#🍥|anime
1 messages · Page 238 of 1


Cktc
That is a very cute style
I have more, just not more of what you wanted 
@wooden coral can you try the prompt swapping trick with those
in a1111 it should be like [Tracer prompts::0.5], [venom prompts:0.5]
I don't know about comfy 
You mean this? [cow|horse] in a field
No, it stops working after half the steps [::0.5] and starts working after half the steps [:0.5]
so it starts with a tracer and ends doing venom
its a fun trick to do some stuff 
I think thats the same thing yes, I do [::numbers] cause I find it easier to keep track cx
But can you do multiple tags inside one like [from this, that, these and that: to this, that, these...:number] ? >.>
I think separating them is better organized, you can do nothing for a few steps or mix them together for a few steps
she'll be the top student
she has a long way to go
There is a custom node for it seems, got it now but no clue how to see if it works 
should be fairly evident
try something like pixel art after a few steps
and after a lot of steps
I cant post my result here 
there should be a higher pixel art effect the lower you go
what happened
are tongues coming out of weird places 
I fixed that by adding the manhood word with P to the negative
Its kinda nsfw
I think it works?
Not really tho
its cool
its very different
since pony is pretty bad at (:numbers less than 1.0) I use that
at least base pony is pretty bad at low weight
I think pony breaks on a111 when using that weight thingy
And thats why I use comfy 
based
I feel like I didnt get it right so far
Yeah I tried a few things now and all look similar 
nvm, maybe
Thats more the effect i was looking for with this
Yep yep
It'd be better without the tongue but still beautiful
mary jane watson as a new spiderman/venom
makes changes and tries again
hmm not evil enough
cool

That's literally my account
I know 
Random robot girl 'sketch'
👍
the syntax is [from:to:when] so you can do it all combined like [tracer prompts:venom prompts:0.5] and it will be the same as if you did them separately.
looks ugly
@lyric zealot nubby called me ugly protect me 
h-how do i fight back against that? When it's fact 
don't worry, even ugly cats get pats
reported for bullying
What's the biggest images you can inpaint? Or, the size of the uploaded img2img is not so important but just the mask size?

I think with controlnet you inpaint the area only
I think you can do that with normal inpaint but it wont know what is outside
otherwise whatever you have vram for
Oh so if I take the rest of the image into context, it will take up more vram?
I see thanks, I'll play with it
How do you know what dan booru tags you need? Like maybe just search up the tags list groups and look until you like what you see?
she'll step on this

