#🍥|anime
1 messages · Page 202 of 1
interesting knife design
might want to try different artist tags too
veg

your picture looks like a noita character
better remove it, if its not working 
dont forget about your daily dose
skibidi toilet
yes i tried the 4 steps and 8 steps lora with animage XL,played around with different weights and the different samplers,only got shitty images and deleted it
ouch
i got some good results with it
but i mostly generated realistic images with dreamshaper lightning
thats one
yea for realistic images is alright i tried with anime checkpoint and looks like it doesnt like it when u add other loras to it
yeah none of the sdxl are good for anime
but im testing out anime lightning model now
basically all 4steps, 2 cfg, no hires, sampler dpm++ sde karras
its important to use that sampler or you'll get messed up image
ahh but that's 1.5
oh right it was this one https://civitai.com/models/266052/asyncs-mix-xl
im trying out sdxl lightnings
ok checking
that's not lightning btw
but thanks for sharing steven, im only trying out lightning version of sdxl now
and then stability ai announced sd 3.0
and we got sd cascade coming too
this is already out
no u can just apply the sdxl lightning lora to any xl checkpoint like lcm
yes, im waiting for a1111 supported release
ahhh okay
tried with an extension tho but its missing some feature i suppose
yeah, i heard about x-adapter .. but waiting for the plugin support for a1111
its not x-adapter is this one https://huggingface.co/ByteDance/SDXL-Lightning/blob/main/sdxl_lightning_4step_lora.safetensors
some users are already trying them out on comfy but im not good with comfy
also forge already added support for lightning
it's just a pixel art wizard
although that is what noita characters are more or less
yep i have that bookmarked but if you check civitai they are already releasing 4 step fine tuned models
oh yea the finetunes are better i tried with lora but it just comes out bad
try the new dreamshaper lighning you might like it
?? autism confetti is good for anime
smol sus you are very creative with images, you should try them out too
pony ?
yeah it's based on pony
i will eventually try them but focusing on lightning now and waiting for sd 3
yes
how does that works? Just get a SDXL lightning based model?
yes, you just use the model like any sdxl model but with few changes in the settings like sampler dpm++ sde karra, steps 4, cfg 2
takes couple of seconds to render 832x1213 image
oh, unless you meant with the lightning lora
can i use it on any sdxl models? or it needs to be SDXL lightning?
weird bug, my upload to discord stays stuck
this lightning version of the anime model im using atm isn't too bad
I tested it on confetti, and it kinda worked with the 8 steps lora version
im not sure about using the lightning LoRA model, steven might know better, but im using fined tuned lightning model
so download lora, put it in prompt, change steps and cfg, use the sampler, then done? 
seemed maybe a bit worse than a 20 step version without the lora
yes
looks awful on animage and when i added a char lora it was worse
there are 2 step lora, 4 step, 8 steps, depending on which you might want to run you'd have to download a huge chunk of file
but lightining trained model is 6.4gb
apparently the official guide said to use a different scheduler, but I don't know if that is necessary, I just tested it on comfy a bit
yes the sgm uniform scheduler
let me try this 
that pretty much sounds like LCM lora... i wonder if its the same process
it uses ADD instead of LCM so it should be better
so say with 4 step version you'd have to download two big chunk of file

