#🍥|anime
1 messages · Page 12 of 1
fuck you
I think I should have removed the big boobs
yes.
senko!!! what model?
Not a model, a lora
i just wanted to know how did you get the lines like that
sorry if this is a dumb question, is there anyway to run my image again with the vae on so it changes the colors of it?
You can drag and drop the image onto PNG Info, and then click send to txt2img
it's normal until you see the face
does the lora itself have the line style?
Yes
You're welcome, glad to be able to help 
so i send to text 2 img and it just changes color or?>
Yeah, then you can re-generate and it should generate the exact same pic but with fixed colors, since now you'd be using a VAE while generating
okay awesome
WAIT WOT? Ufotable Animates Kimetsu No Yaiba? I just noticed 
hopefully it works, i cant seem to do very high resoultions
Fair fair, how much VRAM you got?
im not sure
i think, it said 6000+ something when i run into an error?
mib
or some shit
i may be wrong
You should be able to do at least 512x512 with hires fix
It may take a hot minute tho
Not sure how much it fluctuates depending on how much VRAM you have
Or if its based on the speed of your card, dunno
i clicked send to text2img and it generates a different image also
I had a GTX 960 before, took 10 minutes, I got a 3060 12 GB now, and it takes about 30 secs
what is hires fix?
Wot? Thats weird
i will just use it on new ones
Alrighty :o
thanks for all the help
This
Yw 
i see i see
Lemme do a lil demonstration
Without hires fix VS With hires fix @lament osprey
Better face, better eyes, better details, usually can fix some other details too
Generating normally with hires fix on
SD Upscale also works tho
It is slower I'd say but less VRAM intensive since it upscales doing tiles, instead of like a whole thing
well its not taking tooo much longer
Well, its done in two parts, first normal, then upscale
wait so when generating with this, it automatically upscales?
Also what sampler do you use, and how many steps? Also I'd recommend the settings I set for Hires fix
Yeap
If you want I can also send the upscaler I use
Which is the best one atm I think
which one:
euler a, 20 steps
that would be great, if it works on smaller pc
like
some of the upscalers dont even work form e
me*
says i run out of memory
yea the hires fix doesnt work even on one image unfortunately
You should try DPM 2M++ Karras, also with 10 hires fix steps
danmit
Whats your base res?
i will have a look, last time i have used swinir or something and it was the best i found
DPM 2M+++ Karras is the sampler
oh woops
Oh, well for hires I just do 1 at a time
I think I can do 2 with 512x512 tho
so that just gives higher quality images? that sampler
yea i tried 1 at a time, didnt work unfortuanetly
Well, its the one I like the most, and seems to be pretty good
Hmm, maybe you should try lower res, like 512x512, or 512x768, or maybe like 400x600, I do 400x600 sometimes
Also lemme send the upscaler I use, will take a hot minute tho, my upload speed is abysmal
i'll try that, i did a higher width so i could fit more people in and it looked better
and maybe get a wallpaper
thankyou thankyou!
you like it the most because it gives highest quality?
I think so, though its also the fastest, I usually do 40 steps, but 20 steps works fine too, it will usually look better earlier than the others, but kinda reach like the same conclusions as the others within like the same upscaler category eventually
ah i see
Good morning
nice
thankyou, so how does this work>
I'd say so yeah :o
I think you gotta reload the upscalers, theres probably a button somewhere, maybe in the settings, I know a friend found it, but I haven't had the need to do it :o
thanks
my my
Gardevoir OP 
my attempt
I did it with a LoRA cx
what is lora
It is like, basically, DLC for the AI :o
hmm
what does it do
Tis one I got on atm is this
https://civitai.com/models/17302/gardevoir
Helps make Gardevoir 
ah right
My favorite music genre
The time is 10:21pm. The only online moderator is on do-not-disturb. The online admin Palp rarely shows face and will likely not handle any mod-related duties. Asukii is also rather enigmatic - it is unknown if they regularly moderate or participate in such duties.
In other words, mods offline post dogs
let me load up a dog real quick
its not working, im getting cat girls
success?
a bit more brutal than i was planning
dog
handsome fella
is there anyway to cancel an upscale in automatic1111?
dog is chilling
i think that anime one is about to make my pc explode
and i cant see its progress
nvm worked well
You can hit the interrupt button that appears when generating
that doesnt work for me
found out early on
that and the skip button, seem to do nothing
Oh, weird
ooo nice
any nsfw channels here?
you'd wanna go to Unstable Diffusion server for that
yea myb
no worries, just fyi
Good morning people 🙂
the first one just slaps ! damn, well done here !
multiples are so good, it's just my prefered one
Prepare for Songkran Festival
It took a bit to get my lora working efficiently, once it did i started to get a bit wild with it lol
I can see the wild, and even if I really like it, the blurred one is close to the
line 😉
My apologies 🙏
when you guys train loras, what do you choose for source model?
the model you want the lora to effectively work with?
yep, and the model that fits the best your goal too. Lora will give you the main weights that changed in that model, all the other weights not in the lora are the ones in the base model. LoRA training will compare to its base model, so the resulting quality will be better staying with the same model for prompting
if you use civitai helper does it keep the original model? i noticed there seems to be duplicates of older versions of models when updating
my models folder is suddenly 517gb
💀
such detail
I'd guess that one might be a TOS
I don't want to be the party pooper but I had to there :/
can't really be fun on that case sorry. and yeah, I agree on the awesome
sometimes sd just gives too much boob

