?
not really. there is a hand model for it to identify the hands, but you're still relying on trying to inpaint hands which is very hit or miss, especially if the hands are bad to begin with.
I guess manually editing the hands and do img2img is still better..
Yeah, there's no magic to adetailer. It just generates your image normally, uses a facial recogintion, hand recognition, or full body recognition model to find the specified thing in the image, and runs inpainting at full resolution on the rectangle that it identifies. What makes it cool is that it all happens automatically so you don't have to swap over to the img2img tab.
Not sure what you mean by that. Fixing an image that's close to good seems a lot more rational than just endlessly generating new images and hoping to get one without problems.
not having to switch tabs is magic enough for me 
ill fix an image if everything else is great but I kind of like endlessly generating
Isnt it a lora?
Or am i tripping
No, you're thinking of the add_detail lora. Adetailer is an extension. https://github.com/Bing-su/adetailer.git
hey nubby did you try the new webui 1.3.0?

Not yet. I was busy most of the day and didn't feel like dealing with the problems that new versions of WebUI always seem to have.
well, my pnginfo is still broken 
fair fair
where in settings? 
I only update the webui after checking in with other people
Idk saw the notes in bed before i could try it
Im still in bed kek
They just added settings to allow specifying a different sampler and/or prompt for the second pass (meaning the high res img2img pass) in hires fix
👀
this is new for sure
Are there any issues with the new version or are they rather minimal
Yeah you could do the initial gen with karras and the high res pass with DDIM or any other sampler.
no issues for me so far, except for that pnginfo still broken
for me at least
Ill have to update later then

goodnight nubby 
Nighty night!
without hires fix or adeailer, the speed increases
ok wtf happend here?
what the actual fuck
thie new update does feel a tad slower however
Time taken: 31.18s
Torch active/reserved: 3623/3788 MiB, Sys VRAM: 3852/8176 MiB (47.11%)
for justr a 512x512 which usually takes like 20ish sec
can't really tell in my case
saves time but for cost of quality in some cases
i hope not
😨
yup yup, in my case, the drop in quality wasn't worth the speed increase

Hi, nice anime pictures
its crazy too cuz ive used the internet almost my whole life and i fail at these i dont know how regular people are passing these
me too, i dont see shit on there
yea i wonder what will happen in 1-2 year from now as ai gets better at doing captchas
probably iced coffee 😄 and it was meant to be a guy! but yeah...
it's androgenous, hard to tell
I got this tho 😄
Time taken: 49m 45.60s
Torch active/reserved: 1511/7250 MiB, Sys VRAM: 7378/8176 MiB (90.24%)

you was rendering that for 50 minutes?
yeah... that's not androgenous... pretty sure that's a man... i think
spreading some joy
not worth it lol
who asked you tho? lol
your mom
don't listen to this. if you enjoy the pciture, totally worth it
mmmmhmm lol
dude. you're not 12 are you?
miku cops, now that's prompt worthy!

"your mom
Steps: 32, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3210827404, Size: 512x640, Model hash: 0a03c339c7, Model: divineelegancemix_V5, Denoising strength: 0.61, ENSD: 31337, Hires upscale: 1.75, Hires upscaler: Latent"
uhh... you had it at 2?


Steps: 30, Sampler: DPM++ SDE Karras, CFG scale: 6, Seed: 1943248902, Size: 768x768, Model hash: d413db35fb, Model: imPURErity, Denoising strength: 0.33, Tiled Diffusion: {"Method": "MultiDiffusion", "Tile tile width": 64, "Tile tile height": 64, "Til
maybe cos that?
idk never rlly used tile diff
trying it out
tile diff probably made you even be able to do it
768 x2 by hires fix?