just the 4step lora
8steps but 10 hires fix cos i stoopid
8steps and 4 steps hires fix
so for safetensor that could be any sdxl model?
yes the lora works in any sdxl model
ugly 
cfg 2 might be better
try w/o hiresFix
i mean you could hiresFix but not necessary for quality
thats the lightning?
are you using comfy? also what model
from about a minute of generation to 25 seconds
nice, so you have your new gpu ?
autismmixConfetti
with forge, there's new update on sdxl lightning support
my gpu is still the old 3070 8gb vram 
you're getting better quality than me on such few steps
ahh ok
the sampler i am using is Euler A SGMUniform
that was the recommended, but i found using cfg 2 better
this is for forge update btw
the reason I'm using dpm++ sde karras specifically for the lightning models that im using cause they trained them on it
im gonna go ahead and download the lora in a bit and try out other regular sdxl models
finally
only that
the lightning 4 and 8 step loras seem to work with autismmix_confetti, but I'm not sure it's worth the trade off, especially using the dpm++ sde karras sampler
the 8 step at 2.5 CFG is the closest in quality (for tested prompt) but with the dpm++ sde karras sampler takes 4.7 seconds for native 1024x1024 gen
compared to euler a, 7cfg at 20 steps which takes 4.1s because the sampler is faster, 30 steps (if needed) takes 5.9s which is plenty fast given the much better quality
on top of that, i can use tensorRT and reduce the time to 3.2s without needing to destroy my quality or load a lora
but, like the other similar projects (turbo/lcm) they're mostly good, currently, for lower end cards, i'm probably not the target audience
that said, i did make a turbo merge of SugoiXL because I merged that model specifically for simplistic flat shaded chibi models, so quality overall isn't a high standard aside from clean colors and lines (something lightning loras isn't giving me) and coupled with tensorRT I get 1024 gens in under 2s, so it does have it's uses
now back to playing helldivers 2..
send to me
Dunno where they revealed it 
#📣|announcements
maybe


I'm hyped
looks cool I dunno I want it 
me too
it's cascade? or something totally different? and it's bigger than xl?
weird no official discord channel for it?
Stable Diffusion 3 suite of models currently range from 800M to 8B parameters
quite a range
8 bilions?
insane

sd2 incident 
are you using a dynamic engine with tensorRT? I tried it with xl and it was barely any faster than normal xl
yeah, dynamic engine
for me it's a difference of ~6it/s without and ~11.5it/s with trt
i dont know what kind of performance to expect from different GPUs though, I have a 4090
in the next few days I'll be setting it up on my roommates 3070ti for testing, i should also test my 3080ti
I don't remember the it/s, but on my 3090 I got from ~8s down to ~7s or something similar
strange
but i ilike the og control panel
set COMMANDLINE_ARGS=--allow-code --api --xformers --enable-insecure-extension-access --listen --update-check
any cool new cmd line args i should try
couple more questions if you don't mind. what resolutions was your engine, I think I had mine at like 512x512-1560x1560 (or similar). Also have you tried to use it with comfy?
the command line args rarely change, but if you're exposing your webui by port forwarding, then I recommend the auth flag so it asks for a username and password. Don't worry though, since it only asks it once per device and saves it in memory, so no need to input it every time.
i'm using it with sdxl, so 768 minimum, 1024 optimal, 2048 max for both height and width
on one of my models i also have a second dynamic engine for 768 minimum, 1024 optimal and 3072 max (largest hires fix without OOM)
also no, i don't use comfy, I'm using the a1111 extension
oh airhgt
thats good i hate having to login every time, my router wont save my login info like that
i want to get a big red button on my desk that launches the webui
nice work by all.
slurrp
charming lol
anime?
is it anime I mean ,
going for retro anime poster print
think I nailed it
prompt
by eli otto,, 1girl, long hair light purple hair, curly hair, cute, eyelashes, sitting, white dress, pink ribbon, belt, pink flats, white thighhighs, beautiful 8k wallpaper, outdoors, nature, tree, bush, flower, rustic, [monochrome::0.6] sketch, graphite \(medium\), detailed, aesthetic, beautiful, beautiful color, amazing quality
rating_safe
rating_explicit
rating_questionable
Not rly sure if rating_questionable is used
what does dat do?
I think its used for questionable nsfw in booru sites
rating_safe is only SFW rating tag
@wooden coral

it was a $50 gift card


And how is your week going? 
Got kicked for afk in loading screen
Im going to monoliths now, heard its better there
Yeah you don't get any problems with monoliths
I just haven't had time to play so I just yesterday got to lvl20
But I have vacatation next week so I can play all day
im 57 now after lagon, but loading times are a pain
What class are you playing btw?
Falconer 
Birb
My goal is shadow dagger
I'm trying to do sorc build but the game is like...
Last time i played before 1.0 I played a necro with the buffed up skelly
Last time I played I became purity of elements
turbo xl models work on forge?
I'm doing all the skellies

