#🍥|anime
1 messages · Page 82 of 1
Kalmija was not satisfied enough by the horrible consequence caused by her years of berserk, when she feels the excited climax when creating fleshes by depriving the life of mortals. She is, either, just feel bored on this and decide it is better to sit down to seriously plan how exactly she is going to overthrow the Grand Deity Court, which she declared as corrupted.
very impressive, do you share workflow?
generally, its just generate, inpaint and put in the panels
I don't see a furry specific channel here and have started playing in that space a little bit, so will just throw them here.
Kalmija was represented as one of the highest Goddesses. As the Goddess who controls fire, she possesses a deity ability referred Infinite Flame, an ability allowing her to create fire from nowhere. This kind of fire can be shaped in any kind of stuff, including additional wings, blades of flame, and more, with the area affected burning infinitely unless extinguished by Kalmija herself. The fire is also hot enough to be compared with magma and the sun, making her power extremely dangerous for anyone to challenge.
For a long time, Kalmija was a greatly admired figure not only by the Great Deity Court but also by mortals including humans. That was not until her sudden betrayal of other gods, said to be caused by the "corruption of the Great Deity Court", when seeing her friend Cialie's death which she suspected to be done by Raséal, the Highest God and also her father...
You spelled definitely wrong lmao
custom mix but most is this lora https://civitai.com/models/89551/mix-pattern-yuan?modelVersionId=95332
whoa cool
What's the name for an object like the engine in event horizon? With several rings/discs spinning around
feels nice doesnt it? i also hit 200 total downloads across my three variants of the Tiefling lyco yesterday c:
i do wish people would rate and leave pics more however
TY ❤️
Imma give it a shot in a bit 
The healthiest waifu on the planet
Wow, that's a lot of work. I know this is more difficult than just generating images. Panel layouts, text, coming up with story, and getting consistency in generation results! Must have taken a lot of time!
lovely art
Are there any servers on discord that have a good sized SD community and allows renders with violent content? Like slicing off limbs, bisecting, decapitating, slicing someone to cause a bloodgush with enough blood to sustain 11 people, partial nudity (I don't want a porn server.) Where I could post my art that doesn't exactlt conform to the rules here, cause I keep being afraid of showing something that violates them
not to replace this place here, got too many friends and acquaintances here to ever do that. Just to scratch an itch
hmmmm how does dynamic prompt work compared to matrix prompt?
In Touhou server there was few images with your lora
Lying is love and tasty,right?
Nice stable diffusion is best at making cursed hands so if you're making a cursed image then the hands blend right in
God that's a beautiful hand
waving is cheat word
Yeah
I am the deliverer and deliverance, I sow and I reap, I eat and I feed, I watch the people die one by one and I am watched by them all. They plead for my help and I plead for theirs. For I haven't had access to a computer in 24 hours and it's getting boring to watch peasants die of the plague when I can play a game that made the plague so much cooler. Who am I? That's right....I'm stereotype_emo_anime_character_with_tragic_past_to_excuse_what_he_does_now
anything to improve here? im trying to get truly the best pics
- Brown line between eyebrows.
- I would inpaint ear and eyes little bit.
- Is that supposed to be earring on her neck?
There's hair in the background that seems unconnected, her other shoulder also lacks that golden thing

what?
I have to decide on a movie...not easy with a catalogue of hundreds :<
and I just keep adding more
Stable diffusion images modified with pixel to pixel control net.
After effects used to make the transitions using opacity.
Sound effects ON
Using inpaint to fix hands 
These were all posted by closertodeath in the Touhou server.
They're not, but maybe follow the replies back. Mango asked to see some of the images that were made with their Cinnamuki Lora and posted in the Touhou server since they aren't in that server.
I see
IDK how often she checks them but you can make a discussion on her HF assuming you don't want to join the Touhou server. This is the repo for the model she used on at least one of them that I checked. Specifically hell5-dpep.safetensors https://huggingface.co/closertodeath/ctdmixes/tree/main
just finished ep 17 of psychopass and now I am happy I went through and actually watched it
just join the toho serverino
I can send you invite if you want
The dedication. I feel like you should be awarded for this or something, maybe reach out to stable digest.
very nice
Sweeeet. Glad to hear you like it.
