🍥｜anime

[nova remnant]

Mermaid anime

[wispy canopy]

at some point i should have stopped , but the age old quote Your scientists were so preoccupied with whether they could, they didn’t stop to think if they should

[cyan burrow]

downloading all 1.1 control nets, are there any control net models or pre-processors that aren't in the bulk on huggingface?

[daring ice]

my third creation, is it good?

[wispy canopy]

keep it up!

oh dang 😮 her other wing got ripped off

[desert arch]

Slipped on a banan

[cyan burrow]

Rip the wings of a butterfly emo song plays in head

off

[rich tartan]

"ThAT b** stole my teeth!"

[wispy canopy]

ooff..

[cyan burrow]

add "trapped in a collapsing tesseract, a hypercube folding in on itself, destroying every her in the multiverse"

[rich tartan]

@trim current on rocm you get indeed many naked ones and it is much more focused on other things that im still trying to figure out, in directml you didn't need to type naked into negative prompt, it was somehow better, but on rocm it is much more realistic instead fantasy

How did you feel switching from directml to rocm?

[nova remnant]

i feel sad looking at that

[cyan burrow]

is that the result of the tesseract?

She has a faerie version of herself and maybe those flying there too. She gave birth

btw are there any new VAE files? Aside from SD 1.5 Official or KL-Anime 2?

[trim current]

idk lol i used my own model and the way i do negs is different than you, could be your model being trained on explicit and now that you can actually do the proper res its doing what it was best trained on lol try adding your ngs obvc like nsfw, nude etc in the ngs to enforce SFW. or try another modeel

[wispy canopy]

no tesseracts here

[rich tartan]

yeah, i already did now, that possibly could be

but tf, it still generates nsfw, i'll try with another model then, thank you for that suggestion, possible that is my issue

[trim current]

yee

😄

[wispy canopy]

https://tenor.com/view/i-dont-quite-see-the-problem-here-dont-understand-whtas-wrong-whats-the-issue-lucifer-morningstar-gif-16837823

[nova remnant]

[rich tartan]

I don't want to generate images to jerk on(literally generates only this), i want clean and beautiful landscapes.

And yeah, right away, i wanted to ask you also two questions,
did you ever tried to merge checkpoints with loras?
and also did you merge a anime checkpoint with a landscape/environment checkpoint, what was the outcome if you did?

[wispy canopy]

nope, haven't been able to merge with lora's, but i'm a total noob when it comes to that. And my checkpoint IS merged with a generic model for that specific purpose. go see my last posts in #1071945441081561198 😄

[rich tartan]

ayo, exactly what i am seaching for, seems possible, thank you very much

^^

[cyan burrow]

Try either my Dark Gemini 3 or Eface's Dark Alfa (a spin-off on Dark Gemini 3, merged with his AlfaHentai and now a few more)

[stable crane]

very nice light

[cyan burrow]

then you can make anything you want

and not just waifus

[rich tartan]

Oh, yeah i would give a try

are they on civitai?

[nova remnant]

Are you going for realistic landscaping or more anime style just curious

[stable crane]

it's hard to believe it's AI generated, I am just so drawn to the orange line on the clothes

[cyan burrow]

https://civitai.com/models/6209/dark-gemini-v3

[stable crane]

🤔

[wispy canopy]

yeah, you can see the repeating pattern xD

upscale failed 😦

[rich tartan]

thank you

[rich tartan]

here an example anime/realistic: (i generated that with directml)

[nova remnant]

Hot robo girl

[stable crane]

oh did pika release a new model

[cyan burrow]

some examples

You can do brighter things too

[wispy canopy]

darkalfa v1.1 landscapes 😄 lol

[desert arch]

Inabakumori

[rich tartan]

ah good to know, just gotta wait for download finish

[stable crane]

LottaLenses

[wispy canopy]

just an extra eye on you

[rich tartan]

ayo, could i give that also a try?

[nova remnant]

Lookin at me funny

[wispy canopy]

@rich tartan https://civitai.com/models/64843/dark-alfa

[rich tartan]

Thank you very much

[wispy canopy]

she got out of her cyber-bath

[stable crane]

you can see where her face used to be before it diffused into a mech, and replaced it with a tinier face

[rich tartan]

ayo, that are some good lightings there, what sampler did you use?

[wispy canopy]

UniPC

[rich tartan]

Never heard of, is that alr included?

ah wait yes

i see

[cyan burrow]

more examples

[rich tartan]

never tried out, i'll take a look on that too then, thanks

[rich tartan]

woah, those last 2 mountains, they are freaking awesome

[stable crane]

what model are you using?

[wispy canopy]

darkalfa 1.1

[stable crane]

for the girl in sunset pics too?

[wispy canopy]

everything since the 14th (or somewhere around there) 😄

[stable crane]

nice I'll give it a try

[wispy canopy]

^^

[stable crane]

ohh you suggest https://civitai.com/models/56519?modelVersionId=60938 ?

[rich tartan]

did you made that again randomly and had luck or was this this time with planings?

[cyan burrow]

and this is from a model I'm working on

[rich tartan]

ah, i like the cyber stuff details, keep me updated when it's done, i'll def gonna check it out

[wispy canopy]

darkalfa, and v1.1 were both actually planned 🙂

[cyan burrow]

yeah @wispy canopy got a hold of my model and seized the awesomeness

[wispy canopy]

and i've done some testing, it gained enough intelligence to be able to discern what 'lobsterpunk' should be 😄

[cyan burrow]

to be fair, he made something awesome too

or she? he?

[pastel pollen]

https://tenor.com/view/nachobutt-david-saving-them-for-later-gif-16965627

[wispy canopy]

i'm a he. But as internet concerned: it's an IT 😄

[rich tartan]

xD

uh a dumb question, but can i download those models with a linux command too? They keep failing in the browser download manager

[wispy canopy]

i've been able to wget them, but you'll have to rename them afterwards

(not a linux expert)

[stable crane]

I figure that has worse resumability?

sometimes civitai just fails, click "Retry" in browser

it will continue

[rich tartan]

ah nice, got it, thanks again

[wispy canopy]

@rich tartan

[rich tartan]

i swear literally the most part is true

except my visual

[wispy canopy]

the difference between 30b parameters and 900m parameters

[rich tartan]

just gave "the frieber" into the prompt

tf

crazy that it knew, i like cyan eyes and white hairs

[wispy canopy]

in this case, you could use the extension tokenizer, and see in what tokens that prompt would break down in

[cyan burrow]

Yeah tokens are weird

[wispy canopy]

@cyan burrow gonna do some vertical ones, the others were group pictures (sd duplicates)

but that robot, that feels like a real cryonicus to me!

