#🍥|anime
1 messages · Page 48 of 1
hmm, i fixed it with "crystal eyes" in the negative prompt
anything v3 introduced that shit
this is a belly made of warm ^^
maybe it's easier to make a beam in photoshop than to prompt for it
yeah, beams are notoriosly hard, i suppose there's very little parameters that specify positional relations that clip can use properly
Eyes always hard imo, too many artifacts. Worked a bit adding it on the negative prompt but I don't cut it that well hmm
yeh
junji ito blood cat. lol
magical ice cat
cute
FUSION----- HAAAA!!!!~~~~
that looks so good, what's the model
thanks. im using ghostmix v12 with highres fix
rip the eyelids on the bottom failed 
woow thats a super detailed eye. maybe inpainting could fix that bottom eyelid??
probably, but not sure why it keeps turning a bit red on the edges and so rough
happy accidents best accidents
lowering weight removed burning but made the hand look TERRIBLE
these cats look like they are from an old game lol
@trim current majestic beauties ❤️ excuse my while i go hug my two cats now
nice!
i wonder why animesharp_4x is worse than ultrasharp_4x
each upscaler has ups and downs i suppose, there's no real clear-cut winner i think?
could be a kids series with these 3 magic cats.
i would identify with the pink one!
totally felt the same. that could totally be a series. The girl, the hero and the ... eh... lancer?
SAMURAI PIZZA CATS!
Aaaa adorableeeee! Im glad i inspired these
I was testing my newly found capabilities now im running on linux without the insane memleaks
Being able to hires to 2.5 is very nice
got wsl for my linux needs
I have a spare ssd i use as a dual boot
ooo what gpu do you have?
Amd 6600
I'm trying to get a fine sketch/manga-like pic
any suggestions what I could do to make the outputs less blurry/uneven (idek how to say it properly). or the only cure is to generate in higher res? 😦
||Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 6.5||
adding quadruped to the prompt really stabilizes the cat. you might want to try that if you want to prompt more cats yourself
wow 2.5x on 8GB vram thats really impressive
can you give me ur prompt, i can see what's going on if you want
I borked an install and rebuilding torch 2 atm 
8mins
@trim current https://tenor.com/view/pat-pat-gif-24957682
Fuckin ddetailer
like the full thing?
or just the prompt and negative?
im a newbie sowwy
and it's also an img2img using a sketch
prompt + negative shoudl be good enough
Erryone is and was one
may look dumb asl 💀
1man,battle, Dabi, boku no hero academia, claws, creature, emphasis_lines, eye_contact, eyelashes, face, fangs, fire, flashback, horns, jumping, looking_at_viewer, motion_lines, nib_pen_(medium), official_style, page_number, single_horn, spiked_hair, standing, traditional_media, wristband, (bald lining:1.3)
Negative prompt: bad-artist-anime, bad-artist, bad-hands-5, bad-image-v2-39000, badhandv4, bad_prompt_version2-neg, bad_prompt_version2, EasyNegative, cute, soft, girl, woman, female, feminine, (low quality:1.3), ((blurred)), (uneven lining), armor, disembodiment
I gotta figure out how to install comfyui inside my venv
Am still quite noobish myself to this lol
that's a lot of negative embeddings
sounds reasonable
which are better to keep then? 👉 👈
best? start from 0
better to use a few embeds than write 100000 lines
🤝
fair. but sometimes those can fuck up the thing you really want
I once had used so many negative
i don't have your negative prompts, but this is your positive prompt on my model. pretty damn stable if you ask me
true
so. the takeaway -> you're over 75 tokens. The break happens on an unlucky point, and puts a LOT of emphasis on bald because of that. try to keep your tokens under 75 for now, weird stuff happens after that
if u're aiming for specific character, it best to train a lora or grab it at civit.ai
fire
most anime model doesn't have strong understanding of character
not really aiming rot someone specific
unless it's reimu or miku
getting the right model too helps a lot too, which model are you using?