Last time I played I just did some meteor crit on meteor was stupid and fun
And before that was many years ago


Those weaver will uniques are kinda fun
Got a minion boots but the added mods are crap xd
anime gif 🗿
*anime 🌚

Pretty cringe then
I'll delete it

I will try and find a newer discord version that commands work
Without the UI being garbage to go through
skull issue
It's for blind people now
🗿
Can hardly fit 5 messages in one screen
on PC there's a bug that makes everything massive, resizing the window should reset it
otherwise, theres font size stuff in the settings
I'm on mobile
They removed the font size option
There's a zoom now that I think does the same thing
It looks okay now

And holy shit is the audio button, right where the send message is, admitting
if SD 3.0 is not a hyped botch up, it could replace sd and sdxl
me
guys, it's just me, or forge fu*ked up xl models?
when i try to generate with it, i get TypeError: 'NoneType' object is not iterable
I've seen that happen when using LCM sampler or something is left blank that shouldn't be
or restart it and it might clear it
Does anybody know where in this server I can write prompts to generate images?
bots down
Currently nowhere. The bots are temporarily down. Pretty much everyone in here runs Stable Diffusion on their own PC.#1047610792226340935
Ok. Thank you! I wasn't been here for a long time
yep, just confirmed, the last update screwed sdxl on forgie
is that why sdxl comes out way to bright ?
i don't know, but i can run it without updating
after the update, everything is messeed up
i can generate with therything but sdxl
xl is working for me with latest forge update
I wouldn't think it would fix your problem, but I am running torch 2.2.0+cu121 and xformers 0.0.24 instead of the default versions.




i'm running it without xformers, the last update changed something
why auto always screws with me?!!!!!
should be all stand alone
As long as you have at least torch 2.1, there shouldn't be a problem. I would look at any installed extensions first.
the new copy was without extensions, just fresh from download and tha updated
forge, even if it's stand alone, uses pacages installed on python?
what error message? on auto I used to have wrong color for sdxl but fixed that after set vae to None
a very long error where it repeat all the prompt and in the end TypeError: 'NoneType' object is not iterable

is this svd or anideff?
it's svd
thats unfortunate but if you happen to have log then that may have hints to address
i open forge and generate something
maybe is an oom problem. weird, i didnn't have this problem before
Still on it
and now, when i change to another xl model, forge crushes

forge wasn't so sensitive to xl models before
Did you try opening it without extensions
mh...
maybe they added something that is weighting on the entire gui
the only custon extension is openpose
i doubt that can infleunce the gen and when i downloaded forge again without updating, xl worked again like a charm
Open pose broke UI for me in the past
Try deleting it in the folder and launching

