🍥｜anime

[delicate vessel]

also the 2nd model drew a good rough sketch, hoping it doesnt mess up in the upscaling part

the vibrance tho

lol

ngl the environment is good, geometric scifi structures, but damn the saturation

[native halo]

u could try with another vae for less saturated colors

[delicate vessel]

definitely vae problem

gotta look up some vaes now

[untold glacier]

You can use Nubby's:
https://huggingface.co/nubby/kl-f8-anime2-blessed

[untold glacier]

Also, you're reminding me of some old gens I did in April.

[native halo]

if u want real darkness

[cyan burrow]

more coming

[delicate vessel]

im trying to mimic the artstyle of my inspiration, dk how tf does he/she manage to generate these

insane model

perfect colors and artstyle

[native halo]

looks like meina

[delicate vessel]

he/she even managed to get the weapons on point

[cyan burrow]

Here's the story for my one-shot "In the year of 2099, mining operations had been ongoing for a decade when they strike the deep tunnels. Containing an ancient energy force, dark magenta and pitch black living swirls and movements, it's alive. Mere days later, new years eve of 2100. A great cataclysm created a lunar wide earthquake, opening up a giant rupture, a crack covering the entire vertical surface around the entire moon.

Global natural disasters befall earth and strange creatures rain down on earth in what looks like a natural disaster. A war breaks out between mankind, the demons and each other as the stars of the galaxy are dying and galaxies beyond ours are disappearing. Breaking all the laws of physics. The universe around earth is dying and the war is lost and everyone's memories from before 2100 are fading, until the only memories remaining are of after the darkfall. The war and the darkening of the universe.

The remnants of mankind are spread across the world, as a lovecraftian entity begin devouring the earth. In the final moments of earth, this is the story of people who've only remembered despair and war. Fighting the inevitable. Until their dying breath."

Is that anime/manga enough?

[native halo]

it does when it comes to tentacles

[cyan burrow]

Oh come on...We all know there are very very persistent themes in manga

Like Isekai

or light novels

so it doesnt dictate, it influences it for sure

[delicate vessel]

idk but the generations are insane

[cyan burrow]

If I can only generate the panels nad characters and scenes

I would need to create characters through prompts that I can then make into a lora

so each character can remain consistent

[untold glacier]

Getting somewhat close with Eface's new SDXL model. Trying to up the chaos and retain the cohesion currently.

[native halo]

you could also ask GPT to write the story

[cyan burrow]

I've written a lot of story and world building

What I've written are of the novel though

but I can write a one-shot

prequel

[delicate vessel]

auto1111

[delicate vessel]

all my generations with weapons have levitating weapons, or just the hilt, never the complete wepon on hand

[native halo]

depends on the model

[untold glacier]

It's going to be difficult to find a 1.x model that'll be able to hold a weapon properly.

[native halo]

you can also use a weapon holding lora or controlnet

[delicate vessel]

i tell sd to draw a simple katana, it draws a kukri

[delicate vessel]

i found many for guns but not melees

mostly blazing drive, but now imma try other ones

[native halo]

try with cute mix,camelia mix,darksushi

[delicate vessel]

i just downloaded cute mix

i have cetus mix in queue

[native halo]

yea cetus created this but with the bow lora

[cyan burrow]

why are my SDXL generations like this?

[nova remnant]

prolly cause of vae

[stable crane]

remove vae

[nova remnant]

use this vae for sdxl https://huggingface.co/stabilityai/sdxl-vae/blob/main/sdxl_vae.safetensors

[cyan burrow]

can I switch to sdxl vae?

yeah I have that

thank you

[stable crane]

LCM, 17s

this would take 3min in A1111

though I don't think I've quite hit the same crispness

[nova remnant]

what's LCM?

[stable crane]

https://civitai.com/models/195519/lcm-lora-weights-stable-diffusion-acceleration-module?modelVersionId=219782

[untold glacier]

Yeah, you're definitely sacrificing quality for speed with it. The finer details on a lot of things are lost with my gens when I use it.

[stable crane]

hm non-LCM one, 47s

LCM is probably just best for this: https://twitter.com/krea_ai/status/1723067313392320607
you use a little paint program to img2img fast, then upscale with non-LCM

[cyan burrow]

yes that did it, thank you

[nova remnant]

which mod are you using?

[cyan burrow]

who? me?

[nova remnant]

yeah

[cyan burrow]

SDXL 1.0 base with 0.9 vae

baked into it

[nova remnant]

ahh damn, that image came out pretty good

[cyan burrow]

aye

[nova remnant]

thought you were using a trained mod

[stable crane]

what's a trained mod?

[cyan burrow]

nah, I have one public for 1.5'

he means model I think

[nova remnant]

community models that are trained

[stable crane]

oh model

mod = modification

[native halo]

my hand hurts

[nova remnant]

the moon is calling her

[untold glacier]

Here's the closest I can get in an hour.

[stable crane]

yeah I'm thinking SDXL is good

[iron star]

looks cg and very ai like all of the sdxl models

[stable crane]

it needs a stylizing lora I think

[untold glacier]

You know, that's the second time I've heard this today: "very ai"
What does that mean to you?

[iron star]

it means that it doesnt have the 2d effect that 1.5 have

also the eyes are also not anime like

[untold glacier]

Ah, so you don't mean "very ai" or not ai, just that the style is different than it should be for the type of anime artwork that you're looking for.

I get it.

[iron star]

yea who can forget that very awesome anime that came out in 2000s with that style

[stable crane]

the eyes are my biggest objection but the rest looks anime

maybe some is fixable with prompt

[delicate vessel]

cutemix is really good

[nova remnant]

sdxl but something is off about the image

[untold glacier]

Advent Children?

[native halo]

plastic skin

[nova remnant]

i dont want future to be lifeless synthetic

[stable crane]

how about this one

[iron star]

if you want a corporate lifeless style you can stick with that model

[nova remnant]

i can tell its not right

not the right anime you can expect from sd 1.5

[stable crane]

I think you just dislike the style?

but if someone were to train a good lora on this

there's gotta be at least one out there

[nova remnant]

not the style, i dislike sdxl lol

[stable crane]

more cartoony

[iron star]

yea its not the style it just doesnt have the nai base

[nova remnant]

feed the same prompt into sd 1.5 and you will get sweeter output

[stable crane]

I'll try sd 1.5 now with comfy

should be interesting

[nova remnant]

:p

[untold glacier]

Add this in front of your SDXL prompt:
(Style: "Anime"). (2d art. key art, flat colors. cell shading:1.2) Simple anime illustration of
Neg
disney. realism. disfigured. (photo, photorealistic. 3d:1.3). fat face. chibi. extremely detailed

[stable crane]

a bit diff

[native halo]

[stable crane]

gonna try this anime tuned sdxl model

[untold glacier]

Better in my eyes, that's for sure. You can also try adding "lineart" in between flat colors and cell shading.

