#🍥|anime
1 messages · Page 67 of 1
who?
Smh
sus
Is this going to be streamed
i always lack any good ideas
then you should prolly use dynamic prompts "prompt magic" lol
I have a solution
what that?
best way to learn is to steal
Best way to steal is to learn
borrowing
you mean taking inspiration
no
Fr
i want to learn how to like make specific anime characters but idek how to prompt that
yes yes, "taking inspiration"
I like merging everyone prompts together into a 2000+ token prompt
like Shinoa Hiiragi for example
that's something eface would do lol
you'll need a lora/locon
comfy?
hmmm
whats lora/locon?
Lora is like a filter in laymens
thats also cos i dont want to or know the gnitty details
😅
gosh its lyCoris now
oh i dont even use that i use lohon/locon
do i download that one?
omg man you ARE NOT updated
Is there a new new locon?
pre sure lohon replaced lycoris??


locon deez nuts
wait I think I have alist
....
quick get the magnifying glass!
me still waiting for LoLiCon

it is out there
aight what the flip.
hehe
in the project AI discord
there is also LoKr
damn
Gaming
okay so what of the millions of new stuff is out there that I am missing?
what is pnginfo?
there is something called ADetailer
I think that's the one that fixes faces
I do have that
Adetailer if you havent grabbed yet
didn't use it yet
IDK how to use it
talked about it in the server the other day
i like it
ok
so
its eassy
it maionly for faces
and bodies i think? didnt test
but
oh wait this not updated
select the facefull model in ADetailer, then you prompt what u want, loras workt oo
example
I turned my purple blue with it with an eye lora on top
so it only genned it inside of the face
and not like turned the whole gen into an eye cos lora
is being applied only to the face
ye
me examplaining
is why i dont
kek
IDK man I can't even use controlnet
Forgot to turn on stealth
controlnet is smth else entirely lol Ad is easier
i dont even use CN all thsat much yet
mainly cos svglib is broken on my linux and i cant fucking fix it

can you send in text form?
can't?
ligne claire, flat color,
1girl, upper body, smile, (chouzetsusaikawa tenshi-chan, kangel:1.4), grin, close-up, (indoors:1.1), needy girl overdose, bedroom, gaming chair,
Negative prompt: (low quality, worst quality:1.4), (bad anatomy), (inaccurate limb:1.2), bad composition, inaccurate eyes, extra digit, fewer digits, (extra arms:1.2), negative_hand-neg, bad-image-v2-39000, (lips:1.4)
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 22, Seed: 1139638691, Size: 512x768, Model: cetusMix_Coda2, Denoising strength: 0.51, Clip skip: 2, ENSD: 31337, Version: v1.2.1, Hires upscale: 2, Hires steps: 10, Hires upscaler: R-ESRGAN 4x+ Anime6B, Dynamic thresholding enabled: True, Mimic scale: 7, Threshold percentile: 96, Mimic mode: Half Cosine Up, Mimic scale minimum: 4, CFG mode: Half Cosine Up, CFG scale minimum: 4, AddNet Enabled: True, AddNet Module 1: LoRA, AddNet Model 1: ligne_claire_anime(5f99f766108f), AddNet Weight A 1: 0.95, AddNet Weight B 1: 0.95, AddNet Module 2: LoRA, AddNet Model 2: epi_noiseoffset2(d1131f7207d6), AddNet Weight A 2: 0.85, AddNet Weight B 2: 0.85, AddNet Module 3: LoRA, AddNet Model 3: kangel(6d1391924af8), AddNet Weight A 3: 1, AddNet Weight B 3: 1```
Woops
ngl ligne claire is my favourite sd 1.x model
Fr
model?
just saying you wouldnt want this model now cause it outdated
I am currently using a model that is like 3 I think months old or something
if i remember correctly its either sd2boy girl or wd 1.5 beta2 aesthetic
use older model to make bad image, img2img with newer model (this is the original image)
image got worse lmfao
Some other news:
-
SAG was implemented as native on AUTO, but it is only working on DEV. Stopped working on colabs. It will only work as intended after the release of the latest Auto version.
