#🍥|anime
1 messages · Page 197 of 1
upside down is really hard normally
cinematic film still neko geisha gundam sakura . shallow depth of field, vignette, highly detailed, high budget, bokeh, cinemascope, moody, epic, gorgeous, film grain, grainy
Negative prompt: anime, cartoon, graphic, text, painting, crayon, graphite, abstract, glitch, deformed, mutated, ugly, disfigured, loli,underage,teen,child
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 519892701, Size: 1920x1088, Model hash: 1449e5b0b9, Model: animagineXLV3_v30, Style Selector Enabled: True, Style Selector Randomize: False, Style Selector Style: Cinematic, Version: v1.5.2
Time taken: 10.5 sec.
native full HD, no duplicates, impressive
another full HD result, love the depth of field

standing out 🙂
When main character sees the next character of the story in a crowd
I am merging yutamix aqua and Counterfeit
wait just 1 second, i didnt just create that neon shading that spooderman has. and also, and look how it did the eyes with no added weights to the prompt
eh why there is a black line at the top?
the last 3 have colorize lora, 1st one is with no loras
colorize at :1 power. added a curl to the horn
probably side effect of "cinematic" in the prompt, really wants to be widescreen
sigh,

whoa
Colorize and orangemix blasting your eyes out with coloring
I didnt prompt rainbow hair, the power of the vae and the lora are doing some foolery with the coloring
150 step, idk where the lettering came from it started as a line
btw whats the model?
ehh but I like those cinematic style, I was blown away
I can only find this one
thal looks like it
enjoy
well thats kinda crispy
thats better
even better
some post processing in GIMP
😄
@brittle rampart 
This is a greeting that can be used in a variety of situations, such as “Ohanbanchiwa”, which was created by a first grader. Also, Renge's character is a first grader, but she is smart in many ways. He has a variety of knowledge and has an “all 5” rating on his report card.
What does “Nyanpasu” mean from the main character Renge Miyauchi?
This is “Ohakonbanchiwa” - how. Welcome phrase. It is suitable for many situations. Renge is 1st grade, but she is too mature in many ways, has a lot of knowledge and a report card. It is suitable for many situations.```

really?
interesting
man I haven't touched SD or anything in so long now
(wet:1.8)
wet 
If u can remake the old server ,temporary
need more sunset, beach and sea i hope and no no more?

she disappeared
funya magic
😂
btw. i am not delete it

They are watching!
@brittle rampart Damn beach haters! :p
kawaii haters!
Isn't it hard for her?
What do you mean ?
a lot of horns
probably
What if I do (extremely long horns)
Didn't change much 
Didn't do change much too😁
Itsuki
coot
🫣
its from the other type of anime
What's that 
Swimming anime
👍

these flowers?
yah yah

Awww nice
another variation
u need a lora for that,or controlnet
eyyo can u pls give prompt and models pls
doesnt it show in png info
?
This is why we want Prompt Inspector bot (aka PI-Chan)
i want the peechan too
😔
open the original image url and copy paste it into an online exif viewer website such as jimpl.com
That sounds like lot of work
I mean just clicking reaction under image and getting this kind of dm is much easier and no need to manually check if image has meta
Anime waifus just like walking in water
ill just suck on it
how it work @gloomy scroll ?! not understand...
It's just a bot that can read the image's info and make a react on your message, if you press the react too it will send you the information through dm
But there isn't one in this server

but bot send me just this is
not more info
what mean this is then?
No 
of course no!
oki-doki... I understand that I don't understand anything..
If you want to read more here's link to Nubby's fork he's using on different server
https://github.com/bWm-nubby/Prompt-Dog
u cant get png info from my pics cuz they have zero
yeah you probably don't want to do this (at least not on this post).
This is to report a message to staff (because it breaks such or such rule) so we can take a look at it quicker.
and for a kiss?
yep!
🔫 
I don't think I can make a Discord Bot
I’ve been debating setting up a discord bot on my server to let people gen on my 3080ti since it’s not doing anything anyway
take it
its bit off
Man pintrest is so fucking bad now for finding refrence photos
Enshittification of search engines
I was like who the hell are you

you changed picture I don't know how to read names 
I try my maximum to avoid going there, glad I don't need it ever basically D:
even civitai is better 

Make her with a disguised face
like a mask
for the "Definitely not yue" moment
Is that one you don't have yet ? 
(bikini:2.5)
u can wear it as a mask
And how do you know what I'm generating in a1111?