cetusMix_Version35
your prompt has Dabi and boku no hero academia
SD is creating everything but a particle beam
yeah, but I kinda just expect it to the use the characther and manga as references
idk
So I presume u intend to atleast get closer to that character style
yep
ok, thank u guys
a token works more like "i want you to imagine this"
so in this case you reference all of boku no hero academia, and what the model remembers of it
damn
which... kinda doesn't work like you intend
the idea is there
but, unfortunately, SD doesn't work that way (yet)
thank u 💕
@meager sun getting there
There, took tonights shot
@meager sun , if you want something like the author's drawing, you could either get a lora (probably best option) or try to use something like Kōhei Horikoshi style (author of bnha/mha)
this is my to go, too
art by (x) style seems to work best in my testing
but you're REALLY dependent on model knowledge in that case
what about the fact that i'm generating in 512x512?..💀 💀 💀
lemme go to 512x512 in that case. but this is my output after some tinkering with your prompt
1man,battle, Kōhei Horikoshi style, creature, face, fangs, fire, horns, looking_at_viewer, motion_lines, manga, lineart, black and white, official_style, single_horn, spiked_hair, standing, traditional_media, wristband,
yeah, thank u a lot, buddy
this one you can tell i used regional cos no feathering lol
never used region prompt... only txt2img i think...
so cute 
prompt?
(sle:1.5), mksks style, detailed background, solo, white shirt, black hair, red eyes, long hair, closed mouth, light rays, outdoors, (pink background, cherry blossoms:0.8)
auroraone model
TY
The last two are tiled diff region prompting
That i did
I now attempt background wides
whoops, left batch size at 4, but put size to 8. this is 4x8 by 512x512. seems pretty good actually, i'm using my own negative prompt + negative_hand by nerfgun
🤤 
background wides?
what's the model u r using tho?
alfahentai, my own accidental model
be warned if you want to use it, it's... weird
for example, prompting masterpiece is just a waste of tokens in my model
which i see as a very good point.
but oppinions differ
xd
can someone test out my model? https://civitai.com/models/5374/xynaptix
can you upload a pruned safetensor.fp16?
8gb is a lot man
look again, there is fp16
on what is trained?
it is a merge
v8?
no, what version is listed first?
v1-obsidian
that version
do you know what are the more sensible tokens?
who?
no pruned fp16 safetensor, just sayin'
you, i want to know in what this merge is specialized
general use
very flexible
you downloading? or too much?
i'm downloading the 4gb pickletensor, i don't think you'd put stuff in there
if i generate some homungus monster and die of heart attack, the fault is yours lol
but for the next model, try to upload the pruned fp16 safetensors, that's the thing people download mostly
oh fuk didnt see that is wasn't safetensors
bruh
only 1.2k total downloads on my model, well, for an accident, that was a good accident ^^
XD
and i TOTALLY have no idea on how to improve upon it
please read the description too
anyone know if it's possible to subtract loras?
nah, if i want to improve my model, i'd spend time to remove all the characters possible
i only want my model to know as much concepts as possible, not characters, we got loras for that
why tf would you remove chars?
because if i remove chracters, the model has more "space" for concepts
that's the idea
that idea is dumb
Yep
i love add-detail at -1
Really like the flat color it gives
@merry tendon sorry my man, i'm gonna hit your model with the hardest prompt i have. 😄
don't do me like this 😢
nah, looks like you are getting a passing grade from my stupid prompts!
oooh yes you do!
"Accident"
what do you mean?

because this specific prompt has had a couple of models just fail... make a incoherent mess
girl in front of planet venus, goddess, ethereal, celestial, cosmic, nebula, abstract
testing touhou characters without loras
suwako 8/10
i was drunk, random prompting, got mangled genitalia, tried something in supermerger and... it was a fluke model which actually did a lot right all of a sudden.
looks good, hit it with some more prompts that are abstract
I accidently made a photorealism merge when aiming for a anime with good anatomy. Got the weights backwards smh
I think deforum is working and am excite
i figured out how add difference works, lol xD... i think
how?
Triple sum is my goto
so add difference is a+(b-c)
so in this case, what happens, is it tries to remove all the stuff it learned in c from b. Whever is left in weights, is added then to the original
in my case, my thought was: generic base model + (hentai model - old bad porn model) = might work?
it did.
well, not even bad porn model, f222
Kek
but somehow that removed a lot of the bad things
I hate i know that
will you test more?
f222 kinda old at this point no?
it is. but, as well used as it was, it introduced a lot of 'errors'
which became widespread
what errors?
so, backwards reasoning: if i REMOVE that from what i want to add.. the errors might become less
mangled limbs
@merry tendon this is why your model isn't half bad -> does pretty damn well ^^ lemme hit it some more
everyone does
then,in the same token, i removed AOM3 stuff, really hating everything that model does, for some reason
that's v1.1 of my model
and i think that's why it doesn't respond to masterpiece anymore
I saw a video which mentioned that ControlNet 1.1 Tile can change images to anymore detailed one without changing the structure of the image, but it didn't show how it is done. Anyone know? Example image of before:
After tile:
testing touhou characters without loras
remilia 10/10
looks more detailed
While i wait for deforum. Enjoy this ontopic image
you are not in projectai? wth you doing?
fr no cap ong
oh my fucking god
i never got wtf touhou is, what is it? can someone explain it?
