#🍥|anime
1 messages · Page 120 of 1
Because it doesn't work on anime
I don't know if you are using that tho
In any case I would suggest using adetailer
this is awesome I need to try
this is awesome I need to try

haha
I saw someone doing that today actually, not space images but castles in funny shapes
yess
i bet theres a cloud trained model for similar i wanna play with
leme know if u end up trying it
i wonder what a deforum animatin would look like with that model hmmmmm
https://i.imgur.com/v59puih.mp4 this one almost came out good
animate diff?
deforum extension
have u used animate diff
not yet but was going to
doo it

doesn't run
hmmm
could be a comma in the wrong place these things are touchy , when was the extension last updated
literally just download
like 10 min ago
maybe it's not compatible with SDXL?
Quick little img2img run of a (very) early game screen grab from Cyberpunk 2077 in celebration of the 2.0 patch. Figure photo mode could be fun for setting up image compositions.
I rate it 1 out of arlev aik suncak;
XD That was the "employee number" text on the mirror in the screen grab.
i wouldnt expect it to work with xl but idk
yuuuuuup
switched to 1.5 and work fine
this's real sad
Soon ™️
steals your girls

and he turn into girls
Is that a Succubus?
thats awesome
Does she look like a more 'regular' Succubus, or is she more like a Lilim specifically?
im curious what the result is like and how it works, i thought deforum was overly complicated over my head til i tried it and its not so bad
it's easy to use it's literally same as text to img
but it take sooooo long
it's still not finished yet
and I only started 1
ty
nice, i'll update now
these arent so bad im getting like 5 minutes per video render
https://i.imgur.com/vnwfbJH.mp4
idk what im doing though lmao
the more i tweak settings the more it turns to mush
mine been 25 min and it's at 90%
I think I won't use this ever again
can you use DLSS 3.5 in SD?
Today I learned what lilim is!
there might be some setting to make it faster
this one has a thing where it only diffuses ever x number of frames , so i have it set to 4
so 120 frames gets cut to like 30
then it blends
flashing images go

https://mythos-and-legends.fandom.com/wiki/Lilim Right in here.
not what I expected it to do tbh
looks like a similar concept where it might be trying to blend through seeds or something
careful that preview image is getting kinda borderline on nsfw.
What in the hell
either way I'm deleting this
Blind people!
Jingo is too impatient
here's the odd one old since you prob liked him
30m for 1 gif that didn't even give good results is just not a good trade
what kind of settings is there
it looks like it just needs intermediate frames
and lower fps
I removed the image, the link in of itself still works?
dream about space ship
yeah that should be fine. make sure not to post links to actual nsfw content even without a preview, but since that one was just close to it, I don't see a problem with it. just don't want to see you get a timeout or ban.
does deforum work with SDXL?
it might but idk i havent tried
ah
loras and ti work though
Thanks for clarity.
@gusty palm btw I updated tagcomplete and can confirm that it works as expected now. Not sure if you already updated or not.

I can dm you a website that can help, but it contains nsfw so first I have to ask, are you over 18?

Surely gpus will be priced so low by then
I might actually upgrade soonist as well
inhales hard copium
Reaally
That is pretty insane if so
not sure, but people kinda predict that
or like end year 2024
awesome i havent updated yet
git reset origin head then git pull from in the extension's directory will get you back up to date.
@supple raptor do you remember back then, there was a time when you could just copy the image from web broswer, drop it in discord, and that would include the metadata?
@gusty palm after deforum generate where does it saved and is there a way to change that?
I wouldn't want dem gif meessing up my neatly organized images folder
That was because of the stealth-png-info extension. There might still be a fork of that being maintained, but both parties would need to have it installed for it to work.
ahhh right right
it was that extension 
its been putting deforum folders in my img2img folder , but u can change it i think
it makes a serperate folder per video, and u can have it delete all the individual frames after it compiles the mp4 or gif
by new folder you mean just video in the folder or video and images?
cuz if it's just gif that sounded like pain in the ass
she look horrified at whatever she saw
looks like this, would def require a manual copypaste into a new folder let me see if that can be changed
these are the settings i have on to delete the frames after video compile. looks like u cant change the output directory other than the folder name but its not too messy
i guess u could have the output folder to something without a token and it might put them all into a single folder
thats probably the best way
do you have to change it each time or only once?

