#🍥|anime
1 messages · Page 196 of 1
they'll work best on Ao models such as soushiki, but they work ok on other good models like kakigori
another question if you are still here. Are there any these type of lora for sdxl that you know of?
I don't think so. Ao and one other guy are the only ones I know who have experimented much with Diff Loras and I don't think either of them have made any for XL. Ao mentioned that he might start experimenting but I know he's been busy with work.
yeah believe me... Im already downloading

do we still need sdxl refiner? for these anime models?
sdxl refiner is pretty bad for all finetuned models tbh. it's especially bad for anime.

better off just running hiresfix at around 1.5-2x upscaling and 0.5 denoising with your choice of ESRGAN upscaler.
I see
There is that XL offset example but I'm sure if it actually does anything
you are unsure right? 
Well I just placed it in my prompt and believed it worked but then I started to realise that I still got lot of bright images

But I'm planning to test it soon after done testing couple style loras
let me know the result 

im too lazy to test stuff right now 
go make big dragons


u bought the ticket for the wrong movie
@lyric zealot I'm not rly sure which name to use for this lora since file name is different from what I get when I click from lora tab but the difference isn't that much rly was using Autismmix_confetti
i created a f##king mod support ticket u numbnuts
thats cool but you are at the wrong location
so it's useless? 
Feels like it


no u
Yeah..........
we kinda derailed from the black theme
booba theme
👍
crystal hair is a tag, weird one
getting dangerously close 
I don't know what you're talking about
I am a simple SD user in this server
@lyric zealot try u_u as a tag

yours work?
he fell in a dungeon
btw guys try typing Diao Chan in your anime prompt
yup
no idea what she means
giga game
👍🏻
"pouting chibi witch" 🙂
Mhm. Probably add some words like ' cute' 'gorgeous' 'beauty'
I have a challenge
Make a girl riding a meteor in space

My prompt?

I was trying it
and it wasn't being cute
I forgot serious look at the end of the prompt
lmao
This model tends to make them smol if I type cute
smol like me? 
u ugly
ok...
Just because you are offended, man?
Bruh...
ehhh, we are just joking
we are always like that so....

i was joking too, than i got an arrow on the knee <.<
@signal panther dungeon siege made butt
👍
I don't think this model has any idea about the game, its just doing things
I left 2 years ago
thanks for the lore update 
Try making it supercool
Me trying to proccess what that would be as a prompt
add some ice
you can't, it's an asteroid in space, only a meteor when it enters the atmosphere 😄
dragon siege
oh wait, dungeon siege.. crap
damn, that saga
I was gonna correct myself
lmao
a plot xyz of a lot of artist tag, provided by nubby
here's one pic, if you needed pnginfo
feels like my browser is going to crash every time I move it
yup, so just download it, it's better
download and use your stock windows application to open the png

@lyric zealot what prompt did you have
i literally shared this
I don't have bionic eye to see the prompt inside the image (meme)
I like this artist
👍
perfect 
for lazy people,
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, rating_safe,
1girl, black hair, bob cut, red eyes, maid,
source_anime, masami_chie
Negative prompt: (worst quality, low quality:1.2), 3d, censored, abdominal muscle, artistic glow
Steps: 35, Sampler: Euler a, CFG scale: 7, Seed: 88057677, Size: 768x1024, Model hash: ac006fdd7e, Model: autismmixSDXL_autismmixConfetti, VAE hash: 63aeecb90f, VAE: sdxl_vae.safetensors, Denoising strength: 0.4, Clip skip: 2, ADetailer model: face_yolov8n.pt, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer mask blur: 4, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer version: 24.1.2, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Mask blur: 4, Inpaint area: Only masked, Masked area padding: 32, Version: f0.0.7-latest-44-g7359740f```
the last tag on the prompt is the artist tag, in this case, `masami_chie` is the artist tag
I guess I'll stick to the a1111
comfy kinda going around for me
and lora and other stuff pretty much easy in a1111
Okay time to learn live2d, anyone know basics or smthin bout it?
that didn't last long lmao
comfy is way too complicated indeed, you spent more time doing the wire things than genning
but its kinda ehhh to wire all the loras, hmm I need to try it more when I have free time
that's what she said
hm?
i still can't find the same creativy on auto