[rich tartan]

i gtg go, now, i'll let the downloads run, im back gotta pick up someone from the airport

[cyan burrow]

do you even prompt for yourself anymore @wispy canopy

[rich tartan]

cya

[wispy canopy]

i got the imagination of a goldfish, this is a godsend 😄

however, i've always been EXTREMELY intuitive when it comes down to interfacing with machines, so manipulating the starting prompts to get what i want from it, is not hard 🙂e

[cyan burrow]

I don't see myself as a robot. Not entirely

[wispy canopy]

i can see why it went with it, i suppose i could regenerate a couple of times more

got some cool cryo-bots out of them though 😄

[cyan burrow]

tell it to make it an at least partially organic life form, not augmented with futuristic technology

[wispy canopy]

oooh, LLM decided to get creative 😮

i think next batch will be exactly that!

[desert arch]

Last one, gn chat

God bless inabakumori

[wispy canopy]

@cyan burrow second description. the mini-souls on #3 though ❤️

[cyan burrow]

I aint no woman

[wispy canopy]

that's just my SD model preference when it's unspecified 😛

I'll do a male run too

got any specific mage/magic style you want added @cyan burrow ? i was thinking warlock

ah, just make all neutral gender references male, that fixed it ^^

[cyan burrow]

cryomancer

[wispy canopy]

dang, these are wicked 😮

[stable crane]

it is time to try @wispy canopy's model 🙂

[wispy canopy]

be sure to read the model description, the model has some... quirks 😛

[cyan burrow]

I'm working on HyperVariant right now

[wispy canopy]

(don't set cfg too high, be sure not to just go for all the negative embeddings) 😄

[daring ice]

the minutes from yesterday to today only increased 😦

[stable crane]

spicy

[wispy canopy]

GALACTIC MILK!

[wispy canopy]

are you sure you're not running out of vram?

[stable crane]

Breasts still work though
thank god

[wispy canopy]

😄 😛

[daring ice]

i dont change nothing from yesterday

[wispy canopy]

there might be other things outside SD that might be impacting your performance

[daring ice]

guys, this automatic1111 is better for low pc? dumb question

[wispy canopy]

how do you mean?

[stable crane]

you have to say "better than ____" 😛

[wispy canopy]

damn, is that on my model? if so, take it easy 😮

i know it failed me a couple of times on astronaut suits

[stable crane]

it is

[stable crane]

I'll be trying a random selection of old gens on it

this one has a really cool vibe

[trim current]

You could prolly do a comfyui node to implement the llm like gpt node

[wispy canopy]

more cryonicus' . This time, male heroic edition which the LLM dubbed 'the frostbinder' ... #3 has an actual ice sword?!

[cyan burrow]

heh cool

[wispy canopy]

doesn't work that well in practice... i get the best results on the 30b models, which won't fit alongside SD in 24gb vram

[cyan burrow]

I'm sticking to making my own prompts, it's more fun than letting a bot do it for me

[trim current]

Oh yeah no

Bottlenecks op

[trim current]

Same thing could be said for non ai artists. Its just another tool to use

[wispy canopy]

for sure. but for people with limited imagination, it really does help a lot, and more than often, just that little nudge is enough

[trim current]

I like gpt summarizing and cleaning up my incoherent adhd thoughts

It makes better prompts sometimrs

[stable crane]

your neg vs my neg

[cyan burrow]

I want to learn to draw still and I would use AI to speed through some tedium

[wispy canopy]

works wonders, doesn't it 🙂

[trim current]

Sure does i love it

[stable crane]

it's crazy how much diff the neg can make

[wispy canopy]

flexible model 😄

[cyan burrow]

I'm using my adhd as a tool

I use it to let my imagination give me ideas, hundreds

[trim current]

Unfortunately theres gonna be gatekeeping in any community. Lol

[stable crane]

my imagination is pretty crap

summarized, "but what if it had boobs too 🤔 "

[trim current]

Lmfao

[cyan burrow]

gatekeeping? I have no intention of getting into their little elitist corner of comissions

[trim current]

With a name like LottaLewds who wouldve thought 😛

[wispy canopy]

i have added boobs to black holes. 😄 because i can. and i will do it again if need to!

[stable crane]

you never know when your object needs boobs

[trim current]

Kek

This apple?

[stable crane]

lol I wish it could be more flexible at times though

[cyan burrow]

I'm also very good at writing dialogue and coming up with story and world building, so once I stop procrastinating like a moron and actually do some writing. I can get some work done

[daring ice]

😐

[stable crane]

@wispy canopy I really like the model

[wispy canopy]

any creator (BLATANT RIPPER OF OTHERS) loves to hear that ❤️

[trim current]

Psstt try mine with that prompt https://civitai.com/models/68061 ❤️

Lowkey why im linking

Kek

[stable crane]

5.55gb 😢

[trim current]

Yaaaaa ik

[stable crane]

damn it looks pretty though

[trim current]

The dude posting images since i releasd been going ham

[cyan burrow]

the people flee from the invading demons coming from the sky, but the badass heroine strides towards it

[trim current]

And strix given good feedback. I enjoy feedback lol

[stable crane]

Hades?

[trim current]

Idk them but yea that person haha

[stable crane]

that's nice of them

[wispy canopy]

@stable crane i dunno why it went for green. but given yoru usual character. i love this 😄

[trim current]

I agree

[stable crane]

lol nice

AI knows

[trim current]

Tbh i think its 5gb cos i forgot to save as fp16

[stable crane]

any good suggestion on local LLMs?

[trim current]

I hear wizard is good but i dont use llms that often

[stable crane]

I title this one
MILK

[trim current]

Milk with legs

[wispy canopy]

depending on vram -> wizard mega 13b or gpt4-x-alpasta 30b

[stable crane]

a bit fried, but it really accentuated the milk container

[trim current]

It doesnt like weights i find

If youre weighting stuff try removing

[stable crane]

I haven't downloaded it yet but I will try tonight hopefully @trim current

[trim current]

Oh kek mb

But yeah my model doesnt jeed much weighting

(1.2) at most i use

[cyan burrow]

do LLMs run well on intel gpus?

cause they give a lot of vram for little money

[wispy canopy]

so many jugs in that picture 😄

[stable crane]

it does well with bluearchive emoji lora

some models will get blurry with it

[wispy canopy]

if there is an repository for inference on intel gpu's... i can't recall if there's any though

[wispy canopy]

it doesn't know wariza anymore though 😦 i killed that in the merging process XD

[stable crane]

😦

[wispy canopy]

it recalls it's something related to sitting, but it won't be able to get it done properly

giving it an additional support tag usually fixes it

[cyan burrow]

You can get one of the ARCs with like 16gb ram for 250 euro

not same speed and stuff as nvidia

but for an LLM, maybe

[wispy canopy]

i did a small google, it seems some people just hacked the intel pytorch extensions in, so it does work

[stable crane]

man you guys with your fancy UniPC samplers, I haven't pulled since like Feb

I should do a fresh reinstall

[cyan burrow]

nice

[stable crane]

now iirc there was a fork of A1111 that was further now?