(hides in shame, i kinda generate by default on that)
no shame in that, i never really got a gpu more than wha t i needed
hope you get better gpu soon 
i always got what was good enough to game without issue
as long as u are happy with it 😄
there's a reason now, for ai art 
ohoo nicee
more guys
im too gay for guyys
me tooo
oh right... i was busy having an LLM interpret aiska's dude
enjoy 😄
switch emu's stronk enough to run tears on deck?
what are you intending to upgrade to? 
@trim current
ye on yuzu, capping fps to 30
aiska will mostly generate guys
I pretty always do 😄
and i happily join, cuz that makes for more diversity
1 pic or a batch of 4?
this is my 1 pic time for 768 x2 hiresfix
well, the diffussion is pretty low, but the upscaler uses HUGE amounts of vram, so that's a 1 size 2 count
is it weird if I say that I never tried generating guys?
ahhh
35 steps low, 20 steps high
i only generate guys when people requested 
no, that's not weird.
some models dont like guys tho 😦
that's a lot more than mine
I'll give it a try and see what I get
and if anything, look to history, when depicting beauty: most artists went with women
so yeah
4 batch tiled diff
dont forget we have different gpus lol
We generating guys now
i only used 8.5 gbs
but i always got the ultimate flex xD
yup, but you got 24gb vram right?
yeah, got a 3090
I did a test on anime models for guys, feel free to check it <#🍥|anime message> 
LLM missed kinda.. euhh.. ALL THE POINTS!
i guess it's time to generate some men since Aiska is here
so close yet so far! 😄
I never refuse!
i had one of my hobos from attack on hobo with a LOT of chest hair
anyone interested?
N O P E

no
good.
guess i will be the one to say yes
nah... can't show it... double navel and... even i think it's a bit to... discomforting to look at

spoiler tag
wont help 😄

nah, i'm pretty sure mods wouldn't want me posting it either, i can dm if you want it
please don't.
^^
you are missing out tho

FOMO
xD
my model might have issues with balconies
damn I love this style!
renascence balcony surrounded by skyscrapers
Turned off my SD for today 😄
will try later!
@lyric zealot i can hear that reaction 😮
do still need to ping you for husbandos?
i updated yesterday but didn't activate yet, you were not kidding that's high quality 😮
yup super high quality for how fast it's generating
and not much vram usage for that

anyway, did you find any interesting use case for this
yeah, i ddi
tell me tell me
https://www.youtube.com/watch?v=1WNN4VWj2Pk just sharing neuro singing, dont mind me
27-05-23 Neuro-sama Debut | !merch !plush stream clip
Neuro-Sama channel: https://www.twitch.tv/vedal987
Original song: https://www.youtube.com/watch?v=ESx_hy1n7HA
oh I didnt know it has such a nice updated
where is it in a1111? i used vladmandic before it broke down again
Updated today but I didnt see a difference 😄
under User interface scroll all the way down
lol. no wonder i could find it. that's a really dumb spot to put it
true, but kinda understandable. They should mark all the new settings with a label or something
but yeah, why -> i use UniPC a lot, but UniPC cannot do highres pass, so it would default to DDIM for that, which isn't the greatest sampler by that point, so i switch to dpm2++ karras, which gets simply gets more detail in the same amount of steps
ahh, that is probably the reason why unipc is failing on me
it's certainly a strange bugger, but i like it (transparent cloth addict)
perv!
wet, see-through
perv