I know the prompt you did last night
make more stuff catto
maybe make cat warrior

for these i pretty much set it once and left it
some of the settings in the keyframes, prompt, noise panels though ive been tweaking trying to figure out
booba
I'm messing around with different hairstyles
the other model more anime doesn't seem to want to change them easily 
do mohawk
catto protec
soon..
How long does a 512x768 take on your gpu?
2-3min I guess
idk gotta ask cat to be sure
That image there is 512x768
45 seconds
Hi, would you like to order your out of vram today ?
I don't have xformers or the fancy stuff on my smol amd 
sad AMD noises

cat warrior when
also check DM cat
I put nsfw and cweavage on negatives, images came out worse 
why cant I type the c word reee
all of them
this is a model i wanted to fail.
twin drills
clyde don't like twin drills, duh
bonk
Turn on tiled vae and tiled diffusion
even sadder AMD noises
tiled vae doesn't work on AMD with DirectML
Feeling okay?
is there antyhing you don't have answers to... like the meaning of life
Damn, rip team red 
creating waifu is obviously the meaning of life.
@astral drift my gpu is a RX 580, its preeetty old at this point, I'm surprised it can even do SD :p
then why am i model merging again?! 
so you can create more waifu
when is the last time you saw me post waifu?!
TRUE waifu
oh, shit, page 3 of my image search results. #🍥|anime message
busted
waifu
@oak spoke
in my defence, i have 376 pages of images on this discord server XD
nice hair
hands
not mohawk tho
she be throwing up gang sign
this one is pretty cool
bit too crazy D:
lmao
she was a delibate fabrication for this channel. Her existence is not waifu based at all (she only exists because of BREAK anime; after a lot of non BREAK anime's)
the road to waifu matters not, only the destination
this sounds like a surefire way to go to jail, depending on legal area
you generated that? ^^
I added the text, other than that, yeah
chotto motte ?
chotto matte
chotto matte -> ho, wait up a seccond / please wait
it's japanese
yeah
now do chotto latte
Am I allowed to say this woman's well-endowed?
she packin
very cool, and more cat, looks scuffed 
i have well endowed yarn waifu to cancel it out
hahaha
i'm mixing again 😦
i thought i was done after i posted my model, but i think i have figured a method to improve image quality with no loss to creativity
one hand is enough for you
Such a lovely yarn.
too bad i couldn't use the negatives to go with that prompt (as this one is from my testing prompts)
controlnet was involved?
nah, that was an old sticker
her hipster level is off the charts tho
why am i model merging again 😦
How is this one so fabulous?
That's CoffeeGirl. She delivers coffee with her CoffeeCannon™
Also, that image is ancient. I don't remember when I made the original version, but that one which had a little inpainting was made on Jan 14th
@supple raptor Coffee Girl's Coffee cannon is CoffeeCanon. 😮
yesh
The true story behind CoffeeGirl, is that I found long ago that her prompt was incredibly robust and looked very similar on basically every model at the time, even ones that were very stylized. On newer models, it changes a lot more though.
nubby do you use controlnet ?
The coffee cannon was just a hilarious accident that happened on one of the images.
sometimes
Do you have the reference only model ? Or at least know where I can download it ? I can't find it 
Reference only doesn't use a model, it only uses a preprocessor which should be included with the extension.
big shield it got there
trying model mixes again, hoping to improve my model
more model mixing = less fox girls
that's a weird looking gun
my merge needs to do wewll on ALL of those prompts at the same time
fail one on all 3 seeds -> ur out
dont know what happened
nice twin you got there
new meaning to double head
was there ever other meaning?
🤠
she's happy that you like it is why
time traveling logic

