🍥｜anime

[steady grail]

i got the model right?

[umbral depot]

no?

[steady grail]

oh? what are you using?

[umbral depot]

KingMix

[granite fog]

ha lol good name

[umbral depot]

Came up with it my self :)

[steady grail]

oh, it's your mix

[umbral depot]

mixed it like 2 months ago

[steady grail]

pretty good

how it does with bodies?

[umbral depot]

I can't post that here ;)

[steady grail]

not naked lol

[steady grail]

can even do animals? for example a fox?

[steady grail]

wow, so fluffy

[trim current]

Do a Liger

[umbral depot]

not going to get that at all ol

[steady grail]

what prompt do you use to make the image clean?

[umbral depot]

lol

`(best quality)), ((ultra-detailed)), (official art), Extremely detailed CG unity 8k wallpaper, masterpiece, best quality, ultra-detailed), (fox:1.1)

wallpaper, extremely detailed, intricate, cinematic lighting, Realistic, highres, detailed, Super quality, an extremely detailed cg unity ultra hd, photorealism, ultrarealistic, hdr, `

[steady grail]

would be worth to inpant with them

ty

[trim current]

Do the Mechazord

[steady grail]

lol

yeah, ma robot in a xia xia world lol

the best he can get from the model, is a big middle finger

[steady grail]

i can't wait to end with this image, i want to try many more things

looks like a mechazord

[umbral depot]

close enough

[coral steeple]

can you please share generation data

[granite fog]

here you can try it out, https://civitai.com/models/61493?modelVersionId=65986

[steady grail]

90%

[trim current]

Anyone made character ref sheets before? Any good loras for it? Want to.me a sheet for my oc to model in blender

[umbral depot]

I've done it before but I have no Lora

[trim current]

What tokens do you use ?

[umbral depot]

Character sheet, multiple views, reference sheet

Full body

[trim current]

Awesome ill give them a tru thx

[steady grail]

finally

i can do very high resolution images

cool

[frank peak]

Prompt, model?

[clever shore]

MeiruMix, prompt is hella basic

[brittle jay]

I am literally hurting myself at this point

not going to sleep

but I made such big progress in my game

[tardy bobcat]

does anyone know how to train an anime character with TI embeddings? because ive treid everything and every time it looks nothing like the source images
ive asked everywhere and aparantally noone can figure out how to train anime characters with TI embeddings

[brittle jay]

TIs are overrated and are pretty bad when it comes to training characters

train a LoRA

if you have enough vram that is

[tardy bobcat]

but i asked for TI

there is a reason i want to train TI i can add like 2 characters if i train 2

with Lora you cant

[brittle jay]

you can with lora

[steady grail]

you can load more than a lora

[tardy bobcat]

i know

[brittle jay]

you can even do it with the same lora

[steady grail]

well, they are models, after all

[tardy bobcat]

so does noone know how to make an embedding is what im finding out ?

[steady grail]

never did it

[tardy bobcat]

wbu @brittle jay ?

[brittle jay]

same

if you want help with TIs wait until nerfy comes back from his vacation or whatever it was

he the true master of TIs

but like it is common knowledge that TIs are bad at characters

[tardy bobcat]

duno who that is

[brittle jay]

@ nerfgun3

[tardy bobcat]

i literally did a hapyernet and trained with just 2 images and it worked flawlesly for keeping the character

but for some reason embeddings dont work

[supple raptor]

how are you captioning your images?

[tardy bobcat]

before i train it ?

[supple raptor]

I mean, what do your captions look like. Share an example along with the associated image.

[tardy bobcat]

1girl, bangs, black_sweater, closed_mouth, colored_inner_hair, earrings, eyebrows_visible_through_hair, gradient, gradient_background, grey_background,looking_at_viewer,smile, solo,sweater, virtual_youtuber

[brittle jay]

TIs don't really add information to the model, you are just practically associating a token to parts of the model which didn't have a token associated with them before, so like if a model has knowledge of what frogs look like, but they weren't tagged as such, you could train a TI on images of frogs and you call the embed "frog" then now the model knows what you mean by frog.

If you train an embed on a face then it might get close because it will associate features of faces it already knows with the embed, but it won't be perfect and it is much less efficient than LoRA.

If you don't have enough vram, use colab.

I am bad at explaining

probably misused the word "token" like 20 times

[supple raptor]

Ok, that looks mostly good. I'd ditch the underscores because no modern models use them (replace all underscores with spaces) and make sure there are spaces between separate tags, to help tags being tokenized separately. Settings would be the next place to look, and I don't have good settings to share. If you're not getting enough of an effect, you'll want to increase learning rate or number of epochs. If the effect is too strong or breaks the image then you want to reduce those two things.

[tardy bobcat]

i did undescores cause it was from the generated prompt + it worked when i was making hypernet

Embedding learning rate: 0.05:10, 0.02:20, 0.01:60, 0.005:200, 0.002:500, 0.001:3000, 0.0005
prompt template: custom_subject_filewords.txt
max steps: 3000
save image to log directory N steps: 50
save a copy of embedding to log directory every N steps: 50
save images with embedding in PNG chunks: checked
shuffle tags by',' when creating: checked
drop out tags when creating prompts: 0.1
sampling method: deterministic

these are the settings used

[trim current]

Reee

[real nexus]

daisy my own new daisy lora

[supple raptor]

What kind of results did you get from it? Weak/no effect or overtrained/burned images? Also, how many images in the dataset?

[trim current]

So pretty

[real nexus]

i love it! shes my fave

[trim current]

Reee koboldai keeps freezing my kernel

[real nexus]

somehow managed to get this done between bouts of dying lmao

[tardy bobcat]

well u saw the source immage this is what i got

[trim current]

Wondered why you been quiet, hope your okay

[tardy bobcat]

not even close

[real nexus]

bunch of back to back anaphalaxis hopefully ill be oki soon

[trim current]

Thats not fun :c

Nubby fix my koboldai

[real nexus]

i actually love this someone already downloading and using my lora

[supple raptor]

Yeah, looks like almost no effect. Which model are you training on and how many images in the dataset?

[tardy bobcat]

im using an anime model rn but ive tried the first time with the basic v1.5 model to even worse results and it was with the same settings as this one with the anime model

[trim current]

I brouvht it down to 20 gpu layers pls no crash

Ree

[tardy bobcat]

oh sorry there are 12 images

and ive done it with 2 before with the hapernet and that worked flawlesly so i doubt image count is the problem

[supple raptor]

I'd recommend trying 0.005 learning rate for 5k steps. I think that will probably result in overtraining it a bit, but one of the earlier checkpoints should start showing some actual results and you can finetune your learning rate from there if necessary. 12 images is on the low side, but it should still work.

[brittle jay]

takes way too much time rn and breaks down if LLM gives garbage output but I am happy it at least works

[trim current]

retro, feels like im watching a 2000s clip

[tardy bobcat]

how is it low ? 2 worked fine ._.
ik embeddings are different but still

[brittle jay]

is it really possible to train a good anything with 2 images?