in comparison, auto looks more lackluster
umm I guess
try to use supermerger on comfy
merging is a different thing
or lora previews with example prompts on like 2k loras

with the same prompt, on comfy, to me, things are alvays better
maybe that's because i invested too much time on the workflow and i know how it works the best
probably

btw can you get a good decent image as you think when using comfy UI? like Im asking is there a big difference between a1111 and comfy?
I guess there is probably but need to hear it from a user

i'm not to much in precision, i like comfy because it constantly surprises me
@steady grail https://www.youtube.com/watch?v=PGjJjrjTlNY cool listen 
some nodes are there just to makes results more chaitic
i got jumpscare by the pic
unbeelievable
but when i use ipadapter, usually, it becames predictable enough
idk what you're on about
auto in results is more coerent, but less creative
how i use comfy, is way more creative, but meeting a target, becames more laborious
but honestly, i find ipadapter very boring
hmm I like to invest my free time to comfy now....
@lyric zealot do (white theme:1.5)
now thats creative
nicesu
you take care of it now, its your idea 
i tried, but nothing happened
something something
👍
UwU
lmao
i was able to load xl turbo, but i'm having problems with xl <.<

is impressive how she is drinking the beverage, on the 2° image
xl turbo models, are smaller than classic xl?
another good result
smol, that model is xl?
yup it is
smol has autism
yep, the forms makes more sense than 1.5 once
i can have a textual prompt? i'm testing an xl turbo model
1girl, hatsune miku, drinking, cute, nose blush, looking at viewer, bubble tea,
source_anime, bee_\(deadflow\),
Negative prompt: (worst quality, low quality:1.2), 3d, censored, abdominal muscle, artistic glow
Steps: 35, Sampler: Euler a, CFG scale: 7, Seed: 3378921552, Size: 768x1024, Model hash: ac006fdd7e, Model: autismmixSDXL_autismmixConfetti, VAE hash: 63aeecb90f, VAE: sdxl_vae.safetensors, Denoising strength: 0.4, Clip skip: 2, ADetailer model: face_yolov8n.pt, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer mask blur: 4, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer version: 24.1.2, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Mask blur: 4, Inpaint area: Only masked, Masked area padding: 32, Version: f0.0.7-latest-44-g7359740f```
i don't know why i screenshot that lmao
eheheh

im using automatic1111_forge
KAY...then lets try it also
maybe i should start impainting, more than prompt everything in a single block and letting the model gen magic
it's basically auto1111 webui, but a fork version
Basically a1111 but better
it's made by the person that made(?) controlnet
or maybe the person that migrate(?) controlnet
indeed autism

but for how long they can follow this rhythm?

im still waiting for my live preview to be fixed
a masterpiece lol

alternative paths where you can load models
insert the path and remove @ rem
didn't wanna reply to that, sorry 
luxury
souka thanks
the hell is souka
maybe i should try to load xl on comfy. i rememmber why i stopped using auto
i believe you can only have one ckpt-dir
so only the latest one works(?)

