#🍥|anime
1 messages · Page 53 of 1
shoot it man
oh
when I study I feel sadder usually
but more happy (happy as defined in my weird philosophical definition, not the common one) at the same time
do you know that this work is good enough for unreal engine?
like man I was so depressed after studying schopenhauer
this is amazing wow
do you mean me ?
I don't create games soo I just like create anime animations
happens that with study, the questions just increase, instead of reducing
Do you think I can do the same with both?
I think if you do animations like this, it could actually turn out very well
like long jumps between frames
yeah, that's a good base
yes (ofc depends on what)
you can overlap anything on that body, with the right lora
perfectworld is just very consistent nsfw you have a hard time not get any sexualized generation with that. i tried the v2 version before but it looked very good to say the least. not very much suited for non nsfw stuff tho imo
did you use nude!!! in negative?
you might be able to do it but it's more work
That's more like a superhero, unrealistic build xD
i just say it spits out the same clothing types a lot and spits out random nudity
i did it and the results where very good
okey thanks
when model is trained on a lot of nsfw and very little sfw, it is harder to get one than the other
so you mean the animation is really cool >
very much so
I will see more about lora
I will learn more to make full anime animation :)
using a nsfw model for sfw stuff, it's like trying to cut leaks from a broken buked in the mittle of the sea
epic
if you are focusing on anime, are you working on 3d models?
pretty much accurate 😄 but it works at least for some time until you cant fix it no more and drown
yes
I want to conver 3d to 2d
by ai
soo I used 3d model for frames
and ai turned it 2d
what are you using for the 3d models?
yeah people been doing that for a long time now
till now, i used illusion editor
use koikatsu
it's like vroid ?
I just noticed idk anybody's irl name here
yes
sometimes
nope
5 dolars.
it has not, the game is the most not-copyrighted game ive ever seen
anyway it's just for the base dummy for the positions, how could they know you are using their stuff as base? lol
some guys even export the chars and use them straight up in thier own COMMRCIAL game
well, honey select have so many mods,
anyway, is a solid program to use as a base for animations
and very easy to use
if you think it was born as an hentai game
and here we are again, using hentai stuff for making art
lol
it was
it's just, i don't remember if you can assemble directly a character on the editor, or you still need to make the dummy somewhere else and load on it
what do you mean with dummy?
the dummy, is the base character you use just to get posture of the character
with stable diffusion, we just need an humanoid form without details in the position we want
and than use the prompts to add the details
that's a dummy
@elfin bolt I need help with set-up in paperspace.
that doesent make sense then, why would you need to assemble the dummy in a character editor?
would be so dumb
because assembling a full character in a character editor, require more time a give more rough results
it-master, try on https://discord.com/channels/1002292111942635562/1002602742667280404
it is funny seeing "IT master" having to go to tech support
har har, yea...
@steady grail @brittle jay guys can I ask you for something?
could you help me with animate ? it would take a lot of time if I export all the frames by myself,
me no
I cannot see any guide regarding fast stable diffusion in paperspace
especially when I wanted to mount a google drive into it
Okay
on the character editor, you need to make detailed eyes, air, dressed ecc. ecc., why doing it by hand, when you just give the form from the character editor and than use controlnet the add the details of the character?
why not use collab?
stops literally every 10 minutes
which collab you using?
banned SD
free
sorry, no cash
now sd is occupied with the upscaling, when i finish, let's see what i can do
if you can upscale the frames after I export it
are you sure it won't stop on me?
you can do that atleast
a decent upscaling takes hours on my pc
till now, a 2000x2000 just completed a 5%
you can't even imagine how much would take
I don't have a lot of frames I think
If it does I will ask its creator for help
I like to use rasngen to upsacle
i just have an nvidia gefprce gtx 1650
But he is computer master fr no cap, so it probably won't stop
i'm brutally abusing tiled diffusion to do it
you have the same one I have but I have 1660
a laptop, right?
I'm pc
will do, thanks. Also, how do I use my stuff in gdrive like models, extensions, LORA, etc, into this colab?
shhh....
double my memory lol
anyway, cool model
what do you think about 3060 rtx ?