Boys you are READY? We all wait it! 😉
what is it
Valentine's Day.
finally
nah cant be
spider webs?
😱
Why?!
I dun like them 
its so hot 
oh boy.. umm someone had to much choco snacks
its crazzy !!! 🫣
oo neat, someone posted an animation on civit using my model
oh, i just noticed how the way youtube choses the videos and the models generate images is pretty similar
i feel like the same mechanism is behind the 2 processes.
like a random number generator?
"no I will not share.. mine!"
the topic came up
i totally tried prompting "covered in chocolate" not thinking about what that would generate...
never know what you might get in your chocolate, nuggat, caramel, almonds, cat girl....
😄

i would eat her whole body
not working as well as I intended cx
like more net inside the models we use
they are picked by association
@native halo
its over for the chocolate you wanted
no anime girl there 😡
I shine brighter than you


a custom model merge but using midjourney lora
yeah .. blah dang hands
What effect ?
i know, it's a feeling, maybe it's because the plastic and fake sense it gives
the same feeling i get from teletubbis
oh thanks for the next prompt idea 😄

i can't fathom how could they feed such stuff the innocent childs
the theme is a girl waiting for a furnace inside a house of dirt
take Kirino's heart ( u cannot reject this)
teletubby chipmonk
@steady grail this is the best I could come up with, its not cursed if its in minecraft
ugly anime
name?
guess teletubby and chibi don't work together
oh got it
A society besides the haunted park
I don't watch anime. idk
eh? then wdym by ugly?
ugly real anime character

random ai 1girl much better
👍

okk.. thats scarey
is there a way to create a model without using dreambooth? those classification images and other stuff didnt work for me
What are you trying to do?
create a model
Like fine tune?
hmm achieve? hmm just for fun? I mean image that I want cant create
mostly like skin textures
Have you tried to make lora?
making an entire model by yourself alone is mind numbing
you need millions and millions of tagged images
nope, tried to make textual thing, ehh didnt work
I think lora is what you should try making
from dreambooth?
I think kohya_ss is used for lora training
I see
what bout fine tuning
Or if you have enough Buzz in Civitai then you can use their trainer
no idea, i jsut have a rough idea of what it is, but with my pc, and can't make models
Training NEW model from scratch isn't possible for normal consumer
you probably just want a lora
specify normal, is still possible for someone insane enough
how many buzz?
I think default settings was 500
Not sure how much it can vary
slightly horror
i see no horror, only horny.

lady of the lake
merge of my NOMD and an Epic Relism merge 😄
hmm odd way to hold swords
ok she went overboard over the candy, haha
someone got drunk looks like
what to train depends a lot on what you want to train, a persons face, an art style, a concept, etc, but in general a lora can cover all of those so it's probably the best place to start
I've done a bit of face training and then merged that lora with a custom model merge, it's not super difficult to do but can take a lot of time depending on hardware and what you're training, and if you're trying to do SDXL you pretty much need 24GB vram
i've been thinking of trying out the civitai lora training since i have some buzz, but I haven't decided what to train yet
awww
hmm not exactly from an egg but ok...
shrugs
don't thjink so
o:
you should
From what people said its basically a1111 but more optmized
more speeds and vram stuff
https://github.com/lllyasviel/stable-diffusion-webui-forge same person from controlnet 
thanks might give it a try
@vital raptor give me prompt idea 
I've been using Forge since it was released. A few bugs here and there, but they're getting fixed pretty quick for the most part. Only thing for me not working yet is TensorRT which sucks
hatsune miku at the center of an explosion of ducks 🤷♂️
i'm not creative