-
Control Net now has the reference models addain+attention (style + composition), no preprocessor, you input any image, it changes characteristics, poses, etc. of the original on the fly, but keeps the original subject in there. I've tested it, but I still prefer to train one image models for the same effect and better quality and options. But there are other nice use cases for it.
-
Ashen ported the Rescale Classifier-Free Guidance into an updated Dynamic Threshold extension. Basically, it is touted as the evolution of noise offset—a better and improved noise offset. I haven't tested it myself yet. (Link: https://github.com/ashen-sensored/sd-dynamic-thresholding-rcfg)
And thanks, Crazykazoo and Lizzie; I didn't know about Adetailer
oh yeah those
Nice find, I will test it later 👍
can someone test my character lora?
ok
can someone test my character lora?
why cant you test it lol
because i have a shitty model + im still in the middle of training
When we can create a video of deforum stable deffusion??
ye send
thx my boi
deforum video works in a1111
i got it working on my amd linux
when they told me i coudlnt
Where we can create a deforum stable deffusion video like a tiktoker DK??
I not have computer
Only phone
keyword is anri, + here is an original image for comparison. heres the link: https://anonfiles.com/B0o7Jesbzf/last_000006_safetensors
keep in mind that its still in training
so its not done
just wanna see if it even works a bit
Crazykazoo thanks
@brittle jay but this is for computer? I not have computer only have phone only can create this kinda videos one my Smartphone
oh
kinda impossible then
and if not then maybe youtube could work
Hi
yo
you on it?
first image
not too bad
needs some more training as i said, but looks great
oh I think that was clip skip 3
ye, uniform is wrong
bruh
aurora not happy today
another one aurora
thx for testing, one other thing, could you test it with non-portrait aspect ratio full body?
let me try
thx
aurora does not like my lora lol
i wanted to try this one: https://civitai.com/models/23900?modelVersionId=28562
AnyLoRA Add a ❤️ to receive future updates. Do you like what I do? Consider supporting me on Patreon 🅿️ or feel free to buy me a coffee ☕ Available o...
idk if its good
thats why im trying it with this lora
I also don't know
lol
np
aurora?
hell5-dpep
then fuck
better 👍
same thing, different seed and slightly higher resolution with hires
k, ill let it train a bit more
what optimizer you using sire?
idk wtf that even means
AdamW?
The default one(AdamW8bit)
oh
havent played with the optimizers
I prefer dadaptation
why?
because it's easier to use mainly
what does it change?
and I am lazy
what do you mean with "easier"
"Tired of searching for the best learning rate and guessing cosine repeats? Have Dadaptation do it for you."
basically
k, thx ill try it when i make my next lora
but for settings use this #1092852464052613130 message
can you dm both of these resources?
first one is just explanation of why Dadaptation is good
please both 🙏
wait you mean send the links or what?
@merry tendon bur i see some people ok tiktok use kaiber and use phone is not the same thing?
try use kaiber
or it doesnt work for you?
Flag of the prime minister
Some more news:
- SDXL was available for everyone for the past 5 days or so through the bots. It is now offline because it seems that Discord banned the bot. SAI is doing the so-called RLHF through the bots, where you vote for the best image, and it creates a huge database of human preferences that will later be trained in the final model. When you generate an image in the final model (with RLFH), it will follow these instructions from human preferences, self-correct, and make better generations. The project is really ambitious and very nice, and it already does a lot of generations that compete directly with Midjourney. It will have some type of official ControlNet implementation built in, with a PR on Auto for easier and quicker integration once the model launches. If the community adopts SDXL in the future, it could change a lot about the way we generate anime. The main problem for mass adoption is the requirements; it requires at least 16 GB of VRAM to run locally for the RLHF version (the only one that will be initially launched), but they are counting on the community to optimize it to run on lower specs. And they've said they will launch smaller versions that will run even on phones (but this is for the medium-to-long term). In my tests, I really liked the model; with merges, TIs, LORAS, fine tuning, etc., it could really change everything in the sense that for the first time we will have an open source model superior to Midjourney in every way.