Idk probably says it there. I never used that Collab myself, just heard amazing things about it and saw its list of features.
It downloads most of the important extensions by default I think.
i ready very good things about it
you can even find it for 300$ that's a very good price
it's good for stable right ?
I mean 3060 rtx
yep, the ram looks enough, but i don't know if it may have incompatibility with sd
just ask
I asked they said yes
so, it's a yes
good price, enough ram, should be a good investment in my opinion
on ebay you can even find it at 200$
wait fr?
yeah, but used
well ofc
What model do you use for these?
they are very cute lol
just touched 10
still too early to notice mutations
if i get a mutation, i try with a mask in control net
if i can push it, i'll catch big
upscaling with high denois, would be cool
upscaling without loosing details and sharpness
i'm an idiot, i should have done it from the start -.-''
@ocean reef some dashing gentlemen?
3rd one best
i liked nr1 tho
i like 4th
hi guys, im need to stable diffusion and just finished installing automatic1111, what models do u guys recommend?
browse civitai for your favorite 🙂
there's a LOT of good ones
i'm working with my own model, and a lot of the members here use their own
i hear aurora being used a lot, but i can't make it work for me, then again, i'm an outlier lol
save some money for like 2-3 months and buy a 3080ti 3090 3090ti 4080 or 4090, any of thse have 12-24gb ram, but if i was you i would save for the 24gb ram cards
i see... is it hard to make ur own model?
mine was a accidental mix, some had specific things in mind, so each model has it's ups and downs, unfortunately, no model can be great at everything
this is what we call Happy Accidents - Bob Ross
the skinny one?
supposed to be this character
triggers?
none
I have a 3060 12gb version, it's good
some tokens are more effective on thiss lora?
BUT i recommmend using descriptions for her such as blue hair, green eyes, bangs, child, cute
weight is probably better at ~0.6
i'm working on upscaling it to make a wallpaper
would be funny to see what would happen to this image with this lora
but for now i'm stuck with upscaling
probably would get another style, since pretty much all images from the lora training were somewhat sketches
not many details i mean
maybe lowering the weight
.
i would say 0.2, or 0.3
that's too weak to get a grasp of the character
the last one feels so much vampire the masquerade
art is about details, man
for example, i use many loras for my artworks
all at 0.2 0.3, just to change some details and capture the perfect form
do you understand this is character lora
it's impressive what you can do with 3 or 4 different loras
character loras are the best for altering details
did you train that with dadaptation?
if used well
idk what that is so probably not
i used a kohya colab
using loras like this, i rarely get weird mutations any more
oh cause if yes then the amount of epochs would be too much
did you follow the guide?
yes
like the config?
i was originally doing it locally
most of the time, i even get perfect fingers and eyes
but i ran out of vram
but then on the colab did you use the config?
i used the default config there, they were the same as the tutorial iirc
no
tf you mean no
why do people keep doing this
https://discord.com/channels/1002292111942635562/1091193032273043516 if you can't defeat them, join them
the config only does 5 epochs but you sent epoch 9, so it can't be the same
i am being harassed
exactly!
kek
we saw that lol
working on an anime merge
not you
wow, that's a perfect tails of asteria resemblance
another perfect masquerade resemblance
input_size: torch.Size([1, 3, 2048, 2048]), tile_size: 960, padding: 32
[Tiled VAE]: split to 3x3 = 9 tiles. Optimal tile size 672x672, original tile size 960x960
[Tiled VAE]: Fast mode enabled, estimating group norm parameters on 960 x 960 image
that means it's upscaling at 3.840x3.840, right?
