🍥｜anime

[nova remnant]

VERY SMOL

[terse jacinth]

but it's a comfy workflow and it takes a lot of knowledge of node-based systems to make what i wanna do

gotta kick my own ass and get to it, just lack motivation lately

but now i have my own web-radio ! 🙂

[nova remnant]

[nova remnant]

a what

[grave snow]

now now, now's not the time for buttwinded glasgow techphoto. Instead, now is a great great grandmother Sanders's choice for dogging horse's dick! basteing, mugging, dogging hotspot,venient location

to be specific

[native halo]

#🍥｜anime

[junior quest]

know how to get images like this? png info says its abyss model, but it dunt create styles like this

abyssorange + cetsu? maybe?

[lyric zealot]

if you have the png info, follow it exactly

[junior quest]

yah but it doesnt give

[lyric zealot]

all settings + lora + prompt

[junior quest]

yah yah but nope it doesnt

[lyric zealot]

model is

[junior quest]

yah I got the same one

model hash is also the same

[native halo]

post the whole png info here

[junior quest]

kay

`parameters

(dynamic angle:1.4),(sweat),

smile,half-closed eyes,

well-organized composition,masterpiece, best quality, Amazing,(beautiful detailed eyes:1.2),finely detail,Depth of field,extremely detailed CG unity 8k wallpaper,1girl, lora:GoodHands-beta2:1.0,

(huge breasts:1.4)Doll Festival, japanese clothes,, Junihitoe,(lady's ceremonial court dress:1.4),sitting,(seiza:1.4) (cleavage:1.2), collarbone,
Negative prompt: (verybadimagenegative_v1.3:0.7),(worst quality, low quality:1.5), normal quality, (3d face:1.0),negative_hand-neg, (depth of field, blurry:1.2), (greyscale, monochrome:1.0),watermark,, (realistic, nose,rip:1.4)
Steps: 28, Sampler: Euler, CFG scale: 6, Seed: 922322792, Size: 2048x3072, Model hash: f553e7bde4, Model: AOM3B2_orangemixs, VAE hash: 7ef06b9926, VAE: proto_1_vae.safetensors, Denoising strength: 0.25, Clip skip: 2, Tiled Diffusion upscaler: 4x-UltraSharp, Tiled Diffusion scale factor: 2, Tiled Diffusion: {"Method": "Mixture of Diffusers", "Tile tile width": 96, "Tile tile height": 96, "Tile Overlap": 4, "Tile batch size": 4, "Upscaler": "4x-UltraSharp", "Upscale factor": 2, "Keep input size": true, "NoiseInv": true, "NoiseInv Steps": 15, "NoiseInv Retouch": 1, "NoiseInv Renoise strength": 0, "NoiseInv Kernel size": 64}, ControlNet 0: "Module: none, Model: control_v11f1e_sd15_tile [a371b31b], Weight: 1, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: False, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "GoodHands-beta2: c05ed279295e", TI hashes: "verybadimagenegative_v1.3: d70463f87042, negative_hand-neg: 73b524a2da12", Version: v1.7.0`

[native halo]

well it says there he used controlnet tile so maybe he edited some stuff and then used img2img upscale

[junior quest]

my one and original

artstyle is different

[native halo]

yea i have that aom3b2 and its not shiny like that

[junior quest]

yah yah

[native halo]

did u used euler a sampler?

[junior quest]

yups

[native halo]

yea he probably merged another model or he used a style lora but edited the png info

[junior quest]

I guess those png info from img to img upscale?

[native halo]

yea cuz it says controlnet tile

or he probably generated that img with another model,then upscaled it with aom3b2

[junior quest]

hmm yeah I guess

but still I cant find that model

probably he merged

I looked every civitai model

[nova remnant]

(grabby:1.4)

👍

[supple raptor]

This is from a tiled diffusion upscale with controlnet tile so it doesn't really tell us anything about the original image. Could have been made with an SDXL model and upscaled with AOM and there would be no way to know.

[junior quest]

yah that artist creates grabby

[junior quest]

yups

[junior quest]

yae miko?

