#🍥|anime
1 messages · Page 129 of 1
they should rename to closedai
like wtf
kek
they are also the ones pushin hard for gov to regulate all ai probably to get an even bigger advantage in the market
yeah, regulate everyone. EXCEPT US.
they're a real piece of work in that regard
hm people
@wispy canopy you might know it well
I'm comparing some prompts for quality like "4k", should I pay attention to hands if they are borked cause it always mostly is >.>

nah, they're gonna be borked either wway, so no need to pay attention to it
. o (
)
needs more pointy
i managed to get characters working and stuff without needing to use sillytavern but i might stilltry it
give this prompt silly
dalle or sdxl
1.5
Something like this
chibi girl with white long hair and hime cut with blunt bangs wearing oversized white sweater and azure eyes holding a board written give prompt on it, anime, 2d
A huge advantage of SillyTavern is group chats. You can add multiple characters to the same chat and they can respond either to each other or only to you depending on how you configure it. The interface has a lot of other very nice QoL like using an arrow button to regenerate a response, but keeping all of the previous versions so you can select the one you like best to continue from.
i might try sillytavern i was just concerned about system resources. just the textgenui with a model loaded pretty much filly my gpu and sysram
if i install sillytavern on pc do i get to use it from my phone
silly is just a gui -> it's very light
like ive been accessing the webui from my phone over wifi
crazy
yep, silly can be accessed from your phone too
also wondering if id have better luck/compabitility if i use linux subsystem
ok so silly is like a replacement for the default webui
if i get my access removed for this i will kill mys
multiple characters and stuff would be cool, problem is im only averageing 1-2 tokens a sec
so its like a 50 wpm response
i think i could tune the settings for better performance though
or slightly better probably not much better
Yeah, I wouldn't do any huge conversations, but 2 characters that have low chance of responding to each other shouldn't be bad.
nuhuh false info
getting banned for saying you'll kys is more likely than them saying anything about that image.
oh, can't you just make a new throwaway microsoft account to use for dall-e if you're worried about that?
Quite the sword there.
did you redeem some of your Microsoft Reward points for them on accident?
no
not sure then.the icon for my boosts are gold though. not sure why yours are silver.
its buggin tf out
try to refresh the page I guess
What are you comparing/doing
i managed to get mine running on a free aws ec2 instance
booba:1.3
no, i haven't pulled the latest version though.
It's only working with server reaction, but when i dm the bot, it doesnt give the metadata

testing merges
in DM you'll need to right click the image > apps > view parameters
ahhh
wait, DMs have a channel ID https://cdn.discordapp.com/emojis/1028509125044359259.webp?size=64&name=MomijiWide&quality=lossless
ahhh
just right click the DM with your prompt inspector to get the channel ID and add that to your channels list.
I should use extremely detailed:1.4
that probably work well
no wait

what does that even do
you're also more than welcome to abuse use Prompt-Dog in DMs if you don't have a server that you actually need your own bot for.
yup yup
i was just playing around with starting bot in ec2 instance
might wanna play around with doing stuff with telegram or even ||abusing cookies|| with bard or chatgpt
That's what I'm doing, trying to see if it does something (Its testing each prompt on the same seed stuff)
the above method works, but will be a little funny, after pressing the reaction, nothing happens, but at least you can still right click and get prompt

does the words in () even do anything
or no
If you want your DM with Prompt-Dog added to my channels list I can do that next time I update it. I'm gonna add my own anyways so I'll already be editing the file.
yes, read the instructions. there are a ton of features that you wouldn't know about if you don't read https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#attentionemphasis
on A1111 at least they make prompts stronger
for example wet might not get the image to look wet in some form
And (wet:1.5) will probably start raining 
interesting
You almost don't really wanna have stuff at such high strength tho
@lyric zealot more thick 
but depends, because some keywords do really need strong weight
@supple raptor if a model already has vae and I do that thing to bake a vae in it, does it go bad ? 
nope, it just overwrites the baked in VAE with the new baked in VAE.
do you think I should do that
the colors look a bit dead
another one
I'd try just manually loading the VAE you want to try with the model first.
--opt-split-attention
i had this as a launch arg but i forget why, itook it out and dont notice much of a difference
good idea
here's the description for that argument Force-enables Doggettx's cross-attention layer optimization. By default, it's on for CUDA-enabled systems.
if you're using xformers or SDP then --opt-split-attention will do nothing
am using xformers
ah yeah, then you can remove the --opt-split-attention xformers is much more efficient and faster so that's what you should be using.
yep, that fixed the splotchiness that the baked in VAE on that model had.
fixed the colors too
This is the version of kl-f8-anime2 that I use. it's the same one that I baked into Kakigori. I reduced the contrast by 10% on it which solved the color bleeding issue that the original one has on most models. https://huggingface.co/nubby/kl-f8-anime2-blessed/blob/main/WD1-4-kl-f8-anime2-bless09.ckpt
that's a VAE, not to be confused with model
L
take this to cheer u up
still using dalle?
no this one is breakdo
what that
ok
you love that beach look 
this very good gif
anyways I go slep
GN
🛌
When the computer is turned into sleep mode, it saves the current context in the RAM and it makes sure that the RAM is on.
Hibernate saves the context in the hard drive and as soon as the computer is on, it loads the data from hard drive to the RAM
i go hibernate