[tardy bobcat]

ive watched a vid on TI and they said optimal is around 10-15

[supple raptor]

Typically embeddings are trained on between ~30-60 images. Like I said, no reason it can't work. It will just be more difficult to get good results from it.

[brittle jay]

it would be like overfitted

wouldn't it

[steady grail]

you really did it

you made a game out of her

cool

[brittle jay]

yes

[steady grail]

i love it

you did a very good job

[brittle jay]

thank you very much

[steady grail]

now i'm going to sleep, finished to test sd for today

see ya

after 10 hours of upscaling lol

[brittle jay]

it's now 6:46 for me

AM

I didn't go to sleep at all

and I am in the midst of having way too many important tests in school and stuff

[steel arrow]

im obsessed with generating this character

[brittle jay]

lol

yeah we can see that lol

[steel arrow]

its a problem

[steel arrow]

its surprisingly hard to get a sfw slime girl

[brittle jay]

#1055215351731851354

[wispy canopy]

you need my cursed prompt...

the lyrics to "Meganeko - Milkshake"

Furries and anomalies
The hyperactive scatterbrains
Whit milkshake running though your veins
Unlimited sleep deprivation
Dankest memes and vaporwave
Procastinate at five am
Forever, that shit is my jam!
I come from the future and the
Future will be beautiful
With pizza at the speed of light
And hugs multidimensional
Don't hold anything back, dude your
Fantasies are real to me
So taste'em like a lollipop
Yeah! Go in hard and never stop!```

[sullen hollow]

cozyy

[trim current]

Funny you say that

Its part of the prompt

Glad it captured that

[brittle jay]

did you sleep 9 hours?

[wispy canopy]

5.5: enough.

[brittle jay]

I didn't sleep

[wispy canopy]

again?!

dude.

[brittle jay]

.

#1072220168534642768 #1080946152318443610

to the guy that sent the message

[brittle jay]

wdym again?

I usually sleep, even if for a little

[wispy canopy]

yeah... this... i don't think it's the first time i hear you mentioning (too) little sleep

[brittle jay]

fr

did you see my game though?

[wispy canopy]

nope, at office atm, link me, i'll chekc later ^^

[worn bough]

Hi

[brittle jay]

@wispy canopy the game is not out and it is barely close to completion, I just made a bit of progress in it, which you can see above

[zinc flax]

big booba millennium lady

[brittle jay]

model?

[zinc flax]

good ol v4.5

[brittle jay]

damn

[gloomy scroll]

Without & with add detail lora

-2 add detail

[wheat falcon]

The detail lora is really cool

Tested it yesterday

[wispy canopy]

I like the add detail lora more for the -1 than actually adding details...

It makes anime lineart so much crisper

[wheat falcon]

In General or with the lineart lora ?

[wispy canopy]

In general

#🍥｜anime message this is just the add detail lora and a 'BREAK lineart' in a natural language prompt

Strangely enough, you can use add-detail with -1 and then add detailed to the prompt : get more details in the lineart 👌

[storm sigil]

Как тут генерировать картинки?

[wispy canopy]

And I'm a prompt purist, I already hate it enough that de offset and detail lora are soooo powerful, I just can't ignore them

[pliant epoch]

Just merge it into your model

[wispy canopy]

but then it's applied permanently, right?

[tropic wraith]

@storm sigil

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no public bot on the server that generates images. There is an experimental bot available for early server members & Stable Society. You can obtain the Stable Society role by winning our weekly events #⭐｜pow-info #1087493421209485393! However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610⁠ for more details! You can also stop by ⁠#1025467151206854736 for any issues you experience while using DreamStudio or ⁠#🤝｜tech-support for any problems you encounter while installing it locally!

[wheat falcon]

Yea right, i mean you could merge the loras into your model

Oh lol wpatz already suggested

[pliant epoch]

yes but then you can just swap models instead of activating the lora kek

[wispy canopy]

lol

[zinc flax]

I think you could try this out since it's literally the same as detail tweaker -1
No dang clue which came first though
https://huggingface.co/2vXpSwA7/iroiro-lora/blob/main/test2/flat2.safetensors

[wispy canopy]

elf girl got isekaied?

[zinc flax]

Use -1 on flat2 and you get more details lol

i hope i don't get bonked for this 🤔

[steady grail]

i suggest to use lora:to8contrast-1-5:0.6

hello

i think some loras are good to save tokens

even using many loras at low ratio is funny, it gives surprising results

[gloomy scroll]

She just went home

...Also she's little bit thirsty could you help her?

[gloomy scroll]

Accidents happened and got this

[runic drift]

is civitai shut down again?

[dapper knot]

https://tenor.com/view/kung-fu-panda-oogway-accidents-there-are-no-accidents-destiny-gif-16270886

[gloomy scroll]

Andite is alive again

[zinc flax]

It's always down 24/7

[wispy canopy]

as soon as my stable diffusion is on the new disk ^^

my mobo only has 2 memory banks, i though i had 4.. now i have to wait another day for memory

and added a seperate drive just for all stable diffusion/AI related crap lol

[trim current]

Pls isekei me

[scarlet kelp]

indeed

i wish i had enough money to develope the nervegear

[trim current]

Same

Id contribute

[scarlet kelp]

btw there is a company who doing research on it

maybe we get it soon

[trim current]

Isnt that the guy who made the headset that rlly kills ppl?

Ir was that a meme

I couldnt tell

[scarlet kelp]

[scarlet kelp]

i´m talking about "Galea" its just a short step before the original nervegear and maybe in some more years with success we get the original

[trim current]

ah ok then i mustve saw a meme

https://dailyhive.com/vancouver/vr-headset-deadly

kek

[scarlet kelp]

thats not just a meme

the dev of the occulus said this

[trim current]

imagine buying a headset with c4 charges inside

and some kid on Pavlov tbags and kills you

[scarlet kelp]

XD

imagine u die bc a bug

[trim current]

oh god

[scarlet kelp]

i find all bugs XD

[trim current]

https://tenor.com/view/yes-oh-yeah-baby-smile-gif-13800457

dis guy

[scarlet kelp]

https://tenor.com/view/klee-genshin-impact-klee-arson-klee-genshin-genshin-gif-26207383

[brittle jay]

@wispy canopy so did you see the videos I sent on my game?

[wispy canopy]

Fckn A!

brb, family dinner, have dis in the mean time ❤️

[brittle jay]

is that a yes or a no? in case it's a no here it is again in case you missed #🍥｜anime message

[wispy canopy]

yeah, saw that, pretty damn good idea !

which llm model are you using?

[brittle jay]

stupid pyg7b

[wispy canopy]

you might want to give wizard7b uncensored a shot?

@brittle jay is this you and your waifu? 😄

[brittle jay]

except I have shorter hair

[trim current]

Wizard has a 7b model? Is it 4bit?

[wispy canopy]

yeah, natural language prompts are just crapshot 😄

[brittle jay]

I tried it only once and it worked actually surprisingly well

I will try

[wispy canopy]

https://huggingface.co/TheBloke/WizardLM-7B-uncensored-GPTQ

the wizard series are MILES ahead in terms of understanding/reasoning when compared to pygmalion

too bad you can't run the 13b variants, they're in a whole different class together 😦

aight, bbl

[umbral depot]

I want to run the 13b

We need xformers for llm

[supple raptor]

I use xformers for my LLMs...