2048x
oh ok, that's what i wanted, ty
perfect
bleed my heart chopping some of the image to make a wallpaper
but i have a long monitor
did you found something?
yeah, usual maintenance, to purge the trash

upscaling at such an high resolution, it's very time consuming
suggestions to save some time?
don't upscale at high resolution? 
well, im kidding
it's definitely gonna be limited by your gpu, so yeah
sad story 😦
so many sacrifices in the name of ultimate quality lol
it's already a miracle i can perform at such resolution
.
for now the img is coming out well, 0 rtfacts at 0.4 denoising

can someone for the love of god at least try
who is she?
shes like 13
megumin is 14
ik
thats why i said it


i have like 7700 unread emails
Is that good? 
probably not
Fair 
like 95% are github notifications tho
Oh, then it doesn't sound like its that important 
filter them with -github
lol now only like 50 emails
what type of spam is this time?
automatic1111 notifications
lol
lol
24% upscaled lol
man, are you trying to murder me? lol
after so many test, i should have got the settings right
thanks
that's not how the character is supposed to look
too much skinny, try with curvy details
no
it's the caracter of lucky star?
no
who is this Nemu Fujimi?
Onii-chan wa Oshimai! ?
if i can get the upscaling at 70% without any weird mutation, i can say that the upscaling is done
chibi works as a prompt?
Time to make some Dark Knights
I'm going to try to make my Au Ra from FFXIV.
Here's what I got with it. Most everything at 1 was weird closeup eye shots so it might just not play well with the model I usually use
it was DEFINITELY not made for 2.5d nor 3d
hey, i wanna make a a little Visual novel (only about 5-10 minutes long) to test my capabilities, does someone have a story idea? (chatgpt is shit for this, and it should be about a yandere or smth)
Ola, boa tarde a todos
what?
what do you like? fantasy? drama? noir
Ola! How are you?
idk, just something with yandere, maybe dark or drama idk
is that a self-stab powerup move?
I never claimed that my WoL is smart. She does usually just kinda stand there and nod or punch the palm of her hand during cutscenes.
shonen or seinen?
WoL?
Warrior of Light
what?
Waifu of Love
for teens or adults?still sfw
adults ofc, its a yandere game
mirai nikki is for teen but it's still a yandere story
MEH, downloading new models from HF takes soooo long
downloading a 30b LLM model -> it has both quantized versions... 32gb download 
we can try something, if your eng is better than mine
I created a visual novel engine in Unity for use with SD images I was generating.
Don't know why. I don't play Visual Novels and didn't really have interest in writing one. 😄
But it worked.
actually the web demo might still be up...
teens?
what about the sex scene at the end?
bruh
i don't feel comfortable in developing porn
it's borderline
i dont make NSFW games, SWF only
you make games?
i want to but i am the most uncreative fuck when it comes to the story
Ah, my demo is still up! https://omnia.pics/gametest/
might want to redo this one without the cameltoe... best remove it
you can try little quick arcade-type games that don't require story?
good pic, but well
OH
i thought cameltoe was barefoot
thats not the type of game i like tho
I didnt even notice it
Will post in a nsfw channel instead
cameltoe is slang for visible labia major throug clothes
ive been trying to tune an embedding of an anime character and the results dont match the source images at all ? anyone know what is wrong ?
no sure thing here, so 
i'm running a nsfw model, so i'm VERY keen on checking all my images for that XD
Isn't there a nsfw channel
not in this server, no 
i can post the source images and what i got aswell as all my settings
ask cryonicus lol
Another one
what kind do you like?
wtf is this lol
hahaha
DUDE
is this the project bwm been working on?
Visual novels, dark games(stories), text adventures, retro games with good story, ...
I didn't say it was good! LOL
literally cryonicus
no
No that was a demo for my visual novel engine. 😛
I see
I wanted to see if I could create an easy workflow for generated art to be shoved into a VN.
would've been funny
Turns out - yes.
he has almost no time tho
wdym
he just needs to recover
is it all procedural?
thats what i meant, he currently has almost no time
I mean the waifus
It used basically like text arrays to move between story elements but it got weirdly complex when you had to fork the dialogue depending on whether or not a particular 'waifu' had been collected, decision had been made, or object had been picked up.
haha no - they are generated and imported
Where else can i share art ?
I used some sort of plugin that removed the flat background
I am just working on a visual novel myself where everything is procedural
but you literally would just clone a scene, drop in your background and character images, and write the story
procedural is interesting... so you are generating character images on the fly?