pretty much anything about a character is a tag 
Those tags can't really do much more than characters and poses
just think of something and is probably a tag
if you look up in the site for what you want it can "recommend" you the right word
if you are on auto1111 or forge, then you can use the extension
And on the site you can look up for something like "Tied up"
Then find out all the cool stuff around that
👍
example
onion girl when
Oh I see, I'll play with the tagging system prompts
Cuute
Also thats coot :O
you are cool 
No u :O
Quasarcake style is neat :o
Flexible peaches
adding to this for anyone who might be using Comfy or SwarmUI.
Comfy - use this extension: https://github.com/pythongosssss/ComfyUI-Custom-Scripts
SwarmUI - has had autocomplete built in for a while now. update and check docs here for how to enable it: https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Features/Autocompletions.md
Been trying out pony and not been able to get decent results. Prompts are usually something like this: Positive: score_9, score_up_8, score_7_up, 1boy, black hair, fear, run. Negative: score_4, score_5, score_6
is this a good image?
how many images would one need to create a decent lora on a character's appearance?
i want to create a lora based on Sivil from Macross 7 but there's like no fanart of her so i'll have to use anime screencaps (i'm fine with the artstyle of the anime ending up as part of the lora) and i'm just wondering how many i'll probably need
depends on the char, if its got many intricate details probably more than 20-30
depends, for just a face with hair style, you could potentially get away with 10-15, if you want the whole outfit any everything 30+
as many as you can and with different poses help
she looks like this
but it also depends on what model you use as a base, the closer it can get on its own, the easier it is
i'll either use base pony or the 1990's style checkpoint to train off of
how well does civitai's on-site lora training work?
too ugly
you will wake up without any toes or fingers.
edgy 90s space elf
space vampire
Can't? Or can
Nice have fun playing it because it is gonna become apart of you
i've already sunk more money than i'd like to admit into it
been playing off and on for years
debating finally doing the quest to increase my world level, been sitting at adventure rank 35 for a long time
kind of worried it's going to max me out though 😄
ideally you should try to get at least 50. small datasets make training without overfitting significantly harder. more is always preferable, but quality matters even more. at minimum you want a varied set of poses and angles.
i'm going to see if i can accomplish what i want by using my dataset with ipadapter
the main thing is, do YOU like it?! If yes then everything is fine!
Where did you get the lora from?
Oh, you made me realize I haven't posted that one yet, I'll post it soon
Damn it, I choked on my tea and spilled it!
😄
I thought she brought me new tea... but she shamelessly drinks it herself!
she ate all the snacks too
Genetic experiments in the laboratory of the villainous scientist Boto continue! 😉 😁
I have a question. Can the fisheye lens effect be achieved with tokens or does it depend on the model?
i have done that with several 1.5 models
i.e. is it still a matter of model dependence?
i dont think so,u can probably get it on any model just try it like this (fisheye:1.5)
I've only reached strength 1.3 so far
(fisheye:2.5)
2.5?!
yes,but this lora works better https://civitai.com/models/135838/fisheye-slider
Thank you!
2d ? 3d ? anime ?
colored sketch! 😉
cooked itadori 
you didn't cook it correctly.😁
like how is this?
i'm looking for a more fate style
fate style!?
do you know fate stay night?
i totally in love with the style of ubw and i'm trying to raplicate that style
Understood.
the second one is the chibi version I guess?
if i say you that the cibi version is the top one? XD
and of course I merged the western mix with the realcartoon models. to create my own.
works well 😄
so far
havne't fully tested yet 😄
merge
non merge
hmm
wow ok this is a good model merge someone did
westernbeautiful_v10 on civitai
Migu?
I ffound with coveralls it keeps forgetting the shirt
I mean around here so do the one's who wear them usually , haha
ok somehow this is kinda creepy
I hope fruit just stumbles across those 
morning
morning
its 2:30 PM 
good after morning
ditto
didn't handle sticking tongue out very well
ok whats the theme for the day so far?
any?
it's 8:40pm
@lyric zealot srtuoois
cute I want this maid!
she not for sale!
April O'Neil as Venom

she is a vegetarian 😁
Miku Venom
the kitchen is over! The apple sat there until it took root!
crap! it had to be the Apple Monster! and not a formless something!
cool vase!
Now it looks like a still life!
hemp isn't illegal
Well now it looks like a blueberry girl playing with Robin Hood
These are all experiments by Boto!🤣

hmm odd, yeah i'm gonna retract that one. not sure what that is on the legs but .. oi
Miku's new look!
What an apple a day does to you
I'll stop eating apples!
You dont want blue hair?
Here you go. this is better and looks like a costume!
strawberry-Apple Miku Goddess
O! strawberry!!!!
Reminds me of this old gen I did
Doesnt look comfortable
I'll give half my kingdom for the idea of a prompt!
Half of zero is zero
who knows, who knows... kingdoms are different
adds HR Giger for giggles

SO @wooden coral I guess you aren't seeing my msg's
switches back to a more animeish model
I tend to ignore those, getting too much lately
What do you need
Yeah thats probably the best approach if unsure
i'm banning you
not sure i've seen this angle before, interesting
This is down right incredible
It's a pretty basic prompt, "lying, on stomach, legs up"
close-up and foreshortening as well, didn't use anything special other than that, NAI's image composition is really nice
try it with a tifa 😄
there's actually a specific tag for that pose. the pose
Oh wow there actually is 😭, that makes things easier I guess
your method is also good to keep in mind though because the pose will tend to be slightly more to one side of the character.
I dunno how people know all these tags 😛
Just go to danbooru and explore 
Yeah I use full body instead, I'll try to mess with that tag a bit tho
tags are just easy for me to remember and I've been doing this a long time. the booru wiki is also helpful for finding new related tags, for example if you search feet on the danbooru wiki, you'll find the pose under the related tags, then you can click on the pose to find out what the tag actually means if you didn't already know.
just type booba and see the similar tags
Become one with the booba
NAI is down again ffs 
Anlas compensation never
They really are stingy with the anlas for some reason
Because ez money
@finite herald
wow cute
Makes sense 😭
Though I do hope they have fixed some of the issue I've had back then
Like black image outputs lol
-20 anlas

Strange, I've never had that happen to me
I've only gotten black images a couple of times.
It's no biggie if you have opus
@keen hollow here you go

you should know that fruit can sense every new miku image posted here
No she missed this masterpiece #🍥|anime message