Touhou project?
touhou is a life choice
game series with LOTS of fan content
series of bullet hell games, which people then made a ton of fan content off of (including doujinshis, animation and fan games)
I'll practice playing with tile I guess. It can retain details without changing them too much
why?
the choice to dedicate yourself to an impossible challenge and overcame it
because of the copyrighting of it (very free and open) and idk people just like it
ok i guess
looks like this image is coming out good, i hope to not get another weird face XD
what model are you using?
neverendingdream
there is a discord server called projectai which is filled with touhou fans (it was once a touhou server) and like practically experts in stable diffusion and ai in general
is it public?
oh, i want the link. more bros to laugh with
@merry tendon here's my schizo testing prompt 😄 who did it better 
she zooming
hard fight
oh yeah btw breakcore best
dms
i think your model fits better to abstract art(and styles)(sometimes at least)
That's what the v1.1 was for. to combat the anime overfitting it had originally
i want to feel this prompt is pretty good at testing various results -> it tests booru type prompts, sd type prompts, song lyrics, all kinds of shit at the same time
the good generic models simply fail at lyrics
what were the lyrics?
, well dressed, man, masculine, husbando, short, stubbles, Sniffing out the smell of dripping blood Deep in my heart, my voice echoes get it all now get it all BREAK There's no time to hesitate And under the moonlight that divides the sky Trying to fly without wings sacrifice anything
lyrics on stable diffusion? do you want a sd on lsd? lol
that's one of my most used prompting techniques you know...
mine has good anatomy, but yours has better consistency of the wings(in that pic)
and the other one is: `Want to find a reason why my friends all died
The magic and machines
They stopped and the virus set in
Find a good solution, How to fix this up
You made these crazy worlds
I'm stuck between the two,
(best quality:1.3), (highres:1), (detailed:1.3), (incredible:1.3), (perfect:1.3), (perfection:1.3), (volume cover art:0), (illustration:1.3), fantasy cyberpunk` -> this gets you futuristic buildings
that worked ok-ish on my model
why i feel never ending dreams draws better man than woman?
#3 shows why my model IS a hentai model, after all
from 1 - 10 what would you rate my model?
for my prompting purposes? an 8
let me try it and than i answer
it doesn't go flat-faced with weird prompts
for this image, is left 11%
what does flat-faced mean?
that it simply fails at the task
oh nice
for example, the "gouache by vhils" (the orange/blue draing) fails on your model though, but it does great with the lyrics dude (Wings)
yea, you have some idea on how to improve artist styles?
the improve something, you have to drop something else
tried using song lyrics for the prompt
definitely surprised by the output lol
you cant just "delete" something out of the model
i think so at least
yeah, but have to lose some elasticity if you want to give personality to the model
lora are good for this reason
testing touhou characters without loras
flandre 9/10
maybe is better to have a flexible model and that give personality with loras
an 8, the personality is not there
i did it by re-merging with add-difference, this time adding the difference of the original model(alfamix - aom3)
flandre is a childish yandere
my face when fanon
added some more prompts i know models have trouble with.
testing touhou characters without loras
okuu 8/10 (pretty surprising)
loading the model
if anything, my model keeps contrast nice though.
what model?
obsidian
you mean xynaptix?
yeah
obsidian is the version name
@steady grail it's a good model, it should work fine for you!
testing touhou characters without loras
koishi 10/10 wtf
yeah, but xynaptix looks like a lovercraftian nightmare
does aurora have inbuilt character loras cuz these are perfect
cute
@merry tendon but my prompting is biased, gimme a prompt of yours that you know your model is good at!
oh mah gawd
my prompting is shit af, thats why im asking you guys to test it out
no prompting is shit!