Alice jumpscare
no i dont
true true
I'm not capturing the dream-like element in my pic
and that's my only complaint about my pic
Then again, people just vote for what looks good anyways
true, and what's unique
More on the good looks still 
well, depends tbh. I think it must be unique and looks good. But as far as I can tell, anime style pic usually won't win
so if you are really trying to win, go for a most realistic feel
or cartoonist
avoid full anime

Nah i'll make sure anime wins
I mean, there's already a few winners with anime style submissions
yup
it's kinda rare tho
if you want to at least secure 1 vote
make miku 
Fruit will definitely vote for you

Guess i should've submitted miku last week 
🐐
❤️
how many attempts for the cigarette?
quite a lot, it's not that bad with the lora tho
😭
...
like this




what typically happens to your image the more hires steps you do?

depends, you don't really need to have too many hires fix step
typically you only need about half the steps when compared to normal generation steps
0.5-0.61 denoising strength for me. Higher often starts making a mess of the image and lower can leave the image blurry.
denoising strength, is basically how much to "jumble" the generated pic before upscaling it. If your denoising strength is 1, then it's basically generating a completely new pic without using the previous generation as reference
and so, if your denoising strength is 0, then basically no change to your previous generation
ice cream shop logo
@finite phoenix
Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝|tech-support for any problems you encounter while installing it locally!
oh ok

weird how its called denoising haha

oh you got a cig lora?
your cig lora 
it's a little hard to get good result, so have to nitpick generation
this one has such a cool vibe
I always get so close, but so far
true true, it's really about generating a lot of pic, and praying to the RNG god 
i swear it's because of some negative embedding or something
yeah I'm not convinced the negative embeds aren't corrupting the cig
like, SD is probably trying to generate good mouth, and it thinks that putting a cig over the mouth will ruin it
yup
that's why it's super hard to get mouth generated with cig
it has always been hard generating cigs, even on base models
cigs, swords, tools, anything in hand has a hard time being in the right position
i think it's a concept that's hard to train as well?
quite, yes
in XL model we trust
like for example, there are lots of photos that make it hard to learn. Sword crossing is the best example : lots of photos of that, and it makes SD think swords have a second sword crossing it sometimes
came close w/ controlnet the other day
well things like that can be mitigated with reinforce learning right?
yup, it's annoying and difficult to get right
in retrospect, I should've probably trained 2 versions
1 with cig in mouth, 1 with cig in hand
and maybe with locon and whatnot it'd be better now

this is probably why so many gens have both
new locon inc? 