I also run run Llama-13b 4bit on 12GB of VRAM

[umbral depot]

Hmmm

Their is a gradio for llm right?

[supple raptor]

Yep. I use oobabooga. https://github.com/oobabooga/text-generation-webui

[umbral depot]

Time to squirrel that away to a private discord for later

[supple raptor]

It has 1-click installers available that are supposed to work better than when I tried them last, but I just did the manual installation through WSL because the installers were giving me problems at the time. The WSL setup was pretty easy.

[brittle jay]

ok I only had the non uncensored version

bascially

this is better than pyg7b

and it also runs sooo much faster

[supple raptor]

I'll have to try WizardLM. I tried out Pyg7b last night and it was pretty good, but still not as good as as regular Llama-13b-4bit-128g

[umbral depot]

The one I tried couldn't even give me a cat pun :(

[brittle jay]

since when new pfp?

looks amazing

[brittle jay]

my game is so much better now with this model

[supple raptor]

I've only used the local models as chat bots so I'm not sure how any of them perform at other tasks.

[umbral depot]

I changed profile pics twice yesterday lol

[brittle jay]

damn

should I stop using the word "damn"?

I feel like I use it too much

[umbral depot]

Nah this is a 13+ group chat so you can use it once a day to keep it pg13

[brittle jay]

hehe

[gritty patrol]

i honestly don't consider "damn" to be a bad word

[steady grail]

good gracious! lol

[strong thunder]

I know it is what you like !

[steady grail]

that's a good idea

[rain valve]

f**k insane!

[steady grail]

he forgot the bottles of ketchup

[rain valve]

very wonderful

looks sooooooooooooooooooooooooooo good

[steady grail]

remember guys, we must pretend that that is ketchup

[rain valve]

okay

[sage kiln]

Yeah man I love this tech. I was always shit at drawing but had so many ideas for stuff. First I thought SD was garbage and it was hard to get stuff to look the way i wanted but once I got inpainting and slight editing with gimp it really took off.

[rain valve]

how do you think midjourney

[rain valve]

SO GOOD MANGA

[sage kiln]

Don't know too much about it but I really like running the model on my own hardware.

[rain valve]

oh what's your gpu

[sage kiln]

3070 TI

[trim current]

A shit, damn is a bad word now?

[rain valve]

seems a little low, now i must rent a gpu in runpod and i cant do anything

[sage kiln]

Only thing that is hard limiting is the VRAM.

[rain valve]

Yes

[supple raptor]

Fuck. I'm gonna have to come up with a new vocabulary.

[trim current]

What a fucking headache

[supple raptor]

A real pain in the ass

[trim current]

[rain valve]

[steady grail]

yeah a true gracious conundrum lol

holy moly

[strong thunder]

damn anime models and lora are crazy now...

[steady grail]

found a faulty model finally

[strong thunder]

they all look the same but with a good prompt, it is easy to make something beautiful

[steady grail]

with so many stable models out there, continue to using unstable models is just a sin

[strong thunder]

truth

[steady grail]

ai became so good

maybe i should try to make some content so to buy a new card

[strong thunder]

this 90s anime lora is so good....

and it understands quickly the different angles of view...

lets try an animation with it

[scarlet kelp]

Meow

[scarlet kelp]

that is a masterpiece omg

[gritty patrol]

at least they don't have 3 fingers

i've just discovered how bad anime2 is

clearvae is 1000x better

[steady grail]

wow, that's pretty good

[trim current]

Clearvae link pls

[gritty patrol]

https://civitai.com/models/22354?modelVersionId=62228

[trim current]

Thankie

[supple raptor]

The worst is when one hand has 3 fingers and the other has 5. It's like WTF SD you got the number of fingers right but still managed to mess it up.

[rain valve]

why someone hate AI art

tell me why

[gritty patrol]

"This is a merged VAE that is sharper and slightly more vivid than the [redacted] and does not bleed like kl-f8-anime2."

[trim current]

Lots of valid reasons but blown out of proportion

Waaaay out

[steady grail]

because you need years of practice to learn how to draw

[trim current]

And mostly comes from the uneducated on the tech

[steady grail]

and want to speculate on it, even in a cheap way, if possible

ai is taking away this chance

[gritty patrol]

i'd learn how to draw good hands but AI might do it before me so whatever

[steady grail]

there are models, embeddings and extensions to fix them

[trim current]

I will always believe that ai art wont ever be good automated as it is good with a human driver

[steady grail]

yeah, because the true beauty is in the details

humans can capture the chance

bots can just try

[trim current]

And perspective. Ai doesnt understand nuances like that. Things that us human

[steady grail]

it's that perfect chance that give life to a masterpiece

lizzie, do you know that there is mathematics even in perspective?

[wispy canopy]

10/10, would bomb humanity again

[wispy canopy]

start oobabooga with either xformers or sdp-attention

[wispy canopy]

ah, this is a config.json thing, some model have use_cache : false. change it to true. much faster.

[wispy canopy]

i disabled profile pics -> i long back for the days of IRC

[umbral depot]

If I didn't have profile pics I would have no idea who anyone is

[gritty patrol]

holy shit i just noticed

guys

[scarlet kelp]

that is beautiful

[gritty patrol]

i got +8gb ram yesterday

[gritty patrol]

now i have 16gb

that's ok, but now i can switch models in less than 20 seconds

wtfff

i used to take a minute and a half

[scarlet kelp]

congratz xD

i cant live with lower then 32gb, i have 64 currently xD

[gritty patrol]

WTF

[steady grail]

how much did you spent?

[gritty patrol]

i had 8gb until yesterday chill...

[wispy canopy]

my 64 is on the way XD having all these models loaded causes my 32gb to bottom out xD

[scarlet kelp]

hahaha i feel that xD

[supple raptor]

This is pretty damn good.

[trim current]

saw this today

[wispy canopy]

SD / AI/ LLM has been a pretty damn expensive a hobby

[trim current]

gonna test it later

[wispy canopy]

yeah, it's a pretty damn good 7b model

[trim current]

good to know

[wispy canopy]

think it's one of the best 7b models atm

can't get that MPT model to load

[supple raptor]

I think it's even slightly outperforming Llama 13b for me.

[wispy canopy]

OG llama 13b is already outdated

if you want good 13b: i'm stuck at gpt4-x-vicuna atm, i feel it performs better than gpt4-x-wizard

and they both blow OG llama away

[scarlet kelp]

any suggestions to fix hands?

[wispy canopy]

drop the moon again!

[scarlet kelp]

HAHA

thats would match with klee soooo good

[wispy canopy]

klee? that a specific character?