Pink haired lady.
this one's a little pink too
care to share that propt?
need to test it on my latest merge
you already forgot the chibi prompt I share last time? the one with armor
just remove armor part
literally same thing
🤣
you mean the part where i did 5000 images in between?
maybe?
Powerful stuff, really.
damn, now i just gotta browse my prompts 😮
oh shit... i wish i was shitting you, but it is srsly 10k+ images ago
damn lmao
and that's 10k images on sdxl XD
i mean, that's from 9-9-2023 -> i've had multiple 1k image days lol
"no touch" on her sweater lol
aight, last test had no definitive winner, imma retry, lets remove the armor
removed armor from that propmt, testing on my model mixes, the 4th one was a early reject
her wings got holes interesting
i feel i should test #3 a bit more
2nd and 3rd look basically the same
the models weights are pretty close actually
so now i'm gonna test them on other prompts too, see if either one inches out ahead
and for some prompts... 1.5 is just plain better 😮
yes but SDXL have more future
1.5 will get overtrained easy
theres no porn future for sdxl
there's already handful of SDXL porn model
nude = / = prn

who needs these?
if u could made those dolls u would make a lot of money 💸
yeah, i think so too
you put those up on etsy for $100, and they'd still sell
then again, i think making something like that would take days per doll
if it's handcrafted
Who is this one?
is SD.next better compared to A1111? I'm curious to try it out especially since it has support to run wurstchen
i just find a1111 sluggish and buggy at times, so i think sd next is very nice and sleek
wouldnt hurt to try
thanks, im about to set it up, btw have you tried wurstchen?
no, but just looked into it. seems nice
probably never heard of diffusers before because a1111 doesnt really support them out of the box
if its actually as nice as sdxl i may try it, tell me how it goes
its supposed to be lot faster in generating images with lower vram usage compared to SDXL
yeah, and hires too
yeah but i think there aren't any community models for it
and if there is i dunno
civitai has tons of community models that are trained or merged
wurstchen can only offer their base model
unless i have the concept mistaken ... trying to find out if wurstchen works as a model on top of SDnet
Converting models to diffusers format is pretty simple. It's just a script in the diffusers repo. Using diffusers will make a lot of extensions not work though because it's a different generation pipeline. My experience with SD.Next has not been great. I've tried it a dozen different times at this point and it's had major bugs every time, to the point that it was completely unusable for anything other than plain txt2img generation.
oh wow, i see, im almost done downloading the files, i'll give it a try but im not planning on removing a1111
this is what they are claiming at the website https://stable-diffusion-art.com/sd-next/
This is the script to convert to diffusers. You'll need to clone the repo and install the dependencies https://github.com/huggingface/diffusers/blob/main/scripts/convert_original_stable_diffusion_to_diffusers.py
thanks, help me understand though, i need to convert to run SDXL?
yeah, vlad seems to think very highly of himself. I'm no programmer, but it does seem that plenty of the code is written better than it is in Auto1111, but it breaks constantly. the "suited for users who don't like tinkering." line is an absolute joke.
btw is this the same thing regarding coversions
Last time I tried XL in SD.Next it needed diffusers for it, but I really hope that by this point he's got original stablediffusion/compvis format working for SDXL too. No, the backend is what allows you to run the two different formats.
just a quick screenshot about what the site is talking about
ok that umm. ok then
those hands..umm oooh ummm
ok I think it got worse in terms of clarity
shrugs
Hi guys
It looks cool is this realism
yeah had some realistic and such in there
for details
but still trying for the anime touch
cool!
couldn't settle with SDnext, didn't like the UI, felt kinda wonky how the buttons are
hmm havne't tried that one
tried comfyui some weeks ago too, and now the SDNext .. i prefer A1111
looks at install instruction's.. nope
installation is more or less the same as A1111 but uhh if you like a1111 you can't settle for sdnext...
hmm wonder how sdxl would handle this prompt.... let's see
updating my gpu driver .. lol screen went blank for a bit
which sdxl model are you using?
Does it have any cool stuff that you don't have in a1111
not really no.
only reason i wanted to try out SDNext cause i read on a site that it has support for Wurstchen which works 2x faster and with lower vram use than SDXL
turned out wurstchen is not as popular as SD yet and SDnext didn't feel comfortable to use
cool .. i didnt see that model yesterday when i last checked
ok this is gonna take a bit..sigh

looking good so far
What you cooking this time tora
huh?
What's taking so long 
let him cook ❤️🔥