Smokin'
I spot that pfp and banner pic 
SD decided this one should be lewd
Very nice
lol love your name @mild tree
Im always lurking here for more smolsus


I got a new fan
the key might be arms behind back lol
Yup


Just drink pure nicotine
inject it into my blood stream
Yup
kind of a cool vibe with some neon lighting
I don't miss the morning lung feel though
dash of anyahehface for a stoned look
lol that's cursed xD
shutter shades could be cool too
I think this is a similar example, sunglasses able to go on both so SD decides to do both
Cancer ladies 
Goofy long cig
welp I gotta sleep
Gardevoir op 
you should rename to #1 Biggest "#1 Biggest Smol Sus Fanboy" Fanboy
Biggest smol
shes up to no good
You might be forgetting someone 
DONT EVEN GET ME STARTED ON YOU
@lyric zealot
super cute 

are you gonna make it into a model? 
?
you can already do it on my model
u just have to use this negative TI as positive
lul
lmao
will have to try that later

currently trying to see cyberpunk rtx overdrive mode
on my smol 3070 
tbh
i dont get it
why that one TI produces such cool cute shit
while ever other negative ti i have trained produces cursed shit
for example
i think it depends on what's "opposite" of the embedding right?
hardware abuse 
dog
Same lmao
i remember someone on reddit did an experiment about this, not sure what's the result tho
i think thats me
lmao is it
Overdrive mode? 
the update just came out yesterday
Our recent preview of the Cyberpunk 2077 RT Overdrive Technology Preview was relatively short and condensed - and we actually found that we had a lot of leftover b-roll that was simply too good not to share. John Linneman carried out the capturing and editing for this one, using an RTX 4090 in the RT Overdrive mode in 4K DLSS Performance Mode. W...
like it's actually crazy
Last one, coming back in 3 hours maybe
I heard 4090s can't even run it at native res, need to use dlss
yup, 4090 need to use dlss3
O damn
what's ur res/fps 👁️
I guess that's fine
Cyberpunk 2077 is still like... yknow, game wise I feel it still needs that No Man Sky treament
true
i mean, they improve some stuff, it's not at nms level
but well it's something
I feel like at this point the game has become more of a benchmarking tool rather than a game lmao
Yeah, well, I mean I hope they give it the NMS treatment, or like content mods like the Cinders/Convergence mods for DS3 or Calamity mod for Terraria come out, that like do the devs job for them
True
XD
not saying it was a bad game, but it was really short and not as impactful as other games
Yeah I mean like
Its a game, it could be better tho
If it had like Minecraft modding tools, I feel like it could be even better
Cause some tools did come out, but I haven't seen anything crazy come from that
some mods did came out, not as crazy as minecraft of cos
i guess the engine is just dated
hence why they are changing to unreal for new game
I think some more detail here and there at random spots would be nice, like 7 Days To Die has like buildings that if you climb, there is crazy loot on top, or like there is random loot scattered around on buildings and stuff
It makes more sense cause its a survival game, but it would be good for CP 2077 because it would make you want to explore crazy places more
Maybe one day someone is gonna make a content overhaul mod (like Cinders or Convergence for DS3 as you mentioned), I can see that happening
i want to make her but putting "huge ahoge" (which is a danbooru tag) also adds massive badonkers
Yeee, that would be pretty neat
screen recording, not super high quality but it's something
I think people wanted cyberpunk to be like GTA V, a game where you can roleplay in
300mbs 

for 1 minute clip
I mean, true, but it isn't multiplayer, and 3rd person is awful xD
So it has to be like a 7 days to die vibe imo, like make the player wanna explore and find random stuff
It do that, but it could do it better I think
And I mean, if they make a game in UE5, they got procedural generation they can do, and like AI generated structures and stuff
that's the problem with this game, it was assumed to have multiplayer after launch
this game has potential
they screwed up this, hopefully they learned
and make the next better
I'm hopefully tbh
well, at least we can be occupied with RE4 for a bit atm :)
Yeee, I'm hopeful too :O
Various individual ones
Theyre just all using the same lora
Ok thank you!
can someone help me get her appearance right?
HOLY SHIT
I JUST GOT THE WORST GEN EVER
LOL
my tests using my better sharp teeth lora. Im beginning to question if this was really necessary
generate her without huge ahoge first, then inpaint it?
what models do you all use? i need a new model, mine is a bit outdated.
this server is dead
lmao