i wasn't wrong, they are overloading the gui

Bad cat
i suppose they are burning the steps
i suppose i'll ditch forge and just stick to improve my workflow on comfy
i cant get animatediff to work
basically sacrifices quality for lower vram usage. mainly useful for upscaling higher than your vram normally allows.
nothing, even deactivating all the extensions, sdxl still gives me oom
just to be sure, i try same vae and model on comfy
hmmm
That error you mentioned earlier is not related to OoM.
Ah, there is an OoM above that. I'd try disabling CN for the HR pass since it likely isn't necessary there and will save some vram.
i wasn't using cn and in the last try, i totally deactivated the extension
i suppose it's a chain reaction
hmm. odd that it's still passing all of the controlnet arguments with it turned off then
that explain a lot
I just updated forge for the second time today and xl models are still working as expected on my system so the issue is somewhere on your local environment.
i think the problem is that the gui is becoming too heavy and ican't load xl models any more on 4 g
All of the "One Click installers"/.zip files for these webui's cause nothing but problems for me. I never use them.
yeah, sometimes they are very unreasonable. but the idea of deleting everything and starting from scratch?
i don't even know if uninstalling python, i delete all the pakages i donwloaded till now
The issue is that those .zip files use their own prepackaged versions of git and python and install everything to those folders, rather than using a proper python virtual environment so it makes updating the dependencies unreliable and far more difficult.
so SD 3 is already here?
nyo. only stability employees have access to some early versions.
sadge
one day, should find a way to reset everything from scratch
but formatting the pc is out of question
just make sure you have python git and python 3.10.x installed and added to PATH (don't install from windows store 💀) then run git clone https://github.com/lllyasviel/stable-diffusion-webui-forge.git from the folder that you want forge downloaded to. Once it's downloaded, you simply run webui-user.bat and it will build a shiny new python virtual environment inside the forge folder for you. Feel free to copy over your models, embeddings, and outputs folders to the new forge directory before deleting the old one.
oh, i just need to copy all the repo
yep, that's what the .zip is just with a portable copy of python and git and silly batch files to tell it to use those rather than setting up a proper python venv (which isn't even something that the user is responsible for doing anyways in most of these UIs)
the benefit of the python venv is that you can easily manually update the dependencies that get downloaded the first time you run the UI whenever you want with just a couple of simple commands. Or if something goes catastrophically wrong with them you can simply delete the venv folder and run the webui-user.bat file and it will redownload the dependencies based on the latest version of the requirements.txt file that's included in the repo.
oh
looks like they didn't do a good job integrating turbo on comfy
btw. Pretty sure using a lora was the reason for your problems with forge before. I just remembered someone mentioning that xl loras are broken on forge and have been since pretty much the beginning. I forgot and just tried using the lightning lora and got the same error.
Why do I get an error when I use the translation plug-in?
what is the error message?
traning a lora, and I thought it'll use all of my 12GB vram
but still 100%
can I do smthing with other 7.5GB?

prob reserved memory so a game would crash
nah not game, generate smthing
When entering commands using pika, an error will be reported, indicating that an empty prompt has been detected.
increase your batch size so that you can train in fewer steps
I put 512*512, 10 repeats, 5 epoches
pre lora was a trash
how many images?
hmm about 100 and 500 regularisation images
I still didnt try without regularisation images
So effectively 1500 images assuming no repeats on the regularisation images. that means 1500 steps per epoch at batch size 1 or 500 steps steps per epoch at batch size 3. Each step will be a little slower at batch size 3, but it should still finish faster than training at batch size 1.
yo could potentially do batch size 6 then based on your free vram.
eh!? I watched a video and he had 3090 and he said put 5 so I put 2 cuz of vram
hmm interesting
I'll try dat too
I feel like pre lora got trashed because of too many regularisation images
i generated with lora, before
Idk. Just what i heard. I don't really use loras much on XL. It could just be a certain type of loras, like locon or some other type that causes it.
Hmm. I'm not sure. I've never used Pika
python 3.13 gives problems?
yesh. use 3.10.x. I use 3.10.11
how do I get frocks?
its so hard to get a dress like neir frock
just a frock will be enough
not tight frocks
frock?
send an example pic here to see
cyuute
@dull jackal someone is tempting me with free moneys again
boop
thank god you deleted the feet too,otherwise my muslim friends would be offended
thanks fruit here's miku as payment
What model is this?
https://civitai.com/models/211628/soushiki The full metadata is in the image.
👍
SD1.5?
Well damn
I think I know what I'm adding to my amalgam model if it turns out good then
Soushiki is pretty easily one of the best 1.5 models I've used. DPM++ 2S a or DPM++ SDE will give you a cleaner image, if that's what you're after. I like the more painted look that DPM++ 2M Karras gives most of the time. Hands are more reliable than most models and it's pretty flexible in style with proper prompting. LoRA compatibility is very good as well.
I see
Gonna test out a prompt with it and compare it to my old 1.5 model's results
And then merge it anyways with it having more of the percentage
Man remember when Anythingv3 was the best thing to ever exist
And when models were like
8GB
When merging I'd suggest replacing your current model's CLIP with Soushiki's. Honestly most anime models could benefit from just swapping in Soushiki's CLIP even if you aren't merging them.
hello
im going through serious SD version crisis
and im wanting SD 3 really bad so i can put all the previous sd versions behind
SD has its limitations, sdxl too, and by the look of SD 3 teasers those issues seem to have been addressed
tonight i had a scary trough. maybe my problem with sdxl depends from the last update of the driver O.O
i just generated something with sdxl, but the how, makes no sense.....
simply adding a refiner, i didn't get that error any more
on comfy...
Not enough lamps; should probably hang up more.
I don't know
you are looking at it
I have only seen examples of objects and animals
thre got rid of the oranges
Looking at one picture