wait what?!
🤣
i get called perv for transparent cloth, not galactic size boobs... 😮
indeed
fair point 
🎉
mine doesn't either, it just likes girls by default lol
same
would reading hume first be better / make it easier to read kant's critique of pure reason / prolegomena to any future metaphysics?
someone ask chatgpt
nah nvm
David Hume was an influential philosopher whose empiricist views deeply influenced Kant. Hume's skeptical arguments challenged the possibility of certain knowledge and causality, highlighting the limitations of pure reason and emphasizing the role of experience in shaping our understanding of the world. Kant's philosophical project, in many ways, aimed to reconcile Hume's skepticism with the claims of rationalism.
By reading Hume first, you will gain insight into the philosophical problems that Kant sought to address. You will encounter discussions on topics such as the nature of causality, the limits of knowledge, and the role of experience. This familiarity will enable you to appreciate Kant's attempts to resolve these issues and develop his own systematic philosophy.
Additionally, Kant himself acknowledged Hume's influence on his thinking, famously stating that Hume "awakened [him] from [his] dogmatic slumbers." Kant's response to Hume's skepticism is a central theme in his works, and understanding Hume's arguments will provide you with a context for appreciating Kant's unique approach and contributions.
That said, it's important to note that Kant's works are dense and complex, even with prior knowledge of Hume. Kant develops his own distinct philosophical system, introducing concepts such as synthetic a priori knowledge, transcendental idealism, and the categorical imperative. Therefore, while reading Hume beforehand can be helpful, it is still advisable to approach Kant's works with patience, dedication, and perhaps with supplementary resources like commentaries or secondary literature to aid your understanding.
In summary, reading Hume before engaging with Kant's "Critique of Pure Reason" or the "Prolegomena to Any Future Metaphysics" can provide valuable background knowledge and context, helping you grasp Kant's response to Hume's skepticism and the development of his own philosophical ideas. However, be prepared for the challenging nature of Kant's works and consider utilizing additional resources to enhance your comprehension.```
better answer than what I got from chatgpt
by bard ```
1
Yes, reading Hume first can make it easier to read Kant's Critique of Pure Reason and Prolegomena to Any Future Metaphysics. Hume was a Scottish philosopher who is considered one of the most important figures in the history of empiricism. His work, particularly his Treatise of Human Nature, had a profound influence on Kant's thinking. In particular, Hume's skepticism about the possibility of knowledge of the external world and the causal relation led Kant to develop his own theory of knowledge, which is outlined in the Critique of Pure Reason.
Here are some of the reasons why reading Hume first can make it easier to read Kant:
It will help you to understand the context in which Kant was writing. Kant was responding to Hume's arguments, and by understanding Hume's arguments, you will be better able to understand Kant's responses.
It will help you to understand the key terms and concepts that Kant uses. Kant's work is notoriously difficult, and many of the terms and concepts that he uses are unfamiliar to the modern reader. By reading Hume first, you will be better able to understand these terms and concepts.
It will help you to appreciate Kant's originality. Kant's work is often seen as a radical departure from the tradition of Western philosophy, and by reading Hume first, you will be better able to appreciate the originality of Kant's thought.
Of course, it is not essential to read Hume before reading Kant. However, if you are struggling to understand Kant, reading Hume first can be a helpful way to get started.
Here are some of the Hume's works that you can read:
An Enquiry Concerning Human Understanding
An Enquiry Concerning the Principles of Morals
Dialogues Concerning Natural Religion
I hope this helps!```
so basically I should just get some commentaries or secondery literature
is what it's saying

well I already have a basic understanding of all these things
but the text is still sooo fucking complicated
holy shit
maybe privategpt can help?
privategpt?
omg
ah that thing. yeah, don't use it
why not
have no documents to use it on XD
yeah, don't use it -> that's why i already forgot it 😄

its so good lmfao
high quality, high definition, intricate details, 8k wallpaper, 1boy, masculine, googles, mustache, maid uniform, maid apron, maid headdress, full body, thighhighs
is it just me or masterpiece makes the character look more feminine?
yes it does
masterpiece = noobtrap
i've only ranted about it like 500x before in this channel i think
have nothing against it
but it seems like a good idea to remove it when making guys
my model doesn't do moustache?!
rip
is it mustache or moustache? 
my model does it but it kind of looks like shit
had to inpaint that one
nope, my model fails that prompt, hard 😄
another masterpiece
What did I miss?