[native halo]

yea

[junior quest]

[nova remnant]

miko who

I only see booba

[junior quest]

yae miko

https://tenor.com/view/yae-miko-yae-genshin-impact-genshin-burst-gif-24384706

[nova remnant]

ugly anime character

[junior quest]

https://tenor.com/view/yae-miko-yae-laugh-yae-guuji-yae-guuji-laugh-gif-22927167

merged with different weights

Im good with 4th one

[nova remnant]

compares sizes

4th does seem the best quality

[junior quest]

which one?

[lyric zealot]

@dull jackal #🍥｜anime message

[lyric zealot]

@wooden coral #🍥｜anime message

[nova remnant]

hi smol sus any interesting updates on sd3?

also @lyric zealot last i read about sd3 requirement on a forum it said to have needed 8gb vram

[lyric zealot]

not that i know of

and also

oh right

there's the research paper

here #📣｜announcements message

[nova remnant]

i hope sd3 is more accessible

unlike sdxl

[dull jackal]

Waves, goes back to sleep Goodbye, le spam sammich!

[lyric zealot]

i hope so too

and i think it will be, at least for the smaller sd3 model

[nova remnant]

fingers crossed 🙂

[steady pumice]

pretty

[nova remnant]

i wiped the entire webui folder

waiting for sd3

[native halo]

https://tenor.com/view/mr-peebles-reaction-gif-9154535451992198936

[nova remnant]

worst case scenario, i can reinstall it again lol

[nova remnant]

I can't wait too...

I'll use it on @lyric zealot PC a lot

👍

[nova remnant]

remote access to her pc?

We have like a bot on discord I can use

[native halo]

we live together

[nova remnant]

oh, thats nice

[nova remnant]

that explains the water sounds

Have some elf hair

[nova remnant]

live with who? cat or smol sus? 👀

[native halo]

both

[nova remnant]

sounds terribly fun

https://tenor.com/view/혼파망-피자-당황-community-chaos-gif-16635574

😛

[lyric zealot]

maybe we can get people to pool together their unused gpu

looking at you @vital raptor

[grizzled moat]

Sd3 released ?

Please tell me yes

Me want to give it a try

[native halo]

yes,valve also just released HL3 and L4D3

[grizzled moat]

[native halo]

https://tenor.com/view/real-and-true-cat-walking-gray-cat-gif-14865986020845776272

[grizzled moat]

Valve cant count to 3

[grizzled moat]

What will be the max vram usage tho, anyone tried ?

[lyric zealot]

In early, unoptimized inference tests on consumer hardware our largest SD3 model with 8B parameters fits into the 24GB VRAM of a RTX 4090 and takes 34 seconds to generate an image of resolution 1024x1024 when using 50 sampling steps. Additionally, there will be multiple variations of Stable Diffusion 3 during the initial release, ranging from 800m to 8B parameter models to further eliminate hardware barriers.

Source: https://stability.ai/news/stable-diffusion-3-research-paper?utm_source=Discord&utm_medium=website&utm_campaign=blog

8B is their biggest SD3 model

or at least the one that we know of

[grizzled moat]

Ohhh

Neat

Too bad sd3 will be the last open sourced thing stability will do

[lyric zealot]

Who said that?

[grizzled moat]

I read it somewhere

[native halo]

obama 😔

[lyric zealot]

I almost don't believe that

[grizzled moat]

Well i want to not believe it too

So lets agree its a myth

[lyric zealot]

https://twitter.com/EMostaque/status/1762295574093512749

Emad (@EMostaque) on X

We are committed to Open Models @StabilityAI

We have given over 20 million A100 hours in grants to open AI research & projects across communities.

Stable Diffusion 3 will have an open release.

Same with video, language, code, 3D, audio etc.

Open collaboration is the way 🚀

Twitter · Feb 27, 2024, 1:56 AM

Let's hope his thought still stays the same

[grizzled moat]

Yes

[native halo]

no

[grizzled moat]

https://tenor.com/view/barackobama-talks-press-con-maybe-classified-gif-11864017

[lyric zealot]

[junior quest]

nah, I dun think its audible (or is it?)