i have fallen and cant get up



Beautiful results there.
would prompts care about uppercase letters
nothing much, just curious
proper nouns
i cant seem to get it to work
i posted my error messaged in the animations topic
looking at my output dir u did get it to work at some point in the past and have one output but thats all
RuntimeError: CUDA error: invalid configuration argument CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1. Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.
is it easy to use
it doesnt have nearly as many settings as deforum
yeah it's pretty easy in Auto1111. A little more challenging in Comfy, but you can do longer gens and more stuff in Comfy.
But like this in Auto1111 took me about 5 seconds. 😛
could you be kind enough to send a link
need a new hand
to what?
it's in available extensions, just install from there 🙂
openai would not let this image be generated, i'm sure
ok
.how do i fix animatediff
make it work
Why not
big something goth gf
that something word is the thing openai would be like "oh no, muh censorship"
fashion choice yep
what was your prompt ?
cute vampire girl
oh
actually they kinda look like "cute" bat ears
and not cat ones
probably biased towards cat ears cause it do be like that
tru
my feet hurts
thats peak human evolution
🦶
Red eyed Homura Akemi?
just grab the model and the extention and you're good
this model is super sensitive to cfg
something must be wrong with my install or settings cuz it doesnt wanna work
does it not work with tiled vae or anything like that
some extensions break animatediff try to disable them one by one
just keep it simple
new transformers movie looks great already
lmao
want a ride?
ok jump in
no thanks i got my own
but its a free country
this is the horny police, stop right there
alright
what the hell is goin on with them cars

fine for me 🤷
Girl : ✅
Riding a Car : ✅
AI: good enough ship it
the girl is gone
magic
under where? 😳
car
immersive
moar audi
i wanna ride that
oh yea the car is nice too 🤠
i need a hand
real r8
Definitely calling the police on you

🤑
swear on my YEEZY's that this a real one
no
real recognizes real
car is too skinny its not real
forzw
GTA
this one is thiccer on the back so its def real
the audi driver before ramping off the back of a semi truck:
@lyric zealot hewo
a6 fr

ok
she's isekai-ing
ok
when u drivin your r8 but a pedestrian appears out of nowhere

had no idea my exhaust was a woman
does smoke come out of the mouth like a vape
moar a6
this one is wide so therefore real
why are her legs phasing through the hood
i ran over her