and upscaled.. and done
😔
original was better NGL
still decent tho
ok back to a regular model
yawns
think i'm gonna head to bed. nite.. thud
ok one more.. and thats it. done and wow best one yet. IMO
Didn't go yet smh my head
Good night (real this time)
take the phone with you to bed and make more images on it with remote connection to a1111 on pc
Skulduggery Pleasant, is that you?
i wonder why some ppl are using more than 50 steps on regular images
huggingface wrote about that .. going above 50 steps will not make any noticeable differnce
hey
hey man 🙂
was wondering that cause hyper realism kinda the animated version of photo realism
i guess hyper realism would add more expression not necessarily make analog photography
yeah, it's just "really well drawn/rendered, resembling reality"
yeah
pretty vivid detail
kinda the reason i like sdxl models so much
you must be using a beefy gpu
3090 XD
thats 16gb?
24gb
yeah, it's possible, it's just not the most fun
i was at 10gb before, but then i went looking for used 3090's xD
ahh cool
any chances you might be sharing your merged models at civitai any time soon?
already did 🙂
oh?can i have the link
although i've already been working on a small update xD
how fine tuned is it for anime stuff?
it isn't but just throwing in anime works just fine 🙂
it's because it's not fine tuned for anime that it's creative 🙂
ahh so using prompts anime with weight would do the trick?
very much
cool
all the stuff i've been throwing around here is done with that model
nice 🙂
ooh ❤️
this?
if thats the model you used, you should definitely add that to the gallery
that's on the improved? version, which i'm working on, but the difference is minimal
ahh ok
right noww just trying to see if i can get some more image quality back in there
what's noticeable is how crisp the detail are that catches the eye
i think i can say for sure that the improved models have better visual clarity though
yeah thats what caught my eye
well, that was actually totally the point of tryinng those new mixes
i feel the divide3 has the least amount of artifacts, on the usual
Lmao that censor


Fancy
Do pruned models cost less vram
That's what I'm guessing they are
Some models are 2gb while others are 4

fp16 models
pruned -> removed all the non-weights data
leftovers from training and such
fp32 full accuracy models are ~4gb, fp16 (half accuracy) are 2gb
pruning is just cleaning up junk
I should look for the fp16 ones then
usually people publish those 🙂

Me in PUBG be like
She's cute + one of the most widely known characters so lots of people know who she is.
stop 
i have no idea who Hatsune Miku is
I'm so hungry 
but i googled that name and found out Hatsune Miku is an anime character
ackshually she is a vocaloid (I think)
Don't quote me on that I don't watch anime 
and i only took interest in anime cause for the love of girls

since you are hungry, here is a tasty treat....

no
I don't eat cakes

you dont like chocolate cakes?
They are tasty but I just don't eat cakes :p
oh
the legs midair lmao
intense gaming moments
sad maid noises
idk why but portraits in this model just look a bit sad 
which model you using?
she wrote every book there
lol she is supposed to be cleaning a book shelf
lmao
Don't wanna do the job
hi Smol sus.. how did the competition go?
yeah
<t:1695416400:R>

i think i submitted like 3
it's not the best but at least i tried

nice, do you thing Smol sus, you could win the price 🙂
well, i won the cow once so

prize even
oh wait
actually i got the 2nd most voted, but was chosen as the pick of the week i guess
cow?!
wait what
i should actually submit once
hmmm
cow is basically the challenge of the week (COW)
can i muster up the energy 😮
Yes (maybe)
xD
oh damn so many cool entries
I'll submit some, and win one day
with my smol card you guys will see how powerful I am
it's really interesting, if you guys realised, all black name with the role Stable Society, basically won something like POW/COW at least once
well, for POW, it used to be the top 3 for each week that gets the role, now it's 1? and for COW, it's the top most voted, and a staff pick
good luck and hopefully you get black name soon 

@lyric zealot I wanna try some of those regional prompting things, which one should I try to learn 
i heard regional prompter is currently the top most used
previously, i would use latent coupling aka two shot
is that the one where you draw a little sketch of where things go
yup, or you can divide the region like a scale?
ratio
like if it's 1,1
then split the image left and right, half half
yeah, region prompter, it can do it by attention, which is a lot faster than by latents
plus it has some other nifty features too
not sure if this is still useful but https://badnoise.net/latentcoupleregionmapper/
(two girls:1.1) in the jungle
i wonder about two shots extention
does that let you have precise control over positioning ?
rooting 4 u
what do you mean about precise control?
like placing characters on a canvas
well, you can specify a region where certain prompts will go to
so you can draw a region, put your character there, and another region, put another character
hmm sounds interesting, like controlling multiple region where which prompt should go?
oh wow
yup yup
i should try it out
two shot aka latent coupling does that, but regional prompter also does that, but might be better
haven't really test regional prompter a whole lot, so can't speak of it's effectiveness
well im curious to try that out.. thanks for mentioning