there lot of other examples lol, but so far its a great improvement but very few defects
soon as i get a hold of sd3 im moving away from sd and sdxl
With how good it is
I'm sure the fingers will be even better than xl
it is 🙂
im looking at severals of examples on X
Do we know how much VRAM and stuff it takes
For the better model
yes about 8gb and up

Having fun again this morning?
Who doesn't want free money?
I've had a huge crisis managing my folders
well folders arent that bad but you see when i have to switch between all the models settings it gets tedious
I think we've mostly started solving that problem by numbering the folders. However, my, myself, and I are perfectly happy to report that, as per usual, when we name the folders, occasionally, we just stare at them, wonder how many images we made, what we needed, and then just generate more images instead of actually looking through the folders.
oh right
i have an easy approach to that, i do housecleaning every day im done playing with sd 🙂
hello sunii 
Settings are a bane
I'd never be done
hihi! How are you?
i should probablly get a bigger SSD tho, so that i dont have to delete some of the good images
I...I also should do this. Considering the amount of TB I have...and the fact I need more, just says that I need more. (Or, quite possibly, that I ought to delete something. The more I delete, the larger my warehouse of "planned things to do" grows.)
obviously 🙂
which is why i also upload that image i wanna keep either on tumblr or civitai
That's a good way of doing it, to be honest.
happy cat
I'm really glad comfy just shoves it into one folder. I used to do everything via "here's the name of everything." Now I'm just staring at all of the images by day, and that works so much better. Even if I don't really search by name, at least I can just see it all in one go.
Sometimes I'm glad I don't have the power to generate 9000 images a day 
thats just insanity cx
You have pleased the inner purr within me
bard does pretty good pixel art stuff 
But think, if you could, perhaps you might ALSO have the ability to use the images in the very same day, in some way, to benefit the whole world...
hey I am also a pixel cat, I made it in paint, so smol it doesn't look like big pixels
haha, that's a scary thought when using SSD, writing too much on it can reduce its life span
👻
This sounds like an anime plot line ready to start....
Awesome! Yeah, I would certainly dedicate one--if you can get bigger than that, I would.
not a bad idea, i will check my options
Well now, that kiddo has an adorable flower!!
I have about 4 dedicated to pretty much just art/SD. It makes it so much easier for downloading things.
It is certainly CAT TIME, though
another folder incoming
looks cute 🙂
i mostly generate stuff around female beauty
welp
I misspelled cat