greetings there
no problem just get inside

u can ride this anytime


you definitely lick women feet
Eww
no kink shaming 
who said i was

I need to play around with dalle
ok
what is that angle lmao
kissing her feet angle
rols roce
ahhh i tripped on a rock
She knows how to drive.
ong
you like my new audi?
u right,feet too small
real
never skip foot day
Why not?
Because it straight up dies
Why not run it online?
You're referring to Google Colab? Yeah I haven't done that
I have a 30XX Nvidia GPU, I don't know why it can't handle SDXL
It should be able to
its ok u aint missing anything good
I mean, I can get amazing results regardless
Yeah, its super easy, just go to the modified version of Fooocus, Fooocus-MRE, paste your models and hit play (only 1 cell) it just runs, and all the memory optimizations from Comfy make it work with perfection. Just do this for local and you can run it easily too (local is harder to download and load models though)
I am tempted to get an upgraded setup just for SD alone but I feel like it's not enough of a reason
I've thought a lot about it and I'm not doing it. I'll just stay with colabs and rented computers. SDXL still has problems and it will take time to get where it should be. Many are realizing that 1.5 is still better in a lot of fronts. You are right, there is not enough reason to upgrade to a new setup
@merry tendon @trim trench Do either of you roleplay often sometimes?
this AI race is stupid where each company competes with each other, there should be a commonly accessible repository for everyone to access and build upon that can help the collective mass
the root of the problem is the economic model we have
this economic model have many stupid things, just to split and weaken forces
any way to prolong the skirt...like inpainting?
I have tried several models to inpaint but they cant work well
img2img, and by extension normal inpainting, are not good at adding things that aren't already there because the way they work is they start with the input image and just add some noise to it, then diffuse again. controlnet inpainting, done on the txt2img tab, would be much more suited for what you're trying to do.
ok
why are the foreheads like that😭
Hi very nice anime pictures.
pleae
that is dalle3
don't say it's good (even tho it is)
this makes sad
Thanks
It is not Dalle 3, it is SDXL ^^
omg
no way
frfr???
I mean ye kinda lookss like sdxl
sdxl andd dalle3 look pretty similar
damn
xl is just weird man
I find it somewhat amsuing this even works as a prompt and it follows it
I haven't tested Dalle 3 but I think people are overreacting\overhyping it. I've tried some prompts they've used on reddit comparing SDXL to Dalle 3 and I've reproduced several results or made it better than Dalle 3 using SDXL. Of course, Dalle seems easier to get good results and on several prompts it probably understands better. But today with Ipadapter Plus (instant Lora) and variations SDXL can get really good.
As you've said, they're pretty similar
If you look at the main Dalle 3 example with the girl and the old steampunk man and look at the eyes at a distance, it is pretty bad just like SDXL. MJ or even SD 1.5 gets better sharpness at distances
Of course, Dalle 3 is better than base SDXL; a fair comparison would be Dalle 3 vs the best SDXL models available for each case
yeah, it pits people against each other, there is inherent distrust in this economic model
im amused by its design detail, looks good
ty
Meowning
I'll sleep for 10 years and wake up when AI text2img is solved

Have any of you tried using "AND" in your prompt?
The normal AND implementation is kinda dumb and not very useful on it's own unless you just want abominations. If you're interested in the concept, you should use this https://github.com/ljleb/sd-webui-neutral-prompt
Wait is it really that bad
I saw the example mention like cat AND dog

Yeah, basically the two prompts fight over the latent space and it rarely actually combines them in a coherent way. It can work sometimes, but usually it just makes everything worse. The alternate implementations in that extension fix the problem in different ways.
thanks for this man, animatediff broke today while working yesterday, couldnt figure out what happened
That looks pretty cool I will try later o:
hopefully we will see some changes before that, the problem is technological potentials are delayed by systems that come from century old mindsets
the political figures and law makers are doing a crappy a** job
and most of their contributions to society are non intrinsic either directly or indirectly in defense of outdated systems
Lol I forgot about this server, what's with the ping, what got you to ask? Did you mean to ping me?
THANK YOU SO MUCH FOR THIS
mre? foooocus?
Man, why are hands so hard to generate properly