[nova remnant]

(Style: "Anime"). (2d art. key art, flat colors. cell shading:1.2) Simple anime illustration of 2girls, flowy hair, detailed face, choker, hair ribbon, jewelry, .... and im using dpm++ 3m sde karras

[untold glacier]

And the negative prompt?

[nova remnant]

just some generic ones

[untold glacier]

Use the one I gave if you want a more traditional 2d anime look.

#🍥｜anime message

[stable crane]

other model, still meh

[nova remnant]

whats that style: "anime" btw?

why the quotation?

[iron star]

its "anime"

famous artist named anime

[untold glacier]

My understanding is that it places further emphasis on anime within clip, without actually increasing the emphasis.

[nova remnant]

oh .. interesting didnt know about " "

[untold glacier]

It's what they recommend instead of increasing things to 1.3+ of ()

[cyan burrow]

I thought it just separated it from the surrounding tags to make it its own entity, to be modified by the surrounding text

[nova remnant]

this is what i got with that prompt including neg

[iron star]

i remember this anime

[untold glacier]

Nice

[nova remnant]

but but ....

thats not 2d feel

2.5d feel

[cyan burrow]

argh, 500mbps internet is too slow to get this many models fast...I need to upgrade to 1000mbps

[iron star]

it is you just gotta understand 2d anime

[nova remnant]

i mean 2.5d

i get 2d is simple flat

[iron star]

just understand

https://tenor.com/view/the-universe-tim-and-eric-mind-blown-mind-blown-meme-mind-explosion-mind-explosion-meme-gif-18002878

[untold glacier]

I dunno. That's pretty flat to my eyes.

Cell shaded and all.

[nova remnant]

yes it is

but i was hoping for 2.5d

[untold glacier]

Ohhhh

Sorry. I prompt for superflat, lol.

[cyan burrow]

I would call this 2.5d

[stable crane]

yeah

[nova remnant]

yeah that

like you can feel the depth

[cyan burrow]

and these

[nova remnant]

this

[untold glacier]

I would take out "flat colors" and "cell shading". That'll likely put you in the 2.5 range. And maaaybe taking out "3d" from the negs, but only if you're still getting too flat of a result.

[cyan burrow]

and Frejora says hi

[native halo]

this is the style that u want

[cyan burrow]

moe moe

[nova remnant]

im gonna try this simple prompt ((anime style)), 2girls, flowy hair, detailed face, choker, hair ribbon, jewelry with purely anime based sdxl model

[cyan burrow]

I'm gonna see what I can do with SDXL merging, if my 8gb 3070 ti can deal.

Then make my first sdxl model

[nova remnant]

dont think i need to prompt for anime tho, this model is trained to do anime

[native halo]

another 2d style this is prob what u referring to

[cyan burrow]

https://civitai.com/models/6209/dark-gemini-v3 this here is my model I made much earlier this year

for 1.5

[nova remnant]

see the image? you can see the curvitures

i like that depth

[cyan burrow]

well check out the album/showcase

like these

[iron star]

that looks bad

[nova remnant]

yeah im looking at them

[nova remnant]

i know

[cyan burrow]

the reviews have some good gens too

there's a fp16 model at 3.8gb or so if you want to try it

[nova remnant]

im only trying out a few sdxl

i have bunch of good sd 1.5 mods

[cyan burrow]

okay

I'll have my SDXL out eventually'

[nova remnant]

cool

[native halo]

the only decent sdxl anime model that i have seen is holoanimexl u tried that one yet?

[nova remnant]

the kind of anime i like for example https://civitai.com/user/torqx/posts

[cyan burrow]

many of those are 3d

[nova remnant]

some are

ranging from 2d 2.5d 3d

[cyan burrow]

wtf does IndexError: list index out of range mean

[nova remnant]

aspect ratio?

[cyan burrow]

I'm only trying to load the model

[nova remnant]

not sure then

i've had memory issue before

[cyan burrow]

yeah it's somehow hogging all memory

[native halo]

its ok sdxl is better it just runs like poop,has less styles,less loras and takes longer to render but its better

[cyan burrow]

ok it worked to load it on main

I'm gonna have to see if it's fp32

[nova remnant]

i've tried to cope with sdxl, not possible

[cyan burrow]

if toolkit can only manage it

I mean I can do most stuff on 1.5 just as well if not better

[delicate vessel]

i have decided to main cutemix

[nova remnant]

actually faster, and better

[native halo]

it looks beautiful

[cyan burrow]

trying to FP16 the base model too

thank god for toolkit

[untold glacier]

Sure, that's because you know it well and there are models out there to cater to everything. It takes time to learn a new system, both how to prompt for it properly and the time that it takes to create new models for specific purposes. One can do a lot with the current models out there, and just like the early days of 1.x, there's tons of creativity to be had if you can find the magic words.

The base concepts and natural language in SDXL is better though, hands down. And it's the future of SD.

[cyan burrow]

no it's also cause my GPU can handle it well

[iron star]

its been 6 months already,still coping?

[untold glacier]

When I read "better" I interpreted it to mean quality-wise. If you're talking hardware limitations, then that's outside of the bounds of new innovative software.

[nova remnant]

i keep coming back to it now and then and end up deleting the files

[iron star]

its ok 2.1 is also the future

just wait it will get better too

[nova remnant]

sd has to come up with a newer version that can actually do better than sd 1.5

[iron star]

2.1 the king hands down

[nova remnant]

no way

i think 1.5 is the most popular

[iron star]

well its because its cheap to train in 1.5 if it was cheap to train in sdxl we would have better models

[nova remnant]

are you sure about that?

sdxl engine just cant do anime well

[iron star]

just get an nvidia with 48 or 80gb of vram and train it with 50 million images 🙃

[cyan burrow]

Sure, but the point is. All I can most efficiently train is 1.5 based. The SDXL generative can be pretty slow but I'm sure I can adjust some settings

time to test it out now that I've fp16d stuff

[nova remnant]

Hot bath anyone

[cyan burrow]

Eeh

RuntimeError: Error(s) in loading state_dict for TimestepEmbedSequential: Missing key(s) in state_dict: "1.norm.weight", "1.norm.bias", "1.proj_in.weight", "1.proj_in.bias", "1.transformer_blocks.0.attn1.to_q.weight", "1.transformer_blocks.0.attn1.to_k.weight", "1.transformer_blocks.0.attn1.to_v.weight", "1.transformer_blocks.0.attn1.to_out.0.weight", "1.transformer_blocks.0.attn1.to_out.0.bias", "1.transformer_blocks.0.ff.net.0.proj.weight", "1.transformer_blocks.0.ff.net.0.proj.bias", "1.transformer_blocks.0.ff.net.2.weight", "1.transformer_blocks.0.ff.net.2.bias", "1.transformer_blocks.0.attn2.to_q.weight", "1.transformer_blocks.0.attn2.to_k.weight", "1.transformer_blocks.0.attn2.to_v.weight", "1.transformer_blocks.0.attn2.to_out.0.weight", "1.transformer_blocks.0.attn2.to_out.0.bias", "1.transformer_blocks.0.norm1.weight", "1.transformer_blocks.0.norm1.bias", "1.transformer_blocks.0.norm2.weight", "1.transformer_blocks.0.norm2.bias", "1.transformer_blocks.0.norm3.weight", "1.transformer_blocks.0.norm3.bias", "1.proj_out.weight", "1.proj_out.bias".

and a bunch of unknown keys

I think maybe this model might just be bugged

[native halo]

try to gen this style in sdxl real challenge

[cyan burrow]

the one I'm trying to load

[nova remnant]

Is this iori

[nova remnant]

i can't now, i deleted the files already and back to sd 1.5

[cyan burrow]

now it worked with another model

no

no it didn't

ok moving to comfyui

[native halo]

its barack obama 👍

[hidden lotus]

Can any comfyui experts help me in #🤝｜tech-support? I'm getting really desperate.