[junior quest]

https://tenor.com/view/the-truth-hurts-truth-hurts-the-truth-truth-reality-gif-17959482

[nova remnant]

what makes you say that? Stability policy is heavily focused to remain open source

[grizzled moat]

Idk, i read somewhere (probably dreamed about it ? Since cant find it) that sd3 will be the last img gen model

[nova remnant]

doubt it they will change their moto

[grizzled moat]

Doesnt mean they will change the motto

[nova remnant]

Emad often talks about the importance of open source

few hours ago he even pointed to openai being closed source and laughed at them

[native halo]

not anime 🤓

[nova remnant]

Not Steven

[native halo]

https://tenor.com/view/ergo-proxy-anime-shocked-gif-26754249

[grizzled moat]

Yeah, what i mean is, SD3 will be the last image gen model they'll make

But if someone could confirm

that'd be great

[lyric zealot]

I mean, are you just claiming it on your own, or is there source again?

[grizzled moat]

source : none

i might sound like a sus liar, but i swear i saw it somewhere

[lyric zealot]

Well

Get the source if you can, if not, there’s no point worrying about stuff that it’s not from them

[grizzled moat]

Well let me check X

Also im not saying to believe me, all im trying to know is, if its confirmed or im delusional

[lyric zealot]

I’m almost certain nobody from stability said that, if they did, it would make big news and all AI YouTuber would make those shocking video

Well, I’m 97% sure it’s hearsay

If there’s source, ping me

Imma sleep

https://tenor.com/view/cuddles-cuddle-cat-cat-cuddle-sleepy-cat-cat-sleepy-gif-13127427148831486106

[gloomy scroll]

https://tenor.com/view/gn-peach-cat-peach-and-goma-mochi-mochi-mochi-peach-cat-gif-gif-3231093650662356654

[grizzled moat]

the one who choosed to color a role in black

just want to watch the world burn

[wooden coral]

Deleted A1111 and moved to forge today

[wispy canopy]

good job

starts a lot quicker too, imho

[wispy canopy]

TIL

[gloomy scroll]

Forge loads model when you start first gen unlike a1111 which waits for model to load before starting ui

[wispy canopy]

and some other things too

you know, if my pony based mix does this with song lyrics, i'm not even mad ^^ i mean, it's a friggin' hentai model, i just abuse the fck out of it

yeah, i think i did a good thing ^^

[nova remnant]

i've tried forge,its good but the reason i wiped webui cause im excited for a fresh start with SD3 🙂

[void galleon]

I have tried every single one of them

But comfy gives me stable consistent quality

I do agree in forge it is a thousand times faster

[nova remnant]

its a bit faster not thousand times 😆

[void galleon]

It feels that much faster when you increase the steps to 300

I know it's unnecessary to do that just the fact that it can

[wooden coral]

[nova remnant]

300 steps is highly unnecessary

you wont notice that big of a difference going above 50

at least not noticeable

[void galleon]

Depends on your eyes and monitor. I'm color blind so everything is different to me

[nova remnant]

its been tested by hugging face

i used to run 40 steps then settled for 30 steps mostly which is fairly good for results

[void galleon]

I do agree

I run 40 with comfy and 80 with forge

[nova remnant]

oh wow

guess what i've been doing as of recent

using lightning models that is sufficient with 4 steps lol

exactly 4 steps and 1 cfg

[void galleon]

I've been experimenting with some of those those do look very good

[nova remnant]

very little loss i would claim

detail loss

[void galleon]

I'm just trying to figure out how to integrate one of them into my workflow without losing any of my nodes

Because of updates

[nova remnant]

i figured since i've been using sdxl lightning models with hi resolution canvas i kinda skipped hires fix

[wispy canopy]

is that helldiver frieren?!

[wooden coral]

Hell yes

[wispy canopy]

not lore accurate tho

she's not using the official democracy sanctified helmet

[steady grail]

we need her in the game to win the war

[wooden coral]

Then you wouldnt know its her

[steady grail]

they must make her Canon!

[wispy canopy]

laser cannon got buff -> frieren in a device? 😮

[wispy canopy]

you know she'll be killed without liberty sanctioned protection

[steady grail]

let's make her shot lasers from space

[wispy canopy]

i mean, infinite mana shield cool and all -> but there's more bugs on more planets than she has mana

[steady grail]

we should make that viral XD

how did you prompt this?

there's a lora out there for the armor?