i tried that as well, but only the second one works
oh and, don't mind the --no-gradio-queue
you can only set 1 alt directory, but the ui will still load from the main onece, if there is stuff
exactly that
how do i set, the code box in chat?
mikuu RUN
why
the key beside you number 1
@echo off
`set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=
@REM Uncomment following code to reference an existing A1111 checkout.
set A1111_HOME=D:\Git\ComfyUI_windows_portable\ComfyUI\models\checkpoints
@REM
set COMMANDLINE_ARGS= --xformers --always-batch-cond-uncond --theme dark --vae-dir "D:\Git\ComfyUI_windows_portable\ComfyUI\models\vae" --ckpt-dir "D:\Git\ComfyUI_windows_portable\ComfyUI\models\checkpoints"^ --hypernetwork-dir "D:\Git\ComfyUI_windows_portable\ComfyUI\models\hypernetworks"^ --embeddings-dir "D:\Git\ComfyUI_windows_portable\ComfyUI\models\embeddings"^ --lora-dir "D:\Git\ComfyUI_windows_portable\ComfyUI\models\loras"^ --esrgan-models-path "D:\Git\ComfyUI_windows_portable\ComfyUI\models\upscale_models"^
call webui.bat`

splendid
that's mine
kore wa deska?
--always-batch-cond-uncond is like low vram or smthing?
sokka soka

lmao the bee
oh deym
Why are there bees
ah ah its downloading it again
how about her?
quite relaxed
👍
Hello u

autism?
yes 
too late
splendid
her honey 
Fellow bee enjoyer 🐝

`@echo off
set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=
@REM Uncomment following code to reference an existing A1111 checkout.
set A1111_HOME=F:\AI\stable-diffusion-webui\models\Stable-diffusion
@REM
@REM set VENV_DIR=%A1111_HOME%/venv
set COMMANDLINE_ARGS --xformers --always-batch-cond-uncond --theme dar
--ckpt-dir D:\SD\models\Stable-diffusion
--hypernetwork-dir %A1111_HOME%/models/hypernetworks ^
--embeddings-dir D:\SD\embeddings
--lora-dir D:\SD\models\Lora
--vae-dir D:\SD\models\VAE
call webui.bat`
not working
What prompt do you have for the stinger / tail part?
if you are putting it in different line, then you need the ^
i didn't prompt for that, so 
just bee

Spark hiding in the bush waiting for the bee girls to appear
or bee_girl

power of AUTISM?
eh its pretty
I also have the bee girl with a larger rack but am not sure if I can post it here...
send it in dm to @nova remnant, he will judge it for you

autism mix confetti
nicesu
You can call me a bee expert
👍 let me see that one in dm
not a monster expert?
only a few

Show me the honey pot

I didn't type that
My cat walked on the keyboard
Wait why are her arms floating haha
I just believed they were on her knees for a minute
deym didnt notice dat
downloading other one
Stealth level 100
@lyric zealot know any words to trigger long long skirt?
Not sure, but long skirt might helps?

Anyway, if you don’t have the autocomplete extension
Go get it
It will recommend tags based on what you write
oh shoot I forgot to add dat
wait can we link extention too?
Skirt_
You mean link here?
Or auto1111 to forge?
I don’t think it’s easy to link auto extension to forge
I just downloaded a new copy
what bout copy and paste complete folder?
Wtf when I add "fully clothed" to my prompt it nukes the image. Actual nightmare fuel

I didn’t try that, so not sure
hmm

lemme tru
Not for that tag
wait wats dis modal?
animaPencilXL

Ye I know just thought it was funny/interesting why it specifically blows up on "fully clothed"
is it correct?
Probably. It was the base anime model that came with Fooocus
I dunno something something model weights shenanigans
A bunch of stuff breaks it if you put low weight
I tend to keep everything on at least 1
Yeah same so idk why it happened
Long hair not efficient
Speaking from experience

I see
@lyric zealot go sleep
Yue would like to say goodbye to you
What's this prompt lmao
Cuuute
Pile of cats XD
isnt it showing on png info?