[cyan burrow]

hmmm how do I get a node for comfyui that has the full block chain and not just in mid out like the default?

like does one exist?

[native halo]

i cant enter the pool i have my bodysuit on

[cyan burrow]

bah everyone recommended me comfyui to use for merging and yet it has no good tool for that

[nova remnant]

@native halo a bodysuit is best to protect from the heat

[native halo]

feelin kinda hot rn

[stable crane]

I think I finally understand what "highres steps" is now

after implementing it in comfyui

but now I'm seeing an easy path to:

• generate lowres pose in sdxl
• upscale it in 1.5

[nova remnant]

Quality of output

DPM++ 3M SDE Karras generally produces higher quality images than DPM++ 2M Karras, especially at higher step counts. This is because DPM++ 3M SDE Karras is better at capturing fine details and textures.

For example, DPM++ 3M SDE Karras can produce more realistic skin textures, sharper hair, and more detailed facial features. It can also produce more realistic images of objects with complex textures, such as wood, fabric, and metal.

[stable crane]

I see comfy is missing some samplers

[nova remnant]

DPM++ 3M SDE Karras bit too realistic for anime tho

[nova remnant]

idk looks fine to me, i'll take 2

[native halo]

its ok ill take care of these 2 ladies,anime or not

[untold glacier]

3M SDE can have major artifact issues when using with an anime model. From experience I don't recommend it, though I love the sampler itself.

[nova remnant]

i see, im trying them out for experiments

[cyan burrow]

I think it's funny people were recommending comfyui for merging SDXL all day long

and yet...

there is no real block merge way

[native halo]

just use auto supermerger

[cyan burrow]

there's a reason comfy was recommended, because it doesnt use as much ram etc

on auto, I can't merge, my ram is overwhelmed

I can merge 1.5, but not sdxl and yes I am trying to merge fp16

[native halo]

😔

[cyan burrow]

it shouldn't overwhelmed my ram (32gb ddr4) but it does

[nova remnant]

@native halo

[native halo]

i dont mind burnin my hands

[stable crane]

same prompt and loras, I can't quite get the same character to my gens in comfy

[nova remnant]

do any of your prompts have weight like (mouse ears:1.2)

[untold glacier]

If memory serves, there's something you have to turn on to make like Auto's in Comfy. I can't remember exactly what it is, since I haven't done it myself.

[stable crane]

no

well actually maybe

[cyan burrow]

@untold glacier I've been looking for a proper model block merger for comfyui for the past hour and a half but nobody knows one...Despite recommending I use ComfyUI to block merge SDXL

[untold glacier]

I don't know anything about 'block merging', sorry.

[cyan burrow]

damn, alright

[untold glacier]

I don't even know what that is.

[nova remnant]

@native halo why the nose emote

[native halo]

i love the smell of latex in the mornin

[untold glacier]

Anybody want a drink?

[stable crane]

there's a serious lack of XL anime loras

[cyan burrow]

might be this? https://github.com/YinBailiang/MergeBlockWeighted_fo_ComfyUI

[cyan burrow]

thicc angel

or hawkgirl

[untold glacier]

ᴀɪ•ɪs•ᴏᴜʀ•ᴇɴᴅ

[cyan burrow]

going through image folders of gens I made before June

before I went into treatments etc

though I wasn't very active in May since I had surgery in April

some of them need adetailer, but that didnt exist then

still though, some of it is stylization :p

supercat fires lazer with exploding eye and ends up destroying the world

covered in blood and ash he sits contemplating his next move...

he decided to become an Esper

the hippos watched alien and aliens for the first time and decided it was the most appropriate way to evolve

[native halo]

i would support them in every possible way

[cyan burrow]

simp

[cyan burrow]

be a chat like this guy

[native halo]

i prefer this one

[cyan burrow]

but they be waiting...ready for going to sleep

[keen hollow]

evenin'

[cyan burrow]

you really gonna be that guy? Can you really snore enough to impress these ladies?

hello there @keen hollow

[keen hollow]

yo yo

[native halo]

its ok when all else fails i can always be like this guy

[keen hollow]

i li ke the complexity of this background

trying to find a model that does that right now actually

this reminds me of Rev

[native halo]

thats darksushi

[nova remnant]

darksushi is great for complexity

this one i used is awpainting

[keen hollow]

I'm thinking about maybe doing a superhero series to make a 2024 calendar

probably way ahead of myself 😛

could do it with dalle, but those don't have the res

[nova remnant]

sounds productive and useful

you could even put them on sales

[keen hollow]

stuff like this

nah, not for sale, just to use and enjoy haha

[nova remnant]

like actual paper calendar

[keen hollow]

yeah

but my prompt-fu sucks 😄

[native halo]

make a sexy calendary print it then sell it

[nova remnant]

i think i might think on it myself, fun idea

[keen hollow]

i tried to do one for 2023 but ran outta gas

[nova remnant]

you dont have to rush it lol, you have plently of time

[keen hollow]

yeah just gotta get to this level of output but not with dalle. 😄

[nova remnant]

think you can easily do them with 3D models

and why not even 2.5D models

[keen hollow]

well.. .maybe

[nova remnant]

[nova remnant]

cool texture

[keen hollow]

getting a lot of this crap

lol

[nova remnant]

@nova remnant my prompt was super cool background

[nova remnant]

are you using base model and w/o adetailer?

[native halo]

dont care about bg

[nova remnant]

wait no super complex

lmao

[keen hollow]

yes these are just low-res tests

ignore the mangles low-res issues

I just mean the composition

[nova remnant]

puts the cleevage in the prompt

[nova remnant]

lol ok, i was wondering about that

getting complexity to look symmetrical and meaningful can be hard

also the more detail in prompts the less focus the character gets

[keen hollow]

sdxl can get close with the art style (artist) I was using...

dalle still gets me better with less effort..

[nova remnant]

if sd 1.5 had sdxl context that could have been far superior than dalle

[cyan burrow]

merge block in comfyui

found at last

[nova remnant]

@native halo Do you like this background ?