Procedural story has little interest to me over a hand-crafted scripted experience. That's just where we are. I think procedural story would be really tough.
since my lora always makes lineart style, i have to img2img without it
rn I only wanna make the same character have different emotions and stuff they can do
interesting
You can do something Tamagotchi style?
I did one of my university project games in that style. Back when smartphones were a brand new concept. hashtag-old 😛
abominemution
For emotions, it seems like a scripted branching story would be much more fitting. Emotions based on previous choices. Procedural emotions would result in bi-polar waifus. Do something nice for the waifu and suddenly she's furious with you.
nice glowy purple sword
That'd be a super fun mechanic, imagine being gaslit by a fucking artificial intelligence
im trying to simply inpaint a weapon to be red instead of black and it wont change color, any tips?
Procedural AI created quests etc is a far far way from feeling natural to the game. They'll probably end up following a certain logical trail
would adding this to the dataset mess with hands?
actually curious about this
It'd be funny for the first little while but would get old quickly. Just try having a conversation with a chat bot that has the temperature set too high.
does anyone know?
if youre playing an RPG and maybe end up arguing using your voice with the companion, that would be less tiresome. If it has some meaning. I mean they could have most of the conversation pre-decided but throw in personalized sentences and such
it probably would, but not sure if it'd impact it that seriously
guys help I can't get to swap the color on inpaint 🥲
Inpainting isn't very good for that sort of thing because you have to turn the denoising up too high in order for it to be able to change the color. What are you trying to change the color of?
a shuriken thats on the back
wanna swap it from blue to red
for something that simple your best option would be to paint over it in photoshop/Krita or whatever image editor you like to make it red, and then inpaint over that to make it match the style better.
can i see the image?
oh really
depending i can even edit it
ill dm
For larger changes I would recommend controlnet with the canny model.
Im considering buying Imperators plugin for photoshop, though it does have slow customer support I heard
but it should have just some of the best in and outpaint
my llm doing 1.02 tokens per sec

and I couldn't manage to make stopping strings work in api
so I have to wait until it reaches the limit every time
the limit is 200
I just tell the ai to make them write a reaction
pygmalion already makes characters do that
I just make it do it in a certain way so it's easier to regex
Where can I download LLM etc btw?
Various places
AITREPRENEUR on youtube
The best ones tho need more time to be developed
he provides links
Yeah he just covered one yesterday
I don't trust random locations, people are using one for prompts atm
But like i said, its brand new, so ur better off waiting
he is a funny french guy
French?
he is french
why wait?
Thank you, I'm using an FFXIV Dark Knight lora. I'm switching to Gunbreaker now.
Bc its not optimized yet
can you share it?
there is a version of a LLM that is really well optimised, uncensored and free for comercial use that was released yesterday I think, he covers it on his channel
I stillwant it
^
only 7B parameters so 30B could already be on GPT level
MPT-7B-StoryWriter is revolutionary new LLM model that BEATS GPT-4 with an INSANE 65K+ token limit! These new MPT-7B model sets were created from scratch by MosaicML in only 9,5 days and 200k$ for models that can rival the LLAMA 7B model! Impressive! So in this video, let's have a look at these new game-changing LLM, how to install them on your ...
Its apparently the most capable llm available now
as the google basedchad said it: openAI dont have secret sauce just a good selection of stolen data
Im gonna wait til more people put some work in tho
Yes
define capable
I haven't been impressed with ANY of the local text generation I've tried.
No need, watch the video
im of the opinion that if the model is open source then it doesn't matter where it gathers data from as anyone can use it
did you try pyg?
yes
haha
It's in here https://huggingface.co/joe123yo/lora/tree/main it's a locon so you'll need the lycoris extension to use it. Activation tokens can be found by hovering over the card in WebUI and clicking the circled "i" button.
wizardlm?
yes
already outdated, MPT is on par with better performance
Im trying to get all files sorted so I can install it tomorrow
From what ive heard, they all have very limited total token inputs, which makes less able to remember certain topics beyond a certain point
I want the one that can help with prompts
is it that good?