[supple raptor]

Yeah, I need to try some of the newer ones. I'll grab gpt4-x-vicuna next. Have you tried this one yet? https://huggingface.co/TheBloke/wizard-vicuna-13B-GPTQ

[scarlet kelp]

its a character from genshin impact, she like to create and hide bombs and blow everything up

[wispy canopy]

yeah, that'st he one i mentioned -> i feel it's a bit worse than gpt4-x-vicuna

it listens just a tad bit worse to my initiation prompts, and has a little less context awareness, but that's just my oppininon

https://tenor.com/view/yeah-well-thats-just-like-opinion-gif-18505939

[supple raptor]

Ok cool I'll just grab the better one then. I wasn't very impressed with Pyg7b compared to OG Llama.

[wispy canopy]

nah, once you get used to the 13b models, you just can't be impressed by 7b models

and honestly, i think pyg7b is... just not good at all, it just goes it's own random way pretty quick

[supple raptor]

Yeah, that was my experience too. It hallucinated frequently and was quite bad at getting formatting right for actions.

[wispy canopy]

I. NEED. SDXL. for greater context understanding!

[umbral depot]

I need sdxl so I can train off of it

[wispy canopy]

asked it to make a fitting image for the band two steps from hell

[supple raptor]

Got an A100 laying around?

[wispy canopy]

i fixed our server now we have 8 working a100's again. they're all in use already again

[supple raptor]

https://cdn.discordapp.com/emojis/1025170701025550366.webp?size=64

[rain valve]

[umbral depot]

H100

Took out a small business loan to buy 8 h100 under the disguise of being a start up ai business

[wispy canopy]

i suggested we'd buy a couple today, as all our a100's are in use

[supple raptor]

From what I understand, both h100s and a100s are very hard to get right now even if you have the money.

[steady grail]

a100? a gcard?

[umbral depot]

I am 99% confident you cant get one for personal use right now

[umbral depot]

Yes

[steady grail]

full name?

[umbral depot]

Nvidia a100

[supple raptor]

I mean even for businesses.

[umbral depot]

Im sure the larger ai companies have bought a large amount of the stock

That and server farm companies

[steady grail]

holy macaroni, 48 g

[umbral depot]

You can rent one for 2-4 dollars an hour

[wispy canopy]

yeah, pretty much world stock is running on empty, as ALL major players want them

[trim current]

Isnt that fraud

[supple raptor]

You can get them with 80GB https://www.nvidia.com/en-us/data-center/a100/

[wispy canopy]

we got 4 80gb, and 4 40gb's

[magic wave]

I think my LoRA works, I made LoRA to help make skinny girls

[steady grail]

holy baloni

[trim current]

Told youuu

[supple raptor]

Everything is legal if you don't get caught

[umbral depot]

thats why it is registered to my LLC so they get the blame

[magic wave]

This one is V2 tho xD

[trim current]

[wispy canopy]

we all know where this universe is headed...

[trim current]

AND YOU DIDNT SEND ME IT

[magic wave]

V1 I think was unsuccessful cause I think I accidentally trained at 4 dim / 2 alpha

[steady grail]

if it works, share the lora, i have in mind a dark souls theme, man

[magic wave]

This one was at 32 dim / 16 alpha, I wanted to do 16 dim / 8 alpha, but I guess it trained 32/16 anyway xD

[magic wave]

I can send

[trim current]

[wispy canopy]

yeah. these are unsettling. cool ^^

weird. i'm not mentioning blindfolds

[magic wave]

@trim current Here it

Using it at 0.65 weight atm

[trim current]

very pog ty

[steady grail]

0.6 usually is the best

[magic wave]

[trim current]

[wispy canopy]

i should probably use that one too, else i get big booba's all the time (if i use the prompt 'girl')

[untold glacier]

Best network weighting depends on the model used. A LoRA weight of 0.8 may be far too much for something like Anythingv3 compared to models like AOM2 or Aurora.

[magic wave]

I made that LoRA mostly for big thigh gap and thin thighs, but it should help with big booba too :O

[steady grail]

yeah, but till new, for some strange reason, i always ended using loras at 0.6

[untold glacier]

That's a decent number to start from.

[wispy canopy]

still not really using loras... or embeddings... pure prompting ftw lol

[steady grail]

and don't nominate anything in my presence, that piece of junk wasted so much of my time

every result gave me disgusting mutations that i had to fix with impainting

[untold glacier]

It has its uses. We've moved beyond it in our choices now, but many models have different proper language usages from each other.

There isn't a one-size-fits-all in terms of prompting methods with models.

[steady grail]

yeah, that's very true

so many models got contaminated by that piece of junk

[magic wave]

Skinny :o

[untold glacier]

If only I could get a proper tornado...

[steady grail]

well.... technically.... with control net...

[steady grail]

just now i'm experimenting with it

[wispy canopy]

first attempt tornado, meadow , on alfahentai...

[steady grail]

looks like smoke

[untold glacier]

Yeah. There's a lot of smoke in mine too. Heh.

[steady grail]

not too good in trasnparence

[wispy canopy]

ITS AHENTAI MODEL LOL!

give it some slack xD

[steady grail]

transparence is pretty hard to replicate

maybe with more sharpness?

[untold glacier]

I've been trying for tornados too long to give slack, lol.

[steady grail]

try tornado with sharp forms

[wispy canopy]

perhaps my model does know about tornadoes a bit because i retrofitted generic model knowledge back into it

[untold glacier]

It's likely just in nobody's training data, or I haven't found the right keywords to use with Aurora yet.

[supple raptor]

@wispy canopy Have you made any characters to use with gpt4-x-vicuna yet?

[wispy canopy]

my imagegenerator, which i've been using since sunday? ^^

ALL my images you see are done by the imageassistant

but i have a model which works great with natural language. if the model is booru-centric, it won't work at all

[supple raptor]

Do you use it in Chat > Instruct mode?

[wispy canopy]

transparancy! uh... can't post the other ones because of eh... reasons

[wispy canopy]

chat

@supple raptor

give it a go ❤️

[steady grail]

test transparency with fluids

[untold glacier]

It tries so hard

[steady grail]

i usually push models with such tests

to see what i can and can't do

[steady grail]

no fluids at all

[scarlet kelp]

how can i understand gpt4-x-vicun what is that exactly?

[untold glacier]

I was referring to the continued attempts at a tornado.

[steady grail]

it's a chatgpt like ai

[wispy canopy]

gpt4-x-vicuna is a large language model, like chatgpt

i'm using it to convert my ideas into natural language prompts

[steady grail]

it lack the basic elements to make a tornado

[untold glacier]

No kidding.

[steady grail]

maybe wind moving like a vortex?

[supple raptor]

It keeps complaining that there isn't .json with the same file name in .\characters\instruction-following\

[wispy canopy]

this is why llm + my model = cheatmode

[steady grail]

so warm

[wispy canopy]

llm + darkgemini goes hard too

[magic wave]

Elf :O

[scarlet kelp]

ayo

hot

[wispy canopy]

aurora fails HARD on these kinds of prompts though 😦

dunno why that is

[magic wave]

And still skinny

[scarlet kelp]

tasty

[magic wave]

Probably cause of the clip that struggles more with normal prompting, and works better with booru prompting / keywords :o

[steady grail]

abyssorangemix is going weirdly slow

[wispy canopy]

yeah, i think so too, and i kinda killed the booru stuff when i did the add difference trying to exclude everything AOM3 is

that also fixed the crystal eye horror that anythingv3 started

[magic wave]

Probs :o

[supple raptor]

I figured it out. I just had the dumb

[wispy canopy]

llm+darkgemini

[wispy canopy]

wrong folder? XD

it's for character yamls, not the instruct versions

[supple raptor]

No, it defaulted to to the instruct mode within chat, which apparently needs an associated .json. Switching to cai-chat like I normally use made it work like normal with the characters I've already made.