[cyan burrow]

maybe a small investment into an ARC is worth it

@stable crane the downside with it is that it's even less concerned with conflicts or pointlessly convoluted settings

[stable crane]

so super wild west

[wispy canopy]

i don't think so tho -> you'll be maintining your own pytorch modules for inference

[cyan burrow]

I don't like it, I can't get it to output the images of any quality

[stable crane]

UniPC?

[cyan burrow]

auto started updating again tho

no, vlads auto fork

I don't like it

[stable crane]

ohh

I'll probably stick to what I know

[wispy canopy]

step 1 for vlad -> change ui back to gradio default 😄

[stable crane]

but I'd like to try comfyui too

[cyan burrow]

step 2 would be to remove any pointless settings that arrive without explanation

[wispy canopy]

i like having settings though 🙂

[cyan burrow]

yes but what about a whole page of sliders and check boxes that have already by being there altered how your images are generated and then there's no guide, wiki, faq, nothing

You don't know why it's there

or by whom

or how to use it

the fork throws things in at a speed where they don't have time to explain it

[subtle crown]

risk, always a risk

[daring ice]

wow

[wispy canopy]

mooooh-litary grade 😄

[cyan burrow]

two other variants

[stable crane]

it works very well with my loras so far

[subtle crown]

what model is that

[stable crane]

@wispy canopy's https://civitai.com/models/64843/dark-alfa

[wispy canopy]

got a couple of ideas for a possible 1.2 but kinda burned out testing all kinds of weight merges

[wispy canopy]

KETCHUP 😄

[cyan burrow]

do it the way I do it, intuitively

and then just running through a few prompts I have of different styles, arts, aesthetics, contents

[wispy canopy]

dunno what causes that thing on the noses. you can remove it with glasses in the negative prompt

[stable crane]

perfect veil

[stable crane]

the thing that looks like a scar? or the wrinkles below?

[wispy canopy]

the scar

[stable crane]

oh I thought it looked kinda cool

[wispy canopy]

it's been there since alfahentai xD

[stable crane]

it makes a really good moth

best helmets I've ever seen

[wispy canopy]

😄

[stable crane]

I'll try to post a ton of images for review later on civi

[nova remnant]

Hourglass body oh my

[wispy canopy]

Now that i see this, i'm pretty sure that trying to improve a model's conceptual awareness is just the best way to go

the rest you can fix with lora's and stuff. but you can't add 'intelligence' afterwards

[stable crane]

yeah it seems very good conceptually

it doesn't kill things from other loras

so I wonder what it'd be like to train a lora with this as a base

[wispy canopy]

well.. she kept getting a hat, but it does remarkably well at keeping two people seperate in prompting alone 😮

[stable crane]

solid bots

[wispy canopy]

good thing he's still hooked up, he'd come steal my waifus

[stable crane]

I should make like, a model testing suite script

run through certain prompts, one after the next

I'm having so much fun giving new life to old concepts with the new model

[wispy canopy]

same here, runnign it through old prompts and simply be amazed how much better it performs on old prompts 😮

[stable crane]

it hasn't really missed yet

[wispy canopy]

busty robot blueprint ? 😄

[stable crane]

something like that

[wispy canopy]

REINFORCED HIPS!

[stable crane]

she's resilient

[sullen hollow]

cows in 3023

[cyan burrow]

feel youre in the 90s yet?

[wispy canopy]

LLM take on busty robots:

[cyan burrow]

control net reference pre processors are amazing

[wispy canopy]

yes they are 😄

[cyan burrow]

my hypervariant alpha model and some control net

[cyan burrow]

I think it's doing pretty well

[wispy canopy]

bro, stop flexing on me with my own model 😮 these are gorgeous

[cyan burrow]

even if the fox in the background looks like he just took a solid crap and squeezed a boulder out

What are you guys opinion, is HyperVariant heading the right way?

[cyan burrow]

or is it bad?

[wispy canopy]

what are the intentions for the model to achieve?

[sullen hollow]

awww

[cyan burrow]

To be capable of hybrid art, to gain more consistency than DG3, to gain better overall image quality but retain the prompt recognition and backgrounds

the hybrid art I haven't even shown though

but image quality, retention and consistency etc

[wispy canopy]

biblically accurate angel. fml. that thing is awesome!

[cyan burrow]

I showed you stuff yesterday too

yes nice one

[wispy canopy]

'do not be afraid' . DUDE, I'M FREAKING OUT

[stable crane]

😄

[wispy canopy]

@stable crane how far are you going to push it man?!

[stable crane]

yeah too bad the eye blob didn't have massive wings

though I didn't try adding that to the prompt

[wispy canopy]

llm prompt: love lost

[stable crane]

the hug one is very good

[wispy canopy]

An illustration portraying a poignant moment of heartache and longing. At the center of the composition stands a person, their head bowed and shoulders slumped, clutching a small object close to their chest. Their eyes are closed, and their features are contorted in sorrow. The background is hazy and indistinct, suggesting a blurred memory or distant past. The colors used are muted and somber, with shades of blue and gray dominating the palette. The overall atmosphere is one of melancholy and loss, conveying the deep emotional pain associated with love lost.

i mean, it fails at the exact description, but it does remarkably well capturing the feels

[stable crane]

ohh I have a cool one, trying to get less nipples now

[wispy canopy]

uhh... an organmancer? 😄 intestinemancer?

😄

[stable crane]

lol

that'd be a hella creepy mage

strangled by her floating intestines

also, someone should make that a real wizard in a book/movie. would be real gross in film

[wispy canopy]

yeah, that would be the REAL creepy stuff

[cyan burrow]

an illustration of an amalgamation of slain demons, rising as a kaiju to terrorize and destroy the medieval towns in its path. With long tentacles hanging out of several orifices and 6 spider-like legs but with the face of an expressionless human. The people launch arrows and cannons at the Kaiju to no effect, until the hero arrives carried by the wind and materialized from the blood of the dead. His armor looks almost like a flayed human being and his sword drips of black goo. He stares up at the Kaiju with a grin, ready to make another kill

[wispy canopy]

i really hate that v1.1 butchered hands vs 1.0. but the increased 'intelligence' in the model vastly outweighs bad hands, imho

[subtle crown]

its ok i guess, realistic noses need a lot of work...

[wispy canopy]

i spot it immediatly: too low resolution 🙂

upscale it a bit... 1.3-1.5 should be enough to get the noses fixed

[cyan burrow]

they do? 😮

[subtle crown]

they are too big, its a man nose lol

[wispy canopy]

is this hypervariant? it seems to have some 'indicesiveness' left

[cyan burrow]

what's it indecisive about?