[wooden coral]

Helldivers lora and frieren lora

Should have the meta data

Civitai is down atm, cant link

[steady grail]

damn

[wispy canopy]

here i am, just lyrics prompting

[wispy canopy]

https://tenor.com/view/helldivers-2-freedom-gif-5336740907613633006

but the game really lets you play out stories, friend of my got totally surrounded and no way out: "if i'm going, i'm taking these fuckers with me" -> proceeds to arm a 500kg bomb call and gets killed on the spot. Best self-nade in existance

[wooden coral]

Saw someone jumping into a BT with the 500kg strata in his hands

[wispy canopy]

and for what it's worth, the hmg emplacement isn't all that bad

if you have teammates that can deal with the heavies, you're pretty much a guarantee that no medium enemy can even get remotely close

[wooden coral]

Well their discord certainly is anti AI

[wispy canopy]

i'm not on there

but i suppose, you posted the image there and got bonked?

[wooden coral]

Lots of clowns and vomiting emojis, but also some likes

They are scared of stuff they don't properly understand, can't blame them

[nova remnant]

oh man, i wonder how scared cave men were about calculators

[wooden coral]

Probably wouldn't have been scared at all cause they couldn't really make use of it

[nova remnant]

the sight of it alone would shock them

this is where we are at now with AI lol

it goes like this, AI is trained to self learn and has the collective mass of human intelligence but it's out to kill us

[wispy canopy]

play helldivers, encounter automatons -> seems pretty accurate

[nova remnant]

the reflection of crony systems couldn't be any more vivid

[wooden coral]

Maybe it's more the fear of being replaced and therefore having no reason to be anymore

[wispy canopy]

this

[nova remnant]

but they would also cry freedom

[wispy canopy]

within a copule of years the livelyhood of many artists is endangerd

[nova remnant]

and no 9-5 job labor

[wispy canopy]

as AI can do their art faster, and more consistent

not now, but in the future it can

[nova remnant]

inevitably

[wooden coral]

They don't understand that

[nova remnant]

im extremely optimistic about the potentails of AI

[wooden coral]

And when they do it's too late

Either you dive in now or are left behind in the dust for ever

That's my opinion

[nova remnant]

well the thing about participation shouldn't be an obligation tho

the underlying insecurity is linked to economic survival

it would be whole lot different if money wasnt the driving factor ... hence a change of economy from UBI to RBE even

[wooden coral]

There will still be "real" artists, it will just be less asked and therefore you have to be better than others to stand out, which a lot of self called artists just aren't

[wispy canopy]

i mean, i tried drawing, and i got... mediocre at best in the end

[wooden coral]

It's like when Photoshop was introduced, you can't trust any picture today but people just accepted it and even ignore that today

AI will probably end up the same

[nova remnant]

its going to define a new standards, just like how our current social intelligence permits majority to use home computers

[wooden coral]

I mean the access to it only will become easier

[nova remnant]

yes usability too

[wooden coral]

And maybe schools adjust early enough to teach new generations

About the uses and also what not to use it for

[nova remnant]

the good part of it is that its going to be intrisically intuitive in most aspect

[wooden coral]

It's just a matter of time until the voices against it are silent and adjusted to the new situation

I don't mind the negativity, but I hate that you can't even talk properly to most people about it cause of the extreme stances

[nova remnant]

my understanding of that to happen is for the current economic model to change drastically

[wooden coral]

The market is driven by demand, the demand for anything Ai only rises so the market and the economy will adjust sooner or later

[nova remnant]

its the the need basically, but money is often an interference in accessing needs

[wispy canopy]

"uuuhh... ai bad cuz news says so"

[wooden coral]

Yeah that kind of people

[wispy canopy]

hint: all your favorite media outlets are already having you consume it by the masses

welp, seems i have more mixing to do

mycurrent model iteration is the one on the right

[wooden coral]

Don't tell them social media they use daily is based on algorithms

[wispy canopy]

https://tenor.com/view/jacksparrow-johnnydepp-pirate-spooky-magic-gif-11075885606474700306