[magic wave]

[brittle jay]

does this have png info?

[magic wave]

I've seen too many thicc thigh, now I am on the path of the thinn thigh

[magic wave]

I don't think so :o

[brittle jay]

let me try

[magic wave]

Want me to resend? :o

[brittle jay]

yeah no it doesn't

[magic wave]

I mean, I copied and pasted, so in theory it shouldn't :o

Yehhh

xD

[brittle jay]

pls do

[untold glacier]

Well... at least I can bring about the apocalypse

[wispy canopy]

https://tenor.com/view/yup-were-dead-chip-dale-chip-n-dale-rescue-rangers-were-goners-gif-25849780

[magic wave]

Here it be with PNG info @brittle jay

[brittle jay]

极光极光F16????

[magic wave]

Yeee, thats Aurora

[brittle jay]

lol

waht is that better hands locon?

and also what is AromaStyle?

and also skinny helper?

[magic wave]

I got to test it before it came out, and it didn't change, so I just stuck with it

[magic wave]

@brittle jay This is SkinnyHelper

For skinnier waifus

And lemme grab the other two

https://civitai.com/models/27044
https://civitai.com/models/47085

Though the one up atm is Aroma V7

but i dont think v9 is actually better

I mean, I feel like it might be a bit better

So maybe I could upload it

[brittle jay]

I really like how aroma sensei style looks from the previews

[magic wave]

Yeee, I really like it too, I'm uploading V9 now, since its slightly more stable, maybe

[wispy canopy]

ah, you prompted my example prompt? 😄

[gritty patrol]

yep

[untold glacier]

So close...

[wispy canopy]

these were the ones i got ^^

[brittle jay]

your LoRA looks like how my LoRAs would look if I was less lazy

[umbral depot]

Lol

[brittle jay]

this is pretty much my only good LoRA and it's mid https://civitai.com/models/45491/modaremogoon-style-lora

[wispy canopy]

who here listens to thefatrat?

that 10 minute mashup is fucking glorious!

[gritty patrol]

she's holding 2 planets and 1 moon

[untold glacier]

Looks like I do

[wispy canopy]

doesn't come close to my biggerst boobs yet! (galaxy sized boobs, made for science)

[rain valve]

[wispy canopy]

❤️

fckn love this prompt. fckn love the song that cauased this prompt ❤️

skinnyhelper at -0.5 xD

at -1 it just disintegrated...

[magic wave]

https://civitai.com/models/27044?modelVersionId=66523 @brittle jay It is updated

[magic wave]

I mean, V9 works

But at 0.65 weight

Which still deeply wounds my heart

xD

[brittle jay]

I think I might use it for my game

[untold glacier]

At last.

[untold glacier]

That's probably as good as I'm going to get for now. A decent tornado finally.

[wispy canopy]

you just had to destroy a city for it... a tornado on that size geezz..

gj 🙂

[untold glacier]

Thanks.

[magic wave]

Fair

[untold glacier]

Also I love this

[wispy canopy]

An ethereal figure stands amidst a backdrop of fiery reds and deep purples, with a single tree standing tall in the center of the painting. The figure appears to be made of light, with delicate wings spread wide and a gentle gaze fixed on the tree. The leaves of the tree glow with a soft inner light, casting a warm glow across the entire scene. There is a sense of movement and change in the air, as if something momentous is about to occur. The colors and textures of the painting come together to create a dreamlike atmosphere, full of mystery and wonder. BREAK lineart

[brittle jay]

I see most of the images have good cropping

and eyes are not bad considering how small they are

[magic wave]

Yeee

[wispy canopy]

<3<3<3

i could exhaust all seed numbers on this image and not get tired of it 😄

[untold glacier]

[brittle jay]

you used Dapaptation though?

[wispy canopy]

#2 though.. that gives me malenia vibes though, she's going to wreck your ass in unfair bossfight

[fallen sundial]

hello

I'm using a new pc and I copyed my stable diffusion from my pc to the new pc

but it doesn't work.

even that I installed python

and added it to path

and installed git

[magic wave]

I didn't :o

I don't know how to use Dadaptation

[untold glacier]

It might be overkill (as I don't have the knowledge on how to redirect its looking for python), but I'd delete the venv folder and have it download it again. All your settings, etc should still be retained.

In short, try deleting the venv folder and then starting the webui, it'll redownload everything it needs.

[wispy canopy]

not sure how i feel about clearvae... the vibrant red... isn't vibrant anymore

[fallen sundial]

I will see

it says this now

[untold glacier]

That's good. That means it's reinstalling what it needs.

[fallen sundial]

huh

I need to wait...

[untold glacier]

Just wait for it to finish.

[fallen sundial]

but recently I got this message

I may get this image

sorry I mean error

how to avoid this ? cuz I don't have good interenet to wait until I see if I will get the same error or not

so I want to fix it rn and make sure it won't happen

[untold glacier]

If it shows up again, I don't know the path forward. I'm not an expert in these things, sadly. It should be reinstalling all of that from scratch though in your venv folder.

[brittle jay]

put --reinstall-torch

[merry tendon]

dont wanna advertise again, but can someone test out a couple of abstract prompts on my model?

[brittle jay]

abstract? you mean schizo

[merry tendon]

shi what?

[brittle jay]

or do you mean like abstract art, like in the art movement?

[brittle jay]

like song lyrics, random stories, stuff like that

[merry tendon]

no, i mean abstract prompts like yours

[merry tendon]

yep

[brittle jay]

that's schizo prompting

[merry tendon]

and some normal prompts too maybe

[untold glacier]

That's... an interesting way to describe it.

I'd say non-standard, lol.

[brittle jay]

nah

me and eface been calling it taht for long time now

[wispy canopy]

want me to run my megaprompt on it?

[untold glacier]

Ah, so it's an in-joke.

[merry tendon]

idk what that is, but sure!

[wispy canopy]

half. it's basically feedin the prompt "strange things" in the hope something intersting happens!

[untold glacier]

Upscaled version if anybody wants it. Aight, I'm off to skyrim modding again.

[steady grail]

finally, what prompt did you use?

[merry tendon]

"Aight, I'm off to skyrim modding again"

[brittle jay]

Look at me, I'm as helpless as a kitten up a tree;
And I feel like I'm clingin' to a cloud
I can't understand
I get misty, just holding your hand

Walk my way
And a thousand violins begin to play
Or it might be the sound of your hello
That music I hear
I get misty, the moment you're near

Can't you see that you're leading me on?
And it's just what I want you to do
Don't you notice how hopelessly I'm lost
That's why I'm following you

On my own
Would I wander through this wonderland alone
Never knowing my right foot from my left
My hat from my glove
I'm too misty, and too much in love

[merry tendon]

what model?