Nope, just has a much more coherent text encoder than most models. It does most Hololive characters quite well too.
ok then
I guess this proves that any anime model can do touhou
i'm using aurora
@supple raptor can i ask you for a prompt too, to add to my testing prompt?
Probably because the characters are pretty popular
is this also aurora?
nope
what is it then? that's really good
Unfortunately I'm not at my PC and left my chatbot running instead of SD. You're welcome to grab one from PNG info on my post history though.
A merge I made a while ago. Don't really remember what I used to merge but aurora was probably one of them
aight, fair. thx tho ❤️
not as good
hello
Am i the only one who doesn't get into touhou waifus at all? the only thing i like from touhou are the tokyo active n33ts and demetori music...
i'm pretty sure orangemix was also in it?
what I don't get good hands and good art ?
Nope. Don't like orangemix
it's all rng
that's pretty damn good, actually
how can I install it ?
HOW MANY STEPS DOES THIS MF NEED
I have no dang clue who she is lol
and will it also solve the errors ? like clothes errors
why this weird pose? lol
75 NORMAL + 120 HIRES???????????
"details"
decently close
ignore that💀
me with 30
.
Need more details
could you tell me more details about controlnet ?
she broke her fingers
it pretty abstract prompt btw, should do something like this:
the AI experience
my model fails at 15 steps... lol
it's not model dependent tho?
not sure... some models resolve really easily and for some reason, my model really doesn't want to... i can't go beyond cfg 6 with that model either
maybe it's overfit?
not really, it has a good general grasp of things
Maybe it's not NAI based
except planes lol
i killed that by removing the AOM3 stuff lol
(add difference i want the AOM3 part not there)
why AOM3? it's the culmination of everything i hate in NAI
6 steps on my model:
it's by far the most popular
Is there any character as popular as her? I Wanna try different ones
not really familiar with the touhou universe
cirno and marisa probably
oh, i know marisa. so she should be pretty damn well represented in a model
I wonder what other anime characters can be prompted without loras
giant ass finger 💀
Does semples steps effect ?
Tried a different merge
megumin
i think
Here you, the first page of the most popular characters on Danbooru
It's all touhou
remilia
I know that one
HOMURA IS ONE OF THE TOP ONES???
OH MAH GAWD
I GOTTA TRY
Basically Miku, Artoria Pendragon and touhou XD
Perfect Rem
she is Rem
@supple raptor hey btw, I now can create images in stable diffusion
thanks for your help
"i love Emilia"
Miku is N°1 tho, big a long shot too
Rn I'm creating an image but it looks will take time
it got some of the colors i guess....
Awesome! Glad you got it working 👍
(not even close to homura tho)
didnt you also try out my model?
but I want to ask you
Here's page 2 in case anyone wants it
does it affect ?
still creating, my pc is pretty slow and high res is pretty time expensive
i already love this image that i can't wait to edit it XD
show me the result when its done ❤️
he is so gorgeus
._. it's not high quality drawing
his eyes can see in my soul XD
but members here are making high quality drawings
i'm pretty sure he have laser beam
what the problem I have ?
More steps means the images will take longer to generate. The number of steps needed will depend mostly on the Sampling Method (AKA the Sampler) that you use. Personally, I recommend DPM++ 2M Karras with between 15-20 steps because it's fast and generally usually it will give good results.
people on reddit man
cfg and denoise too low
so I should reduse cgf ?
this ai?
Of course
use it at least at 7
their cfg is fine
oh my god it even got the armpits
GPU GOES BRRRRRRRR
with at least 40 steps and denoise at least at 6
@cyan burrow put your model to the test too, using my inane testing prompt 😄
usually i get the best results with this parameters
you mean the cfg ?
2º most popular character
darkgemini takes the cake for sure
the most important part
i should test karras
😢 nooo my model is trashtier 😦
nose
i can already see a mistake
never tried darkgemini, by those results it should be better than aurora
a creepu one
ok mine fucked up on that one
i need to impaint 😦
cuasou
creepy
what's the problem?????
but hands are pretty good
just the body and clothes are good
everything else is bad
these are my settings
what should I do?
aymen, i suggest you to download open pose
what model?
what is this model
can you add my model to the comparisons?
hmmm... if i feed that massive prompt to an LLM... (1000+ tokens) hmmm 🤔
sure, which one?