you should use my new trexmix_V3 mixture
link?
anything V5
no
why not lol
anything v5 is anything v3 just renamed
no it's not

x d
model?
25% my model, 25% persona style lite, 50% anihex
can you put it up for download?
I mean its probably faster you downloading all of them and merging it, my upload isnt the fastest
nope, dont have nearly enough space for merging. and idk why but merging on my side is like 30 minutes for 1 single merge
one crazy new model is coming 
it'll be here in ~30 mins https://huggingface.co/Wpatzz/Spam/tree/main
thx bro
ghost stories English dub is ducking wild
i get this error when upscaling, any ideas?
try other upscalers
Good evening
something i got from my pow ideas
u think i havent?
okay sorry for trying to help
Making cyberpunk gens after playing cyberpunk?
have u made pixiv yet
made but didn't upload shit
do it bro
too lazy 
Uploading on pixiv?
well i'm too lazy to do that
i tried to improve my drawing with img2img but i'm getting bad results what do i do?
looks great tho
it's an unique style
it's mine
what's the before?
that IS my art pre-AI
oh lmao
Prompt harder 
a more detailed / generic artstyle
not sure what you mean by that, but using img2img, go with a higher denoising strength, prompt accordingly
or
you can try using controlnet
with canny module
or the sketch module
after inpaint fixing
img2img made her bratty
i mean, the quality is way better
but it's no longer the same character
the character i made has the spiral ahoge and some white point stuff on the ears
the thing is that, with high denoising strength, the details like the white pointy stuff is lost
this is how popular she is
autocorrecting name, <1k results
good signs only fr fr
this, close but yes.
#🍥|anime message that @simple oyster is a lot less, more towards the line than anything
entered the matrix
@lyric zealot your stuff is better than like everyone on there lol
well, i mean, it's really just cos i'm spamming this server like it's my own server 
https://civitai.com/models/37715/crackedfracturedshatteredetc-skin for those who wanna give it a try
its a shame noone else sees your shit
tho
not that im complaining...
good model?
now make the model do super nsfw stuff, and you are gonna be rich!
it kinda does
sorta
a bit
i swear if u post consistently on pixiv
and set up a patreon or smth
you can make some pocket change ngl

well the thing is that, I'm really too lazy. Just generating for fun
and, i think you severely underestimated how well people got with SD
or AI art in general
true
yes
i havent gone on pixiv in a while
i installed the stealth info, is it activated by default?
yours is better
should be yes
at least imo
what's that
well, I really appreciate that thought tho 
makers of waifu diffusion
well i can't really link their discord, so go here and scroll down https://huggingface.co/hakurei/waifu-diffusion
same vibe 
super cute
@pastel bluff this is the anime channel
@magic wave femboy enough? 
for me, if i encounter that error, i just restart my SD
alr lemme try

oh.... 
first try
a little girl
the first one was super nsfw
@nova remnant
Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝|tech-support for any problems you encounter while installing it locally!
Yeah im not entirely sure but most of the dataset was of near naked characters (bc thats usually the context of cracked skin) so that could be influencing it
well not really a big problem tho, just have to put nude in neg
this is amazing


it will devolve into ruining him again with the LoRa...
oh my goooosh
so cute


😭
oh kayyy 
That’s just cute
ABORT ABORT!
looks to much like a child since im gradually raising the ahegao lora weight
yup definitely ABORT THE OPERATION!
put mommy milkers in prompt, might help 
will do 
Need more femboyness 
Oof™️
teach me the way senpai 
oof,
(1boy:1.3), (otoko no ko:1.2), (flat chest:1.3) 

alr thanks
im trying femboys too
Fair xD

Im doing eldritch stuff
nooo its nsfw
forgor bout neg
The boobage :o
back pain simulator
why are you #1 smol sus simp
cuz i am :)
new kordhell song, wheeee time to feel angry
NO
lmao smolsussers
hes mine