[brittle jay]

yours

[merry tendon]

nice

[stable crane]

cool!

[merry tendon]

looks sick!

[brittle jay]

I live in a house with a tin roof and
Every time it rains, I can feel my brain
It's moving back and forth
Upside down, east-west
Feeling and remembering everything you do
I was up last night tossing and turning
Couldn't get to sleep and I slept thru the morning
Need to clear my head and get out of the city
All alone in the jungle you'll find me
Close your eyes and hear my secret
Deep, deep loving, hear my secret
Hear my secret, hear my secret
Sometimes I can hear the tremble below me
Pounding thru the floor, my body so restless
It lives in the basement below some boxes
Makes me feel so strange and so thoughtless so
Maybe someday this roof will cave in
You'll find me on the floor looking at the stars
These walls are made of brick, plated in gold
But I'm still here growing so old

[wispy canopy]

this is not my model, is it?

[brittle jay]

xynaptix

[wispy canopy]

my model has a fascination with doors... lots of them

have to redo my mega prompt 😦

[merry tendon]

my model?

[brittle jay]

bruh discord is blocking the lyrics

[stable crane]

long time no see

[brittle jay]

I missed you

[merry tendon]

[wispy canopy]

@merry tendon megageneration on your model

[brittle jay]

proof

[wispy canopy]

wish i could post a spicy pic celebrating you rreturn, but i have none prepared 😢

[merry tendon]

is my model good(in your opinion)?

[wispy canopy]

it's not bad per-se, but it has trouble grasping some concepts

[merry tendon]

for example?

[stable crane]

I like 2nd row 1st pic

and mecha girl

[brittle jay]

next time be prepared

[wispy canopy]

4th picture -> it simply doesn't know what to do

[brittle jay]

what was the prompt for it?

[wispy canopy]

gonna do a non-upscaled version, taking alfahentai, dg3, and xynaptix

[merry tendon]

i agree, you need to gave some luck with shizo prompts(but probably works better on my model that some other popular ones)

[wispy canopy]

@merry tendon 1) xynaptix 2) alfahentai 3) dg3

[sly rune]

New anime tagger extension. https://github.com/7eu7d7/ML-Danbooru-webui

[steady grail]

it's an interrogator?

[wispy canopy]

best gun goes to dg3

[untold glacier]

Judging from a single seed?

[wispy canopy]

nah, i know it's consistent, but i'll humor you 🙂 x3 coming up

[untold glacier]

[wispy canopy]

dg3 in general has incredible concept awareness

[untold glacier]

Yep. Definite improvements with Gemini comparatively. Are you just saying "gun"?

As part of the prompt

[wispy canopy]

futuristic-looking gunslinger, dressed in sleek black armor adorned with wires and high-tech gadgets. Instead of a traditional revolver, the gunslinger wields a powerful plasma rifle that emits bursts of bright blue energy. The background features a mix of futuristic cityscapes and old western landscapes, showcasing the fusion of technology and tradition in this unique character.

[untold glacier]

Technically CLIP takes "traditional revolver" and tries to gen what "instead of" means as well as the rest of the prompt correct? I know it's straight from your LLM but it doesn't have the ability to neg gen anything in the regular prompt area right?

[wispy canopy]

ooof... fail mark on alfahentai

nah, i haven't done a trial on negative genning, but that's actually a great idea!

i'm pretty sure i can prompt it (get the LLM to do both parts)

[untold glacier]

Not what I meant but glad to help lol.

I meant that anything that's included in the prompt is what SD tries to gen. You can't say "Instead of" and expect it to understand what that means in context.

[wispy canopy]

yeah, unfortunately, SD it too dumb for that

even 7b models have trouble with "instead of"

13b models are pretty good at it though

[untold glacier]

I wonder how accurate your LLM can get in negative prompting though. It'll be interesting to see.

[wispy canopy]

i'd just nudge it in a direction so that it would add stuff in the negative prompt it doesn't want in the image

not sure how i'd go about initial generation

and it would need to adhere to a VERY strict format

and that's where 13b models struggle a bit too

(when not used with a lora or smth similar)

[untold glacier]

How much vram is needed for the LLM you're using--Vecuna, right?

[wispy canopy]

13b models =~ 8gb

but you need 1gb for os and rest

and then you need 1gb + for inference + (long term) memory

[untold glacier]

Yikes, right on the edge. I wonder if my 3080 could handle it.

[wispy canopy]

so that's why i discarded my 3080 xD

nope. you can't

[untold glacier]

[wispy canopy]

i was in that situation, and IT FRUSTRATED THE FUCK OUT OF ME

[untold glacier]

I knew when I bought it that 10gb wasn't enough. I knew it

[wispy canopy]

you can get it to gen up until 800 tokens and then you oom

this specific prompt goes to 1000 tokens with a bit longer lyrics text

it does lyrics pretty damn well though

gimme an artist you like, well known

(music artist)

[untold glacier]

Eh, porter robinson

[wispy canopy]

good choice, lets see what it does!

[untold glacier]

Probably need one with more lyrics though. Lemme think

SYML?

[untold glacier]

Lol. Yeah, that's pretty Porter.

Having a whole anime short is pretty helpful in that regard, heh.

[wispy canopy]

honestly, i think that's pretty fucking amazing, actually

it can't see

it can only learn from what it 'read'

but it did dthe same for thefatrat, two steps from hell

fitting images for the general style the artist has

[untold glacier]

I can imagine that a lot of people have talked about shelter over the years, but I see what you're saying.

[wispy canopy]

when i feed it lyrics, it takes a single or two concepts, and just makes an image with that

[steady grail]

works with 4g vram?

[untold glacier]

Alright, I'm going to have to mentally distance myself from this convo, lol. I want it too much and am too broke for a new GPU.

Can't wait until we have a decent LLM that can with with less VRAM.

[wispy canopy]

sorry my man ❤️ sd / AI is the new crack and you know it 😦

[untold glacier]

ᴛʀᴜᴇ

[sly rune]

Won't be sure unless you try.

[steady grail]

asking because i'm looking for something like interrogate click that runs, this time XD

[wispy canopy]

why interrogate?! JUST THROW RANDOM SHIT IN PROMPT!

[steady grail]

because i like improving old images

[wispy canopy]

even before LLM i did that..

i mean, my true native creativity comes down to girl

lol

[steady grail]

control net it's too hot

[wispy canopy]

when you just scream something at the llm and this quality/creativity is the result.. honestly, it's VERY hard going back to normal prompting...

the worst part? the LLM got the intent, and made a perfect fitting image...

SD failed hard

[sly rune]

How do you tell ChatGPT or whatever to spit out a prompt for Stable Diffusion?