You know that there's more seasons, movies and ova's right? 😛
yeah ofc
also cryonicus
you still got like 2k messages more than me
how tf
is this like actually good enough to be used without needing a lot of fixes? like from the example you gave it like changed the white into a whiter white for example and overall changed a bit of the color-scheme.
what loras?
Typically it works quite well. It can be combined with controlnet reference to make it even better. Prompts that don't force a different colorscheme than the default of the model work best.
G'night! 
weird accident
Thank you, It took me 2 week 
Should all be in png info. I'll take a look later if not.
The gat
if i wasnt training atm i would be using that lol
Model?
buttons & thread I'm still working through beta with it tho its not released atm
Looks pretty dope tho :O
wondering why my model style isnt consistent
changing every day
Alright, I'll keep an eye out
Jit? :o
Oh thats fair then :O
@pale monolith yes I got banned
Wait why on a alt on discord tho :o
my account got disabled
and I also didn't want to lose my old prompts
Oh damn :o
A certain beta model 
This one needs a little tweaking, might fix it up when I'm at my PC.
Oooo, its looking good tho :O
I guess it was Aurora and not my lora merge that doesn't like landscape images
Damn I love my girls to have a mutated toe
I feel that!! Landscape images are always generating cursed body parts, especially 3 legs
It's depressing
Well my lora merge isn't rly helping here
I'm on mobile so fixes are off the table right now. Here's some better toes in the meantime
I honestly gave up on feet, but that image is really awesome
Did you merge it with the Elaina The Wandering Witch one? Style looks close especially the hat
Nope
I spent my whole life gaming and this is the first time an activity has consumed more time for me than gaming
(AI art making)
my rtx 4090 hasn't seen many games since I bought it
those cuda cores are working double overtime
1070 
Doesn't even take that long
28 steps?
20 + 8 highres fix
Thanks, that one had a fair bit of manual work done to it. Minor fixes on the feet, lots of fixing for the tail, and almost a complete recolor of the background.
you mean 8 highres fix steps
Yea
512x768? 2x highres scale?
4 seconds for me, I don't want to sound like I'm bragging I am just genunely curious how much of a difference this 4090 makes compared to others
I always forget to change upscaler to remacri
There is some videos that have comparisons
Good seed
I took out a big loan for this card cuz I became obsessed with AI when I saw it everywhere
I was like I NEED to get into this
imagine being such a weeb you know every character Rice is posting
thigh-tao
((masterpiece)), ((great quality)), absurdres, 8k, long hair, bangs, pink hair, yellow eyes, chibistyle, chibi, hitchhiking, hailing a cab, standing on side of road, confused, @_@, shiny hair
dunno how
lol
lol it's putting a chibi anime girl in realistic photos
hot
did you try other vaes?
i think mine was set to "automatic" that time
its either automatic or "none"
I don't like
very bad
use wd vae
what vae is good?
vae is for colors right?
this one
you could say
alright cool, i'll try again with the Ai prompt
basically it takes the data the ai gives and transforms it into pixels
keep same settings and seed and stuff so you can compare
owki, i'll do that later, i'll try with vae's
you know how to use?
yeah, i just put the VAE in sd/models/VAE right?
ye
aight
alright, this will take a while, i generated that Ai Hoshino pic in 30 minutes lol
You know I don't use NAI vaes
then why washed out and has very little minimal texture on some stuff?
what vae is it then
I definitely wouldn't say it's washed out. My version of kl-f8-anime2 has a .9 multiplier applied to contrast which is still a lot more than NAI has. The texture you're talking about is probably the slight grain that you get from using anime screencap in the prompt which is intentional for the style.
reasonable
Interesting sign above that car
by washed out I meant the contrast
hhaa burger cab
was trying to get that
ahh
and to get that chibi I used kkw-chibi inversion embedding
😄
doesn't always work
I forgot what lora I'm using lol
I should be in bed still. woke up hungry. had to eat , now i'm up. sigh
haha switched models 😄
When you forget to loser the denoising strength of hires fix and the initial generation makes anatomically correct hands, but the highres adds 5 more fingers
nah, im using 1660 ti
maybe got issue with something?
tried generating the Ai Hoshino pic again im running out of memory 🤣🤣
ohhh no, i see it on the boot menu saying "no xformers"
how can i see that?
running sillytavern kobold AI with 13b model and stable diffusion at the same time
lol
stop the flex
I cry now
boot menu????
sigh....