This new one is a big deal bc it way outclasses others thanks to its amount of tokens
much faster, you have to install a few things in oobabooga to make it work
I'll have to try MPT I guess.
Not cuz of the ammount, but rather the quality on just 7B parameters
Whats the point
OH MY GOSH.. TELL ME ABOUT IT. Several of the prominent "AI youtubers" do this. It drives me insane.
He's a youtube, gotta get that 10 minute mark my dude
30 minute videos that could be 3 minutes.
.....I want to install LLM, the one that does prompts
thats what I want
thats the point
Then make it 3 minutes. Search for what u need in the time bar
its already very dumbed down
Lol
Depends what you're trying to do, but setting up a chat bot even Llama13b-4bit-128g does quite a good job. If you're looking for something that's actually smart like GPT-4 then I don't think we're anywhere close to that for local models yet.
if you think this is the case you should try MPT, it can even code at 7B parameters
I'm sure it can do some things well, but I doubt it has anywhere near the range of knowledge that GPT has simply due to size. I definitely plan to try it though.
Id use it right away if there an optimized version
ok uh which one is it that does the prompt thing? which version?
Tsukiko, Witch of the Shadows
Actually i think it was optimized wasnt it?
You're right it doesn't come close, but we are getting there, I wouldn't be surprised if something equal to GPT comes out in the next month or two
Like the google employee said, they got no real secret sauce, just a bunch of parameters, i believe in open source
I tried downloading a few "characters" too, but they would quickly just start meandering and confusing things.
It was neat, but has a ways to go for sure.
(Also kinda slow on my 3060 😄 )
for all most people want LLMs for ||(erotic roleplay)|| we are already there which is incredibly promising ||(and dystopic)||
Different models interpret prompts differently so it's usually best to fine tune the character prompt based on the model that you're using. You also need to find the right parameters for the prompt and model that you're using.
Can't wait to have a robot GF
That weights half as much as a car, and I die of crushed pelvis
Is there a way to get the prompt of a picture i made?
PNG Info tab
png info
ty
^as long as you saved it correctly
actually its 20$ for AI expert knowledge /s
i guess auto save is saved correctly haha
This lil tab, and you drag and drop stuff in, and it should show all the info, + you'll have buttons to send all that straight into one of the other generation tabs
Remember to set the seed back to -1 tho, cause it also sets the seed as the same of the pic
With default settings, yeah
hey i see u dont have this
can you explain how you get rid of it
MF HOW MANY TABS DO YOU HAVE
XD
mf has 2 rows of tabs
I have too many extensions
how
that i dont use 
Isn't that only for img2img? 
Is it? 
you'd need to modify the css.
Whats the best way to add a background, img2img or inpaint?
Not really, just let it download the models once and then it will finish interrogating within a few seconds every time you click on them.
still to long for me xd
thats what she said
I'm surprised this sign isn't up 24/7

I believe they are R or something, or M, that's 16 or 17 right? well no matter not important
What am I walking into
ratings board discussing
R is 18+? weak
in my countri m is 16+
Wut?
and usually have to do with gore, violence, drugs, alcohol and implicit sex
I think Swedens highest rating is 16 on games, (except for hentai) and 15 on movies )except pr0n)
something most of the time ignored by both shopkeeper, buyer and buyers parents
btw what was the best llm called that can do SD prompts?

Honestly
I think the answer is all of them
Are u saying its kinda ironic theyre called large language models despite having such weaknesses?
Not really
youre a large language model with more computing power and a bag of meat to move
Thanks i hate it
yw
This large language model is a product of thousands of years of training and limited data
thousand, thanks to quantum computing
U can certainly try
Might as well cut the foreground focus out first and then just paste it onto something
oh really
which one is it
Be warned
Its not that good
sure
to not get weird results
for example, the black back ground can influence the color of the new one
its fine i want it dark themed
I was referring to DNA and evolution
you can try to strongly influence it using tokens, but why botering?