[wispy canopy]

that's why i'm so excited for SDXL, if it just has one level of order of context awareness better, that would make images so much more controllable, so much more ... amazing

[wispy canopy]

step 1: i'm not using chatgpt

i've created a character in oobabooga (a LLM web-ui) that spits out natural language prompts

natural language prompts don't work on all models, the more booru-oriented a model is, the worse it will perform with those -> my model performs very well with them(but in reverse, my model SUCKS at booru type prompts). I'm super jelly of cryonicus, who has made an actual model, which honestly, just flatout beats my model xD (the model darkgeminiv3)

though weird, i personally have a hard time prompting lyrics for a song i'm not actually listening to

[untold glacier]

Btw... is VRchat really hurting for business or something?

[wispy canopy]

i still have my rift on my desk... never tried vrchat

[untold glacier]

I've only ever used it on desktop. Some fantastic worlds to visit. Been a while though--didn't know they tried to bring people back to it with spam emails though, lol.

Also just started using the infinite image browsing extension for SD. It's life changing. I can never go back.

[wispy canopy]

you spam email me? that's a hardskip for ANY company from me

"you forgot about us?" -> "probably not, but now i will"

[untold glacier]

[untold glacier]

Top right is

Love the contrast

[wispy canopy]

tell @merry tendon ❤️ that's his model

[steady grail]

i love the first alfhentai and the last gemini

[wispy canopy]

i'll re-do this prompt, upres scaling

... getting fucking fried being in the exhaust air of my pc

the prompt the LLM made was to the lyrics of the song A million miles away from the movie belle

[cyan burrow]

my model does best when given instructions on what art style to use and what type of illustration it is (regular, digital, 2.5d, 3d etc)

[wispy canopy]

so that's why it sucks when it isn't given one... (not a criticism)

[steady grail]

i should try alfahentai and darkgemini

but first, i need to free more space lol

[wispy canopy]

i haven't tried all anime models (waaaay to many) but these do tend to have a focus on being able to work with natural language

[cyan burrow]

since it has so many styles in it with its versatility

defaulting is something it has a hard time doing

[wispy canopy]

which in return, my model does

[steady grail]

it's a funny way to prompt

very cool doing wallpapers with them

[wispy canopy]

@cyan burrow i've been thinking i want to try a merge of our models, see if i can get my "default image quality" working with your model context awareness and concept grasping

[cyan burrow]

once I recover enough that my brain works and I can sit up long enough, I will work on getting those loras into it

[wispy canopy]

my model works worst on booru prompts, actually lol

and masterpiece literally adds nothing, except big boobs

[steady grail]

good, i hate them

uahahahah

[untold glacier]

Just tried the same tornado gen and seed with darkgemini, no more tornado anymore. Lol. I wonder if there's an anime model out there that can do it.

[wispy canopy]

i've had people guess the masterpiece vs non-masterpiece images, and generally they kept picking the non-masterpiece as better image output

[steady grail]

try to look for tornado on booru

the best would be to study how they trained the tokens

[wispy canopy]

i think a very few, my model has some weights from generic models, so it has a little understanding, but full anime models... don't think they ever put in a single tornado image

[cyan burrow]

@wispy canopy mmm I dunno, start out with it as private.

[untold glacier]

Definitely depends on the model and prompt. I've had prompts that would NOT work with masterpiece in it. Especially if you're trying to do something outside of the normal bounds of anime, like a charcoal drawing or something.

[cyan burrow]

you can always try tempest, vortex of air, tornado, storm etc

set things up

and dont use best quality, masterpice

[brittle jay]

wd1.4 tagger better

[wispy canopy]

masterpiece fucks that up. masterpiece fucks EVERYTHING up. there is LITERALLY no reason to prompt masterpiece : it's just a placebo saying "i can't generate enough detail for the prompt, so make it better: THE ANSWER IS BOOBS"

[brittle jay]

probably

[steady grail]

or tempest

[untold glacier]

[wispy canopy]

obviously, and the first one to try , if i'd get it to work as wanted, will be you!

[steady grail]

well, that's how i would answer

[untold glacier]

Well, it used to be great in the early days. It's still a qualifier though.

[steady grail]

what would make stable diffusion better? boobs!

[wispy canopy]

mandatory on nai, discutable on anything, then i dropped anything

and switched to generic models

now i'm back on my own model which is a bastard hybrid between hentai and generic models lol

[steady grail]

i know that feeling -.-'

[cyan burrow]

that's the thing, it really isn't much of a qualifier, perhaps it was before if the dataset contained a lot of masterpiece tags

[untold glacier]

I still see some benefit on Aurora, depending on the prompt, but I haven't done specific testing on the subject.

[steady grail]

i try to avoid any model with anything as component

[cyan burrow]

but right now, it's a hinderance

[wispy canopy]

VERY booru-centric model -> it'll work on that

[untold glacier]

True enough

[steady grail]

it mess even with booru prompts

[wispy canopy]

i've done testing on a LOT of models, but masterpiece will increase breast size or introduce sluttiness 9/10 times

and honestly, i'm SO. FUCKING. DONE. with that

[cyan burrow]

there's a billion and a half hentai models

[steady grail]

you are doing an image, and every 3 4 results on 10, have weird mutations that makes ask yourself, and now, what i did wrong?

than you change model and find that nothing is wrong with the prompt

[cyan burrow]

no if the artifact or distortions are that rare, I don't ask myself what I did wrong

[wispy canopy]

i did download the skinnyhelper, because i have a hentai model which prefers doing k-cup boobs

[steady grail]

it's just anything that is messing with you like every time

[wispy canopy]

but honestly, skinnyhelper is only needed because of masterpiece and the placebo bullshit for image quality

since i switched over to natural language (ditching booru contamination), i've not had oversized boobs

[steady grail]

that model makes me so mad, let's just forget about it

[cyan burrow]

absurdres is also pointless, what exactly are you telling it to do when it's still stuck in a resolution and can only do so many pixels per inch?

better to give it the max PPI it could do

[wispy canopy]

YOU FUCKING CALLED IT OUT BEFORE ME

please, kill absurdres already

absurdres = MORE SLUT = better image

[steady grail]

just change it with bananares

it changes every pixel with bananas

not really, but would be funny XD

[wispy canopy]

seriously, i have to increase my anti-nsfw weight if i even TRY thinking of absurdres

[cyan burrow]

using 4k or 8k etc can help a tiny bit cause the information on images in the dataset with 4k or 8k is greater, so the amount of details it can pack into your res and PPI will be greater.

[steady grail]

just add nude!!! on nega

or (nude:3)

usually works well

[wispy canopy]

4k, 8k, hdr, high quality, this stuff can help for sure, but those are associated with natural/photographic images

[cyan burrow]

I don't want hentai reintroduced into my model, I've almost gotten rid of it 😛

[steady grail]

even in hentai models

[cyan burrow]

HDR and 12bit helps

can also do 12bit YCBCR for color mode

[wispy canopy]

don't think enough images are tagged with it though

[cyan burrow]

probably not

[wispy canopy]

you don't want to know my default neg... it has multiple anti-nsfw for my model... it's REALLY horny

[steady grail]

i noticed nsfw don't works so well

that's why i stopped using it

[cyan burrow]

nsfw also kills violence, combat, blood etc

[steady grail]

just a waste of tokens

[wispy canopy]

, cleavage, sex, nsfw, ecchi, slutty, leotard, add hentai if needed

[untold glacier]

Lol. Just what models are you using for that to happen?