But my time with Nico is worth it
scratch that, the command window when i start SD***
"no xformers, proceeding without it"
ill install xformers, maybe cuda too, i tried installing cuda before but i dont think it worked
edit webui-user.bat
it should install automatically when using a1111
when you scroll down what do you see there
wait, let me start it
looks like i have it
Cries in gtx 1060 6gb
cuda workking
for xformers
right click on webui-user.bat
edit
then
do like set COMMANDLINE_ARGS=--reinstall-xformers --reinstall-torch --xformers
then let it fully run once
then all the next times just do set COMMANDLINE_ARGS=--xformers
ok ok, do i put it on any line?
no
it's like instead of an already existing line
wait do i modify a line to "set COMMANDLINE_ARGS=--reinstall-xformers --reinstall-torch --xformers"?
yeah
the one that says "set COMMANDLINE_ARGS="
or you can also just add the arguments
i added this line lol, there's were no lines starting with "set COMMANDLINE_ARGS"
wrong file
this is webui.bat
not webui-user.bat
haha
I wonder how it will react to the --reinstall-xformers
since you don't have it installed I am guessing
i'll just run this webui-user.bat right?
alright cool cool, its installing now
hmms and changes to a non merged model
stop that
i tried using imgtoimg with this as the reference image, cool stuff SD
Well I am going to poop, ping me if you want
I'm gonna go back to lay down for a bit
hha, poop emoji
Negative prompt: an6,kkw-extreme-neg,
Steps: 20, Sampler: UniPC, CFG scale: 7, Seed: 1205129846, Face restoration: CodeFormer, Size: 768x432, Model hash: a49140c6c5, Model: mistoonAnime_v10
anyhow last one then back to bed fora few hrs.
just got xformers now thanks man
should I change anything on the webui-user.bat after installing this?
Very nice
Yes
Remove both the reinstall thingys
Just keep --xformers
alright gotchu, this is gonna make image generation faster right?
tell me how much time it takes per image?
sets up a bunch of batches, then goes to bed
no man
wait im getting an error
just do the infinite generation thingy
what
NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.
uh
i already enabled the upcast cross attention layer to float32 setting
let me restart the webui rq
oh if restart doesn't fix
then try doing the arguments they suggested
but that will make stuff slower
yeah ima do that rn
no wait restart may fix this
ok ok i'll restart hopefully it gets fixed, stuff like this always happens when updating stuff lmfao
yeah i got it on top of the webui.bat
on top lol
tell me how much time though
ok ok, running it rn
hopefully it wont run out of memory, i think thats because of the VAE
it completed it around 1:46 minutes
not including the highres fix tho
very very nice
what res
uhm
for that resolution
still kinda slow
yeah, good improvement tho, no complains here
512 x 512 used to be 1:46
maybe i can generate 512x512 in around 30 seconds now lol
no way, it displayed the error again while performing the hires fix
stable diffusion is too hard man
ill do the command line arguments
which ones
it told me to do --no-half commandline argument
on webui.user.bat right?
same error?
yeah
NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.
ask someone at #🤝|tech-support
--no-half will probably fix it
but it will make it slower
and like more resource intensive if I am not wrong
aight aight thanks G
Model?