How did that happen x3
i know, i copy pasted it from the repo 😏

idk what u mean

😄

Rip Hatsune Miku
wtf
thanks for suggesting
upps it to 2048 x 2048 as s test
1.5 model or XL?
apparently you can do XL at 1024x1024 on 4gb with forge
but i'm amazed it's actully doing it
it ignores them if you do, so it doesnt matter
yeah I did from original bat file and seemed to work, cause it didnt' before I used them
xformers works, but i'm not sure it's needed
omg nice ! can u share prompt and model what u use?
before I readded it just crashed
actually, with small cards i think xformers flag is needed because of a bug currently
Fruit mod moment
Very cute fantasy Chibi Miku baby, innocent, adorable, epic detail, soft vivid colors, sprite, spirit, little fairy wings, surrounded by floating hearts, simple background
Negative prompt: easynegative
Steps: 20, Sampler: Euler, CFG scale: 7, Seed: 2051749860, Size: 1024x1024, Model: NOMDER, VAE hash: f921fb3f29, VAE: counterfeitV25VAE_v25.pt, Clip skip: 2, RNG: CPU, TI hashes: "easynegative: c74b4e810b03",
wasn't it for the Luffy picture one
60% render on 2048 looks like a nitemare wow
added a bunch of new stuff i'm not familiar with
huh only using 3.3 gig out of 4
nice
but htis is chibi girl not one piece :D?
😄
got it from heroprompt: Monkey D. Luffy, one piece anime, X-shaped scar on his chest, Straw Hat, symmetrical body, portrait, cinematic lighting, high contrast, artstation, character concept art, 8K
what is that heroprompt?
I used my own model merge
hmm seems I need a torch and xformers update
huh it's telling me I don't need to do the medvram anymore it's automatic
interesting
yep
hmm thats diff
in 31 secs.. dammmn
keep getting odd errro: 'NoneType' object is not iterable
yeah.. thats hlpful
i mean, you could share the whole error
it's hard to figure out a generic error message with just NoneType, that basically just means it got a 0
uhh that is the error. and to much crap in cmd window
hmmms
ok changing model seems to have worked
and vae
seems it handles models and vae diff.. hmms
gotta admit it is quicker
added some more samplers too
hmm eular a turbo?
whoa
12 secs
16
secs
loving this
ok thats not the enterprise...
casa
What portrait size works well for XL models?
Here are the SDXL sizes. 2:3 or 3:4 likely work best for standard portraits and the like, depending on the model.
Additionally, if you use ComfyUI, I recommend Mikey Nodes for a super quick+easy starting empty latent for SDXL sizes. I haven't stopped using it since I got it, and that's what the above pictured dropdown menu is from.
downloads some xl models and vae
hmm thta dont' look like miku
even does xl very quickly
whoa
ok doesn't like lcm
faints
she looks like Cara 
oooh wow
por que
fumu... I think today is a free topic.
splendid
Today I decided to update Forge and after the update, imagine, the server didn’t start and started swearing at xformer
SD
Im clueless

try removing --xformers from bat file?
try. and have other error. but if write the --xformers parameter like this "--xformers" . All right. and Server started!
precisely using "
in version 12 several additions were integrated
"DynamicThresholding" and "MultiDiffusion"
These are the ones I know about and have used once
looks like if you remove --xformers you have to delete the 2 xformers folders in venv\Lib\site-packages or pip uninstall xformers
I didn’t know that deeply. but if I delete xformers it will not affect all work? not slow down?
Forge doesn't need xformers, so it shouldn't change anything
yeah I guess, and training images are also little bit different? I think
but only way to know is to test it
pip uninstall xformers
Should I do this in the folder with Forge installed?
no real noticeable difference for me in speed
it's easier to just go in venv\Lib\site-packages and delete the xformers folders
otherwise you have to activate the venv then do the pip uninstall
is unnoticeable only for very big vram
that's possible
hello guys
the author claims forge can do faster and less vram without xformers though
I think this is a bad idea. because I have this set VENV_DIR=%A1111_HOME%/venv in the bat file
hi hi
yeah, if you're sharing venv with a1111 then i wouldn't do that
i wouldn't share venv either though 🤷♂️
saves 8GB or so sharing it i guess
maybe he did something to replace xformers
Maybe then it’s enough to just remove this line?
they did, the entire memory allocation code was rewritten
considering how he is fixing and fusing everythings
i expected it
really a big work
yeah you can do it that way
good to know, ty
oki.. then trying it!
what bout this?
simply removing xformers from the .bat, is not enough to deactivate it?
yeah I guess
there's a bug causing it to load it when it's not turned on, so it errors
it wasn't meant to be used at all, so i guess it wasn't tested very well
wait a sec...