OMFG this is pfp worthy
The High Priestess
The Empress
The emperor
that one was tricky
The Hierophant
The Lovers (mike and miku)
hmm the chariot card is tough.. hmmm ponders alternates
Yay tifa! She needs her red shoes though 😝
Yay tifa again! Also no red shoes. "The pose" must be barefoot
I think it can be with shoes as well, I was just too lazy to change my prompt 😭
No worries.
Maybe I'll try it later. Gotta get it in my weird 3840x1600 ultra wide res
Strength
The Hermit
The Wheel of Fortune
that was tough
Justice
Hanged Woman
can't believe th at worked on first try
You don’t really need "the pose" from what I've tested, "lying, on stomach, full body, legs up" works just as good if not better
Did some Inpainting on her shoes and skirt so it's not really accurate but good enough I guess 😭
Miku 
yep
thx
Temperance ?
The Devil
the tower is proving difficult
guys
if i have a image from instagram, how can i see what model/lora the person used?
i did but no response from the guy
couldn't figure out the tower so I did "The Joker" instead
both are chaos
or change
This little one wants a hug
guys
if just basic upscaling then go under Extras tab
The Stars
if you generated the pic before hand, and know of the model that was used, you can try doing img2img
how?
^ this way will change the image slightly
ye
You got the red shoes and eyes so pretty good!
wat model?
select the same model, write the same prompt, and just put the image you want to upscale in img2img, then resize by however much you want
The Moon
but keep the denosing low, like maybe 0.1 - 0.2
tPonynai3_v55
The sun
oh, ok ty
i use autismmix
but i like the little realistic touch

might be possible on autismmix as well
let me try
k
should help with the eyes
or should i say the face
auto detects the face, and auto inpaints it
the original

should be
?
I thought they were doing i2i
hires fix?
ye its all on img2img
It's automatic i2i after t2i
oh ok
okay i didn't know what i was saying then lmao
💀
Judgement/Judge?
i don't do much inpaint, so unless someone more knowledge come and help
if not, i can give you my low level tips
i just want better eyes mate
tho now i'm gonna use some other tools and actually put what they are on them 😄

one sec. let me start forge so I can grab a couple screenshots to help
i see nubby 
and post on my site 😄
I guess I can stop Forge start up then

no this one's a better, World
Here's moon for you
wat i cant see ur parameters
you need this extension to see all my generation
Stored in alpha layer
oh...
or i can just give you the proper metadata here as well

might be the prompting then

this is t2i right?
ye
what are the base resolution are you generating at, and are you using hires fix?
1024 1024
no im not using it
then try using it
do something simple like, hires fix at 10 steps, upscale by 1.5x, denosing 0.4
then use some upscaler
like 4x_fatal_Anime_500000_G
how do u use those upscalers?
this is what i usually use as my upscaler, but there's others the people will recommend
shareX? 
Yesh
^
there isnt a ESRGAN folder
Cuute
what ya think?
Imo it needs something like this behind that text
Are those words generated as well? 
Here you go. The important parameters are the ones I circled in red in the last screenshot. Inpaint masked should be default and will make it so that the area that you draw your mask (see the white area over her face in the first screenshot) is the area that will be changed in the image. Only masked makes it so that the image will be cropped down to just the area around the mask, then that is upscaled before inpainting happens, and finally it's downscaled and stitched back into the original image which allows for more detail in the inpainted area. Denoising strength controls how much the image is allowed to change, I find around 0.5 is usually good for faces.
yes over on cooltext.com 😄
hmm good point
ahhh, not on stable diffusion, but still good 
oh ok ty
can a mod just pin this @wooden coral
@lyric zealot even after all o that the eyes are still not detailed enough
show example 
well, you eyes are definitely more detailed than before
but to get better result, here is where you need to use your adetailer
since your subject face isn't that close up, adetailer will help a lot
Not gonna get better at that resolution imo
so too high of a resolution?
Actually my browser probably trolled me
xd
the eyes are too realistic for the image style
Yeah browser decided not to load actualy image but compressed version
then it's probably prompt issue or model issue
1024x1024 is good for a starting resolution on SDXL. I usually do 1.5x upscaling with img2img at 0.5 denoising (this is the same process as Hires Fix does on txt2img). Then I also run adetailer to inpaint the faces to even higher detail.
we don't? 
i blame fruit. we should withhold all miku posting until it is returned to us.
it's gone with miku-mash
There's always animeai
oooh
totally agree! 