A mix I made, kinda forgot recipe. Better than Kenshi.
also test the good old anythingV5
_ANE_Pruned-clip-fix
https://huggingface.co/Surficialruby/RandomMixes/tree/main
download goes brrr
can you give me the link ?
meinamix
nah, that one is overfit -> not even bothering testing that one
it was my goto model for a long while
open pose?
install from url https://github.com/fkunn1326/openpose-editor.git
just all the cns
i stopped with anyv3, because of the overfitting
what is cns?
silly
??
doesn't he just want to make his image higher quality? like from sketch to good image?
i lovely extension that helps to stick with the pose
controlnets
is that image done?
Victory!
he is maintaining the denoise low so to not lose the pose, losing creativity
92%
fro, here?
yep
I saw this thing, haven't tried it but it might be good for @fallen sundial
how do I download contronets?
what do you do in the prompt to get this style?
i think it's his lora mix, i'm still jelly about it
then download models from here https://huggingface.co/lllyasviel/ControlNet-v1-1/tree/main
once you install the expansion, go to installed and press the orange button, apply and restart
extension*
Should be in pnginfo
oh, right
my discord is broken or smth, i can't get pnginfo from images
no
to install the extension find it in the available tab
I have contro net
this is controlnet https://github.com/Mikubill/sd-webui-controlnet.git
just press like ctrl+f and type control
BREAK 1girl, solo, (open hair, long hair, straight hair, white hair, hime cut, blunt bangs, azure eyes:1.3), pointy ears, white_eyelashes, dark, plain choker, (realistic:0.8), fangs,
BREAK (oversized_clothes, puffy clothes, white hoodie),
BREAK happy, blush, twitch emoji, (👍🏼✌🏼:1.3), portrait,
Negative prompt: (nipples, open clothes, cleavage:1.3), verybadimagenegative_v1.3, (low quality, worst quality, bad anatomy, bad hands, normal quality:1.5), artist name, animal ears, hair ornament, (mini skirt:1.2), (leotard:1.3), holding skirt, lifting skirt, clothing pull, animal ears,
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 580766668, Size: 512x512, Model hash: 80d5cce368, Model: aurora_v10, Denoising strength: 0.5, Clip skip: 2, Hires upscale: 2, Hires steps: 8, Hires upscaler: 4x_fatal_Anime_500000_G, Discard penultimate sigma: True```
then after that go to the link I sent and download the pth files of the controlnets you want (maybe tile? I saw someone getting good results with it)
a spooky image
Niji just changed over to V5 of their model and enabled free trials for the weekend in case anyone feels inclined to abuse it try it out.
what model?
then put these pth files in stable-diffusion-webui\extensions\sd-webui-controlnet\models
xynap
do you have link?
oh nice
It's just in the nijijourney server?
is there a tutorial ?
video
what's missing is a mic in his hand and a big ward saying Enka
nijijourney server?
just put it there
they probably did
weried image..
they didn't
the model was almost 4GB
if you do this it's always 2GB
can you send me the original image?
no
you have to select no-ema
for it to be like that
yes
.
bitch that's what i showed
it's better now but still not high quality
I thought you meant use the extension lol
I still don't get it the reason that I don't get high qualiity
what's your denoising strength?
0.5
increase it to 0.75
btw I don't have pose editor and I still don't know how to download it
but I would like to make animation so I want to look like the original
Im waiting till i can train to pull my 10k images i made with mj and train a cherry pick lora
???
really differnet image .-.
?
no access
I mean I want it to look like the original
Been messing with that Detail Tweaker Lora, I don't know how this thing works but...it certainly does change things up for the better. lol
then do 0.55 or 0.6
Her eyes are empty like my soul....lol
don't care for eyes
Id uh take that down before mods get upsetti lol
nah
it's normal
also it did not get tagged by the bot so it's safe 👍
why are you deleting the gif
Fuck you discord mobile -_-
ok
Trying to tell kazoo that he has a skill issue but instead it's me having the skill issue 
visible version where
tf did you use? camouflage
Sorry her tits arent triple d for you 
i prefer flat.
wwho asked?
why your model does this to me? lol
wait what happened?
their gifs decayed
Nah, they embedded fine. Discord just refused to send them in a reply to you so I hid my shame and failure.
refused?