Did you try different textual inversions to see if they help ?
I've played around with them, I mostly have all of them on but then they started to go really weird on me, tried some on and off and man idk
Seems like nothing works perfectly
Right now I'm trying aDetailer
But it only enhances the weirdness
@past junco can you show me your settings, resolution and stuff
I do however think inpainting might be a solution
Yeah sure just give me a sec
I'll grab a random image output and it's text file
(I'll spare the file, just the needed info inside it)
(((anime style, cartoon style))),
(best quality, masterpiece, colorful),
(graffiti background),
((detailed clothes, detailed printed clothing, detailed face))
((four late teens boys), (tall boy), thin boy, realistic proportions, cute face, cute fur, neko, cat boy, cute animal ears, cute clothes)
(happy, joy, emotion, expression),
(depth, perspective)
(detailed and colorful hair)
(full body)
((detailed hands and fingers, anatomically correct), (holding hands and hugging, group photo, group shot, dynamic pose))
Negative prompt: (worst quality, ugly, monochrome, greyscale, dull),
(((disfigured, disformed, mutated)))
(missing or disappearing hands and fingers),
(less than two arms, more than two arms, less than two legs, more than two legs)
(blended or merged upper and lower legs, blended or merged bodies and legs, merged feet),
(ugly arms)
(large adult)
(text, logo, watermark)
aid210 bad_prompt_version2-neg By bad artist -neg easynegative EasyNegativeV2 lr NegfeetV2 ng_deepnegative_v1_75t verybadimagenegative_v1.3
Steps: 100, Sampler: DPM++ 3M SDE Karras, CFG scale: 7, Seed: 3578527912, Size: 1024x1024
VAE: klF8Anime2_klF8Anime2VAE.pt, Denoising strength: 0.9, Token merging ratio: 0.1, Token merging ratio hr: 0.1, NGMS: 2.5, Hires upscale: 1, Hires upscaler: RealESRGAN_x4plus_anime_6B
(I spared some of the unnecessary stuff)
And lmao I know, I should not be trying holding hands
Apparently holding hands is forbidden with AI generation because of the monstrosities that it produces
no monstrosities ... just nooddle fingers
does it put their fingers through a meat grinder
shiny
Basically yeah
holy wall of text
must be painful
I've taken out hand holding, particularly I added "interlocking fingers" to the neg
@past junco I don't think its your fault
When you're doing images like that, with people "far" away
Oh yeah
details like hands and eyes will be botched almost all the time
yes its very hard
Especially the hands and face
Eyes are hit the worst
That's why I'm trying to use aDetailer
It's kind of working
Kind of*
If the eye is extremely bad
It doesn't help
adetailer does "zoom in" and re-do things like inpaint automatically
I did have some success with inpainting eyes manually
I might try that again, but to me it's always a gamble because I don't know how to correctly configure my inpainting settings
Exactly what I mean by eyes and faces being super weird at a distance
Also, whatever you do... don't try to do group shots with aDetailer, it takes a long time
Those are very cool
wish my pc could handle this much regional prompting
@past junco have you tried the restart sampler?
I have gone through extensive X/Y on samplers and this was the one that I found to work the best
But
I didn't use all samplers
Alright, I'll try that
I used to do around 30 and 40 steps, until I did some much needed optimization tweaks which sped up the process
Just need to wait for this one to finish and I'll try that
Redo this one with restart
I wonder though... if the sampler would still keep the style
:D
Alright
samplers don't change much in the image
they change how they are made
(I have no clue me dumb
but its the engine of sorts that makes images, they all try their best to do the image correctly)
I've noticed big differences in the samplers I use
Some aren't very good at details, and some pick the wrong things to be detailed
But mhm
Restart as the name implies, it restarts the image in the middle of making it
and makes some details better
I didn't know what all of them were like
it might help with hands
I just looked online to see what people recommended
DPM+ seems to rule
I chose performance over speed
The only thing that gives me OOM is 1024x1024 with upscaling, and SDXL
Sometimes I run into OOM with absolutely minimum, that's when I know that I need to restart my laptop
Holy crap this one came out great!
If you ignore certain people in the background
is this restart
oh
On too many faces
You using the hands model too ?
I'm using textural inference for hands, that and maybe some extra neg stuff for bad hands, bad fingers, etc
(I don't know why they weren't in these prompts, I usually have them)
I'll try doing what you said with the earlier image
This

what
No clue what I'm doing, I'm trying that neural prompt
Should I leave aDetailer on?
Mhm alright
magical tree forests, eternal city
AND_PERP
electrical pole voyage
AND_SALT small nocturne companion
AND_SALT
electrical tornado
AND_SALT electric arcs, bzzz, sparks
Are you satisfied with your answer

It did something
It kinda messed up the guy on the top left, I liked how he kind of looked before, and besides his hand I'd say all hands look better
Not perfect, but better
restart kinda of does 2x the work
it restarts haha
and takes longer per step to do things
So I said half the steps
which is what people were using with restart
Yeah I figured that was why
And I believe I was wrong, the steps doesn't change the guy in the top left, I think it was just the way it samples
from what I know
samplers can reach the conversion state or something like that
where the image won't improve after certain steps
like it did what it wanted
menacing 6 fingers
Lmao
now die @sharp swan
It's taking it's time with this one
Is it just me or does her left eye look different to her right eye?
its all scuffed
I can't do much upscalling
also just trying to understand what this thing does
This is just crazy
I LOVE IT
also thick thighs by luck
I should prompt for more
I do love me some Yugioh
Better, but the girl on the very far right has a really messed up hand
You know, as a kind of artist myself, trying to draw something like this even for a professional artist, a group shot like this would take a long time
You drew that?
yes
That's cool
Aw
I've been told that tracing is really bad... but it's the only way I can really learn and practice
My crappy art...
tracing is like bad for learning yes
No
It's great for practice, but only if you practice without tracing too. So you have to do both to really learn
I have a lot of artist friends
And uh
the only thing I got from the original for my paint cat was the circumference of the eyes and where they were placed, the rest is all done by me
I can really feel the weight of AI art generation...
cause I wanted the eyes to be perfectly round and cute idk 
The eyes are great