Almost good hands
while we're making demands. we need miku role returned as well
Looks perfect to me
broo where anime eye
man NAI has some great eyes lol


pickle you say?
pony/animagine could never lol
1pickle, waifu
We need someone to get a highest tier nai account, then hook it up to a discord bot
For “free” generation
:D only eyes tho those are way too realistic for the anime-like style
I think my friend tried that but the discord bot broke in the process
ooh
grinz evilly and cackles .. PICKLES!
Tbh, in my head, this idea works flawlessly, but when implementing it, there’s probably gonna be problems over problems
If their api doesn’t change, then the bot should work just as well
But someone will need to pay for nai account, and hosting of discord bot
And I’m too lazy to do that right now
prob would need to get permission from server too
I know one server that has that kind of bot and it at least needs queue since to my understanding you can't gen multiple images at the same time with one subscription
damn bro
got the almost perfect pic
with 1 issue
trying to use inpaint to remove it
doesnt seems it wants to
there's this also
but it tried on one leg
yup, queuing system should be fairly easy to do
kek
ye sadge

just try like 0.2 img2img without leggings on prompt and it should correct
diffences?

You can choose color to paint over area
ye cat doesnt work
try like 0.35

problem is that they don't want you sharing the sub and watch for suspicious behavior. in a small server it would probably be fine, but in big servers they'll just ban the account when they notice that it's being used 24/7. as far as bot hosting.... i already have two instances on oracle cloud running for discord bots. just saying
got this, txt2img
Here's fake Yue image
then what's the point of them having api? 
theres a nai bot on another server im in but people can only use every 30 minutes 
yeah i think that was their intention with it. the vast majority of my gens were via api in Comfy since it allowed me to to local hires fix instead of spending anlas.
ooh
makes sense i guess 
@nova remnant
0.7 denoising
smooth legs
biteable
make someone biting her ears
🔥
no u 
:3
mine is just hot and ** t h i c c **
I swear my images generate slower on some days...
||that's not what my girlfriend said|| 
||AYOOOOOOOOOOOOO||
color looks weird lmao, i love zerotwo's horns
it hides between the hair
@lyric zealot 
i love her
well, whatever model im using, at least have some slight knowledge of her
pls tell me i can remove the double headed hat
meanwhile I'm generating balls 
i think you got a bigger problem of having two right foot
licks
what the
Maybe I need to gen some fang images since I can't find good ones 
trying zerotwo?
it's in the prompt, but not really trying to gen her, just getting her style slightly
oh ok
analyses the thickness of both

hmm almost
why not both 
no idea
umm.....
ball
too many wings 

now I feel bad because I pinged them 😛
come on 3060, gogogo!
ugh when you want it to be fast it's like 'hang on, i'm tired'
then it generates a garbage one and you gotta start over
@keen hollow sd3 made tifa very emo 
not sure that's even tifa anymore 😛
lmao
I just tried one wearing different clothes
but its bad at her exact appearance
how can you do so many of those real skin/clothes weird gens 
wym?
they seem normal to me
lmao
There we go
probably cause i actually post them here
most ppl probably just dont
also i suck at prompt
so :D
i think it looks good apart from her third knee
they aren't shadows lmao its the leggings appearing as shadows
yeah lmao
oh yeah it's missing a leg
on mine i mean
remove her leggings
no what
why stop there
yeah...

@nova remnant this is what u like?
I'd let her be my sofa 
xd
me 2
for me its too big

i would be her sofa 
AYOOOOOOOO
There, u can download it if u want
I fixed it
tf was fixed

BRUH
mini tie
good tie not looking at tie
ayooo wat
❤️
bad fingers but....
taalive
upscaled
but i kinda dislike seeing the bellybutton 
Cute! 
well is the one with metadata
it was the sports wear one
I take nap now
yes
set whatever you want first, then go to Settings and all the way down in Others > Defaults
Now for him to not die next week
xd
Don't remind me.. 
wat
@zinc flax
can u explain to me some of ur prompting?
also wat model
idk what's there to explain though lol
the /
That's to separate the series name from prompts
im too lazy 
its so simple
wym?
@nova remnant for u ❤️
not da smokin 😭
Welp, someone might be able to explain it better lol
oh ok
@noble flame are you confused by the metadata of comfyui generation?
the prompt

holy cow THAT'S INFINITE/10