`parameters
flat color, chibi, girl on cat on giant cat, sleeping, simple background, girl, long hair
Negative prompt: (worst quality, low quality:1.2), 3d, censored, abdominal muscle,
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 10, Seed: 145265264, Size: 1024x1536, Model hash: ab4ea1ebbe, Model: sugoixl_v10, VAE hash: 235745af8d, VAE: sdxl.vae fp16 fix.safetensors, Clip skip: 2, Version: f0.0.12-latest-110-g15bb49e7`
I think its about year since I was told that prompt inspector bot was in development
what bot?
Prompt inspector is discord bot that checks if image contains prompt info and adds reaction to image which users can click to get private message with image prompt info
Oh seems pretty good
can someone explain to me how pony works? Still lil confused for that model
In the oven
like the source_4 etc stuff is there somewhere like a website or smth I can read up on understanding prompts more
and how they work?
Put this in your prompt and you good
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, source_anime,
Just special quality tags for Pony model
Although if you use pure pony model then you may need furry stuff in negative unless that's what you are after
And rating_something
whats rating_something do?
Something being safe, questionable or explicit
Ohhh
If you wanna force that
and whats the number ranges?
It's one prompt
ahhh
I think it's these 2
rating_safe rating_explicit
alr thank you
Basically SFW and NSFW
There is questionable
is there a website i can view that has the list of all these kind of prompts for Pony?
Didn't specify that since I have gotten quite questionable stuff without it
Well, with it, you'll get questionable stuff
@supple raptor Send Elp
There you go. (Deleted old one to fix stuff)
That's pretty much all
deym shes good
source_stuff not very useful from what I tried
what does source_stuff do exactly?
May be challenging
Have you tried turning it off and on again
Sadly I don't have access to NAIs servers
"The queen and her trusted worker"

是不是不能用了?
你是什么意思?
请问是不是不能画图了?
当然是可以的。
另外,你可能应该在这里用英语问。
ChatGPT is legit such a good translator 
大哥,我英语不好啊,我找不到画图的频道了
这里没有人会说中文(我在翻译)。如果你想找中文的绘画资源,你可能应该去别的地方看看。
我用翻译器翻译标签生成图片,但是现在的频道没有了,我不知道在哪里可以画,大哥你知道在哪里画吗?我用的是手机软件
Stable Diffusion 不是手机软件。你可能在找 Midjourney。那是在另一个 Discord 服务器里!
不是,我以前就在用这个,大哥,我的意思是我用手机版的discord,现在找不到这个里面图片生成的频道了,是不是被撤下不给我们用了?
Was there ever a channel in this discord where you could generate images?
Or is this guy just looking for midjourney?
是啊,我也喜欢,这个好用,还可以生成视频
Source_anime or source anime? Does it matter?
Very nice wings
i used the tag "dragon fairy wings" if you want to try and recreate it but I did do some inpainting to get them flame like near the bottom
Took a lot of inpainting and fine tuning but I really like this one
cute
Jesus, that was a lot of Chinese characters above
对不起 🙏
Source_anime, source anime and anime all seem to work
Slight differences
Bing Chilling
没问题
This goes hard ngl
@lyric zealot heyy can i ask you a favor can you turn this
Into this style?
Art style of your don't mind doing it
Cuz uhh my laptops confiscated for abit
drop me some prompts I got 1080p yutamix/mistoon merge runnin
biden skibidi toilet fortnite burger sussy mogus israel gaming palworld
say less
less
fatire burger 🍔
you meant img2img? 
spicy 1girl,solo,sitting,giant snakes around her body,1 hand up,holding apple,beret,fishnet legwear, very long hair, blonde hair,
im playing game right now tho 
🫴 as you wish
Give me a prompt, and I will generate a 16 step version, and a 150 step version, and you have to guess which one is the 150 step
Here is a 16 step/ and a 150 step example
ping
/daeam a flower