Their is an extension tab
If it is not in the list you can go to the url tab and paste the github
extension tab, second tab "available" and than load from
on the search bar, look for the extension
There you go
oky ty
Click installed and restart
Or close it all and restart if it has dependencies
hoping this way of instilling didn't mess up the extension lol
never used any extentions
on the extras tab
you have the box, remove back ground
cmooon someone has to know which language model is best at prompts and learning prompts?
Try the llm channel
i think it's the last time i try to upscale at such an high resolution
i finally touched 50%
uhm its kinda cropping and blurring some parts of the actual picture, any tips?
bruh a1111 api docs literally don't work
never used it, maybe changed the resolution
in the example it perfectly removes it so there must be a way
which was the resolution of the image?
1024 1024
yep, you didn't change the resolution of the result and it did get downgraded to 512x512
share the result
does anybody know how to revert back to a an older version?
yep
I tried git checkout and then some commit hash thingy
I got fatal: reference is not a tree:
someone explained it to me in tech sup
pretty!
What do you mean by "prompts" you want it to make SD prompts for you?
yeah, like i've been doing
You hse vicuna dontcha?
Kek best of luck
Why didn't you just download it?
lol
the ram? XD
Download more vram duh
My gpu can’t even handle sd 🥲
Which gpu
yeah, turns out, that after all these years... these MANY years, it has been a scam!
which gpu has 24 gb ?

3090, 3090ti, and 4090
My current anime mix am working on
Is alr nice, idk if i should keep working on it lol
man there are 2 thousands dollars
yup yup, the one you been working since a long time ago

Nyo diff one
why not?
oh is it? 
got my used 3090 for €750

they said 2 thousnds
is used gpus okay?
you can get a 3090 on ebay right now for $700
I gotta save up abit first lol, who knows by the time i have it ready ill find a better deal
I dont even use my oled switch in months been debating sellin it
@supple raptor kind of, I give it a prompt, tell it to improve it and see
and you wont get 3090s in Europe for 700 dollars
if some fucker didn't overbid my initial bid, i would've XD
you got lucky
3060 12gb is a pretty capable card. Though it's only half the vram. 12GB is still plenty for Stable Diffusion currently, but when SDXL comes out, that might not be the case.
I havent found a single one
what is that sdxl ?
to be honest I have 3060
my brother has this gpu
so let's say I can use his pc
Really, any instruct model should be pretty decent at that. The instruct prompt that you write for it will be the most important part.
stable diffusion, but bigger.
the one eface was using was particularly good with my model
does saving really worth it ?
and is this it for me then? 1.5 is the only SD I will ever use?
I don't know If stable worth that saving
Only you can make that decision. That depends on your financial situation and how much enjoyment and/or profit you will get out of using it.
like I don't know if I can make money from stable
since people hate ai and would say I steal from artists and you can't make money
I have a 3080 12GB and would have saved for just a little longer to get a 3090/3090 ti if I knew then what I know now.
I thought they said they would releasea smaller version of SDXL too
I like to use stable for making anime animations
like this
but can I make money? I don't know
The only reason I have a 3060 is because I bought a PC that's what it came with. It's probably the best GPU for the money
u only wanna use it to make money? If you enjoy it you should just do it
I wana make money and enjoy doing it
I really like doing aniamtions with it
I got a 3060ti because I thought it was best value, but in a gaming sense
I haVE 3060 IN HOME
then keep doing it lol
Yeah, I think they said they will be releasing one that can run on 8GB of vram, but we don't know what exactly that means. Is a 512x512 image going to be on the verge of OoM with the smaller SDXL with all current optimizations? If so then it wouldn't really be worth using over our current models if you're limited to 8GB no matter how good it is.
my brother will buy a pc has 3060
what about copyright issues and problems with ai art ?
If it's for personal enjoyment then it's not like anyone can really do anything against you
People sell AI art, no problem
If you wanna make money off of it, maybe wait for the dust to settle although I seriously doubt you'd get in legal hot trouble for using it
I won't make it personal enjoyment, I will make it a business
do some research then
I live in iraq so I don't live in usa or Europe countries
I mean business by making animations
It varies depending on your area, AFAIK it isn't illegal to make and/or sell AI generated art anywhere currently, but you very likely will have difficulty protecting your work from other people taking it and selling it themselves unless you've done significant creative work on your own outside of the AI.