[nova remnant]

you see our current social system is based around the concept that to obtain something we have to earn it, and with AI outperforming human dexterity and intelligence is concerning for many as society's status quo is challenged

we are literally at a crossroads of utopia and dystopia, just that we have to make the colletive choice

[wooden coral]

We all know where we are headed lol

[nova remnant]

my guess is a speed bump

[wooden coral]

If everyone would make rational instead of emotional choices we wouldn't even have to discuss this

But unfortunately we're humans

[nova remnant]

tue but its not that overly simple, a country's economic strength is at stake here

[wooden coral]

That's why we need super earth and no nations

[zinc flax]

Earth 2: The Sequel

[nova remnant]

also large scale automation is extremely important considering how society is set free, we haven't reached the technological excellence just yet

[zinc flax]

I feel like AI art isn't that scary. It can't make a truly original concept e.g. character design, clothes, etc

Though I'm still impressed by stuff like style loras lol

[wooden coral]

It's just something new to unite against

[nova remnant]

what they fear about Ai is that when it reaches autonomy and able to function w/o human governance

[zinc flax]

I doubt it'll go that far.
inb4 jinxing

[wooden coral]

If AI will erase us one day, it's probably because we couldn't stop fighting each other while destroying our planet and eliminating us would be the only valid choice

[nova remnant]

AGI is that

[nova remnant]

lol that's a legit argument tho

but dont think super intelligence will find destructive ways of doing things

[wooden coral]

Things are but people aren't

[nova remnant]

old customs, habits, dogma

[wooden coral]

AI moves way faster right now tho

Industrial revolution took a century, we don't have that much time now

[zinc flax]

I didn't expect AI to be growing this fast tbh

Back then I thought the first ever NAI was impressive

1 year later and it's even more impressive lol

[nova remnant]

imagine that into weapons system and politicians will lose sleep lol

[zinc flax]

imagine if this didn't happen lol

[nova remnant]

there are certain things need to roll out for system changes, like blockchain, decentralization and things will start to fall in places lot smoother than to worry about Ai

right now we live in a system that makes society fear the system itself

just imagine these two sides, 1) Ai freeing society from mundane labor 2) AI stealing jobs

i mean I understand the fear of losing jobs, but that's not due to Ai being harmful, instead its the flawed economic model

but hey, there is also this aspect about financial control over society

those are the systemic design flaw

my uncle got mad at me when i told him to remove a predatory economic model society has to be set free from labor

he is a retired economist 🙂

yep

not on personal level but by systemic design, thats the critical part we have to understand

i think Ai is a blessing in many ways and its unavoidable

yeah, i wouldn't want this opportunity to go to waste

AI will definitely augment society 's depth of understanding and i'm sure eventually good will come out of that

[supple raptor]

The real issue with AI is who will have control over and and access to it once it becomes something that's actually useful and things are not looking particularly great on that front. Doesn't help that the biggest players in the game have spent the last decade fear mongering to make sure the general population is so resistant to letting it be accessible.

[nova remnant]

a hint of shift from materialistic world view maybe

[nova remnant]

the definition of Ai is not that it is a tool for use, but to function independent of human control and limitations

thats the power of collective customs

imagine how things can change when the collective understanding expands

[nova remnant]

cant force realization

[steady pumice]

yeah

granblue fantasy lora

same I haven't played the new event yet

need to before it goes away a few friends told me its pretty good

[nova remnant]

sometimes you have to leave that up to how life plays out, while doing the best you can

[cerulean bane]

Has one of you guys played around with making Game icons? if yes what checkpoints do you use?

[steady grail]

it's not faster to make them manually?

[cerulean bane]

probably, but i was still wondering

i mean there is Game Icon Institute on civit, that one is pretty nice for some things

[stable crane]

bing create probably works nicely for this too

[magic wave]

New styles trained today

[magic wave]

Wao das really cute :O

[supple raptor]

AYAYAYA

I'm playing in comfy with NAI api > local hires fix with kohakuxl-delta for unlimited free hires gens.

[steady grail]

that's good

[magic wave]

Fair :o

[steady grail]

ehy smile, that lora is sdxl?