[wispy canopy]

what it wants to do with the stick

[clever shore]

:O how’d u get that style, it looks straight from an anime

[cyan burrow]

that's cause I was very unspecific

so it really doesn't know

[cyan burrow]

these are pretty consistent for example

[wispy canopy]

i had similar outputs in my model, regarding the eyes and striping -> that's something i attributed to a hard fail -> it's hallucinating hair in early stages but can't resolve it in later stages and just makes something up

[cyan burrow]

the bloodier (red) her eyes get, the more excited her vampire side is

oh you mean that?

[wispy canopy]

yeah. if you see that, the merge is a fail

[cyan burrow]

yeah I think I resolved it with the latest model

[wispy canopy]

it can NEVER do that

because it will do that everywhere

[cyan burrow]

only it doesn't

[wispy canopy]

hallucinate shit in early stages, but can't resolve in later stages

remember what i said about the stick?

[cyan burrow]

that was 2 versions ago though

[cyan burrow]

(insanely detailed beautiful young woman with two tone hair black and red ends:1.10), medium length flowing hair, smooth pale skin, genesis blue eyes, (beauty filter:1.096), (black skeletal wings coated with a net of fleshy tendons:1.12), (she holds a staff with a beating heart at the top:1.156), at night under a blood red moon, (Disaster Art Style:1.10), (Digital Art:1.14), (digital illustration:1.15), (Dark Fantasy Art Style:1.13), (Pseudorealism:1.20), dim lighting, soft lighting, (2.5d Art Style infusion:1.2), Cowboy shot, eye level, Ultra-Wide Angle, (masterpiece:1.132), (best quality:1.132)
Negative prompt:neg_hands, (low quality:1.3), (worst quality:1.3), (lowres:1.3), (error:1.3), (fault:1.3), (undefined:1.3), (crude:1.3), jagged, spiky, ugly hands, extra arm, extra hand, split arm, missing finger, extra finger, three fingers, four fingers, six fingers, (disfigured:1.3), (unclear:1.3), (indistinct:1.3), merged fingers, (bad anatomy,:1.3) misplaced hand, misplaced foot, (text:1.3), (signature:1.3), (title:1.3), big tits, major tits, massive tits, big breast, major breast, massive breast,

Steps: 25, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1152534537, Size: 768x768, Model hash: e82a2dadaa, Model: HyperVariant_alpha_0.3, Clip skip: 2, Version: v1.2.1

try it yourself

sometimes it's the prompt that confuses any model

[cyan burrow]

and has to be fixed

[stable crane]

cropped

[cyan burrow]

and the eye thing, I think it's trying to make her cry blood

you got the same confused result

[wispy canopy]

it's a very ... laden prompt. which is pretty cool, but some things will not make sense if you respect SD's current boundaries

[cyan burrow]

I know, which is what I said

that it's the prompts fault

not the model

[wispy canopy]

imma remove all the weights, and stuff i feel my model has no use for

[trim current]

https://tenor.com/view/jesuschrist-jesus-christ-kid-giacomopisano-gif-19088356

[wispy canopy]

perfection!

[wispy canopy]

so, what did we learn? Don't prompt stuff the model won't understand

[stable crane]

"shrimp tank" .. lol

[wispy canopy]

more perfection!

[cyan burrow]

I've always known that, it's a very old prompt

that I just decided to test

[wispy canopy]

in this case -> i know masterpiece made it go wanky, because i've basically just killed that bullshit in my model

why would you need a BULLSHIT prompt to get good quality, when that should be default?

[cyan burrow]

I removed it too when I just rewrote it

[cyan burrow]

like I said, old prompt

[wispy canopy]

😄

and i'm pretty sure disaster art style is... hard to grasp

[wispy canopy]

food is fighting back!

[cyan burrow]

it used to work, but the context has to be more apocalyptic

[wispy canopy]

turret has a turret 😄 that gives me metal slug vibes 😄

[wispy canopy]

@cyan burrow changing aspect ratio brought it back. lemme run daam on this specific seed and prompt, see if i can isolate what block and word is causing it

[wispy canopy]

failed jellyfish hair attempt?

[cyan burrow]

indeed

[stable crane]

I like her graphic T

[subtle crown]

#🍥｜anime message @wispy canopy no hyperrealism version

[cyan burrow]

whatever I just did has cursed me with these faces so now I spread the joy'

[stable crane]

4th pic is real cute

[wispy canopy]

I identified the issue on my model on OUT3 and OUT4

might want to check it on your model too. and be blessed with your cursed generations ❤️

[wispy canopy]

she's a jellymancer 😄

[wispy canopy]

'Tanked' Almost 😄

[wispy canopy]

Impressive playing cards ❤️

[stable crane]

you can see the spidergirl influence

[wispy canopy]

she's hiding the whole spider body in the bakcground 😄

you know... the scene where the shadow drops and she's actually a friggin 10m spider body thingy

[cyan burrow]

how'd you identify that?

[wispy canopy]

DAAM -> just run the whole prompt and see for each word which causes what in which layer -> OUT3-4 cause that wierd hair/eye thingy

[silk narwhal]

anyone successfully managed to make a composition like this with t2i?

[cyan burrow]

that says nothing to me

explain in DM?

[subtle crown]

girl got skills

[stable crane]

not quite, but close with latent couple

[silk narwhal]

???

[lilac totem]

bark

[stable crane]

leaves aren't occluding though

but splits the prompt into 3 regions, girl in center, leafy stuff on sides

similar

[rich tartan]

would be cool to add an "blur areas" option

back btw^

[lilac totem]

prompt exist

inpaint exist

[silk narwhal]

ah..ok..so basically just masking and it will generate where the masked areas are based on the mask color

[rich tartan]

i mean with RPC

[lilac totem]

[lilac totem]

result i got off your prompt

[subtle crown]

this is https://civitai.com/models/65986/ruvistyle

[lilac totem]

i dont got the lora lol

jesus

[stable crane]

oh you need to enable latent coupling and define the regions too\

[lilac totem]

have you ever tried a anime pic with 150samples

[stable crane]

yeah but mostly it's a waste of time

[lilac totem]

dehek

[stable crane]

it can change things on samplers that don't converge

but like, DDIM after 30 is very similar to DDIM after 90

so might as well stop at 30

[silk narwhal]

Honkai vibes

[lilac totem]

nah

[wispy canopy]

illuminati confirmed

[lilac totem]

illuminati is a triangle

[lilac totem]

some good pics i made yesterday

[stable crane]

every once in a while

[lilac totem]

gonna do a 150steps

[stable crane]

@wispy canopy it likes the nose lines

[stable crane]

and the "glasses" come up again

[lilac totem]

the hand is gonna smear

i forgot anime vae

[wispy canopy]

it's really annoying, amiright?

i think it's a thing that was in alfamix

[silk narwhal]

better thanks for this 😄

[lilac totem]

the hand got fine

[stable crane]

nice

[silk narwhal]

not sure why every single model loves to do this weird watery drip down randomly and the next few are regularly sharp

[wispy canopy]

model understands what a passionate kiss is ^^

[lilac totem]

hahaaahafidsgdgf

[rich tartan]

ah, training another one or still testing the darkalfa?