[keen hollow]

a lot of the 1.5 models have context for supergirl as a character, but not necessarily that artist style. but, I could u se another artist... just need to find a way to get to the right style 🙂

[native halo]

i like big backgrounds

[nova remnant]

dancing keyword doesn't play out too good

getting weird outputs consistently

[keen hollow]

closer...

[nova remnant]

how are you planning for the actual images? saving these seeds?

[keen hollow]

right now I'm just trying to figure out a style tag I could apply across scenes/characters

dalle gets me stuff much easier... check out this guy

i'm confident I can get there in SD, but gotta find the right styles

can't keep using local sdxl either... too slow, driving me nuts 😄

[native halo]

whats the name of the artist

[keen hollow]

the prompt is literally just supergirl, red skirt, comic style, desolate city, art by J.C. Leyendecker

[cyan burrow]

dall-e is also pay only

and they outsource their "improvements" to Nigeria, Kenya etc

cause its cheap

[haughty kindle]

heyy!!

[nova remnant]

think this one works on both sd 1.5 and xl https://civitai.com/models/132426?modelVersionId=155630

[keen hollow]

ill take a gander

i may pop on the subreddit and see if I can find someone that' sshared something similar already

[nova remnant]

full body and full body portrait ... the latter gives character a closer focus

[supple raptor]

it just adds text to your prompt so there's no reason it couldn't work with any model. I suspect it isn't particularly effective on most anime models though.

[nova remnant]

yeah logical, they are both text based except it also craetes button for a1111 which doesn't affect sd engine

[native halo]

best thing u can do is find an artist lora

[supple raptor]

Yeah, it definitely looks like a convenient way to use it. If I used XL models I would use it.

[native halo]

idk about comics so cant tell u which artists are similar to the one u posted😔

[nova remnant]

i extracted some interesting ones and put them on my style menu for easy click and select... ddnt want to use waste space in a111 with buttons

[keen hollow]

Greg Capullo would be good, but it probably doesn't even need to be a comic artist

[supple raptor]

it'd be pretty easy to turn this into a wildcard list too https://github.com/ahgsql/StyleSelectorXL/blob/main/sdxl_styles.json

[keen hollow]

I mean, JC Leyendecker wasn't 😛

[native halo]

there was another artist that i tried to find on civitai Etienne Louis Boullee but looks like no one made a lora about him 😔

dead game

something like this 😔

[native halo]

yas queen 👑 💅 💅 💅

[supple raptor]

I can see why it hasn't been trained. SD would likely struggle with.

[native halo]

he used to do massive architecture and as @nova remnant knows i like it massive

[cyan burrow]

let's see if comfyui can do the job

[nova remnant]

cinematic indoor setting, dimly lit room, soft light filtering through curtains. 1girl dressed in a flowing red dress, short blonde hair, detailed face.

easy technique to get down to thigh shot is to mention bare legs

[keen hollow]

ugh i give up

[nova remnant]

are you trying to compile the calendar in one siting?

[keen hollow]

no i'm still looking to get the style

[nova remnant]

those list would have it

those = that

[keen hollow]

naw

[nova remnant]

did you look under marvel styles?

[keen hollow]

i probably need a model and lora combo, with some really good prompting

[nova remnant]

yeah that too

there are some good mods that does marvel heros pretty well

but hands suck

[keen hollow]

frustring because I can get stuff like this with minimal effort in dalle. 😛

well... that didn't work...

try that again

[nova remnant]

dalle doesnt' give you as much control

[keen hollow]

maybe I can use these and some img2img or upscaling or something

[nova remnant]

can you share the prompt i want to see what i can do with it

you dont have to if that's part of your dev plan

[keen hollow]

pff no problem, I always share

<character>, comic style, desolate city, art by J.C. Leyendecker

that's all ya need in dalle 🙂

[nova remnant]

wonder woman, comic style, desolate city, art by J.C. Leyendecker

just trying out on random mod

[keen hollow]

with SD, I'm trying to hone in, but not quite getting there...

[keen hollow]

much different prompt of course

[nova remnant]

lacking context a bit

[native halo]

u prob want that Brian Bolland style

[untold glacier]

Just put the prompt into an LLM and have it expound on it.

[native halo]

this was brian style

u probably can get it if u download the pencil sketch lora

[keen hollow]

dalle is just nuts... i can take this straight as is ... never seen SD do a decent wolvie 🙂

[keen hollow]

this isn't bad.. SD? this could be usable

[native halo]

naw that just Brian own hands,tried to find a lora but there isnt one

[keen hollow]

ah

david finch, greg capullo, heck even a decent jim lee might work

[native halo]

heres one similar https://civitai.com/models/8949/jim-lee-dc-comics-marvel-style-lora

but not the same

[keen hollow]

got that one

didn't have much luck

[native halo]

try to combine that one with this https://civitai.com/models/55047/pencil-sketch

[keen hollow]

i need a background though

still not getting much closer in SD yet

[nova remnant]

how do you know SD knows all the artists?

[keen hollow]

there's lists out there with artists in the model

i dunno which comic artists are in there, but it could probably be many other artists frankly

[native halo]

try to add this at end of your prompt (destroyed city,ruins,debris:1.5)

[nova remnant]

wonder woman, desolate city, city ruins, debris, art by Leinil Francis Yu

[keen hollow]

how'd that do?

[keen hollow]

the lighting in that model with the reflections is neat

you got a bad case of samey-face i think though 🙂

[nova remnant]

https://civitai.com/models/84476/awpainting?modelVersionId=156202

looks like she is visiting palestine

[native halo]

😔

[keen hollow]

that's that 'uncanny valley' is it?

trying to lean a little more into the illustration side

[nova remnant]

..........art by Mark Brooks

[native halo]

gal gadot really loves that place

[nova remnant]

too much debris there

[nova remnant]

im getting her outta there

[keen hollow]

face changed, nice. coherence on the character is rough

[native halo]

there was a comic checkpoint but looks like the author never updated it https://huggingface.co/ogkalu/Comic-Diffusion

[keen hollow]

yeah it's hot garbage

[native halo]

😔

ok you should try another thing take the dalle images then outpaint to see if SD can pick up and copy the style

[keen hollow]

yeah i might lol

gotta find a model that melds with it

[nova remnant]

she fought enough crime, she is on a coffee break

[native halo]

outpaint + ip adapter + reference idk if your gpu can handle all of that tho

[keen hollow]

yeah probably can.. 12 gb

[keen hollow]

dang ... cap is mad haha

[native halo]

now i know why they call him cap america

this one came out similar still needs adetailer for face but better than base model

[keen hollow]

still has a ways to go

superman .. I can get stuff like this...

[nova remnant]

cityscape, daylight, far view, dynamic angle.

[nova remnant]

Wonder Woman standing tall on a cliff overlooking a vast landscape. Her cape billows in the wind., far view, dynamic angle.