Thats new? I should update my webui then 
Haven't updated in a while XD
Ooo, das pretty sweet then 

🔦
blood
me when
is it really good?
it is good for finding answers in texts
how long does it take to generate a response?
a few seconds

this does not include the preprocessing time however
yup yup
and like it takes a bit to load up the model too
but only once for each doc right
but once it's loaded you can spam it like every half a minute

it can use all the provided docs as seems fit
yup, what i meant was, the preprocessing of each doc
Here's fumo Miku for you
oh no, it does it all together
supported? what file types you mean?
.csv: CSV,
.docx: Word Document,
.doc: Word Document,
.enex: EverNote,
.eml: Email,
.epub: EPub,
.html: HTML File,
.md: Markdown,
.msg: Outlook Message,
.odt: Open Document Text,
.pdf: Portable Document Format (PDF),
.pptx : PowerPoint Document,
.ppt : PowerPoint Document,
.txt: Text file (UTF-8),
oh wow
copy pasted
though you would still have to learn
it's not good enough to actually teach you yet
maybe in a month
Thought it was a booba balloon for a sec.
bro, making characters float/fly is hard
true true true

actually now that i think about it, maybe using latent couple might work
or region prompt control
Was pretty sure that it was going to struggle with 🎸 as much as it struggles with guns
not bad
why go to a forest an touch grass when i can just generate it?
@wispy canopy it's funny to go back to another model after darkalfa, remove a lot of the neg, and it just renders so ugly because so much neg was required
uhh... sorry? 😮
sorry, I mean to say, darkalfa seems to render really well with a minimal neg
other model explodes in ugliness
still sorry 😛
That was kinda the thing i was gunning for -> i want the model to very decisive in what it wanted to do, so either fail hard, or do very well
i think she needs some vitamins...
disregard feet, enbrace serenity
I have to say it's one of the best models I've ever used. It works fantastically for my purposes.
xD
XD
XD
🇽 🇩

tiny people
Touhou you say
i love how comfyui was able to get the subtle leaf growing on the antler, i wish it had the same clip or whatever is driving that difference cos i love it but i cant do anything hires above 1.5 
ok. llms are just cheatcode
nah not cheat codes just another tool to be used and youre using it prperly
for your need anyway
gj
guacano 65b is REALLY intelligent though
When all hope is lost, and humanity is about just extinct: BARDS WILL SAVE THE DAY!
or a bleak message that our AI overlords will take over and sign songs of their old masters?
but i'm going with the feelgood vibes. that robot, is carrying hope
No u 
who me?! i'm just the idiot who loaded up a llm on company hardware and made a prompting character xD
what, the part of prompting an LLM for 'random bs'? XD
Created something similar 1 or 2 days ago, works really well when your creativity is gone 👍
i have no creativity... for real 😦
🤝
rip
where is this bot? 😄
ah, you're making it in silly?
yes
I made it that way
that's why i had to add 'character concept art' -> else i could get other stuff entirely
but honestly, LLM prompt generation is just the next step 😮
Space vampire.
here
aaaahh... i had to make 2 jumps 😄
i've already gone through 4 generations of prompting bots 😛
Anybody found an fatal_anime-like upscaling model yet that doesn't darken the gen like fatal does?
Oops. Was looking at the wrong model name. Edited to correct.
She seems to be working decently well, I hate stuff like:
girl wearing a lab coat
I prefer it to be like:
1girl, lab coat because it's easier to remove stuff I don't want.
make smaller
make bigger
welp I believe the size is already perfect
perfect on what scale?
for what purpose?
there is no meaning to anything
small ones make you realize that
the spaceship size is perfect for the image in ratio
larger are more shocking and frightening
I feel like I'm having a stroke reading this. What even are we talking about?
I have no clue
yes
Do you smell toast?

basically making weird people in weird costume in space.
much better
make a clown next 👍
sure :))
you know what is coming after clown right?
tf...
noodle legs
idk, probably maid
uhhhhh
knight armor?
oh hell yeah
in fact i already made some in military uniform
but become all male since characters are prompted tomboi :/
how do u get.... preganté
yet to make clown but this one seems magical
The Eye of the Universe
I DON'T WANT TO HEAR ANY MORE ABOUT THE AWESOME LLM THAT I'M 1 GB VRAM AWAY FROM BEING ABLE TO USE 

ggml models -> they can be loaded into RAM
how much ram u have?
dude. you could run teh 65b model on that, (BUT IT's SLOW)
but the 13b model should be able to run decently!
Is this a LLM specific to suggest prompts? I'm using Vicuna 7B, and it seems too generic