what sampler? first is 150?
DPM++ 2M SDE Heun Karras

why do you think the first is higher steps

send me more prompts, I have some of my own models I am testing out
some with and without vae's some with loras some that are merges etc
is it not? looks a little more detailed
not sure tho
yes the left one is the 150 step
yes it eats away at any detail and makes it uniform
but also because why i asked for sampler, cos if it's euler a
then it might introduce more of these when generating at high steps
that dont even look like ai
good shit
that's from @glossy eagle #🍥|anime message
What other samplers you recommend
o
yo guys
I havnt done any 12 step testing with 3M
what are you talking about?
i usually go with the karras one, not super sure the name cos im not in webui now
ah
but now with sdxl, im usually on eular a
ykw I will do a lil test
we talking about high steps vs low steps, and samplers
i usually do some streps in euler and complete in karras XD
I will gen with 5 different samplers
that's only possible in comfyui
But i need a prompt
except maybe you can do that with refiner(?)
my typical prompt, 1girl, black hair, bob cut, red eyes, maid
then add whatever quality tags you want
for a little while my testing prompt was just a katana upright in the sand
usually, 16 20 steps are good enough to get a base
by doing multiple,s you add details
if 16 is good, try for 32
of 20 is what you like, go for 40

no we must do 150
i was playing with 150 steps when i have the tensorflow thingy
since it almost 5x my speed
for what i experienced, 150 is not always better
ok 3M Exponential 150 step Hires. fix esrgan anime
over 60 80, becomes pretty meaningless
still in my experience, exponenciaal makes images more luminous, but more coarse too
euler a hit the wrong dies i guess and it crashed my pc but I restarted and here is euler a
what res are you doing?
960x540 then I use hiresfix to get to 1080p which is 2x
and I only use 3 hiresfix steps
not sure if im doing it right
There have been a few more anime SDXL models out in the past couple months. Are they better than the "general" anime models out from before or are they just the usual blends that aren't anything special?
i find pony pretty cute
animage release a new version that looks like it was polished with oil and has a light source that cant be removed from each generation
i really really like autismmisSDXL
sparse noise
I have 3 models on civitai the 2 pentellion ones are pretty outdated now
I see. I'll write all these down and try them
so much so that, im now just generating on sdxl based model now, and seldom go back to sd1.5
the mineral merge that I use is on there
the issue im having is that, i can't use all my old lora 
it's incredible how colors are more fluid on xl, even without upscaler
orangemix safentensor is a crazy vae
all colors look same
if you really want to blow your eyes out with colors, use orangemix vae and Colorize lora
i use the vae nubby posted that one has a higher contrast/saturation
I think that is the only way I can get to 150 steps and not lose any coloring or lines
I use 3d anime characters in blender then use img2img and 1.5 is best for that (Mistoon Anime) but SDXL models have a more detailed output. But they are less coherent because it changes more things like the eye direction
one of the better ones

wai wai wait
btw what the fuck is this button
copilot?

any one else have some prompts I can test
that looks liek your start menu or smth
im on windows 10 btw
no it's free
here is a watermarked gen I made
I'm win11 so that might explain differencfe