One thing in my tests that I've observed and liked is that it can do very high-quality direct generations of artist styles and images that would require at least two merges to get on SD 1.5 and 2.1. The artists are back in full force, and when you input an artist in the prompt, it is stronger than everything else. You can input styles and aesthetics, and it does recognize a lot of them.
(continues)
(cont.)
Example: If you input "art by Van Gogh" in a voxel art style, Van Gogh will have the preference and can eclipse or even nullify the effects of voxel art. Sometimes, if it diminishes the effects, voxel art could then function as a quality improvement modifier. If you then decide to input "candycore" into the Van Gogh voxel art prompt, you could see traces of the candycore aesthetic, but it would be diminished. When you adjust the weights, it sometimes helps a little, but most of the time it does nothing. It depends on the artist you choose, and this is not an exact science and requires a lot of tests. If you choose only the aesthetics, it gets the preference, but you lose the versatility that artists provide. One thing that I don't like is that when you change positions or go away from portraits, there is still a lot to go, as the fused hands and feet and triple legs are still there.
- SDXL can do really good anime eyes on the base model, but it needs some prompting tricks. The bot has quality limitations, and SAI staff have already confirmed that samplers and some other type of undisclosed CFG improvement will improve quality even more. People are struggling with photorealism (the Photoshoped Face effect), but SAI staff have said they are working on improving it.
why no one talking man?
i interrogated clip on the picture and it gave me this prompt: a drawing of a man with red eyes and a black nose with red eyes and a white background with a black outline, Aya Goda, detailed eyes, concept art, vanitas,
pictures with that prompt come a bit close to what you want i think? maybe vanitas and aya goda play a good part in that style of eyes
turned out good
i only have one official prompt of those, and it's actually 16 prompts selected at random 😄 -> Testing all kinds of prompting styles
haha
I knew it
why A in name?
i think my name is now genius, AMegu2Bro = (A = Anri) + (Megu = megumin) + (2Bro = YoRHa 2B + Bro(because i say "bro" often))
damn
i change my name whenever i want to add smth
xd
so crazy
@wispy canopy i was looking at your model and pruned fp16 is the same size as full fp32?
that doesn't make sense?
nope, that would be a fuckup on my side xD
wait what.... 3x the same
they are the same file
2GB and same name when downloading
i think you did an oopsie
even worse, the file names i can see (internal) are actually the right ones
i think it's civitai fucking up again
that also confused me
no biggy, i'll just reupload
whut?! toolkit pruned the models 😮

my bad my dudes
seems this dark tech fucked mee over yet again xD
right now don't have sd started, i'll fix later
fk wha
toolkit pruned the model when i added in the wd 1.4 vae
oh, that doesn't sound good
is the full model now gone?
and Eface, the characters are fked again (this time 0.5 denoise at upscale)
somehow this generated a nearly whole new picture, maybe it could be the landscape lora messing it up
they seem good, but looks like someone slapped them with water on the picture
Good day, everyone.
When the clouds become cotton candy.
but i somehow start to like that design
i don't have a clure why tf it generates a earth ball, but i'll try to fix that
accidental naked smug
has anyone gotten able to do 2 hair colors to work fine in 1 character? Like shoto for example. I have no idea how to word this.
kinda tried some time ago
how did it end up?
when I tried it worked
hello
hi
^^
Hello
multicolor hair works
couldn't get one too
Does someone knows why i cant use bot?
i cannot force the thing to give me half red and half white hair. It always brings me blue (not to mention ugly,)
gave me blue tho,
Is (blue hair) in the negative prompt?
no but that is environment color
yes, in that case its only red hair...
Ugh, easy to get addicted to generating.
you should be able to
as I did up there
let me try with half white and half red
I could get even more colors XD
Black and red hair. The streaks aren't quite what I was imagining but
Maybe the model you're using just doesn't have enough of a reference for multi-colored hair?
i think this might be the case.