i love everything searcher
you probably have to delete the xformers folders since you started it with --xformers, so it got installed
the url is webui_forge_cu121_torch21\system\python\Lib\site-packages
but i am not have a Venv folder in install folder Forge
you are installing venv elsewhere
maybe is better to uninstall it and just reinstall it when you use it on auto
it's better to not share a venv, it's supposed to be seperated for this reason
delete everything and start from scratch, that's my first way of solving things
if it comes to that, i start from scratch and make sure it's working, then i delete the old folder
Radical!
i agree
just delete the ven
you waste 1 hour or 2, but at least you are sure to fix it
don't press it
it's a scam
clicks
I put a warning
i already have nitro anyway 🤷♂️
i don't, maybe i should press it, after all XD
lol
RESIST
let me forget that stuff, plz XD
like, at the window?
looking at viewer in the negatives seems to help
I see
Grabby 
shes doesnt like to take eyes
magic
little girl, petite, short black hair, purple eyes, cat ears, choker, standing at window, (looking in:1.2),
negative: (looking at viewer:1.1),
@vital raptor why it shows dat warning mark?
Click it?
oh its fine now
i don't understand how is possible for this score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, source_pony, to influnece a model different from pony
everything affects every model
yep, but that token shouldn't sum to anything
you could just put hundreds of periods in the promopt and it will change something
the tokenizer will split 'score' and '9' if it doesnt find a match for 'score_9'
'9' at least is likely in every model in existence
oh, right
even 'score' will get split into individual letters and combos (core) to see if it can match it
time for work, be back in a bit
see ya
Then it finds "ore" and your girl is now mining
I didn't know it worked like that
I can see dat
looking to the side is an actual tag
it didnt work
but yeah now its fine
ohh side eye will be good?
sugoixl
whats dis modal?



Itaiiii
lets go re-install Forge!!! https://soundcloud.com/hyperpotions/chaodancemix
FREE DOWNLOAD:
https://www.dropbox.com/s/sqhtscnxtidyayj/Chao Dance Mix.mp3?dl=0
As a thank you for all your support and love for this mix we made for the Sonic 25th Anniversary Party, we decide
Cute song 
yeeep! i like it too
Guys, who knows how to correctly specify folders for generation results in the settings if they are not in the WebUI installation directory?
I think you just need to set these
@gloomy scroll thank you
There is also folder for grids
I just set them all to the same thing except the one for saved images
And I turn off grids
Not when you generate a batch of 100 xl images with 2x hires

So. I'm ready again! and already without xformers
@vital raptor give me an idea of a discord name

discord will just assign random shit to people who haven't updated the #numbers names
and I can't have Cat Bot
Very good system
4gbcat