Is that Expmixline_v3


Not at my computer so can’t share prompt
I need to try out other models since starting to get bored to just using Aurora
I still need to explore Aurora more
My prompting style is really bad for Aurora tho
Imma go nap now
It’s like 8pm, so should nap before waking up at 12am later
8pm best time to nap
epic gamer moment

I just generate until they come out good
Oh yeah I can usually get good hands, it's just that 1% I get an image I really like and the hand is all
lmao
Thank the SD gods for inpainting and img2img
112 min tile upscale
and her muscle is kinda weird
she's super cut
That sounds like, way too long O.o
deff not worth the time, dang
I'm still confused how its that high :o
What do I put in the prompt for this hair length?
Shoulder length hair
Medium hair
Shoulder length would probably work best
Not working
When I set "medium hair" it is very short, when I set "long hair" it is very long.
What is this type of weapon called?
scythe
Thanks
scythe gun
What would be your way to integrate that weapon to an image? I have this image, i have cut in photoshop and i am using inpaint with Denoising strength of 0.3-0.5, how would you do it? Also i am trying to use reference only from controlnet
With the hands it'd be faster to either cut it from the reference pic or paint manually then inpaint with low denoise afterwards, for the scythe you will probably have to inpaint it part by part with a low denoising.
Ok thanks, I think I have more or less integrated the scythe, now I'm making the hands.
scary
Uh
that curve
Any top anime models that aren't on Civitai?
hmmm I'd suggest using the reference image as a depth map in control net int txt2img or depthmap in inpainting img2img and masking where the scythe should go then turning up the denoise to 0.5-0.7
making a threshold inpaint mask thats basically the second picture but just the scythe would be good
alternatively going from the 2nd picture use 0.1-0.3 denoise on the scythe bits to slowly warp it into an original design without losing coherency
https://segment-anything.com/demo
also this is amazing if you want to snip out objects for these things
@finite herald sakura miku for you
and CinnaMiku
👋
hi
!!!?!?#?#?#?#?
Tysm for my life
downloading rn
hehehe
Tysm !!

I love cinnamiku too
One of my fave hairstyles she has too 
Thank you very much, I will try it for the next time.
Aurora or DarkAlfa.
And here's Miku getting ready for work, with an extra long toothbrush, lol.
Very difficult to get this particular prompt to work, unfortunately, so this slightly wonky gen is the best of the bunch.
We need a brushing teeth LoRA.
Easy image quality that has a good context awareness! This model was mixed with the intent of having great image quality, and good context awarenes...
Best natural-language-prompting anime model I've used, wheras Aurora is the best booru language anime model I've used.
Is this darkalfa?
I believe so, lemme check.
Right now I use aom3 and I've been using it for quite some time, but there are probably better options, that's why I want to change a little bit.
Yep, it's DarkAlfa.
I used a few models before moving to nearly only these two. My main issue with AOM3 was that it didn't do well with LoRAs. At all.
aaaaaaa
I use a ton and constantly switch around. Mains would be AnimepunkV10, NyanMix, and Flat2DAni tho.
darkSushiV20 is in the running for my new rotation

You use Clip skip: 1?
DarkAlfa works best with clip skip 1, Aurora works best with clip skip 2.
I have never used unipc
I use clip 1 on some models
I don't use UniPC. I usually use DPM++ 2M SDE or 2M Karras.
This way I get the same sampling method on upscaling via hires fix. UniPC doesn't work with upscaling and defaults to DDIM.
just got my starfield controller... so why not a kind of starfield esque gen?
Me when finally playing.
I absolutely love this style! Do you mind sharing the model/loras that you used?