copilont bringing back the glory day windows 11 pls purchase a key watermark 💀
this is how it looks
yeah that's your start menu dude
HUGE
click on the "Ask Copilot" bit
in Win11 it's a dedicated sidebar
now gives me prompts🫠 pls...
Edge is a good browser tho
yup so it's stupid on windows 10
I switched off gx to waterfox then to brave after the whole screaming dude thing
i did try a bit
you launch opera gx and he scream OPERA GXXXXX
i was also playing with gemini img generation
I thought Opera was dead
LibreWolf better
I just found that brave seemed faster, and a little less cluttered
im back to gaming so, 
gl hf
How can make Forge use Controlnet from A1111
lets refrase that its a little to...
is fixed
still no like
which face seems better, personal op, left
that is a detailer
overall face shape is better on the left though
it gives depth to the eyes which I guess is dependant on the gen
especially side shots then a detailer will do some anti-magic
yeah its hit or miss with some models too
here I would say a detailer worked its magic
fyi the clothing im using is black red kimono
i've noticed if you're just doing a closeup shot of a single person, adetailer kind of sits in the 'completely optional' category, assuming it's a model that can do a decent face
playing with dal e while testing xl
adetailer is still good to add prompts to the face, without compromising the whole image
but, if the subject is further away, or if you have multiple faces, adetailer really does some work
i love to use adetailer to add random expressions the the various faces
they are very common
i just found out you can prompt ^U^ expression
or OwO expression
eheheh
looking at viewer
it only is good until you get a dude.....
ayo
aight that send me to sleep
see ya
with ipadapter for the overkill
Yes
cute, but it's burn out
Umm last can you img2img normal human thanks appreciate it one img is enough
maybe a lower cfg or reducing the weight of the dapter will fix it
nah
it's a bit oversaturated, but not burned
or maybe an img2img without ipadapter and lower denoise
thats because of cdtuner with saturation
oh
i don't have a problem with it, just sayin
k
the composition is almost perfect because a real artist illustration was used as input
real art 
is this a lora or just alone prompt?
lood art 
It's autism 
It knows a lot of characters already
I don't know her name
Do you 
⚠️ Don't be fooled by the new logo. They are lurking underneath the coat.
okay the best one so far
PERFECT TYSM AGAIN 100X TYSM i can write 1k things great about you
you're welcome 
@lyric zealot how is dat game? good?
good, really good
but i mean, i've play helldivers 1 so
i kinda know how it works

I tried but have to fix something with the controlnet I think
Draw a picture using the numbers 14 13 19
there you have it
need bloooood
she likes snakes
sd base finetuned by themselves
ofc

@native halo
you ever felt your face before
She refused to let her go. Sauce?
Serial experiments Lain
you don't know lain? heresy? remember!
LALL
make this char 
What about now
What even is that
SMT 
@native halo is pink a good color 
i prefer golden color 🗿

but not as wise as him 🙏 🙇♂️

make her invisible
And answering them "Yes yes you are totally correct"

Now you make me wonder
What does the AI do if you prompt for an invisible girl
nothing?


yeah
hahah... no
lora actually the model knows it
lol 😄
they should unironically bring Clippy back
oh good one!
nice vibe
cute expression 🙂
its pretty fun to do "anything theme" in the prompt
that was pink theme:1.5 I think
:p
I'm too stupid not to laugh at these xD
if only I could somehow get this clean of clothes colors on my good model
hmm I see whta ya mean
NAI doesn't want to co-operate
changes models, and settings but leave the prompt alone and adds some extra :D
waits
i foundsomeone with a messier workflow than mine
i must say that this deserve some credit lol
UwU charming
@near elbow have you tried pony/autism yet ? 
1 meina 2 mistoon 3 yuta-aqua 4 lsx's mineral merge 5 OG Counterfeit 6 Versamix
pony is on my list as well this is animagine currently
counterfeit is... specialized, if i remember it from SD 1.5 days (or is that still 1.5?)
"cute succubus in black sweater"
3.0
its really gooood 
she says i will have to try it
bit different to use than other models
how so?
Like its a completely new model on how it was trained o:
Idk how to explain, it doesn't work like the others xd
sounds complicated
it's more like a base model, needs a lot of prompting tags, but it seems to be a lot more flexible because of it
natural prompt into tags at the end seems to work really well too
mecha pilot
but I'm too lazy for that 
🙂
the easier a model is to prompt (fewer words), the more specific it's training probably is
but the absolute best part about pony is the anatomy
feels nice to have good hands most of the time
what i loved about SDXL is how good the prompt understanding, compared to SD 1.x
out of these 3 which one should I attempt a merge with autism
*Nom
neko crowd
ponders a prompt.. hmm any suggestions?
Did you have ADetailer face model on?
ADetailer 7 years later
no, with normal SDXL i skip it... this image could benefit from it though
super minimalist version 🙂
I know the feeling
anything with "neko geisha gundam sakura" can't be wrong





@nova remnant