dalle 3 might also be offering regional prompting i think
i don't think so. I think it understands the context more like SDXL, but you still can't tell it to generate the young woman on the left side, at the specific area
the AI does that for you, by randomly placing anywhere in the whole canvas, but will be able to know it's haggling with the grumpy old vendor
basically, if it's just txt2img, you will not be able to get regional prompting, you need to be able to specify an area, and tell it in this specific area to have something
for it to be regional
i suppose it's like the attention mapping from regional prompts
ahh . im not sure about dalle 3 specs, they just released it and i briefly read about it on reddit
unless you put it in corrodinate format
dall-e 3 will be closed source, iirc?
yup, attention mapping is probably the closest
yes part of paid Api
this whole text to image gen tools started not too long ago.... has it been one full year yet? i think it was around dec 2022?
not sure if it's old or new
doggo
I thought this was a captcha at first glance
Dalle Initial release date: January 5, 2021
true true
chatgpt came out nov 2022, stable diffusion aug 2022....
holy crap.... i wonder what's to come in the next 1 or 2 years
text2movie 
dont think AI can replace a human mind and girls
well, that's basically the turing test
quite an intense stare
i wouldn't be so sure 
how so?
the thing is Ai can have collective power of human society
and it can stun a human mind with possiblities, but Ai can never attain what it means to be conscious
well, just can't be sure of the future in this case, not speaking with knowledge about the subject or anything
well, might be true, but there might be stuff you never know that can attain that
so 
she got a selfie stick lmao
character.ai exists... well...
Ai is technically a sophisticated object that will never have the capability to understand consciousness

meh ppl back then also say human will never fly

well, you are explaining how AI is at our current level
not at all the same
when the time come AI can do anything
exactly the same
you think based on your ignorance of the present
while we look into future
in fact Ai will do wonders that a human mind will find magical, but yet consciousness will remain illusive to Ai by the laws of physics in this universe
nope
human brain are object too
not talking about brain
talking about mind and consciousness adn those are not objects
not at all the same thing
it's the product of object

cave man think fire are impossible to reproduce
this is a fundamental confusion with many interpreters
you're thinking with caveman level of thinking
again you are mixing up objects with mind
are you saying it's not possible to "code" a "soul"?
mind are product of object
cant create soul
are you saying that?
you can't prove existance of a soul
i mean, im getting confused
in fact science has no objective understanding of soul or mind
soul doesn't exist
perhaps we are all AI ourselves, we just don't know it yet
just religious nonsense
matrix do be like that
mind does and if you trace that into experience there is something much deeper

you are just denying based on limited thinking
that was the hard problem of science for decades
well, to be honest, what does that got to do with anything, it's not like if my thinking changes, then stuff will happen or not
so to each your own thinking

and scientist back in the day can't create light bulb
there you go again with mixing up ojbects and mind
like I said mind are product of object
without human brain there's no mind
or are you gonna make up more delusional stuff
that mind can exist without object

just like how religion are product of human
religion aren't object
but without human religion also doesn't exist
here is the thing ... AI will be capable of sophisticated and abstract phenomenon both in terms of physics and mind based on underlyng physics nature of the unvierse, but it will never be able to create life/ mind / or experience consciousness
you and your narrowed mind thinking
you want it to do so but it can't happen, thats something beyond Ai
nope
how do you find that narrow thinking?
beside how's your silly subject have anything to do with ai art
I'm sure #1008652627204132904 will love having you
you have some fundamental fallacy coming from strictly materialistic view
@lyric zealot
I'm typing breaks and stuff and things happen, im dumb
break? BREAK?
although i'm no smol sus
yup, like eface said, you got to use BREAK
the complication comes with one side leaning towards religion and science needing to adhere to objective proofs, and i would tell you either of those sides are messing up clear thinking and human potential in relation to mind and materialism
what does BREAK do?
I think I figured it out maybe
im dumb even reading the guide page
Ai is basically a sophisticated algorithm that will supersede human linear thinking .. but that has no connection to consciousness
eface is gonna explain to you 
as that's the original spec (you coulndn't go beyond 75 tokens)
imma go cook dinner 
ahh
so they figured out, well just send multiple groups of 75, weighted by average
so you need to BREAK every 75?
no 🙂
it's good to read the a1111 feature page 
it will break automatically
oh so you don't need to put it in prompts?
however, somtimes you have something that's 40 tokens, and something else that you'd rather send in a fresh set of tokens, isntead of the already existing 40
yeah, it's totally optional
I see
i use it a lot, but that's because of my prompting style
yup, this is accurate
i had it happen on the word blue
the image was INCREDIBLY blue