Its quite rare. If I will fail will have to feed it all fan-art of multi-color anime characters,,,
Or you could use inpaint
oh, right... they had a tool for that.
Looks like this will be my third sleepless night learning stable diffusion 🙂
it also depends on prompting
with some models
by prompting red hair and then white hair you can get more red hair and white hair only as some stripes
but idk usually it works
can i ask for the prompt you did to make this one?
I am using SXZ LUMA model, not anime one, have not found one yet for anime
even as shit it looks
Negative prompt: Ugly, Morbid, Extra fingers, Poorly drawn hands, Mutation, Blurry, Extra limbs, Gross proportions, Missing arms, Mutated hands, Long neck, Duplicate, Mutilated, Mutilated hands, Poorly drawn face, Deformed, Bad anatomy, Cloned face, Malformed limbs, Missing legs, Too many fingers
Steps: 30, Sampler: DPM++ 2S a Karras, CFG scale: 7, Seed: 4208038873, Size: 1024x720, Model hash: e4bb82907c, Model: EimisAnimeModel_2-0
also you can use as I used :
but that's not needed usually
at last i got the hair needed. At least quite close to what was needed.
Don't look at her hand lmao
Just realized my model knows how to do rainbow hair
Never thought to try it until now
not all models do 🙂
but rainbow hair IS amazing yeh
im looking at them
red white hair. EZ.
Luffy gear 5
cutoff?
not even
how! !!!! (emoji as on your profile picture)
red white two tone hair, black jacket, miniskirt,
i was able to get consistent hair gradient
interrupts the generation....
(blonde hair with \/(pink hair gradient)/)
tbh idk if im using that bracket properly
notice that i NEVER use masterpiece, best quality, 1girl
i will rant about it a 1000 times, but they are noobtraps imho
masterpiece will increase bustsize 9/10 times
it failed the assignment 😮
yup, I was trying to figure out how to make my Aasimar paladin's boons normal size, took my 30 minutes to realize that
when i want nice celshaded 2d style on my model it likes masterpiece, hd, hdr for whatever reason
use flat-shading, or flat-shaded
but without those quality tags its nice anime not
and 3d in neg prompt
oooo i will next time, on ER atm jus messing around
one lora on my todo list when 3060 comes in, a nice WoW and ER lora, curious how that would turn out
damn, red white two tone hair is pretty weak as a tag though 😦
well, it's 5 tokens... there's going to be little relation left over...
yeah, it looses the white as soon as you add something to that
the current shortcomings of SD
nope, can't get it to survive 😦
Yeah white and [other color here] is rough. I'm just gonna stick with inpaint for white hair
where do you find pictures sets to train models ?
Oh I know this style
cheating just ruins the quality of the game
why do I like this one so much? just why? why?
this is so cute
Because she is a mommy cow
no not yet... it seems toolkit did prune my models unintendely so i'll have to re-create and uplaod the full models tomorrow
Oh wow!, now I got m3. 😆
m3? :O
damn well it still looks great
Y3ah m3, I p3rsonally lik3 to us3 3 in plac3 of e. 😄
Oh, thats pretty funky, its a bit confusing :o
oh y0u d0 l337 sp34k?
It's mthree why are you so surprised?
hhh, now you got m3 too. 😁
also i can get boobs surprisingly big without loras
Yup, z0m813 g07 17 700 🤞
Damn, I wanna to learn this thing
opposite of the booba spectrum
nope, Ama thickness addict now, since @magic wave 's posts. So don't turn off the BOOBA spectrum on me 😄
no thin is cute
well this one's cute, the pose is 🔥
Saviel, the Steward of Heaven and head of the Archangel Council
Fair XD
👍
I played around with one a little but but it just wasn't my thing. I forget what the site was even called lol
Character.ai I think?