I wana try making a webtoon and if it works okay then it could be used to make lots of content really fast, way more than highly experienced pros.
Like you mean from YouTube money?
SELLING"*
Oh, commission sales?
I already have an channel and I can get money by donating
yes
To be honest also making animation in stable diffusion is hard
you need to spend your money on a perfect gpu
if you made the pose for control net and then a sketch /scribble over it for some details and then primitive 3d objects around the scene that the AI converted into a full image. Nothing would be in that scene as it is if it weren't for me. Same if I train a face and character look and design, that would be mine
you mean since I don't use original artists' work
does anyone have any idea as to why there are blacked out characters in github?
Huh? I'm talking about instruct LLMs. nvm. Discord jumped to a different message when I clicked to see which one you replied to.
I think the copyright stuff said that you couldn't really protect the output although you could protect the methodology?
oh, cause I thought people were talking about images, animations etc
hard to say until it goes to court
a discord classic
Yeah discord had a skill issue and jumped me back to the conversation that you and I were having about LLMs being used for writing prompts when I clicked on it.
Fiverr has people selling animated AI stuff, you can check that out and see what it's like
could you give me links ?
https://www.fiverr.com/search/gigs?query=AI animation&source=main_banner&ref_ctx_id=36c7608d02273a1ec18b224e3a3474ec&search_in=everywhere&search-autocomplete-original-term=ai animation I just typed in AI animation
They said you may be able to protect things like a particularly creative prompt, but not the images output from the prompt. You can also protect a finished product such as a comic that uses AI generated images in it because that requires human authorship to assemble them into a comic. If you came up with an AI that did all of that for you then the comic wouldn't be able to be protected. In both cases the images themselves would still not be protected, so nothing would stop someone from taking apart your comic and making their own, using your images.
thanks for the clarification
so you mean this something is normal ?
Why not?
I take it back, recepmengi is sick lol
the first one I looked at was omar37
I was like 160$???
for this??
@supple raptor but if I posed it a certain way, if I used a specific depth map, texture map, if I used a specific Lora of textures or clothes I'd made myself
what then?
and what if it started out as a drawing that was then trained into a lora and used by AI to generate images? Is my character no longer copyright protected?
If it's lucrative, might as well suck up the money from AI
cause then every author is in for a rude awakening
While its profitable
elaborate?
True. Customer is always right after all
I draw a character, I train a model in making my character
the AI makes my character
do I no longer have ownership then?
you would have ownership of the character given they were distinct enough, but not any image outputs I believe.
Ultimately, none of what I'm saying applies to you unless you're attempting to register a copyright in the US. Based on what they said, no amount of finetuning models using any method has any bearing on whether or not the output a copyright registration for the image would be accepted. Maybe using a depth map could be enough to claim human authorship, but I don't think so based on the language they used.
you dont have to register a copyright for it to exist
shrodinger's copyright
you do in order to defend it in court.
the moment you make something, its your copyright

also, we can just lie and say an AI didn't do it
which is the only time a copyright matters
lying is bad
Then you're opening yourself up to pergury charges
in court we need only show that we made it first, as such registering it ANYWHERE is fine. Having the bank deposit a copy to never open except in the case of a court case
Meh, I won't say it's not AI made
I just won't say it is either
There's nothing to say you have to say how you made it, unless it's in court
when you register the copyright in order to defend it in court you have to claim authorship which means you are claiming that the AI did not make it.
People who know how to do it often undervalue their labor. But I think if people are willing to pay, let them. Only a small amount of people actually know how to use AI
no, I don't need to go to any official instance or agency to register a copyright
that's a trademark
just 30 mins in deforum
boom ez 100$
but I am too lazy
🤨
nah actually i agree with that
I might see if I can get the ball rolling once I've recovered on either a VN, a novel or maybe like a pseudomanga/novel that I eventually release on patreon or something not as shitty
most people aren't gonna be downloading shit
Join the club buddy 😏
GN
Goodnight 
Maybe if just a few people get together, it's possible to make a decently bad animation studio, but still decently
cute!