[lilac totem]

return torch.group_norm(input, num_groups, weight, bias, eps, torch.backends.cudnn.enabled)
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 2.47 GiB (GPU 0; 12.00 GiB total capacity; 8.26 GiB already allocated; 0 bytes free; 9.55 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

[wispy canopy]

not testing. i put 1.1 online 😄

btw, @rich tartan -> LLM models associate your name with freedom/exploring 😄 tried it a couple of times, but it kept going to that theme 😛

[silk narwhal]

time to save up for more and faster RAM :/

[rich tartan]

wtf, what are you running, a linux server?

[lilac totem]

:o

[lilac totem]

2

2000

2100?

[rich tartan]

oh interesting

[lilac totem]

pls use xmp

gives 3200

or buy ddr5

[silk narwhal]

win 11 and cannot use xmp due to mixed sticks

[lilac totem]

gives 5600mhz

[lilac totem]

it can

just dont mix

[rich tartan]

pf what, i thought that was win10

[wispy canopy]

uhh... this is just my pc? XD

[rich tartan]

dam the task manager def didn't got a bit of change

[lilac totem]

[lilac totem]

bro got 3 ethernet cables

[wispy canopy]

wsl -> internal network connection to ubuntu machine, openvpn tap = network tunnel

[rich tartan]

holy sh* how much drives you got

[wispy canopy]

a few.

[cyan burrow]

I have two ports for ethernet on my mobo

[rich tartan]

i got 4 and i don't have as much as that

[cyan burrow]

one 1gb and one 2.5gb

[lilac totem]

why does a motherboard even have 2

[wispy canopy]

Eface does not delete. Eface does not forget!

[lilac totem]

bro got a 8tb drive

[stable crane]

damn yo

[lilac totem]

childpo

https://tenor.com/view/child-po-kung-fu-panda-panda-kung-fu-gif-27359038

[cyan burrow]

I don't know, maybe for NAS?

[lilac totem]

bro works for nasa

[rich tartan]

holy

[rich tartan]

what did you download so much?

[lilac totem]

2 are usb

1 is hdd from 2009

[wispy canopy]

anime, games 😄

don't watch them, don't play them

[rich tartan]

oh

[wispy canopy]

but i just want them, in case apolypse happens

[lilac totem]

i suffered by the health drain of the 990 pro

i got my pc beginning this year

[rich tartan]

but they took terrabytes?

[lilac totem]

my ssd health is on like 90%

[wispy canopy]

worst thing you can do to an SSD is make the %free go below 5%

[lilac totem]

what

%free?

[wispy canopy]

ssd's can only be written so many times

anything that is 'written' is safe

[wispy canopy]

but free space can be overwritten any amount of times

[silk narwhal]

like this 🙂

[lilac totem]

let me download magisk

crystal*

[wispy canopy]

so if you have very low free space : that little free space gets overwritten MANY times -> SSD's don't like that

[lilac totem]

https://crystalmark.info/en/software/crystaldiskinfo/

[rich tartan]

Chill, it is enough to download 100.000 times an os

[lilac totem]

💀

[rich tartan]

what the fk?

[lilac totem]

hm

[stable crane]

I figure you replace it before it fails 99% of the time

[lilac totem]

turned on:

200times

1085 hours

[subtle crown]

RIP

[lilac totem]

hdd made in 2009

[sullen hollow]

gut

[wispy canopy]

milky mommy military mech butterfly?! now you're just styling dude

[rich tartan]

huh, did you use your ssd as extra ram storage or how did you come with 200times at 1%?

[lilac totem]

what

that wasnt mine

[rich tartan]

oh

[lilac totem]

look at the image of the link i sent

[subtle crown]

the newest drives are really well made, compared to older models

[wispy canopy]

swap memory = ram (just EXTREMELY slow)

[lilac totem]

usb hdd

[rich tartan]

now i understood

[lilac totem]

what

[rich tartan]

im on linux, fk

[lilac totem]

i wanna do a benchmark rn

[lilac totem]

wine

[wispy canopy]

any OS has swap memory -> it uses swap memory when it runs out of physical ram, or notices some programs are not being active, and puts that memory on slow storage

[silk narwhal]

SSD is fine 🙂

[rich tartan]

can it read ext4 formatted sdd?

[lilac totem]

i have 64gb lol

[lilac totem]

idfk

[wispy canopy]

same here. and still swap is used -> unused memory pages get swapped out regardless

[stable crane]

@wispy canopy it seems to make nipples wrinkly

[lilac totem]

https://crystalmark.info/en/download/

[wispy canopy]

yeah, i know. noticed the same. i dropped the hentai part for a reason, to get better regular results 🙂 (same as why genitalia aren't as good as they were)

[lilac totem]

https://osdn.net/projects/crystaldiskmark/downloads/77936/CrystalDiskMark8_0_4c.zip/

benchmark

[stable crane]

cool

[wispy canopy]

goddamn... why does asticassia have to sound as good as it does

[wispy canopy]

dude. just. ... DUDE

lol

[stable crane]

I'm so hungry now

[subtle crown]

i say go for the eye first

[wispy canopy]

extra chewy

[subtle crown]

and moist

[rich tartan]

gonna wait 2 min to check ssd health

[wispy canopy]

oh @stable crane you haven't even seen the best feature yet!!!

[stable crane]

oh?

[lilac totem]

C

[wispy canopy]

rainbow hair, it understands!

[stable crane]

wow

impressive

[wispy canopy]

i noticed something similar with elysium v2 -> it understood lobsterpunk

i think it's the phenomenon called 'emergent behaviour'

nobody told it what lobsterpunk is, because IT DOESN'T EXIST

[stable crane]

(Double exposure effect:1.3) of (female portrait :1.2) (silhouette:1.1) superimposed against sakura trees, soft colors, negative space, photoillustration

[silk narwhal]

O_O

[wispy canopy]

however -> it has seen the term enough, to know that somethingpunk means, it's not punk, it's supposed to be applied as a style

[stable crane]

let me try some lobsterpunk too

[wispy canopy]

@sullen hollow my model can apply lobsterpunk. please don't.

[gritty patrol]

how did clyde not nuke this

[silk narwhal]

bribed with the gold

[wispy canopy]

it's wholesome familiy oriented gold hoarding dragon lady

[stable crane]

yeah it knows this is just how dragon girls are

[lilac totem]

D

2009 hdd

[lilac totem]

usb hdd old one

[lilac totem]

newer usb hdd

😎

faster than my internal hdd

idk why im sending this here

[stable crane]

strawberry seed freckles

[wispy canopy]

THE ALFAHENTAI GOGGLES MAKE THEIR RETURN

[tropic wraith]

FAQ: Why are my images blurry?