[steady grail]

once you say you don't want nude, the model can just castrate itself

[wispy canopy]

ANY descendant of anyv3

[steady grail]

yeah, leotard and cleavage are useful too

[wispy canopy]

which is why i'm SO FUCKING GLAD i killed it my model

i get the quality, without the fucking bullshit 😄

you are WASTING TOKENS ON QUALITY which should be default in the model

WASTING. TOKENS.

[untold glacier]

I need a vitriol anime reaction lol

[steady grail]

i suppose it's time to purge even base64mix

[untold glacier]

NOPE

I'm holding on to that until I die

[wispy canopy]

best to start over humanity again.... we failed, you know it....

[wispy canopy]

why not husbando ?

[ocean reef]

making character for RPG session for a friend 😄

[wispy canopy]

"for a friend"

i hear that one 100x too manytimes lol

[ocean reef]

hahaha

well, I'm the only one who knows how to use this sh!t (and is interested in it) from my friends circle 😄

[wispy canopy]

still, that as character concept si pretty damn cool

[ocean reef]

and since we play a lot RPG together, why not make it more original

[wispy canopy]

u in a rpg/dnd group?

[ocean reef]

yup

for our playable characters I make full body, for others just torso view 😄

[wispy canopy]

i played 1 dnd game ever, and that dm even made comisions for our player characters

talk about dm dedication

[ocean reef]

It's fun to bring our characters to life, since I cannot draw that well 😄

[untold glacier]

I made this for my friend's D&D char a few months back. It seems such a thing is more common than I thought, lol.

[wispy canopy]

can you give me a character description, (either here or DM) i wanna give it a go (from that campaign)

[ocean reef]

absurdres, (masterpiece:1.4), great composition, (full_body:1.2), (portrait), 1woman, solo, (dark_skin:1.8), (Indian:1.4), (mature:1.4), (futuristic:1.1), (cyberpunk:1.4), ((long black wavy hair:1.4)), brown eyes, futuristic:1.2, finely detailed, breathtaking, cinematic lighting, daytime, dynamic lighting, simple background, depth of field, CG, black leather jacket, dark pants, (henna tattoos on hands:1.4), multicolored eyes, lora:GoodHands-beta2:1.6, casual clothes, lora:eyesgen:0.4, intricate iris, outdoor, lora:add_detail:0.6

oh, Ive noticed that I've put futuristic twice xD haha

[wispy canopy]

oh no. "i spot masterpiece, i die inside"

[ocean reef]

[wispy canopy]

well, first, lets see what my top3 favorite models have to say about it

without the appropriate loras lol

[ocean reef]

hehe

my other friend wanted femboi character, soooo she got this, was happy with it xD

[wispy canopy]

oh. it took 2 minutse for 1/3

[ocean reef]

hah

[wispy canopy]

upscaling time

😦

[wispy canopy]

i can TOTALLY imagine, that's a femboi dream there

adding andrenogous can help a lot for fembois

[ocean reef]

well, we are going to play cyberpunk so yeah, no1 cares about sex 😄

[wispy canopy]

cyberpunk dnd?

[ocean reef]

Cyberpunk RED

DnD is not only RPG to play 😄

[wispy canopy]

i know 😦 but dnd is the only rpg i played

[ocean reef]

I've played Neuroshima and Warhammer

now 1st time Cyberpunk 😄

ofc DnD too

[wispy canopy]

alfahentai why u have to suck this much!

[ocean reef]

omg

so bad

[wispy canopy]

ikr!?

[ocean reef]

but gemini and xynaptx are nice!

[wispy canopy]

my ego is crushed

[ocean reef]

well I did those with DarkSushiMix

[wispy canopy]

the first time alfahentai failed as bad

[ocean reef]

had to add fire to negative coz it got hot xD

[wispy canopy]

it's a full booru prompt... isn't it? (<- didn't even bother to check before copy pasting)

it is.

[ocean reef]

and what does booru mean? xD I just type it all by myself! xD

[wispy canopy]

booru is a propmting style, saying "i'm just gonna throw tags in here"

[ocean reef]

ah yeah

I got story ones from chatGTP only

edit them a bit tho

[wispy canopy]

ok, i think i got my LLM to understand the prompt

lets see what i can get out of it 😄

[ocean reef]

[wispy canopy]

(my LLM understood.... SD doesn't)

[ocean reef]

sad

[wispy canopy]

A masterfully crafted portrait of a mature Indian woman wearing a black leather jacket and dark pants, set against a futuristic backdrop featuring cyberpunk elements. The woman has long, wavy black hair accented with intricate henna tattoos on her hands, and multicolored eyes that seem to glow in the sunlight. She stands alone, looking confidently towards the future with a hint of mystery in her gaze. The artwork should be finely detailed and feature great composition, with cinematic lighting that enhances the overall impact of the piece

[ocean reef]

lemme try this one too! 😄 haha

[wispy canopy]

i mean, come on, this is just making any form of translation EZ mode...

[ocean reef]

yup, like a lot

[wispy canopy]

it did capture ALL of the things in the prompt and translated it to natural language

[ocean reef]

Great... Ive close opera... xD

[wispy canopy]

you closed the browser with the SD ?!

[ocean reef]

yup...

[wispy canopy]

think about the $0.40 you wasted on energy xD

[ocean reef]

🤣

[wispy canopy]

we go pure natural language?! Alfahentai is the best!

[ocean reef]

yup

[wispy canopy]

but this is why i like natural language prompts: it's able to set the scene

[ocean reef]

yeah, sometimes the result is rly good

[wispy canopy]

and i'm still jelly about darkgemini3 - > for like 10% comprimise on the chracter, you get 90% more environmental stuff

contacted cryonicus already, imma be trying to make an ultimate merge on those 2

[ocean reef]

ohoo

gonna be sth good 😄

[wispy canopy]

i need his model's context awareness .. BAD, it's simply better

[ocean reef]

I got those with your prompt

had to add (dark skin:1.6) tho, coz they were all white 😄

and yeah, no wavy hair

[wispy canopy]

pretty damn good, you know

[ocean reef]

I like the style of this model

sometimes it's way too dark tho

now trying nabimix, and with preview I can see it's more artistic 😄

[wispy canopy]

i think EVERYONE wants a model which is "best at everything"

[ocean reef]

impossible

thats why I have a lot of anime models with different styles

[wispy canopy]

lol, clyde blocked a sexy tummy

[ocean reef]

xD

[wispy canopy]

(i dont even care about the breasts here, the tummy is 100% the sexy part)

reposted with some extra borders to fool clyde

[ocean reef]

haha

[wispy canopy]

but i REALLY like that style (eface model merging time?!)

this be good shit!

[ocean reef]

looks nice!

[wispy canopy]

.... is darkxynalfa a thing?

[ocean reef]

imgs from your prompt with nabimix model

and my prompt (got ignored)

[wispy canopy]

it just skippped the dark skin?

[ocean reef]

and hair xD

and long haaair

so yeah

artistic model 😄