2d artists actually have it good; they can fix problems with AI art that most AI users can't. Like in the pic of the girl with hands up, the elbow looks like the back side of a knee, and fingers aren't too bad and easily correctable. They should use their skills with AI instead of complaining imo
I think the problem with that is that it comes from an AI user perspective. AI and digital art are related but they don't really have transferrable skillsets (well as much as skillsets can be transferred lol). From my personal perspective, AI is fun but just not what I want to do
Not to mention, you'd have to match the AI in aesthetic/rendering skill already or focus on adapting to an AI in order to correct it's mistake
then it's like, what sounds more appealing to you: Drawing something you want to draw, or being a error corrector for an AI.
Not to mention the difference in expressive power between drawing manually and using AI, a manual artist is more capable of making miniscule decisions in their work but gives that up to an AI when they use it
of course, macro decisions are also arguably given up, things such as composition, light sources, certain background details, clothing
pose
well pose has controlnet now lol
you are multiple artists?
Yes. But if you generate what you want from the ai, it's basically what you wanted but faster. Stuff like ControlNet gives one the control, as does multi region prompter, etc. As for things that require marching the AI, this is mitigated by simpler styles. That's why I don't generate gradient shade styles, but simple 3-5 tone styles
But you're right, a lot of the fine control is gone, the technology isn't there
is http://127.0.0.1:7860/docs working for anyone?
IMO it's again, sort of an AI perspective thing. "it's basically what you wanted but faster" makes sense when you've never touched pen to paper before but makes less sense when you're used to getting what you wanted
that's your local instance, no one else across the internet can access it.
what is )multiple artists) ?
(I think you need to launch with --api)
oh mb I see
you said "artists" not "artist"
oh I didn't notice
you aren't running automatic1111
oh should I run it to check for you?
if you can, but add --api in the bat file
where exactly ?
That said, I do think it can be utilized in it's own right. I've always been wowed by AI backgrounds, and since I hate drawing backgrounds but want them it makes sense enough. Ultimately, I'm not sure if I'm ready to take the plunge or even give up that much creative control
the better solution would probably be referencing since I'm always wowed by AI backgrounds but disappointed upon zooming in
edit it
so ?
add --api after no half
also I am pretty sure no half is like bad or something
just makes stuff slower
Woaaah that is very cool!
anyways, I'll just go off videos and random pieces of stuff I find on github instead of the doc of the api cause it is not working
export COMMANDLINE_ARGS="--xformers --theme dark --no-half-vae --api"
what this?
open pose, add an accessory to the top of the head like hairbow, or add mid shot.
zoom out a little 
can't open pose because it would defeat the purpose of the procedural generation, maybe I could add hairbow
maybe try upper body
already in prompt

still
I don't wanna see her from the side
profile will turn her to the side.
no
try that 
but it removed a lot of the upper body
increase weight on upper body, keep portrait but maybe even decrease the weight of that
I tried upping the weight on skirt and now it cropped her head again
it's like it's a tradeoff
70% and still no mutation
outpaint?
I want it automated
is outpainting not automated? Just keep the prompt and add more to the top
and outpainting takes too much time anyways
hard to use in api
and takes extra time
mm i see
while just using the LLM and sd itself takes up a ton of time
anyways what I did is just take her outside
now I can get stuff like this after removing the background
it can even be used in a game
wow, weird style, but i like it
I feel exactly the same about, is pretty interesting style,
um, it's my own mix model, it was mostly prompts that got me this style.
looks like the style of suda51
oh
the upscale is at 81%
i'll upload it if you still want to try it out.
I want to use it for my animations
you may don't accept so no need
unlees you allow me to use it for my animations
i already i can't see pixels any more in the preview
it's fine, the issue is i dont know if my model would be good at animations
they are absolutely good as a base
did you guys finetime any anime models? if so how did you handle the faces? in my tests the faces always turned out very messy
nu
she got isekaied by the mystery machine too?
She got there by herself
the most important thing is the poses
i'm starting to have a plan
the style is so like perfect world