dat graphic t-shirt
holy shit fr
cute
you took her lollipop! 
https://www.pixiv.net/en/artworks/107101823
If i can finally post on pixiv you can too !
it takes a lot of effort 
i know!
with the help of ai im at a 1/month rate so far lol
pathetic tbh
At least I have "good" art stored up 
but your gens are higher quality so you don't need so much touch up either though
lmao fr
appreciate the thought! 
well, I'm too lazy to do any touch up
unless it's for pow 
i think this doesnt pass right?
@wind girder
yet hornier things fly unnoticed 
that's why i'm asking mod 

but the perfect fingers 
o i used the ngs you posted on that lasst night
smth about badhands and artists, went and downloasded a few ti lol
ahhh the normal neg prompt i used for holokukiv2

i been usingg w Based64Mix
is it working well? 
Like nothing shows up?
ye
Discord removed metadata
lovely
you need to have the stealth metadata extension
Yep
betterdisc or auto
install this extension in your webui
webui bet oki
But if the person making the gen isn't using it there still won't be anything
Alri i used to do this allot when I was a member of midjourney and want to see if I can have a similar workflow.
I would use my selfie as starting point for some if the characters. Would img2img achieve this?
(I also have a lora training on my selfie for this purpose in the future)
To get stuff like this
definitely. You can even use controlnet for that
I never used controlnet yet 
Yep, made from school of athens 
it's a super strong tool, might wanna get into learning that 
Its on the list of things to learn, been procrastinating cause i have a feeling It will overtake me for a few days like learning sd was 😅
well, maybe start small with controlnet
like there's quite a few modules in controlnet
slowly get to learn each one of them
the one that probably works best in your case will be canny, depth and/or sketch
Sketch is the one that makes w.e into an outline right?

well
actually canny as well
but it's kinda different
I don't really know how to explain tho
is okie 
https://github.com/lllyasviel/ControlNet some examples here
goddamnit ofc the beach one i did has clyde feeling smth 
A little messy but the style 👌
This is a pretty good guide to get started with controlnet https://www.reddit.com/r/StableDiffusion/comments/119o71b/a1111_controlnet_extension_explained_like_youre_5/
thank you ❤️
nubby is such
, doesn't speak much, comes in, drop tips and leaves
this looks like allot but fun 
https://www.youtube.com/watch?v=N7OVagiblPo what did i just watch...
i want this so bad if they manage to do it
lmao saw that on kickstarter
looks great, but is it really good tho 
Basically phone keyboards but real?

True 
like, wtf are maglev switches...
it's the next "in switches" is what they said
I mean
That looks like a 1100$ keyboard
It is cool, but probably like as expensive as your graphics card from the sounds of it
XD
Exactly xD
I wasn't there.. if you want a check by the mod team, please use self report, anyone of us will check it.
And like the crew here told you too
the crouch was quite explicit. I could have seen so many camels in that pic
This is better
cool
time to sleep
the most expansive tier
good night
sheeeesh
Self report is like doing ⚠️ on one of your messages? 
Nighty! 
yes, or right clicking +> report to staff
It's tagged as self report for us so we look at it in this respect
noted noted

if it's self report, then it's basically for checking and will not get into trouble right? 

self reports, gets perma banned
XD

this is getting more and more abstract 
yo smol....
and back on track 

hahaha
I'll need to revive my graphic tablet tho, I'll probably need it and spend a few hours in Krita
and that is? 
🤖 
Inb4 smile makes a femboy alice
XD
i just realised there's no timestamp for when pow submission ends
and any other timestamp for that matters
It's at the very top of #1023999442338201721
yeah i ran into that before too
i think it was because it was slightly different format than cow?
for me it was because i didnt scroll up
or i was just high

lmao
did you report it 
Same for me when pow ended
yes
Asked when is the next pow when it's literally at announcements 
am i tripping or is the time slightly wrong?
Wat tis 
mlady
I was super sure i took a long time to generate before submitting, but the time difference between my submission and opening of submission is only 5 mins?

JUST TESTING PURPOSES

for self report











😭😭