Oh fair, I'm looking to figure out how to get it working locally, I did it once, but now it just does not want to work xD
And tech support at the pygmalion discord is ass
I run Oobabooga for running local LLMs
I want to do that too
And I tried to do like a fresh install and stuff to run pygmalion 13b
But it just doesn't work 
Did you install the stuff for stuff for GPTQ so you can run 4bit models?
That doesn't install automatically? Though also I got a OOM error on llama 7b, and OOM on gpt4xalpaca 13b, which I mean is already bad enough cause they worked fine on my previous install, and pygmalion 13b I just get the green loading message then says "Done! Press any key to continue", and it crashes, no error
isnt that occam's fork
wait
im mixing up
thiong
im talking about kobold sry
A\ny of the 13b models will use almost all of my 12GB of vram durring inference and I haven't gotten CPU offloading working (only tried once when I first installed to be fair). The GPTQ/4bit stuff is all extra. I'm guessing you used the one click installer? If so, you should have a batch file that will launch a CMD window with the venv already activated. I don't remember what it's called. Anyways, once you have the venv activated you want to follow the instructions here to add 4-bit model support. The Old CUDA branch is the one that I use as it is about 2x faster on my 3080. https://github.com/oobabooga/text-generation-webui/blob/main/docs/GPTQ-models-(4-bit-mode).md
Hmmm, I'll try that then, though I was able to run 13b before, and now I can't even run the llama 7b one, so something is definitely funky 
I should probably give kobold another try now that I know a bit more about things. I stopped using it because I didn't have 4-bit models running and Pygmalion 6B was the best I could run which was boring at best, and struggled hard to follow context.
in my user ya
You should probably also install xformers. From venv the command is pip install -U xformers
Medium Lizzie, sadge :O
Will do :o
Lizzie grew up right before our very eyes 🥹
kek

Nice, I'll check out Occam's fork. I had a much more functional chat bot that ran on Kobold's API so I'd be happy to switch back if I can run the 13b models on it and get it performing as well as Ooba is.
True 
bne sure to check the first link on the model u want see if its working or not if u wanna be sure
i dont use it all that much but it works
Thers also SillyTavern now which seems kewl
man why does comfy understand prompts better
only downside is my hires workflow hates me and takes like 4mins for x2
Yeah I haven't checked that one out. I've used the regular TavernAI, but it was not much better functionally than just chatting right in the Ooba WebUI.
Fair, I mean they're just UIs
if you figure out how to make it a prompt helper lmk lmao im too lazy to set it up
Though SillyTavern has some fancy features
Apparently you can connect it to SD :O
Yep, Ooba has an extension for that too. I just don't have enough vram to run both at the same time 
Yeah, Eface does tho, he likes flexing XD
I'm interested to see how the group-chat thing works in silly tavern. If it can run multiple characters from a single api then that could be really cool.
That would be pretty dope :O
I could finally let my tsundere and yandere bots abuse me at the same time 
Damn that is a powerful team :O
The yandere tried to stab me the other night after I told her to go away.
Hot 
Gm chat
ive been away for far too long
Just like that one spongebob episode
Note to self: never prompt SD to generate a picture of Spongebob
spingebill
Spunch bop
OMG i can not un see it
What even is that
SD + Photoshop
amazing
breakdomain new anime model is actually GOOD! https://civitai.com/models/72675/breakdomainanime
Very nice chen
Morning
Yoo smol sus
pretty! what model do you use?
that's ExpMixLineV2
beauty
Ligne claire?
Ligne Claire? :o
Ohhh, fair then :O





Good thing your OC is very easy to make :o
(Palpatine voice) Do it
Please next time don't use caps it's annoying when you try to scam others by shouting
Coffie
This is so cute! A firefly in a jar would be a good alternative to the lamp
Tree waifu
Hello guys, if u guys dont mind, can i ask about the command prompt for this art's quality? Or maybe is actually something different input other than command prompt?
it could be the model they used
but usually putting ((masterpiece)), ((ultra-detailed)), ((best quality)), in the positive prompt is good, and in your negative put something like (low quality, worst quality:1,5), (normal quality:1.5)
also high res upscale option thingie is really good it will help a lot
Usually, im using these command prompt, but didnt get any result like the example.
even the high res upscale option?