😆
4GoodBigCat
you guys are not good at this
JoenVidia4gb
but to the point. after reinstallation. I don't have a Venv folder again...
No grabby
but im run this is now
GrabbyCat
Bro I just need an ai which can generate food for me after i throw grains in it
At this point this is a must to be worked towards
When we can get images which just text. Food should be a option too
Dark Elf, cute, gorgeous, beautiful, shy, full body in view, thin legs, slim, dynasty warriors, seducing a boy, My Hero Academia art style, (Attack on Titan art style:1.3), (glow, god rays, ethereal, dreamy, heavenly, otherworldly, dream-like, breathtaking, captivating, divine), (multiple views:1.2), (cartoon-style bold line work:1.2), vibrant colors, cel shading, dungeon siege, nagog, stealthparameter, xcom alien, reinhart
Look what i found in the database
...
You can see the results upper
a oh shit moment
yep, no bad result
skirt, 1girl, solo, hair_ornament, photo, lace, nail_polish, flower, gloves, black_nails, hair_flower, lips, blurry, looking_at_viewer, white_hair, breasts, bangs, black_skirt, bare_shoulders, long_hair, choker, gothic, makeup, blunt_bangs, fingerless_gloves, 3d, tattoo, lipstick, depth_of_field, leopard_print, black_eyes, blurry_background
mixes the other prompt up and presses gen
1girl, solo, striped, black_choker, white_shirt, choker, shirt, sleeveless, vertical_stripes, lips, collared_shirt, purple_hair, looking_at_viewer, hair_bun, ribbon, black_ribbon, upper_body, realistic, green_eyes, sleeveless_shirt, nose, hair_between_eyes, breasts, black_neckwear
scary
yep, scary
adding art by H.P. Lovecraft will do that
solo, long_hair, upper_body, lips, lace, 1girl, closed_mouth, looking_at_viewer, jewelry, portrait, brown_eyes, white_hair, facial_mark, female_focus, bone, headpiece, parted_lips, skull, blonde_hair, red_eyes, ring
switches to XL model to try that
turns off stylepile
doesn't really look anime. looks 3d'ish
tho that could be the midjourney lora
why does that prompt say white hair with brown eyes and blonde hair with red eyes, plus closed mouth but parted lips (not closed)
o.O
seems a little redundant
trying to fight the bias of the model?
i agree
or negatives for whatever the bias is
very often is even just a problem of leak from the precedent tokens
this is what that prompt does for me (without the extra garbage) but i changes the hair/eye colors
cute
i think it stuck with the prompt a touch better
reminds me of lowercraft
girl with long black hair and purple eyes, upper body, lace, closed_mouth, jewelry, portrait, facial mark, bone, headpiece, parted lips, skull, ring
for me this is
i so dead tired from studying comfy workflows
i'm inverse enginering a workflow, but it's totally a tangled mess
all those underscores bug me
but i can already see some very good improvements
SD experts when they see someone put an extra coma on their prompt
eheheh
actual experts know extra commas don't matter 😏
too bad i can't study the workflow properly, till i don't disentangle all of it
for what it's worth, this is the prompt as given, without me editing anything
GIVE ME THAT, deym she looks great
this one?
oh I almost fell for her
okay thanks

only one stool
wait wait wait, those letters look gooodd
goth rave 😄
lol
I know
wait.. ramen in a melon? what?
Yeah
A cool prompt it is....
now it's the New Hokage Tower
or
maybe
these last three are XL
hmm looks like th ere was a battle
switched back to 1.5 and hmm english instead of japanese?
here this is better
yeah better
Lady uzimaki
turbo models can even be used on a 4gvram
the speed is pretty much the same as upscaling with a 1.5
around 15 min
and the results are not much different from a normal sdxl
it's not a model. it is a sampler
Naruto as a girl, orange Ninja outfit, long strawberry blond hair, ninja headband with line through symbol, portrait, elegantly detailed, 4k, unreal, art by Yoshiyuki Tomino, medium breasts, setting:Battle, background:Village of the Leaf in Japan, soft colors, black outline style, elegant, very detailed Cel-Shaded With elegant shaded detail's, soft shadows, illustration
Negative prompt: easynegative, mangles hands
you can use without the right model, but usually is coupled with an xl turbo model
like the pony one
really like what it does, but i'm not able to do 1 step gens, contrary to what they say 😭
civitai is down
nooooo cutie 😭
Deviant same... 😂
same host?
the world is ending?
all the waifu became sentient?
i hope in a waifupocalipse
cos why not 
naruta lol
more pls
grinz evilly