In order to ensure a safe experience, the DreamStudio website has a NSFW classifier that will detect and blur any potential NSFW images. While in most cases the classifier will appropriately identify NSFW images, there may be occasional false positives due to the nature of how these systems work. We will continue to work on and improve the classifier to make false positives less and less likely! You are not charged for any images that are blurred

[wispy canopy]

don't be a ...

[sullen hollow]

does anyone know if other inpaint models than this exist or is this good as it is?

[wispy canopy]

@stable crane u still on my model?

[stable crane]

ya

[wispy canopy]

you really making it shine man 😮

[stable crane]

thanks 🙂

[wispy canopy]

dude. you're fucking flexing on me now

awesome ❤️

[stable crane]

lol

it's a great model

[wispy canopy]

LOBSTERPUNK MECHA! REJOICE

srsly, no joke 😄

[rich tartan]

[wispy canopy]

stuff like this would win art contests pre-AI era... hands down

[rich tartan]

0% ssd

[cyan burrow]

Now they're paranoid like during the red scare, anyone could be a commu--I mean synthographer

and good night to you all

[rich tartan]

gn

dam, uniPC not available for me

[wispy canopy]

still not updating? 😮

else create new directory and git clone new repo there

[wispy canopy]

is this even a single prompt? 😮

[rich tartan]

since i need to use custom pytorch build for my gpu, UniPC seems like one of the features not supported

[stable crane]

yeah 😄 as opposed to? img2img?

[wispy canopy]

sucks to hear

[wispy canopy]

(inpaint, img2img, controlnet, regional prompting)

[stable crane]

oh yeah all just single txt2img

[wispy canopy]

you putting it to work man

[stable crane]

(Double exposure effect:1.3) of (female portrait:1.2) (silhouette:1.1) (superimposed:1.2) against (swirling clouds:1.4), long white hair, white complexion, by Dan Mountford, by Dan Hillier, soft colors, negative space, photoillustration

[rich tartan]

now testing both models

[wispy canopy]

illuminati waifu

[rich tartan]

hm, she doesn't seem dangerous

[wispy canopy]

you fell for it. they never seem dangerous!

[rich tartan]

ah dam

[wispy canopy]

one finger away from perfection 😮

[rich tartan]

@wispy canopy what's happening here? It got a little too bright

[wispy canopy]

i think that'st he old v1.1 version

[stable crane]

it's even good at interpreting what other models render as a schizo prompt (left yours, right some other model)

[wispy canopy]

i reuploaded v1.1.1 about 12h ago

the vae is borked -> either manually load a vae, or re-download

[wispy canopy]

yeah, it's REALLY resilient against prompts that are 'pure garbage' to others

[rich tartan]

used the download link from the "Download (1.99 GB)" button

[rich tartan]

you got vae?

[stable crane]

though it struggles with jellyfish hair

[wispy canopy]

https://huggingface.co/hakurei/waifu-diffusion-v1-4/resolve/main/vae/kl-f8-anime2.ckpt

[rich tartan]

ah this one ok

[wispy canopy]

you know rainbow hair is already a problem for MANY models 😄 jellyfish hair is... problematic?

[stable crane]

oh let me try the rainbow hair

[wispy canopy]

blueprints to the perfect waifu!

[stable crane]

it's nice

hm that one maybe looks too sketchy

[wispy canopy]

girl, jellyfish hair uhhh... that jellyfish though

[rich tartan]

i got the ckpt file, it acts like a checkpoint

[wispy canopy]

put it in vae folder 🙂

[rich tartan]

ahhh ok

[wispy canopy]

which model did you download, if you report a borked output, i'm pretty sure i did upload a wrong file AGAIN

[rich tartan]

this one

74900

[wispy canopy]

pruned fp16, lemme check 🙂

what's the hash?

[stable crane]

remember that guy who always uploaded pics like this? I wonder how he's doing

[wispy canopy]

i feel personally attacked by this

[rich tartan]

ok wait, i mixed them

[stable crane]

was it you??

[rich tartan]

dark_gemini is the one with the hash you also got

[wispy canopy]

no. but i do feel attacked because lots of my outputs look like that

[stable crane]

the subject matter or the artstyle?

lol

it was always these battle girls

[wispy canopy]

i've never made a darkalfa with hash 0fd1....

[stable crane]

I think this is mostly a prompt I got from him at that time

[rich tartan]

i guess thats dark_gemini

[wispy canopy]

back turned towards viewer

[supple raptor]

An "embedding" AKA: textual inversion, is a method of training the AI by feeding captioned images into, but rather than making any changes to the actual model the information that is learned is stored in a very small python file. This file is often referred to as an "embedding". There's a much more in depth explanation that I'm not going to get in to here. To use them, you simply put the file in your embeddings folder and the filename of all of your embeddings loaded by SD and appended to your model's text encoder so that you can simply activate them by using the filename (without the file extension) anywhere in either your positive or negative prompt.

Negative embeddings are the same thing, it's just the name we've given to embeddings that are specifically intended to be used in the negative prompt. Typically negative embeddings are intentionally trained with very poor quality images.

[lilac totem]

pls kiss

uwu

jk

[rich tartan]

Thank you for explaining me, now i got it

[wispy canopy]

you got these copy'paste ready? 😄

applauds nubby's attentivenes

[supple raptor]

I really should, but no. I type them every time.

[rich tartan]

this tastes indeed good

[lilac totem]

[wispy canopy]

too little steps

[lilac totem]

ye

[lilac totem]

20

[wispy canopy]

if that happens on 20 steps... there might be something wrong

[lilac totem]

i used unipc

lol

[rich tartan]

ngl but that looks like you fked up a drawing and made the paper to a ball and expand it back

i like it

[wispy canopy]

@lilac totem which model?

not all models like the same settings

[stable crane]

welp I better head out

[wispy canopy]

that late?

and thx @stable crane , seeing you generate that stuff on my model really gave my mental a boost ❤️

[stable crane]

nah I should just do something more productive 😂

[silk narwhal]

last 10 steps the model choked :/

[stable crane]

np, glad you enjoyed it! it's a nice model, I'll try to remember to post a review later, I should be posting more in general to the internet

[wispy canopy]

as i fucked up like every model upload -> is your model VAE ok?

[wispy canopy]

absolute upskirt 😄

[wispy canopy]

i see this, i get the urge to get you to prompt THE CURSED LYRICS {2-3$$ BREAK $$ This is for my slime boys out there 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!}

[supple raptor]

if weebs had been allowed to design the washington monument.