Can you explain what the exact model like the example? Something like Euler A?
also another tip do you use like textual embeddings?
no a model as in the model to generate the image theres a lot out there
Basically a good model, and good prompt + highres fix
thats a sampler
Ah, i see...gotta try it myself. Thx for the info btw
Euler a is good but a lot of people also use dpm 2++ m karras
UniPC is good as well
so you could try between those two or expierment with a lot
So far, this is the result from what i generate
either dpm2++ karras or unipc, those are the best atm imho
which model are you using?
Euler A(?)
thats a sampler
Sorry, i kinda new for this stuff
Lemme see, maybe i can take a look at the model im using
a model is something that generates a certain style, it could change how the image looks by a massive amount
not could -> it will 😄
i can always insert a reference to my model and settings, that usually gets pretty decent images by default
yea it will actually lol
what are you using to make your images?
are you on computer or mobile?
Both, but mainly computer
I love the figurine look
do you see this? this is the models, so you could create different ones with what you choose
i dont really know how to use the website but thats the models we were talking about
Ah i see, so these are the models.
Then what model should i use for that example?
maybe abyss orange mix or anything V3/V4
darkalfa obviously 😄 lol
That too xD
Anything/AOM for basics i guess
Btw, this one use AnythingV3
nah, anything is pretty much obsolete at this point...
you could also try counterveit also if you want to try a different style
For people making previews for their loras 
this is what i made with counterveit just simple
Looks sick
counterfeit is pretty good too yeah
defo one of my favorites recently
The hands are kinda decent on this one
and dark alfa ofc 😭
Hide hands if it aint good 
Beer ad xD
energy drink 😛
Isn't doomguy's pet a rabbit
Thx for the info guys. Gotta try those advice. Appreciate it!
if you ever need more help dont mind asking
I guess Martian hamster is larger lol
Oh right, I forgot that as well
What's the name of that switch game again?
Animal crossing?
Right
But I think Isabelle is a dog?
Yeah
🎵 Listen to the EXTENDED VERSION (Spotify, Apple Music etc.): https://thechalkeate.rs/doomcrossing
➡ Follow us on Spotify: https://thechalkeate.rs/spotify
We couldn't walk past the most unexpected game crossover of 2020. With "Animal Crossing: New Horizons" and "DOOM Eternal" released the same day, the adorable duo of Isabelle and Doom Slayer a...

Goodness gracious
since the beginning it was trash
shhh.. i agree, but i'm just not saying it 😛
anything too overfitted for that
added jesus to my prompt. i'm pretty sure he's coming back to punish the sinners personally
slaughtered man
first one very cool
I would wear that hoodie

generation addicts
what happens on the 24th
New event on Arknights
It's a gacha game
how much money have you spent
I'd say the most free-to-play friendly but maybe I'm biased
50$? don't know for sure
Can beat everything without paying cause there's literal stage guides on youtube
you mustve got lucky and got who you wanted
Also waiting for Arknights Endfield which god knows when will be released
Probably, that's why i play too much gacha games
i could never not after genshin trash game
Also the progression isn't shit with full on rng stats so i lasted till now
Getting the right stats on artifacts is a pain
Their new game also has that
I played it still
Story's pretty good it's just that the progression is the same as genshin
Maybe 1-2 months i'll drop it if nothing interesting happens
Anyways
sir
click on this
No
Is my fucking tempest class out yet reee
how can you guys resist the gacha temptation
Metaslave
If it's not meta don't pull it
hope you get someone you dont want
Which game are we talking about again
No youre rude to me why would i add you lol
Dude was talking about Arknights not lost ark
LOL
Does it have gacha
mmm
somewhat
Its like
Pay for cosmetics and pay to not rely on rng
Dont need to spend money but it helps lol
Mainly cos youll be buying upg mats
And upg your wep is some bs rng thing that eventually gives you pity rolls.