[nova remnant]

Most of the time, heterochromia doesn't cause any problems to your eye health and it does not affect your vision either. It's often just another characteristic passed down to your genes through your parents.

[untold glacier]

Unsubscribe from Eye Facts

[pure oyster]

Anyone using an AMD card that is able to use high res fix? Apparently 24gb of vram is too low.

[native halo]

it sucks on windows u have to install it on linux because drivers work better there

[pure oyster]

Yuk, I don't feel like making vms

[supple raptor]

you don't want vm. you want dualboot. which amd card do you have?

[pure oyster]

7900 xtx

[supple raptor]

Yeah, you're going to want to run linux for SD.

You'll get probably 2x performance or better on linux with fewer problems.

[pure oyster]

I was able to make high res fix work using certain augs, but it is still only giving me 1.6 iterations per sec

still 2.2x faster than my 1060 6gb from my last machine

[supple raptor]

Even on Linux, you'd be better off with Nvidia for ML stuff because AMD ML support is bad in general on their consumer products, but it's infinitely better on Linux than it is in Windows.

You're going to want to look into using ROCM which is something that you're only realistically going to be able to do on Linux.

[native halo]

how many it/s u get on your current sd installation?

[pure oyster]

oh my I apparently do not know how to crop

[native halo]

thats so bad,on linux u can get like 20it per second so a 40 steps image would take like 2secs to complete

[pure oyster]

I was expecting "non-optimal speed" knowing it was optimized for nvidia cards, but wasn't expecting "no speed". Nonetheless, it looks like I'm going to need to find guides on dual booting.

[brittle jay]

on what gpu?

[native halo]

7900 xtx

[brittle jay]

it has 24gb gddr6

damn

on what kinda image do you get 40it/sec

there are a ton of factors

sampler, steps, resolution

[native halo]

512*768

1.5

[brittle jay]

what 1.5

hires fix?

what sampler? how many steps

[wispy canopy]

good prompt 😄

[brittle jay]

actually so good

[supple raptor]

protip: don't try to directly compare CTD's models with other models. Her models really only look good with her neg prompt and other models usually don't look as good with her neg prompt.

[wispy canopy]

cute eyes -> mission success

ok, i promise to be nice from now on

[zinc flax]

What will she cook

[wispy canopy]

not this chicken!

[zinc flax]

Attack of the Chicken

[wispy canopy]

will he steal your hear, or your gold?

[dull jackal]

That's adorable

[dull jackal]

Both

[wispy canopy]

i feel like a yorkshire terrier has a lot more charisma though 😄

[dull jackal]

I'm def seeing more of that vibe, more trustworthy there, hahaha!

[wispy canopy]

pretty sure he's the charmer of the party xD

[dull jackal]

What's his role? Other than clearly getting everyone out of any haphazard situation, such as why his friend is stealing from the King's treasury, by smoothtalking?

[wispy canopy]

the boss!

[dull jackal]

I can approve!

[wispy canopy]

and obviously, it's to steal from this guy!

[dull jackal]

Suddenly, all I can think of is Beastars.

That's beautiful.

[wispy canopy]

i really love my model mix 😮

it does so well whatever the F i throw at it

[dull jackal]

Do tell!

I love the way the mane flows

[wispy canopy]

https://civitai.com/models/148871

[dull jackal]

Lot of great 2d/3d presence

[wispy canopy]

but it's "just a generic" model, which happens to be quite robust in image quality, and has pretty good image comprehension too

anime, An illustration of a confident-looking person wearing a crown made of flowers, sitting on a throne adorned with rainbow flags. They have a smug grin on their face as they hold up a trophy with their dominant hand, while flipping off the camera with their other hand. Behind them, there is a backdrop of a majestic mountain range, implying that they rule over a powerful kingdom. The overall tone is playful and flirty, with a touch of arrogance.
it got so close (the hands, as always)

[dull jackal]

That's really pretty! Love the color and background! Beautiful, and hilarious XD

[wispy canopy]

tho this is probably the thing the model does best. It's not an anime speicfic model so i feel quality is a bit hampered when trying ot make it do anime, nor does it do really abstract stuff well either

[wispy canopy]

wow, very nice style ❤️

[cyan burrow]

thanks, been trying to learn comfyui and testing out merge between 2 sdxl models, without merging yet

[dull jackal]

Nice composition

[cyan burrow]

gonna see if I can finally get them into a shootout

[wispy canopy]

felt when it came out, just needs a caption and can be used as an ad, just like that

[dull jackal]

Love these, too

Lot of thick ink kind of style going on

with clothing

[cyan burrow]

yeh, trying different combos of settings and such, got lucky

[dull jackal]

I haven't updated this in awhile--there's still a lot I have yet to put in, but feel free to check out this: https://docs.google.com/document/d/1BxdWqfBJ3QPggHnBCBx3QIkUpHnBWShd_s3zEt2dTLM/edit#heading=h.po8dv7lcq48s

[dull jackal]

Well, I'd love to see more!

[cyan burrow]

looks like this atm 😛

[wispy canopy]

@dull jackal you made this?

[dull jackal]

Oh, yeah.

[wispy canopy]

i think i can make it really proper with regional prompting

want me to give it a try?

[dull jackal]

By all means, read through it all and do as you like

[glacial bane]

#1100170365604483202 isn't responding to my prompt

Anyhow how yall been doing

[wispy canopy]

prompting.

done with merging for a while

[dull jackal]

I can give you some negatives to use as well

[wispy canopy]

my model generally doesn't need them, so if they're quality related, i can go without those

but if there's stuff compositionally good for the negatives, by all means ^^

[dull jackal]

Well, we'll see how it turns out

[cyan burrow]

a little different

it's a nice noir style

I prefer the more dirty style though

[cyan burrow]

not bad at all

[wispy canopy]

@dull jackal ... SQUEEEE

[dull jackal]

THAT IS SUPER adorable!!

hahahaha!

[wispy canopy]

well, you kinda prompted it 😛

now trying to do the vixen prompt in your prompting book

[glacial bane]

@wispy canopy i need help why can't the bot respond to my dream

Its stuck

And if i try again it says pending

[wispy canopy]

i'm no admin here 😦 i can't help you

just a regular user, just like you

[glacial bane]

Damn welp thx for trying to help though

[wispy canopy]

@dull jackal and 2 nsfw ones xD

[dull jackal]

Hahahaha! Well, side prompting!

Looks nice!

Love how the fireball turned out!

[wispy canopy]

LOL THE BIRD PROMPT! LOVE IT

[dull jackal]

Mahahahaha, yeah!

There's so much I still have left to add/do/update. The bird just kills me with the wings.

[wispy canopy]

sunny, you got a belly dancer vixen in there as well... not sure if ican post those here

spicy stuff, and i've been given a timeout for less 😛

[dull jackal]

Haha, they all should be good to prompt--they have been on my end, at least!