Kept sending as a normal message instead of a reply.
better
ez
Great friends
could probably do same res
but it'll take a bit of time
I am too impatient for that
the model don't give me a microphone 😦
i can't make a enka singer 😦
i love yuri
enough is enough!
what should I watch to get what you have and learn?
okay
aitrepeneur got some nice vids but not all are the most updated
okay
wow
i suppose this model don't recognize objects 😦
fingers are not the best
i placed a microphone in his hand with photoshop and than inpainted it, but the model deleted it, with this prompt holding a fucking microphone!!! on his left hand,
can always fix later
just upscale then inpaint only masked
how to fix them later?
lower denoising?
yeah, i can, but it was a test
i think i was enough explicit and still is not enough
man I love this so much https://huggingface.co/spaces/skytnt/anime-remove-background
???
?
He sure ain't cooking
bruh this is exactly the games each of my female friends played
except like TLauncher
and star stable and tales of something
and osu
i suppose that's how my dream will die XD
i couldn't make a microphone with this model lol
that microphone is so out of place lol
this neko chan needs a microphone!
tf is this goofy ahh gif
i get this error when i try to img2img of the mic
ValueError: Coordinate 'right' is less than 'left'
the model really refuse to see the mic lol
show me in pvt
@brittle jay @cyan burrow @merry tendon more comparison.. sorry for delay, had client crash and keep getting LoL invites
bitch crop it correctly!!!
bruh my model sucks fr
stop calling everyone bitch
is x/y/z... it fails at the resolution lol
me after comparison
it's satirical
does aurora have a mental breakdown at chibi art or am i doing this wrong
pink girl shopping #2974389274

I would increase weight on chibi and reduce weight on SLE
sus
thanks
I usually do (chibi:1.3)
why do people see my png info but i cant see theirs
She saw some tasty blood
fuck me ig
lol
blood? that ain't lookk like blood to me
a bit better now
She's just drooling while looking at her food
But if you want something sus
I would prefer to say my model is just a bit more abstract, than say it is bad
why is it white then?
oh my goodness gracious
what cfg is that?
7
then idk
i realized, this is 1.5
maybe "cute" is a strong keyword
Try (negative_hand-neg:0.8)
or chibi is idk
oh my god this is the best one
(KHFB, AuroraNegative:0.8)
why is WizardLM better than pygmalion 7b in roleplay wtf
the ring
maybe :3 is causing it
nope
that's a tag i've never had trouble with
and now the issue is gone wow
Restart webui
it was this
which is weird since i've used this tag before
it is just like more consistent from what I tested but idk about how interesting the stories and reactions it makes are
tell me if you continue
cause I might use it for my game
Try \:3
continue what?
issue came back
i found the exact issue
:3 at the absolute end of the prompt fucks it up
using it as the penultimate tag doesn't fuck it up
testing/ role playing
then my thing is broken or some shit
(\:3:1)
?
I noticed nerf-kun talks way less recently
I have only few embeddins
I believe nerf is at the moment travelling
bruh wtf how
oh
I have 27
Left like yesterday or today
oh yeah
@gritty patrol did you try this?
yeah, it works
Do you use dynamic prompt extension?
I want to do emotes but I have no ideas
you already made quite a few amazing ones
they should add them to this server fr
I have made like 6/14 of this server's stickers
just because miku though
Fruit is easy to bribe 
she should add your other ones aswell
@cyan burrow i sent a dm to you
I wanted this but I guess Fruit didn't like enough
like this one good
But I want to make emotes using this character
she your oc?
Yeah
also this one you made is nice
(original)
Sorry, I don't swing that way. 
so cute...
what is your model?
Only the best
I saw emotes I've been summoned
What happened
Aurora 
I'm trying to make emotes for my oc
Wait, I think it was Aurora
⚠️ goober alert
Oki have u tried using my anime punk model ?
Nope
Try it with my settings it should give good emotes
I have style I like but more like can't think what emotes I want to make
Ohhhhh okay well
@real nexus This is style I'm using atm
Checked to be sure, it was indeed Aurora.
aint no way the smug one isnt edited
Dude you're killing me. Upload them in batches so we get a nice single post grid
It is only slightly I thickned the mouth line is all
im so mad rn. processed a larger batch and almost every picture is unusable cause of double features because i prompted something double 
They are discord stickers I cant
looks weird
Ohhhhh. 
armless
I apologize.