Thank you!
I really love showing off different kinds of designs--character designs, and fashion are one of those passions that I have. I like people being able to express themselves in whatever way that they choose. I actually have a survey about this, so please feel free to fill it out: https://forms.gle/57FbJD1EdU4UYNxK6
You...you could put a cat in the cart
/the portrait of anan etching of her head with a banana in it, detailed crosshatching, is shown against a black background, in the style of abstract expressionist ink, ilya repin, socialist realism, jenny saville, realist detail, lucian freud, monotype/monoprint
Please see #1047610792226340935 for more information about the bot! Thank you!
This is beautiful.
I copied ariamperez prompt and used the retro style I made 😄
hmm choices look very elaborate, some of which i havent even thought of
Nicee, thanks
my own prompt
The whole motion/retro style is super fun! Love the hair and the part over there!
Lots of motion; super fun!
Just pick whatever feels good to you! No need to worry about the output!
Love the off-white aspect! Someday, I'd love for my hair to straight up go black, and then have all these white streaks. You know, so when I get old, I can look like I stepped out of a fantasy novel.
LOVE THE TEAL
I love this hair
Frankly, all of the hair
I still have the pink in some of mine--gotta dye it again.
THIS IS PERFECTION
uhh i dont think i can fill up all those detail ...some are confusing to me, and im guessing that survey is for females, which im not lol
It's for anyone, really! Can you explain which ones are confusing to you, so that I can make it easier for you?
well its complicated to choose what kind of lips i like lol, i kinda appreciate all females
Now ya gettin' the color! Perfection! Love all the zooms and vector look there. Super cool!
yep vector art and screen print go well together
That's actually based more on physical appearance/genetics. If you Google, say, "Types of lips," https://www.google.com/search?sca_esv=95acdf9b02a04aa8&rlz=1C1VDKB_enUS975US975&q=types+of+lips&tbm=isch&source=lnms&sa=X&ved=2ahUKEwirlsP6mcSEAxWFl4kEHc3rCkIQ0pQJegQICxAB&biw=1920&bih=953#imgrc=wojW2PReyI4pgM You can see various types of lip shapes. The same applies to noses, etc.
sunny lol if you dont mind i'd rather not fill that survey, im struggling to choose , i dont compare girls
Oh, there's no obligation, to be clear!
thanks hope you are not upset
lowred the sample steps and the cfg 😄
I'm good! The purposes is actually to make it easier for people to make characters, and finite details that they might be interested in. I'm all good!
yeah i understand that but thats what puts me in struggle, cause i love girls not a girl
Love the braiiid!
thanks
throws in a clothing wildcard
good thing about that bottom part .oh boy
that arm tho...
raise cfg lower steps
thats without any kind of hi-res fix
restart sampler , before it was euler
munches on breakfast
cart house
This is pretty much my character in ever rpg
And was in rl, with the blue. Love all the braids
kewl deal you are welcome to use any of the images
PURRfection!
ty!! They'd be fun to do some paint overs!
Now you're getting the full palette color!
ok, i'll stop upgrading forge
haha switched original style before the edit, it has orange in it, gives me the fruit instead of color DOH!
and sometimes it loses it's mind
-adds a bunch of styles and presses generate-
turns on random mode
turns it to 11
Momo at the beach 
looks like lightning models are still too much for a 4gvram
sdxl lightning
For business inquires, commercial licensing, custom models, and consultation contact me under juggernaut@rundiffusion.com Juggernaut is available o...
its just the new turbo/lcm type of lora that exchanges detail and quality for low steps to gen faster
unless pony is an exception, i can run turbo
oh one of those
for what i read, is a condensed version of sdxl
turbo and LCM at least can have decent quality if used with the right model, lightning had pretty bad results for me
just condensed differently from turbo
ended up only slightly faster as well
still, on 4gvram, the gui crushes
oh, maybe i don't have enough memory to load an xl model adter another
Very ark, much knight