[cyan burrow]

furry

[wispy canopy]

it's sunny's prompt, i just yoinked it

[cyan burrow]

I managed to set up comfyui to run test gens on a merge without merging

using the full block chains

00-11

[native halo]

i would take care of her in every possible way

[wispy canopy]

don't be exclusionist, you'll take care of this fair lady too, right?

@sunny, they asked me for the prmopting book on another discord server when they saw the images. you have fans!

[native halo]

my fair lady 🙇‍♂️

[dull jackal]

Aww, thanks!

I adore snow leopards. ❤️ And fashion, haha.

[dull jackal]

I'll have to update it.

[wispy canopy]

still going in your prmopting book. that's a lot of furry stuff in there tho

OMFG THE BEAR BOXER PROMPT IS WILD!

[dull jackal]

There's a human section! But like I said, there's still a lot that I haven't done, and I'll be adding more to it when I have the time.

[wispy canopy]

did you ever try prompt alternating/editing?

[dull jackal]

On a sidenote, though, here's this I've been working on: https://youtube.com/playlist?list=PLbJAuQIEql3YuacjyEedREPnwgljH7CST&si=KOLPdTq8vY6iYrau

[cyan burrow]

old gen, ramen girls

[wispy canopy]

The main focal point of the design is the mask, which represents deception and manipulation. The eyes are hidden behind a veil of feathers, giving the impression of someone who hides their true intentions. The intricate patterns on the mask also represent the complexities of life and the choices we make.

The overall design features a mix of organic and geometric shapes, creating a contrast between chaos and order. This further emphasizes the internal conflict experienced by the character. The red robes signify passion, anger, and intensity, while the white cape represents purity and protection.|shattered reference of mind,  fractured reflection, delirious, shards of a fractured and cracked face, floating broken shards, refractions, abstract]```

[cyan burrow]

the guy who sold it to them and spiked it with MDMA

[wispy canopy]

@dull jackal i can HEAR eye of the tiger playing in my mind, and it's such a mismatch!

[dull jackal]

I love that song, haha!

[cyan burrow]

I'm seeing Yogi Bear or Steroids

[dull jackal]

You could always try a tiger

Instead of bear

[wispy canopy]

it's already running that gen!

[cyan burrow]

its a bears life

[native halo]

he truly has earned the right to bear arms

[cyan burrow]

hahahaha I remember when this was the best midjourney could do

oh good times

gotten far in a year

[wispy canopy]

yeah, AI image generation went from techdemo, to silly, to thats cute, to ai took er jerbs, to whatever it is now

i mean, in a year or two, this tech will probably be regulated like hell

[cyan burrow]

well I don't think so, maybe the commercial side of it

but pandoras box has been opened

it cannot be closed

[native halo]

yes joe bedon will come here and stop me from selling a picture

[wispy canopy]

@dull jackal what where you smoking/drinking? 😛

[cyan burrow]

hospital sent me home with bandaids etc total value at 300 euro or soo, maybe a bit more

[native halo]

14th century mischievous entities when u dont leave a piece of bread ,3 strawberries and a glass of liquor under the barn every sunday

[dull jackal]

Well, neither. Instead, I'm eating chicken cheese "enchilada flavor" soup.

[cyan burrow]

including collagen (suprasorb c) that I put inside the surgical scar opening that they had to open as a 5cm diameter, 8cm deep hole after I got sepsis

but it closed almost entirely in just a month

managed to soak through 3 drymax soft pads + tegaderm high performance adhesives a week though 😄

I suppose it's more funny for me than disgusting since I'm so used to it

[pallid inlet]

🎵

[cyan burrow]

mmm Rem Ram

I should play the re:zero game

[keen hollow]

mornin'

[wispy canopy]

@dull jackal very cool prompt!

[keen hollow]

rainbowdog

[cyan burrow]

genetically modified and born in a lab to walk in a pride parade, later it dies and its acidic blood bursts out and melts through the floor before a strange alien like parasite burrows into the nearest person. Within 24 hours, a chest burster alien with rainbow colors is born.

As it goes on a rampage, nobody dares to kill it. They might be called bigots

and that was how civilization ended

not exaggerated or silly at all

not nonsensical either!

I hope you can see the jokes in there :<

[dull jackal]

Thank you! It's very sparkly. And rainbow-y. SDXL is so great!

[wispy canopy]

my use case for SDXL -> throw in song lyrics in my format, watch pretty pictures happen!

[dull jackal]

There's a lot of variability in it.

[keen hollow]

what's your sdxl workflow?

I was trying to get some comic-like styles last night for a while and sdxl was the closest I got. But the generation time was prohibitive.

[wispy canopy]

as in, how do i get song lyrics to those images?

omg, i'm surprised this image isn't more cursed than it is 😮

[keen hollow]

nah, which sdxl model, comfy or Auto, starting/ending res, etc.

[wispy canopy]

brightprotonuke v1.2 , auto, 8 steps with lcm lora

[keen hollow]

my sdxl images always hang at like 98% for a while too... dunno why

[wispy canopy]

1024 start and end

[wispy canopy]

check your vram, most probably the vae step is taking your gpu memory over the limit

[keen hollow]

hmm maybe

12 gig card, ill watch it

[cyan burrow]

hmmmm a 3090 on ebay for 260 dollars...."perfect condition"

doubt

[keen hollow]

"made from leftover taco shells and bandaids"

[keen hollow]

dang, bing fixed its token glitch

last couple days it wasn't using tokens 😛

[wispy canopy]

can't get it like that 😮 whta's the secret sauce?

[keen hollow]

dalle is always the secret sauce

[wispy canopy]

yeah, can't compee with that one

[wispy canopy]

this is when you say "oh shit"

[cyan burrow]

I wonder if KatVR would be worth it

to run around realtime infinitely...In games like Skyrim, Fallout 4, Starfield, Cyberpunk 2077

[brittle jay]

ah rly?

didnt knnow

link to her embeds?

[untold glacier]

Lol

[untold glacier]

If you're running --no-half-vae on Auto, that's why. That, and having to run everything at fp32 for the vae doubles the time for generating SDXL images when I'm on Auto. Super dumb.

[brittle jay]

man tht thing in the mouth always happens

very annoyying

[wispy canopy]

my cursed lyrics prompts on sdxl even worse, that 'mouth drip' happens all the time

straws are cursed too

[zinc flax]

Which thing lol

Was doing a pocky challenge thingy

[brittle jay]

oh

like there is always a thing in their mouth

sometimes their fingers are the thing

and the fingers are like streched weirdly

or some random objects

[zinc flax]

Very cool

[wispy canopy]

it's a VERY harsh prompt though

[zinc flax]

At least it gives something cool lol

[wispy canopy]

i would say creative {absurdism|surrealism|realism|fantasy|symbolism|abstract| } , {2-3$$ BREAK $$ __songmaster__} (had to remove image: nipple was visible)

[cyan burrow]

kingslayer ❤️

I love bring me the horizon

[wispy canopy]

case in point

[wispy canopy]

heh. such a banger song!