me when finally playing
oh wait
wrong one
||highly detailed portrait of halo, (perfect face:1.1), blue eyes, tartan scarf, white hair, (by atey ghailan, by greg rutkowski, by greg tocchini, by james gilleard, by joe fenton, by kaethe butcher), gradient yellow, black, brown and magenta color scheme, grunge aesthetic, graffiti tag wall background, (sexual:1.2),
Negative prompt: (bad-image-v2-39000, bad_prompt_version2, bad-hands-5, EasyNegative, NG_DeepNegative_V1_4T, bad-artist-anime:0.7), (worst quality, low quality:1.3), (depth of field, blurry:1.2), cropped, lowres, text, jpeg artifacts, signature, username, blurry, artist name, trademark, watermark, title, (tan, muscular, loli, petite, child, infant, toddlers, chibi, sd character:1.1), multiple view, Reference sheet, (Brooch, Ring, Earrings, Necklace, Bracelet, Carcanet, choker, collar :1.2)
Steps: 50, Sampler: DPM++ 2M SDE Karras, CFG scale: 7, Seed: 250174699, Size: 512x768, Model hash: 8ac3e79e96, Model: nyanMix_230303Absurd2, Denoising strength: 0.5, Hires upscale: 1.5, Hires upscaler: 4x-AnimeSharp, AddNet Enabled: True, AddNet Module 1: LoRA, AddNet Model 1: add_detail(7c6bad76eb54), AddNet Weight A 1: 0.75, AddNet Weight B 1: 0.75, Version: v1.3.2||
@supple raptor
there it is
Thank you!
Which one do you think is the best model
I've already stated my opinion if anyone else wants to join in. Really though, they have different purposes. For example, PastelMix is going to do things dramatically differently than flat2dAnimerge.
unipc is nice in some cases
Model, not sampler, 
I've always wanted more of a neutral and flexible model, so that's why I use Aurora/DarkAlfa. I can do many different things with them, and use many different networks while keeping great quality.
What model?
I like to gen the first image in AnimePunkv10 then upscale in Flat2Danimerge to get that look
SD upscale?
Also, if you want prompt/model/etc. info from images that people upload here, you can often find that in your PNG Info if you're using Auto1111's build of SD. Open the image in a browser and download it, then drag the image into your PNG Info tab.
and i use 4x_fatal_Anime_500000_G in SD upscale
yeh and i always just have png info intact
Yes I know
But i am on mobile
Thanks
Is there an anime inpaint model? There's this pesky inpainting I'm failing repeatedly and, once I changed the model to Realistic Vision 2.0, I got a reasonable result on the first try. But the image is not realistic, so it's less than ideal
I've had to make my own. https://www.reddit.com/r/StableDiffusion/comments/zyi24j/how_to_turn_any_model_into_an_inpainting_model/
Though I can't guarantee quality from any resulting models. Inpainting is still a game of chance.
Thanks, I'll try
me watching starfield direct thinking of all the adventures to come
The less than 5G means it's going to take a while to download everything needed 
Super old generation from when I was first learning how to upscale
Love the style!
I decided to not wait and used RealisticVision2.0 (both inpaint and normal) over the sd-1.5 while the 1.5 is downloading. It's already better, but I have my doubts
Background looks pretty clean, what model is that
Easy image quality that has a good context awareness! This model was mixed with the intent of having great image quality, and good context awarenes...
Thanks, I'll check it out
interesting tie / zipper(?)
Civitai goes down mid review

Oh good, it was just a hiccup. Now if they could just import all my image data automatically...
what's "and prep"?
@untold glacier FINALLY the downloads have finished, I started to use it and the results seems to be good, so far
But which would you choose?
Ya gotta tag me otherwise I may not see it until hours later, lol. Info here, mainly when talking about the extension: #1087494110358798547 message
I'll take a look
right
Same. I'll take german chocolate over white icing any day.
CinnaMiku x Yuki Miku
whats that model?
AnyLora
ty

I don't remember Lain looking this good
I still need to watch that.
SRIE
GNL!
didn't like where it cut off so I used Adobe AI to un-crop
controlnet does #🍥|anime message
You can also add reference only as a second controlnet if you want


i have never gotten into ControlNet

I think there's a pruned set now, but ppl keep making new controlnets
this is a cool vibe
gn 👋
Good nite 
🛌
Got a pretty good naruto from SDXL, although it looks like a 3d model from a game

time to crush my pride and actually learn all be extensions and control center so I can make more precise images that I'll ultimately need for the proof of concept VN
I just took a shower and while in the shower, I had an epiphany. I thought for a bit and realized that the best, or rather, my favorite ingredient/sauce is mayonaise, and that is mostly because I could not think of a single savory food that would not go well with mayo.
pure bacon?
i mean... that's just ruining the bacon at that point.
but. you have a point. mayo is good.
would work great with that
DAYUM couldi have the parameters for that?
no no
bacon + mayo good combo
NOT IN MY WORLD!
rice maybe
rice by itself isn't savory, i suppose?
hmm ur right
nah people eat them together all the time
Fr? Never seen that
@brittle jay i have like 5 varieties of mayo stocked usually... so uh... yeah, mayo is kinda good 😄
Black and red flames engulf his fists as he lunges at his enemy
A furious attack like that of a dragon
He is the end.