I don't draw much, I've done it here and there for years (I'm old) so I can do stuff but slow as hell :p
Apparently adding "bad hands, bad fingers) to neg prompt can really alter an image
Aw, I too also feel like I'm getting old
But I'm not really
yes, even a single space in the prompt can alter the image, SD will kinda of "remove stuff from it's brain" before thinking on what your image will look like based on the negative prompts
I've noticed that sometimes it completely ignores a neg prompt
And I think I might know why, inherited or associative prompting
I can't help but see problems in generated art
Help me lol
a 4090ti might help you with that
Stable Diffusion is trying it's best
and me too
the right most person Lmao
There's a lot of stuff happening with him
Including his legs
But the main issue
The hands in the middle

how long do these take
They are taking a few minutes each now, for some reason
I think it's because IT KEEPS ADDING MORE PEOPLE
I just want four guys
That's all
Right now it's giving me even more
I want to see it but like
After this
I'm adding "((people in background, more than four men))"
I love this, gives me DBZ and Akira vibes
i just think of random words and add it to prompt

Have you thought about Dynamic Prompts?
doesnt exist
Ah I see
Yeah I got a lot of wildcards but haven't taken the chance to use any of them
Except for negative prompt ones
bentley ganmgin



Also, that some nice interior for a car
very familiar
evem maor beln tgy
ok
goodnight 
Goodnight beautiful people
i am not but 👍
girl, fox ears, white dress, sky background
AND_PERP
wings
AND_PERP armor
s/r to replace and_perp into the others
does it replace both in my prompt ?
or only the first
should gives you an idea
smort

@lyric zealot its replacing both
I'll try a simpler one
maybe when im more interested in trying to understand that, will i go and read it up properly
its kinda of like regional prompting
I had another examples like "Girl, AND stuff and sun rays"
I believe AND_SALT does the stuff where the prompt isn't working as much in, like in that image above the wings are behind the character where it doesn't matter as much
AND_SALT with the sun rays in that other prompt would place sun rays behind her
and the other one would place it on her
but no idea other than that
complicated
Uhh so you trace your ai art?
Well i dont quite know but
I think artist would say you would have MASSIVE SKILL ISSUE


Is there an anime clothing guide anywhere? I'm trying to create an image with a women in a large gown dress but nothing's working.
I don't know of any
Someone should make one I think
If I knew good prompts I would of that's why I asked.
and If I discover anyone I will compile them
true 
there is
search for boru tags wiki or something like that
there should be one for clothing
probably nsfw stuff
Roleplay related reasons actually, is why I did it.
Are you a roleplayer?
how's everyone today?
im still baffled with dall e 3
oh?
doing great

but in all seriousness, it's actually cool tho
no seriously, i wish SD were at this level man
yeah those are neat
i like how it's able to get text right more than sdxl
and those hands, and holding things...
the "model" is probably like huge huge
the only "huge" in here is the anime tiddies

I am but I still do not understand what got you to think to ping me?
I stumbled on posts in the server that say you were, is all.
Oh, I mostly do erp though, you do that kinda stuff?
No kill that one
Looks like a vtuber

You did tabletop RPs too?
Because the answer is yes I sometimes do ERP stuff.
AI good
Non AI bad

well, tbh, i don't even watch her/know her
just know that she existed because the author of the model used her as example

I'm lazy
What is it doing
since it's hosted on the same network
the one where you change model after certain steps
No just erp, no table top stuff like DND. If you do erp I might want to DM and see where this goes if you want to?
I have so many images I've saved and they're all disorganized and... ugh
Plus, the stuff I saved early on is not really on par with the quality we can get today
And what are you achieving with it 
hold up, let me pull up the grid image i did earlier
sure do
8000 of em? 😐
the only lora i was using it's for the character
that's really not a lot compared to others tbh
are they sorted or in any kind of order?
oh well... so you just keep everything
I was curating for a while but it just got out of hand
Sent a direct message.
well, for some event, i might create a new folder for stuff generated for that specific event
example for the old easter event that some of us did
i have a whole folder structure of themes and characters
but it's just kind of a mess
and it's got a lot of old stuff... like -- do I really need to keep this?
Which one do you think is best in this
im currently using 0.6
i think it really depends
my refiner model adds really thick outline
and more anime-y
but the base model that im using, can have some semi-realistic stuff
so, im mixing it


Ball of power


Are you regional promoting
nope
i'm too lazy 

I know how you can improve them!!
not with regional prompter 
With this one trick in your prompt you'll get much better images
(Thick thighs, showing legs)

👍
Hahaha
@lyric zealot how old are the gens you're going through
how long has 1.5 existed for o:
maybe like 6 months back
or something
probably for like a year now, i guess