[cyan burrow]

check DM

[wispy canopy]

@cyan burrow extract the prompt 🙂 it's the verses of falling in reverse as my songmaster format

[cyan burrow]

nice

[keen hollow]

I do not run that parameter.

Hey so, you're all the experts. A few folks were trying last night to help. Was trying to replicate this style (from Dall-e 3) in SD. Couldn't get there... my prompt-fu isn't good enough. 😄

[keen hollow]

a bit cyber-stylized but I like it! Coherent enough to do the known characters?

[wispy canopy]

is rendering right now

hope something good comes out

(what i think it could be a bit)

[cyan burrow]

yeah its cyberpunk noir

and manga

[keen hollow]

I wanted to try to make myself a calendar with popular characters.. The Dalle output is great but not high enough res. I can generate much higher quality in SD, but don't have the model+lora+script skills to get there. 😄

[wispy canopy]

i don't have the model (wwell, that's what i keep saying), loras (i don't have loras except lcm) and script skills... what skills? XD

[keen hollow]

I can get some cool stuff in SDXL but coherency is just a little short...

pff y a'll got skills

[wispy canopy]

@keen hollow

[cyan burrow]

https://civitai.com/models/30711/m4rv3ls-and-dungeons-new-v40-comics-and-fantasy-model

[keen hollow]

not portraits 🙂 Interesting poses with interesting backgrounds. (I've been using "desolate city")

[wispy canopy]

no biggy, we can prompt for that!

[keen hollow]

good luck 🙂

[cyan burrow]

https://civitai.com/models/123841/comiccraft https://civitai.com/models/194518/cosmic-web-art-style-sd15 https://civitai.com/models/92729/comic-magazine-covers https://civitai.com/models/20294/comic-babes

[keen hollow]

ill check 'em out

good grief civit is a field of landmines

[cyan burrow]

they were all really easy to find

[keen hollow]

'comic babes' is just 'generic anime model' as far as I can tell.
the magazine one is odd, a bit too stylized to specifically that.

[cyan burrow]

You judge before you've even tried to prompt something in it

[keen hollow]

the 'comic web art' one.. not sure what to do with that.
comic craft might be worth playing with. not dalle-level, but maybe? maybe combine that to some outpainting.

[keen hollow]

that is a pretty good one

[wispy canopy]

better ones are coming up!

think i got it down 😄

well, for a bit

my model is certainly not dall3 level, but it can deliver decent results wiht the right prompt engineering

[keen hollow]

my dalle3 images are nuts...

last night we were thinking it might be better to just try to outpaint with SD or img2img if something can be found close enough

[wispy canopy]

dall3 is a couple of H100's

no consumer grade model/gpu can even compare

[keen hollow]

maybe a sketch lora or something on these to add some grit

[wispy canopy]

nah, just gonna up some weights... it's my sdxl model, it'll handle it just fine 😄

[keen hollow]

im using the sdxl niji se model for some tests

the wait is just brutal 😄

gotta get the face cleaned up

and add some grit 😛

[keen hollow]

dalle comparison

[wispy canopy]

you'r asking me to compensate for €100k hardware with some prompting skills by an amateur xD

[keen hollow]

you got this

😄

I wonder if I could do something with this. https://civitai.com/models/5009/jasgh-phrasm-inky-stylized-drawings

[wispy canopy]

stil, i feel i'm doing a pretty good job -> with just my generic model

[keen hollow]

feel like i'm getting close to something interesting too..

the fists are derpy

but ive seen worse

[wispy canopy]

dalle?

[keen hollow]

yeah

i've never seen SD get wolvie's claws right 😛

[keen hollow]

try adding "raining"

[wispy canopy]

sure 🙂

@keen hollow unless resorting to lora's or speciality models, it's hard getting closer to dall3 output, this is as far as my skill can take it

[keen hollow]

it's pretty close.. closer than I've gotten, especially with the backgrounds

[wispy canopy]

or is it?! just increase weights on "comic style"

@hallow sequoia (full color:0.1) (sketch art:2.0) drawn (comic style:2.0), (etched ,screentone effect, screentones, :2.5), {batman|the hulk|superman|wolverine|shazam| wonderwoman}, desolate __locations*__, dynamic pose kinetic action, [simon bisley|richard corben|stan lee|todd mcfarlane], raining

yup, wolverine claws still scuffed

[wispy canopy]

sd or dall3? i feel that's dall3

[keen hollow]

sd

crystalClearXL_ccxl

[wispy canopy]

dude, amazing!

[keen hollow]

supergirl in the shadows, red skirt, dynamic pose, desolate city detailed background, lora:Pencil_Sketch:1, comic style, inks and shading, raining

[wispy canopy]

ah. you used a lora

[keen hollow]

she's a bit too bright

[wispy canopy]

CHEEATER

xD

[keen hollow]

yeah im running a batch with the lora against a bunch of sdxl models

then gonna try without

[wispy canopy]

then again, i'm trying this on a generic model, biased towards realism, lol xD

[keen hollow]

needs more sketch-style on it

sdxlNijiSpecial_sdxlNijiSE is completely wild btw... you can get some really neat stuff...

[stable crane]

does comyfui just use embeddings if you supplied the path to them? can't tell if they're applied

[wispy canopy]

(in my grading scale, this image is not even worth considering) -> so much SD artifacting going on

not saying mine is perfect, but when you get SD noise or overcooking as output -> that's NOT output you'd want

aight, i'm done with superhero prompts 😄

time to do dumb stuff, even if no one is around to witness

[merry tendon]

guys, im making a LLM rp bot, can someone create an image for it?

its sfw btw

[wispy canopy]

got any idea's of image elements you'd wnat to see?

[merry tendon]

yep, so basically the bot is called "Storyteller - The Kingdom of Ashes", and would like a portrait format image of a kingdom. Idk make some variations and i can take one. Maybe with a castle, streets or smth

let your imagination flow

[wispy canopy]

llm rp bot -> it's basically as good as it's initiation prompt, right?

[merry tendon]

sometimes, really depends on many factors

[merry tendon]

i should add that its the middle age and no humans should be seen, should be from the outside, also i prefer anime style

sorry for not mentioning earlier

[wispy canopy]

nah, should've seen it coming

[untold glacier]

It's hilarious that you mention that, as I just made that yesterday.

[merry tendon]

made an llm bot?

[untold glacier]

An image exactly as you described.

[wispy canopy]

so, it's not a storyteller image, but simply an rp setting image?

[merry tendon]

oh lol, maybe darker would be nice

[merry tendon]

exactly

[untold glacier]

I'd have to fire comfy back up. I'll also leave you with this for now, though it definitely has humans in it.

[merry tendon]

humans in background are ok

just not right infront of the camera

[wispy canopy]

generic aesthestically pleasing medieval town vista, anime style inspired, by ghibly 100% prompting skill bois

[merry tendon]

the "storyteller" part is to tell the user and the ai that its not a person

[merry tendon]

edited it a bit to show you the style im going for

[wispy canopy]

aight, lets see if i can prompt this shit 😄

[merry tendon]

thx man, dont have sd installed sadly

[wispy canopy]

that done with sd, compared with RP?