Upgrade has pity?
mhm
Yup lol
worst thing ever created
Damn at least there's a pity
Yup haha cos it gets super dumb rng when youre getting into the +20
But rng on upgrade is ass in general
😄 😕 😞
The game is fun but progrssion without money is fucking ass
Buy im cheap and ion fuck with that pay2skip
Ree ima boot up lost ark now today when i get out of bed i why did i check this
Nice shoes
ugh... why did i have to click the supermerger tab...
whats that
it's a extension, where you can do model merges (a bit more than the default gradio setup), plus it can do the model merge in memory, so you don't have to save per se
Youre drunk
What model are you using btw
lol no i probably should not 😄
Damn looks very different from the last version i used
yea i was using the old one for a long time i was not expecting it to look so different
i like both tbh old and new
Hmmm, I'm gonna try to prompt and then photoshop/control net manipulate a silhouette decapitation from a rush attack
Asking SD to make a single good reflection is a pretty tall ask. I'm sure you could overfit a LoRA to be able to do it, but I doubt it would have much flexibility.
Yeah I do wonder
getting stuff like this, trying for it though
Damn that looks sick 
Fog
Yeah, but none of those reflections are right.
yes in there lies the challenge
you don't really need it to reflect, you only need to see the back of one and the front of the same woman, positions could be controlled by control net
I mean, the fact that they're not right is cooler :o
hey would someone be willing to vc with me for help? I’m new to this and I rlly need help, I already have it installed
specifically for anime images lol
sorry I no longer teach for free
aw man :((
dm me i will help
omg thank you :))
True, true. I just wanted lamp bc she a moth 
it kinda repeats the edge forever, but not really...Ah but I get so many weird images from this
Does anyone know how I can have a higher rate of images with the character inside the frame? I create a lot of images, and most of them have the hair outside the frame
like this
If you use a anime model "full_body" and "cowboy_shot" probably work, full body is full body, and cowboy shot is waist-up
my model would understand "full view of character" or anything else in normal language 😛
more or less
Yeah, but it wouldn't work for all of em, or at least work the best for all
Though specifying pose is also pretty good
I usually just do like (standing:1.1) and that works 
https://civitai.com/models/6209/dark-gemini-v3 is mine though, but I'm working on an even better one called Hyper Variant
Fair fair :o
Now that I know I'm probably not dying for awhile yet (cancer tumor removed successfully, again) I'm motivated again
wee got my first review on my model
nice
Ayyy, nisu 
Also found out my CyanCapsule LoRA helps in making furry
Fair, it be like that sometimes :o
I usually don't like videos, only ones I really really like, and this one coincidentally is such a video
ez
I very occasionally just take a week long break
not even meaning to do so really
just too much school stuff and me doing other projects and stuff with other kinds of AI
or anime
I rarely take a day break 
you sure you have no burn out? how much time you been doing this in a row?
well in the beginning I was on it everyday all day for like 3-4 months or so
Well, I got SD a while back, and I've only had a few days of break :o
what do you mean by a while back
I used MJ for a bit, but not too much cause I didn't have the muns to throw at it 
no stop?
I don't think so :o
this is highly sus 
I started sd like a month before you
Maybe near the start for a bit since it wasn't as good as it is now
yet you so much better lol
Am I? xD
probably because I don't grind as much
I mean, whenever I can nowadays I launch up SD and generate random stuff xD
my most recent cow submission speaks for itself
I don't even know if it's not nsfw 
I mean this is practically the best I can do
I mean I think thats kinda dope cause it looks like it was made by a person and not actually AI, but das fair :O
Fair, though I mean, I don't think I put that much effort into my AI stuffs
it was very hard to do because I had a very specific thing I wanted to make, and I accomplished it
Fair fair, I vibe with it :O
Better in what way tho? 
based on this
less overfit and follows prompt better
Fair :O
Got a link? 
ctd the all powerful goddess
made it
I mean, if CTD did it, then I'll have to try it out 