[silk narwhal]

is on default settings soo..idk

[wispy canopy]

is not my model... right? many model merges have a broken vae, make sure to select a separate vae in your settings

[silk narwhal]

am messing with latent coupling so ..guess that messes stuff up

[wispy canopy]

very much so

[supple raptor]

Yeah, latent couple can get some weird results at times.

[wispy canopy]

Even the SD image replies with what i should do with the cursed lyrics: NON! (DON'T DO IT)

[wispy canopy]

how many AND's are you using?

AND is a great idea in theory, but has MAJOR drawbacks (which why i stopped using them)

[rich tartan]

i'll head out from counting

[wispy canopy]

CURSED. LYRICS. DO NOT EVER USE

[supple raptor]

AND with latent couple is handled completely differently than AND without latent couple

[rich tartan]

you trained your model on furries too?

[wispy canopy]

it's a generic model. 😄

[rich tartan]

dam, how can so much fit into one 2gb model

[wispy canopy]

is it? i felt i had the same slowdown regular AND usage had...

[wispy canopy]

and nobody can even scratch the surface of what is truly possible in an 2gb model

[rich tartan]

seems like the key is optimizaion and size

[silk narwhal]

4 LC AND's

[wispy canopy]

don't use AND if you don't have to (my model, removed the 'AND')

[supple raptor]

The main problem with AND normally is that it causes images incoherency in images because it causes the model to try to put two things in the same place. Latent Couple resolves this by defining the area that each prompt will apply to. As far as speed, you're right and there's nothing latent couple can do to resolve that, because AND still results in the model effectively diffusing multiple images at the same time.

[rich tartan]

^bro is literally automatic1111

[rich tartan]

interesting

[wispy canopy]

AND prevents DPM adaptive from working at all. i stopped using it altogether... the idea is ok... it just doesn't work imho (exceptions for extensions)

DPM adaptive will NEVER resolve with AND

[rich tartan]

what about if i tell: thing a FUSION thing b? would it go the same way as and?

[wispy canopy]

AND is a special keyword

fusion is not, as far as i know

[rich tartan]

oh

[wispy canopy]

same as BREAK -> special keyword

[rich tartan]

wait what, for me it works

(it doesn't break down)

[wispy canopy]

for me it doesn't -> it'll got to 900 steps and simply won't resolve ever (because it keeps switching between 2 images)

[supple raptor]

Yeah, it's a fairly useless function on it's own, unless you just want to make ugly attrocities. I've never been a fan of DPM adaptive, not sure what causes it to fail when AND is used though.

[rich tartan]

hm cringe, i love adaptive

it gives good results, but takes long

[untold geyser]

how can i install the ia?

[supple raptor]

https://github.com/AUTOMATIC1111/stable-diffusion-webui

[rich tartan]

DPM Adaptive results

[wispy canopy]

i liked it too for a while 🙂 dpm++ 2m gets 95% as good results at 20 steps

and sorry, but to go for 150 steps to get 100/100 instead of 20 s teps for 95/100, that's too big of a tradeoff

and at 50 steps, UniPC just slaps dpm adaptive just around. not even a contest

[rich tartan]

wait you try to use 150 sampling steps?

[wispy canopy]

dpm adaptive regularly does (time your results)

[rich tartan]

hm, im confused

[supple raptor]

dpm adaptive ignores your set number of steps and just generates until it's "done"

[wispy canopy]

dpm adaptive uses as many steps needed -> minimum 90 steps usually

[rich tartan]

oh

that's the cause it takes ages

[wispy canopy]

so it's VERY expensive, computationally

[rich tartan]

^^

[wispy canopy]

so it does get good results, but that computational cost is just not worth it

when other samplers (any of the dpm++ or unipc) line just plain beat it at 30 steps

[rich tartan]

interesting, but i think the rough shading from dpm adaptive is unbeatable

[silk narwhal]

aallrighty then .. disabled LC

[wispy canopy]

uhh... 😮 WHAT HAPPENED?!

[rich tartan]

@wispy canopy everything except the characters got good, that was such failure, nearly got a masterpiece

[wispy canopy]

upscale should fix this

[rich tartan]

which one?

[wispy canopy]

any highres, x1.5

[rich tartan]

ah, i used 1.4x

kk i'll try out tmrw

[wispy canopy]

oof, denoising 0.8 is REALLY high

that will introduce artifacts for sure

[silk narwhal]

LC back on..with 20 steps instead of the 35 i use normally on DarkAlfa

[rich tartan]

can give sometimes good results ig

[rich tartan]

what would you use instead?

[wispy canopy]

0.55-0.58 for dpm2++ samplers, 0.4-0.5 for ddim

[supple raptor]

I'll stick with my 2M Karras, 2S a, and SDE go to's with occasional UniPC usage thrown in. BTW thanks UniPC for nearly making me unable to post this.

[rich tartan]

kk thank you

then gn guys, i'll head out

[silk narwhal]

yea..after 20 steps it chokes,at least on my gpu

35 steps

[wispy canopy]

@rich tartan dpm++ vs adaptive vs unipc

adaptive took more time than both unipc and dpm++ combined... that's why i claim it's not worth it

[wispy canopy]

i'm at 30 steps...

can u give me the prompt

full prompt

[silk narwhal]

1girl, green hair, ponytail, green eyes
AND sparkling river,
AND medieval town,
AND blue sky
AND 1girl, green hair, ponytail, green eyes,crying
Negative prompt: bad-artist-anime bad-artist badquality bad_prompt_version2 negative_hand-neg ng_deepnegative_v1_75t
Steps: 35, Sampler: Euler a, CFG scale: 5, Seed: 2543947178, Size: 512x840, Model hash: ceb28f2a01, Model: darkalfaEffortless_v11, VAE: clearvae_main, Clip skip: 2

Used embeddings: bad-artist-anime [53f1], bad-artist [2a38], bad_prompt_version2 [afea], negative_hand-neg [b740], ng_deepnegative_v1_75t [1a3e]

[wispy canopy]

NOOOOO

the embeddings!