Negative prompt: negative_hand, easynegative, negative_hand-neg, verybadimagenegative_v1.3, (low quality:1.3), (worst quality:1.3), (lowres:1.3), (error:1.3), (fault:1.3), (undefined:1.3), (crude:1.3), text, watermark, logo, words, word, trademark, copyright, title, header, footer, subtitle, written, name, jagged, spiky, ugly hands, extra arm, extra hand, split arm, missing finger, extra finger, three fingers, four fingers, six fingers, (disfigured:1.3), (unclear:1.3), (indistinct:1.3), merged fingers, (bad anatomy,:1.3) misplaced hand, misplaced foot, cleavage, sex, naked, nude, unclothed, nsfw, ecchi, hentai, slutty, leotard, (text:1.3), (signature:1.3), (title:1.3)
Steps: 50, Sampler: UniPC, CFG scale: 5, Seed: 1798, Size: 768x512, Model hash: 15ee56b992, Model: darkalfaEffortless_v10, Denoising strength: 0.5, Version: v1.2.1, Hires upscale: 1.55, Hires steps: 25, Hires upscaler: 4x-UltraSharp```
I'm using a model that is good with natural text
this us
Oh no so booru prompting? Thats cool
that's an LLM prompt, pretty sure
even better 😄
electro wizard from clash royale but anime
hahaha i can see that
pretty cool ^^
the colors r fire
cute
sakura apple miku?
why do people type BREAK in the prompts?
BREAK is a special keyword
it has to do with the token limits that the thing parses or something?
you notice in A1111 it counts in x/75
im kinda new to SD so idk what that means lol
basically, your prompt is chopped up in sets of 75 tokens
cherry
if you have a long prompt that exceeds 75 then itll count it as a new thing altogether so it might break up a prompt you want to be together. if you BREAK then it's kind of like restarting the counter back to 0/75
so if you write more than 75 tokens (usually words/syllables) it gets chopped up in 2x75 tokens
those 2x75 tokens are weighted average
that's where BREAK comes in
sometimes, if you get for example, 90 tokens, you'd get a distribution of 75+15 tokens
with BREAK, you can 'break' early into the next set, making it 45+45 tokens (45 tokens BREAK 45 tokens)
so with the prompt i gave, that makes it will apply a whole set of 75 tokens, just for the image style
damn, so basically i use it if my prompt is too long?
I think someone should pin nubbys explanation of BREAK
yeah, then it gets applied automatically
or something like that
when it's too long, it does it automatically, in fact, everyy 75 tokens
not 100% sure what's going on in this one
OC :o
thicc
Yes 
THAT'S AWESOME!
Thank you Nicoooo
I put much edit in my generations
U could check out my pixiv i got many beautiful images
Also Yeah cat in the mountain sounds terrible
AI art gave people who can't draw a way to express their imagination visually
seeing so much unique stuff
Exactly
You have no idea how much hate i get in art servers
For no reason
I just share my visual ideas
YEP!! I got shit on for posting ONE ai art image in some visual novel server I was in
4 people jumped me
I told them to f off and left
a few people use AI art for bad reasons and now we're all bad?