⚠️ this bitch is not armed
If you don't thicked some lines they won't read well when smol
except they are stickers, not emojis
goober
This one supa cute
Haven't heard that one in a while
oh mah gawd
This not rly emote like but here
I DID NOT PROMPT FOR THIS LOL
STOP I want to make my own emojis now
That's a very innocent hmm
Do it do it
AI: what?
Supa cute
I WILL
but in a little bit I'm not at my pc rn
I want my AI to talk back to me like the CollegeHumor Google guy.
what lora?
_ANE_Pruned-clip-fix model
For reference: https://www.youtube.com/watch?v=Cxqca4RQd_M
Support CollegeHumor by signing up for DROPOUT: https://signup.dropout.tv. Tons of exclusive content, ad-free, for only $5 a month (that's like 17 cents a day).
Every episode of your favorite web series, all in one video! Starring Brian Huskey, If Google Was A Guy answers the stupid questions we asked, still ask, and will always ask.
FOLLOW T...
i saw this one, so fucking good lol
Also should have pnginfo
I'm just used to throwing your pics into SD and being disappointed with the lack of info lol
shhhh
I'm lazy often
my discord still strips pnginfo or something cuz i can't get pnginfo from the images you guys post here
I have to ask though, do I see 109 artist in some of the recent stuff?
floor
But if you want prompt you can ask but sometimes I'm testing lore style combos and doesn't fee like sharing those
Yeeeeah I figured.
Part of lora merge
But prompts I can share
i prompted for a black background...
Ooh, they made an animated quarantine edition.
I believe it was 109 that gave shape for eyes I liked
I just love the shading so much.
There's a natural softness there that I can't get with other artists.
This is probably worth fixing for sip emote

they're the same picture.
Good old nerf boba chibi
my model (latest version), my model (older version), aurora, xynaptix obsidian, hell5-dpep
y no alfahentai?!
You reminded me that I need to do more testing with this lora
all my output is done on alfahentai
it's a generic model capable of hentai
last nsfw image I made was in 6th of march and it was just to test nsfw capabilities of models
and has a ... mildly strong preference for women, but still pretty good at a lot of things
Another good one
How could anime models have preference for women 
GASP
functional armor. mind you
Night & day

and gundams are the most aerodynamic things in existence!
daaaang, prompting with llm's is sooo much fun ❤️
heyoo ppl
she is kinda scary
i just found out half of the images here are also posted on touhou project ai server
Since there is people who only follow one server
ofc
because these are the best sd/ai servers
understandable
and for anime fans
misplaced ahoge?
this is literally me
Here's old image
fwedwick nietzsche
music: rannar sillard - choir ambience
twitch: https://www.twitch.tv/dkooters
twitter: https://twitter.com/dkooters
merch: https://www.sodarncutebrand.com/
patreon: https://www.patreon.com/dakooters
here is older image
after I went through all the effort of googling ytb to mp4
november 22nd
October 15th
september 5th (dalle2)
I started at the start of october
aug 27th, SD1.4
wasn't my earliest image here, but one of the first good ones.
lmaoooooooooooooooooooooooo same
I was scared ok.
#1010577750077210726 message
it do be like that
actually the contest I won with ai images. I made the images in october 13th
original dalle2 then img2img in sd
yeah, that was literally crazy
well done tho
just after I discovered nai leak
oct 13. and now
tech went crazy hard
nai leak made whole sd boom
its not possible to keep the same dynamic of an image through img2img. if i have to use even a little bit of denoise it ruins the colors and makes it unsharp. even tho the inital image came straight out of the text 2img
i think its because theres no upscale
bad settings
screenshot settings
If the color is problem you can always try this
where is this setting?
same seed number?
just add to quick settings img2img_color_correction
I have it at quick setting list
accidentally made abstract art
ohh that actually might be a reason.. nope.
ok maybe i can get the seed from last generation back somehow.
if not im going to mess around with the quick settings. thanks guys 🙂
You don't want to use the same seed for img2img.
just use my model lol
i edited something out in photoshop and wanted to fill in the background seamless with img2img. so same seed would be fine no?
can you just screenshot settings?
Anyone have good deforum 3d settings i can use as a basis to learn?
Is that truck-kun behind her?
Wonder what kind of world she'll wake up in 
"i've been re-incarnated as an AI waifu uWu"
island
10/7 idea. would watch
steins gate s2
?
part of that season is literally "i've been reincarnated as an ai-waifu"
wiat what
I need to rewatch it now