😛

use my negatives

i think it's deepnegative that's fucking you over

and you can keep clip skip on 1, i've never had to drop it to 2

yeah, it's borked

[silk narwhal]

this is what happens at steps 21+

[wispy canopy]

sparkling river,
medieval town,
blue sky
1girl, green hair, ponytail, green eyes,crying
Negative prompt: (low quality:1.3), (worst quality:1.3), (lowres:1.3), (error:1.3), (fault:1.3), (undefined:1.3), (crude:1.3), jagged, spiky, ugly hands, extra arm, extra hand, split arm, missing finger, extra finger, three fingers, four fingers, six fingers, (disfigured:1.3), (unclear:1.3), (indistinct:1.3), merged fingers, (bad anatomy,:1.3) misplaced hand, misplaced foot,, cleavage, sex, nsfw, ecchi, slutty, leotard, (text:1.3), (signature:1.3), (title:1.3),
Steps: 35, Sampler: UniPC, CFG scale: 5, Seed: 2543947178, Size: 512x832, Model hash: d9538e0883, Model: darkalfav11-correctedvae, Version: v1.2.1```

for real. don't use AND unless you REALLY, but i really mean REALLY have to

BREAK is almost superior in every way, but in the end, they're ALL ways to get around the 75 token problem or multiple subject problem

[silk narwhal]

trying to learn this Latent Coupling extension and that adds AND to the prompt

[wispy canopy]

do you want your blue sky on the ground? -> if not, it doesn't need any specific latent coupling, just throw it in the main prompt

same for the sparkling river, or medieval town

does the medieval town only apply to the left side? -> if not, it does not need latent coupling

[rich tartan]

Dpm adaptive has something own i dunno, while uniPC also looks good ngl

[wispy canopy]

i knwo it might sound harsh, and latent coupling has uses, for certain (i've used it to great effect with duality prompts), but it's really not fit for 'generic' images, increasing the computational power required for 0 extra image output

[silk narwhal]

ok

[wispy canopy]

for the time i had to use for dpm adaptive, i have done 3 extra images on unipc, that's how much more time it takes for too little extra effect

[rich tartan]

Haha, that's the downside of it. But i only use it on images where only one person is or that needs much detail

[wispy canopy]

that's why i have unipc on 50 steps -> beats dpm adaptive hands down, in ANY contest, and is still faster

[rich tartan]

Dam i wish i could use it too

[silk narwhal]

aaannd Cuda Launch_Block failure O_O

[wispy canopy]

the idea behind dpm adaptive is very good. but the time for convergence is not

any dpm2++ will beat dpm adaptive at 30+ steps

(that's why dpm adaptive is the old generation)

there's a reason dpm2++ took over

they're simply faster.

[silk narwhal]

just wait for dpm3++++ adaptive

[wispy canopy]

can't wait!

if it can determistic resolve in 5 steps, i'm their bitch!

[rich tartan]

??

Inform me

[silk narwhal]

https://media.discordapp.net/stickers/980741427325698058.png?size=256&name=LaughCapAm

[rich tartan]

Jk?

[subtle crown]

i always use ++ 2m k

[wispy canopy]

i used to, too. until i realised it can't do transparancy realy well (unipc can)

[supple raptor]

Yep, that's what I always default to. If it doesn't work well with a prompt, then I try some of the others.

[wispy canopy]

i like the aesthestics of 2m karras more, but transparancy man...

[rich tartan]

Ok, i'll really go sleeping now, gn

[wispy canopy]

gn

[silk narwhal]

https://media.discordapp.net/stickers/968000868244946965.png?size=256&name=MarchNap

[true latch]

[wispy canopy]

new user? 😮

[silk narwhal]

3 days old 😛

[wispy canopy]

that's new

wonder how long they will stick around 😄

[supple raptor]

8 day old account with a very sus username (not the display name)

[wispy canopy]

autoban just cause? 😄

[supple raptor]

https://tenor.com/view/ban-hammer-gif-19698184

[wispy canopy]

... why did you have to remind me of that series...

[elder smelt]

Hi, is the bot down?

[gloomy scroll]

Bot?

[supple raptor]

TBH I'm not even sure what it's from, just picked a gif.

[wispy canopy]

sword art online

[gloomy scroll]

Now that I think of it I have still couple seasons to watch for SAO

[supple raptor]

Ah, that explains it. I watched one of the movies quite a while ago, but that's it.

[wispy canopy]

keep it that way, unless you are mentally 5yo...

it's a mary sue powerwank fest all the way, with ... little redeeming factors

[supple raptor]

#1047610792226340935

[gloomy scroll]

Can I finish it if I'm missing few screws?

[wispy canopy]

sure you can. just skip skip skip skip skip skip until something relevant to the plopt happens. now you skipped 4 seasons worth and missed nothing XD

the art and music direction are gorgeous, don't get me wrong

but the story couldn't be a more bland boring powerfantasy wankfest even if it tried

... the best part of the series is when the protagonist was a fckn vegetable... that goes to show how bad the series just is...

[wispy canopy]

you run this through a pixelizer, and that's snes rpg feelings right there ❤️

[wispy canopy]

thx for giving me nostalgia vibes ❤️

[supple raptor]

Haha no problem. I've been having a bit of an itch for (slightly less) retro gaming lately. All the Tears of the Kingdom talk makes me want to play through Ocarina of Time again.

[wispy canopy]

i really need to purchase chained echoes and give it a go

... as with crosscode

[supple raptor]

[wispy canopy]

that parkbench has some serious planks going on 😄

which model is that, it seems to have some logics in there i... uhh... need to borrow

[supple raptor]

Long park bench so that even the tall hoboes can sleep on it.

[desert arch]

Man coffee messed me up real bad

[supple raptor]

aurora with my super secret Lora mixture that definitely isn't in the PNG info of like that last 50+ images I've posted XD

[trim current]

That reminds me to install that on the deck next week

[wispy canopy]

ah. lora mixture. not native model 😦. thx for clarifying

[stable crane]

cursed

[supple raptor]

Her friend left her

[wispy canopy]

whatever you do. DO NOT EVER PROMPT THAT ON ALFAHENTAI, especially with no NSFW in neg prompt 😄. unless you want acid in your eyes

[trim current]

TROM THR NEF!!

I love that movie!

[wispy canopy]

got milkshake though

[stable crane]

I'm getting some wild results

lewd milkshakes

[trim current]

Looks like a acai berry

[trim current]

OwO

Imposter milkshake

[desert arch]

Oh gm btw

[trim current]

Gm

[desert arch]

Can't sleep the whole night

[supple raptor]

@wispy canopy here's the same image without any LoRAs. Would probably look pretty good with an upscale.

[stable crane]

nice hat in your pfp

[desert arch]

My favorite hat

[wispy canopy]

in aurora you said? imma go see which blocks i gotta plunder for the loot (not today though)

[desert arch]

The bench keeps getting longer

[supple raptor]

Yep. I'm using AuroraONE_F16 specifically.

[wispy canopy]

auroraone?

[supple raptor]

Now even taller hoboes can sleep on it comfortably. Or 2 average sized hoboes.

[supple raptor]

That's the full name of Aurora

[wispy canopy]

not regular aurora?

ah, in that case, i already got it.

it seemed extremely bad with natural language prompting, so it seems it has some concepts which i do not yet have in my model. i just want the conceptual knowledge, i don't care about the art

[stable crane]

it's so good w/ bug related things

[wispy canopy]

thicc mommy bug ❤️

[supple raptor]

Why not just merge then transplant the text encoder over from the current version of your model onto the version merged with Aurora?

[wispy canopy]

it's not just the text encoder