stupid and unreasonable
I know u were lucky it was atleast 4 people the whole server jumped at me and continued even with personal insults
Oof
if only artists know how much of an advantage they have still
it won't take them hours of inpainting, photoshopping and img2img to fix a 7 finger hand
That's right, I believe even if i put 2 hours editing work i still can claim the image as my own
U have no idea how many issues i fix by hand in photoshop
the law is so vague you can claim it as your own if you inserted the word "cat" in it
lol
Im glad someone finally gets me and everybody here will do too i guess
That's why I joined
don't worry about it, name a time in history when a technological movement just stopped and disappeared
people like us will be ahead
I checked out your bio i would kove to join your server
U got 4090 me jelly
you're not locally hosting?
people with lesser video cards are making better art than me, hardware just saves you time
I want to check out your art tho
it's just mixed into a single channel of art spam on my server. I have been thinking of making a pixiv and posting the good stuff
there's this game called library of ruina
I was screenshotting each character and throwing them in img2img to generate a high res anime version
yep
go to civitai and download some models
trying to find the original
this is the original
I think the img2img was pretty impressive
I got models, using aurora right now, switch to anything v5 ink the new version and AOM3A1Borangemixs
306 loras loaded
221 of those are available on my bot
But i really made this style mine
And even made some OC and always using them
that's great
Rina Suzuki
She gomma be the protagonist od the Manga im making
most of my AI art experience has just been playing the gacha
but I had considered using it to make a visual novel
Ahahaha it do be like that sometimes
But i try to learn evrything as much as i can
So i can become a real ai generating expert
I am a very good character writer and a C# programmer, now with AI art I can fill in the gap I needed to make a visual novel
That's perfect
Some chatgpt on top
And tou can make miracles
You*
Suddenly github became my favorite website bahahha
someday ai art will be able to draw perfectly without requiring editing and be able to draw on transparent backgrounds and text AI will be able to run high quality models on cheaper hardware
I even downloaded deepfake ai webcamera and i can be whoever i want now BAHHAHA
that means AI will be able to create video games that never run out of content
Let's just start with smart chatgpt npcs
Who never says the same thing
That will be a game changer
Waiting for the first game to come out with that
That's lit i ve never tried kobold ai yet
not even a 4090 can run a very high quality text AI
I can 'load' a 30b model but it is literally at 19gb vram/23
the more messages you send the more vram will be required
cuz apparently AI memory isn't a thing
you are just submitting the text back to the AI for it to read
so it ends up being very cluttery over time withotu the proper settings
it will have to trim its memory
I can use a 13b model or a 20b model as the highest quality and chatgpt is like a 160 or something b model
I see
but it's kind of like AI art ina sense
I had to edit Nico's profile a bit and add a c ustom world data to get her to deviate from her canon world
then you learn what to say to the AI to get better quality results
I'm sure people will be lining up to buy my cursed R18 art with 30 toes, 3 legs and 2 fingers
I wish referenc eonly hits me with an error I gave up. Guess I'm the only one in the world with that error
whats the error
dont use it on image2image
use it on text2image
I actually get perfect hentai results
Don't know about you
Check that out
Introduction:
Welcome to my Pixiv profile, a gateway into my creative universe! I invite you to join me on an extraordinary journey filled with artistic exploration and inspiration. Here, I pour my heart and soul into my creations, expressing my unique perspective through visual storytelling.
Within these digital canvases, you'll discover a div...
it's so imaginative
Those are lovely
How is the model called
Is all ai?
Can i see?
Who wants to start the GoFundMe?
Very nice drawings
thank you very much ^^
that's my model mix, darkalfa https://civitai.com/models/64843
Thanks its awesome
if you like natural language prompting, you really should give it a go
is this a bleeding heart?
alcohol ink makes for such pretty pictures ❤️
Not what I was expecting but I dig how it turned out
ah, that model simply doesn't support it...
which model is that?
It's a custom mix
I love what it did with her hair though, I'll have to just play around with the prompt some more and see what other kinds of results I get
removed all quality tags, only left in the hands stuff, so it can get maximum creativity!
💙
i'm... pretty sure this is NOT angelic actually 😮
Doesn't work for me 😔
model diff
yeah, i did kinda mix my model to be able to this kinda stuff
ngl, still pretty cool!
70% legs
that's the good part, right?
Bill Cipher from gravity falls