(to be fair, reading is MUCH more engaging with the imagination as images)

[untold glacier]

Best I can do atm.

[wispy canopy]

@untold glacier that feathered lady you posted on the brightprotonuke page ❤️

[untold glacier]

Oh thanks! Yeah, I love that one.

[wispy canopy]

seems more and more people are liking it 😄

hope it'll break 2k downloads again

pretty sure it'll be used by others for merging

[untold glacier]

That's awesome. I'm glad people are picking it up

[wispy canopy]

nearly 500 downloads.... i mean... pretty sure it'll die a quiet death as darkalfa did

but for those who know 🙂 it's a great model

[stable crane]

I'm loving it

I have to post some pics on the page

so you get buzz

[untold glacier]

D O I T

[wispy canopy]

... do i have to make a post for each image? XD

[stable crane]

no

actually not sure how it works

it's wonky

[wispy canopy]

i mean... i've got like.. uhh... 800 images today alone? XD

[stable crane]

I've been mostly loving it for reaslitic

[merry tendon]

im back

[untold glacier]

That sums up Civitai

[merry tendon]

my internet go bye bye

[stable crane]

bless whoever included kintsugi in training images

[wispy canopy]

probably a couple of dudes who used dall3 pprompts, which thaevil1 trained, and then merged by my model xD

[stable crane]

comfyui is key for the model though I think

at least for me, for the speed / RAM

a bit more anime

[wispy canopy]

comfyui is WAY faster than a1111

[stable crane]

I'm feeling like I understand the process better now too

having to connect nodes

[untold glacier]

It still takes sooooo long to change things though

[stable crane]

but if you get into a groove 😄

just click "Queue" and then cancel the bad ones

[untold glacier]

If only people would fix things with Auto's extensions

[stable crane]

I find it much easier to cancel bad ones on that initial preview gen

and I like that I can actually queue up another while the first is running

change prompt, queue, change prompt, queue, multiple times before gen is done

[wispy canopy]

lewd carrot, in all it's nudity

[untold glacier]

You can do both of those with auto, just by changing settings or installing extensions, but I get what you're saying. There are many things that are much, much faster with comfy. Workflow ain't one of 'em though, lol.

[stable crane]

I gotta figure out how to save shit as webp in comfy

because this is too big

[wispy canopy]

"i don't think you'll fit nii-chan" -> that's enough for today xD

[stable crane]

oh cool

[merry tendon]

how does this look guys?

internet was out again btw

[untold glacier]

Dark, I guess? It's hard to see.

[stable crane]

too purple

[merry tendon]

thats the point

[merry tendon]

its not about my ST theme lol

[untold glacier]

You're one of those types of people that slaps a reshade on skyrim that super crushes the blacks regardless of if you can't see in a dungeon, aren't you? Lol

[merry tendon]

yes

[untold glacier]

https://tenor.com/view/kaguya-sa-love-is-war-shirogane-pain-painful-torture-gif-17311051

[untold glacier]

Made a couple darker, the proper way... The images are properly vertical for a card--the previews here are square and cut them off.

[merry tendon]

the first one is fire

gonna edit it a bit, and the use it on my card, thanks for that bro ❤️

[untold glacier]

Alright, last one. Glad you like 'em!

[bitter latch]

pig-kitty

😄

[wispy canopy]

"DUN DUN DUN" -> HIT THE SONG

now it conforms to anime. @dull jackal perhaps add "kawaii anime" to your prompt book 😄

[untold glacier]

Sorry, but I'm not sure I'm going to be able to stop posting images on your model page lol #🍥｜anime message

[wispy canopy]

keep 'em coming!

yorkshire terrier!

[dull jackal]

Adorable!!

[wispy canopy]

kawaii is a bit more "neutral" in terms of image composition as chibi is

[dull jackal]

I'll be sure to add more of the anime stuff to the book--it's already something I've been working on

🙂

[wispy canopy]

chibi goes stickerlike very quick, and is often female-gendered

[gloomy scroll]

Welcome to 2022

[wispy canopy]

that a 2022 SD image?

[merry tendon]

used the first image, what do you think?

wrong image

[gloomy scroll]

Prompt is but image was just done

[wispy canopy]

"gib prompt"

[gloomy scroll]

@wispy canopy

[wispy canopy]

the OG image?

[gloomy scroll]

Yes

[wispy canopy]

not even bad.

[dull jackal]

I certainly have more male prompts to add! Will do so when I have the time!

[dull jackal]

Beautiful!

[wispy canopy]

can't wait!

[gloomy scroll]

These are from time when I used Eppinette's ouroboros

[gloomy scroll]

Oh yeah I used quite few embeddings

[wispy canopy]

no problem, "my sdxl model will handle it"

[gloomy scroll]

The thing is that Ouroboros doesn't rly "handle" the prompt

[wispy canopy]

i might have a backlog if i have to try all kinds of 2022 variations of prmopts of 2022

[untold glacier]

[merry tendon]

wrong image

[untold glacier]

Win explorer

[wispy canopy]

2022 fest. LETS GO!

[untold glacier]

So it's purple after all, lol. Looks neat.

[merry tendon]

i like purple

[dawn jasper]

@nova remnant How goes life?

[wispy canopy]

i'mpretty sure he's rushing poe for free loot boxes 😄

[gloomy scroll]

Free stuff!

[keen hollow]

cool reflections

[wispy canopy]

can v3 pull of a full anthromorphic tree waifu? 😮

[short coral]

lmao, I can try

I might need a visual reference tho

[wispy canopy]

somethign like this, only female waifu oriented

sexy female groot, i supose?

pretty hard to get done on my generic model

[short coral]

will try it out

[wispy canopy]

i love pushing models to fringe cases tho ^^

[brittle jay]

man I swear

how does nai always do this

[wispy canopy]

@brittle jay they be cheating

making theyr own models and all xD

[brittle jay]

without nai stable diffusion wouldn't be close to where it is today

they single handedly revolutionized it and the sd anime scene

when the first nai model got leaked

actually

[short coral]

tbf, not on entire purpose

[brittle jay]

hehe

true

actually @short coral I have a question. do you have any law advisers or something so you know if it is legal to make your models closed source (image generation)

[short coral]

we have lawyers yes lol

[brittle jay]

so how legal is it?

[short coral]

why would it be illegal?

[brittle jay]

https://huggingface.co/spaces/CompVis/stable-diffusion-license

[short coral]

if we were doing something wrong Emad probably would've told us himself by now lol

[brittle jay]

I see

cuz I remember like right after the leak there were some conversations about this

was very big controversy, I mean still to this day

[brittle jay]

I hope goes well cuz I rly wanna see that

[short coral]

sloowwwwwly getting somewhere lmao