[magic wave]

I like using just Pony more cause of natural prompting, but I do need to get like a good offset lora

And yeah

I've been making Pony LoRAs

https://civitai.com/user/LUUNA
Tis my Civit page if you wanna check some of the ones I have up

[steady grail]

i was trying to make 1.5 lora with kohya, but i felt lost

[supple raptor]

Pony just doesn't work as well for upscaling nai images so far. I could probably make it work better if I wrote separate prompts for the upscale, but I'm lazy so I just share the same prompt for both passes.

[steady grail]

learning how to gen with comfy was hard, but cooking a lora is a pseudo science

[steady grail]

how much do you want to upscale an image? lol

[nova remnant]

@supple raptor

todays pic

[steady grail]

it can already gen at 1024x1024. with high res fix, it become 2048x2048

[nova remnant]

Now they can train me instead

[magic wave]

Oh I mean like, Pony to Pony, well, AutismMix, since thats the thing I use

[magic wave]

Tru

[supple raptor]

I'm genning 832x1216 then 2x upscale to 1664x2432. The problem with Pony for this isn't the resolution, it's that prompting it is very different from prompting NAI so my shared prompt workflow just doesn't work well with that. Also, Pony is absolute garbage at dark images and I'm genning dark images right now.

[steady grail]

ohhh

[supple raptor]

aifeetpics2

[steady grail]

yeah, on ccomfy works weird

i run pony only on forge

and 1.5 on comfy

[supple raptor]

pony works fine on comfy for me, it just isn't suited to this specific task.

[steady grail]

on forge i could get a pony, on comfy, no matter what, i only generated nekomimi with pink hair

weird

[magic wave]

Proceeds to prompt the same as I did for Aurora in 1.5

https://tenor.com/view/escanor-gif-21501934

Though it do be bad at dark images, that is factual

[supple raptor]

sounds like possibly not using a workflow that utilized both CLIPs. If you just load an XL model in a workflow made for 1.5 and adjust resolution it isn't going to work right.

Here's a nice, simple workflow for XL with hires fix and face detailer (workflow is embedded in the image).

[steady grail]

oh, i didn't know

that's cool, ty

[magic wave]

This is "darkness" with a lora for noise offset, and its still just normal brightness

[supple raptor]

btw, if you don't already have comfy manager, install this so that you can have it fetch any missing custom nodes for you. https://github.com/ltdrdata/ComfyUI-Manager

[magic wave]

Well maybe this wasn't the best example cause of the style lora, lemme remove that

[supple raptor]

yeah darkness on pony is as bad as early sd1.5 models.

[magic wave]

Background aint even fully black

And ye, I was thinking about doing a offset LoRA, but I don't know how one does that

[supple raptor]

want a popsicle?

[vital raptor]

thanks 😄

[supple raptor]

FlanClap

[vital raptor]

lol "drinking purple slime"

[junior quest]

gimme gimme infos

[junior quest]

https://tenor.com/view/herta-doll-puppet-hammer-honkai-gif-25556598

[nova remnant]

its pony

If you want foot focused images

You type the secret foot focus tag

👍

[supple raptor]

is gud tag. i approb

FlanClap

[native halo]

feet u say?

[nova remnant]

lick

[junior quest]

yah but those expressions and face?

are you using dat lora for anime?

[magic wave]

Thats just regular Pony

[junior quest]

it doesnt gimme anime like images

and those big eyes

[supple raptor]

it has hundreds of artists that you can use to influence style. just gotta find what you like.

[junior quest]

face

[native halo]

what is that wobbly mess of a background

[nova remnant]

Its literally just pony

[junior quest]

dunno I didnt type anything for background

[native halo]

https://tenor.com/view/hacker-gif-19246062

[nova remnant]

use the extension for tags, will help you

[supple raptor]

that art style would look weird if the background were much cleaner. steven just doesn't realize that more than 1 art style exists i guess.

[native halo]

post one with a detailed background here then to see

[nova remnant]

@junior quest go look dm

[native halo]

something similar to this one

[nova remnant]

https://tenor.com/view/ace-attorney-phoenix-wright-objection-capcom-gif-20965907

[supple raptor]

congrats. you proved my point.

[native halo]

no,post a pony img with a very detailed bg

[supple raptor]

lora:add_details:2.0 on AOM adjacent model is certainly an artstyle. there are many others and that's a good thing.