because it was the first word on the second set of 75
yup yup
earlier words are weighted heavier than later words
like in the example here, it will not know that it's red eyes that you are talking about
stuff like that, you can use BREAK
{2-3$$ BREAK $$__songmaster__} BREAK anime
this would get me 2 or 3 prompts from my songmaster file, and then have another tokenset just specifying anime
does the sequence of word ordering take precedence even with weights?
it's so-so
you'd have to experiment
sometimes weight, sometimes position, sometimes just specifying the same token double
hmm .. something to test ... in most cases it seems weights take more emphasis regardless of word positioning
usually yeah, compositionally is more the idea you'd have to go with
putting something earlier in the prompt usually means compositionally things change
i guess that's logical to meaningfully structurize a prompt
@lyric zealot
elp
It's doing things right but they look weird, like it isn't applying quality prompts/negatives to the whole image
@nova remnant quality prompts here ADDCOMM what the image should be ADDCOLL thing on the left ADDCOLL thing in th emiddle ADDCOLL thing on the right
that's what you should write
will try that
no ADDBASE ?
it says that on the extension
oh, addbase fogot that
the first addcoll should be addbase
my bad
wait so how is it exactly
quality ADDCOMM
lamp ADDBASE
waterfall ADDCOL
girl, portrait ADDCOL ?
I need one more before the lamp 
the overall image ?
cat...you might wanna read through this link, it covers a lot https://stable-diffusion-art.com/regional-prompter/
In Coloms mode, the image is first split to rows with ADDROW or ; in Divide ratio, then each row is split to regions with ADDCOL or , in Divide ratio.
In Rows mode, the image is first split to columns with ADDCOL or , in Divide ratio, then each column is split to regions with ADDROW or ; in Divide ratio.
When the flip option is enabled, it swaps the , and ;. This allows you to obtain an area that is rotated 90 degrees while keeping the same ratios used in Columns/Rows.```
from the regional prompter github
im used to using BREAK tho
but haven't really try regional prompter much, so yeah
requirements for good quality monitor for SD ? - heared about colors need to be the best ?
monitor in no way affects SD
:)
Is that a Combat suit, or cybernetics?
They are all fabulous.
Why do the latter two seem taller?
@lyric zealot This trio is adorable, really. ^.^

In their GitHub or something so someone smort in the codes take a look
Hmm when I try to gen something it gives me an out of vram error and if I do --lowvram instead of normalvram
I get something "torch not compiled with cuda enabled"
Yes

I think i lost the link for a guide to that
I was trying to merge using the default one but it's so boring it hurts
I don't do much merging nowadays, so can't speak about it, sorry 


#🍥|anime message found something from nubby
ty
Is this the niji geometry thingy?
yup, that lora is definitely in used
How are they all fabulously expressive or expressively fabulous?
Also how come they all get along with each other so well?
Imma be honest, i have no idea how they get along so well together, since their personality isn't similar
Did you check out Jujutsu Kaisen or Madoka Magica before?
Is the one with the pink hair making like Yuji Itadori or Madoka Kaname in there?
it's an OC, not based on anything
Miku Hatsune, what happened?
How do you ask the AI to tell you the prompt for an image >.>
Is she better now, or has she gone Fallen Angel on us?
img2img interrogate CLIP, or the other for more anime prompting
she might have 
hi there's no bot here, you might wanna go to #1100170312106127410

@nova remnant So far so good?
👍




carroots