ouh I thought you deleted it
Grabby/10, it's on the side and looks big enough for a good grip
I mean cute
👍
Best for what? Maybe it's the best for specific things, but saying something is best in a broad sense kinda short-changes other tools that do well in their own areas.
I prefer ComfyUI, for instance, since I can stitch together a workflow that goes from beginning to end without me having to take any manual steps once it's set. I can also do things that are difficult/impossible to do in A1111 without extensive work.
The downside is that you have to understand node-based workflows and therefore you have to understand a bit more of how each node works for input, function, and output.
a1111 best big generate button
but comfy is faster tho
but I dun use comfy
ofc faster isnt the only thing we need to consider
have u seen the videos where its like a dancing e girl with an ai filter. are those as easy to make as they seem and is there any recommendations i wanna figure that out
hmm never tried that
i wanna get into making simple videos or something but animatediff pretty much was very slow i did a few yesterday
ive heard that about comfy and i kinda tried it but
i couldnt really 'get it' even though i use blender and am familiar with nodes
#🧣|comfy-ui wut bout this channel?
my a111 workflow is just so practiced at this point ><
mainly id only switch if i could get higher quality and better resolution
or .. hmm easier videos
i got the mov2mov extension but i need to find some short gifs or something to practice with before trying it on something real
what bout YT? tutorials or smthing?
well true
i dont mind reading wikis and stuff
i prefer documentation and manuals to videos
i think i just need to experiment with animatediff and deforum a bit more or maybe try another extension til i find one that works for me
im botting up my server rn gonna try deforum again
booting rather
might just make some pics but i wanna try a video andhopefully make a lil bit of progress each day
u can also use high res fix with animatediff
well the first two came out alright
idk what i changed but they all just deggenerate like this now
my try just with high res fix at 1.5 and 8fps and 24 frames max
animtediff?
yea
i might just go back to that, these ones i just posted were deforum
i like the animatediff look better
which motion model were u using
i found a v3 safetensor thats only about 800 mb that i figure might be less memory expensive
damn
how long did that take
i had one animatediff run that took like 20 minutes last night
ok ya i hate this
20mins at 768x960 resolution 3 secs at 8fps
on comfy is probably faster and better but i dont wanna install another webui 😔
tru
how much vram do u have
mine dips into shared i think is why it tooks forever
and did u use tiled vae and stuff
do u know if dynamic prompts and wildcards works
12gb i dont have the overflow to ram setting enabled so 768x960 is highest i can go
idk i just tested it with normal text2img and only extension i enabled was cd-tuner
if u have latest nvidia drivers its on nvidia control panel settings
💀
i guess its avant gard
I know that banana
It's Naruto!
huh?!
a cute cat
i get better results just making a batch with low variation and then blendeing the frames in gimp
use a char lora that way the shape of the face and outfit doesnt change
oo truu
Trained 7 new styles 

Search for perchance image gen is fast but not great quality
ah
yeah
have tried it out
but
im looking for quality
not speed
Well, image generating takes hardware
Hard to find good free options
aw
:(
i have the worst laptop ever
how bad is it? 
In the pony diffusion discord they have a bot that runs... Pony diffusion

intel i3 processor and integrated graphics card
Sometimes it's slow and doesn't work but if you're like me doing gens late at night it runs pretty good
ive tried running some local stuff on it
and if it was sentient during those moments
i think it would have been screaming in agony
I feel your pain, my old PC that my family use for YouTube basically, has an i3 too XD
yeah :(
I3 6100 cpu
Ancient stuff
yeah
alr
thanks
Try the pony bot
ill try lookin for some sites
It's pretty good, can make lots of Booba
i mean

if it has private threads
sure
Only mods can see what you gen
oh
alr
dm invite
okay
leonardo?
Holy
discord have a sfw policy
we can edit dat right?

yeah :(
the only anime model i managed to find is animagine xl 3.0
but the hf space has horrendously long queues
Rating safe
it's good
👍
yeah, became a sponsor of discord and ask them to stop suppressing nsfw stuff on their servers
idle like idle
I dunno how to explain it
you know when we just chilling doing nothing important or smthing Im placing idle
thats my defenition for idle

notifications can still see in idle mode I guess
It could be anything
I think we should generate Booba
That might solve our doubts
ehhh???
@wooden coral #🍥|anime message #🍥|anime message
it's alright
It's almost like we are the moderators of this channel, just this channel 
i'll report the channel on youtube too
Jesus christ 
time to look at the moderators application form 
what'd I miss
oh the youtube spam?
yup