[native halo]

does XL has an add detail lora?

[nova remnant]

[nova remnant]

Why detailed background when you can have detailed foreground

[junior quest]

im impressed bot didnt notice

[native halo]

https://tenor.com/view/oh-my-goodness-gracious-meme-template-humping-hump-thrust-gif-26373779

[nova remnant]

I don't know what you're talking about

I'm innocent cat

[junior quest]

ofc

[junior quest]

those are cactus right

[nova remnant]

Yes

The one and only

[junior quest]

Yes

[supple stag]

Are you using 4x-ultrasharp.pth?

[nova remnant]

I'm not using hires

[junior quest]

yes

[native halo]

heres an XL one with good bg

[nova remnant]

https://tenor.com/view/kaiba-yugioh-gif-11477966

Booba

[vital raptor]

backgrounds are overrated

[native halo]

we need the big guns here

[nova remnant]

Grabby

[nova remnant]

grabby

[vital raptor]

grabby you

that said, some XL models do backgrounds just fine, pony is more character and style trained, not so much for backgrounds

[nova remnant]

how dare you, I'm the grabber

[nova remnant]

those other models have influence from real photos, right ?

[vital raptor]

yeah probably

[nova remnant]

Thats why they look kinda realistic on clothes and stuff, pony is just art, for the anime style part of it at least

[junior quest]

know any models like this? (pastel look?)

[nova remnant]

is that nai

[junior quest]

wut is NAI

A model?

[vital raptor]

what is pastel look?

[junior quest]

hmm that messy look?

[vital raptor]

pastel is a range of colors

[nova remnant]

that looks painterly/shiny

[vital raptor]

oh, pony can do that

tons of styles in pony

just have to find the right artist or mix of artists to prompt

[native halo]

no pastel is like this one

they are often messy and add a bunch of particles to the outputs

[supple raptor]

pastel is an art medium the color range was named after it, probably because traditional pastel art had softer colors

[junior quest]

hmm its a great style

[junior quest]

hmm it looks 3d

[native halo]

yea the meina one doesnt look flat

[native halo]

this one looks like nai

[junior quest]

what is NAI? I heard bpout it, dunno whats it

[native halo]

novelai v3

[vital raptor]

i know that much, but nothing posted is even near pastel medium, or color, so i was wondering what was being asked for

[nova remnant]

genshin impact

[junior quest]

deym it, hopes disappearing

[supple raptor]

218xl, mecha-umbra, nai, and pony can all do this sort of art style with the right prompt kohakuxl-delta can get close as well.

[supple raptor]

imi does a lot of similar styles #🍥｜anime message

[junior quest]

got it

the artist say its a sdxl model that released in oct, there is no name

[native halo]

id say its not pony because the window blinds look properly lined up and not messed up

[native halo]

not a lot of sdxl anime models released in october,try to search in civitai

[nova remnant]

@native halo a message for you

And you @lyric zealot

[native halo]

https://tenor.com/view/joe-biden-thumb-uncanny-canny-mr-incredible-gif-13123031614149167572

[lyric zealot]

stooopid

[supple raptor]

it sure is impossible to get backgrounds in pony

[vital raptor]

yeah

[native halo]

such details im flabbergasted

[lyric zealot]

all i see is feet

[supple raptor]

this is the correct response

aifeetpics2

[lyric zealot]

[vital raptor]

i guess if pony/xl can't do backgrounds you like then stick with 1.5

it's perfectly usable

[native halo]

xl def can,it depends on the model

[junior quest]

for 25$ you get to generate only 500 images? hmm kinda sucks

[native halo]

well,trainin is expensive

[supple raptor]

you can generate unlimited normal res images on nai for $25. you can hires fix ~430 of them.

[native halo]

gotta milk that ai money while u can

[junior quest]

I see I see

me thinking how to get those models

stoopid

[supple raptor]

you just flatout can't get their model aside from using it on their site / api unless you're hackerman and can pull off NAI Leak 2: Electric Boogaloo

For what you get, the price really isn't bad, though I refuse to pay for ai gen at all. I'm only using it because I got a gift code for a free month from one of the people that works there in another server.

[junior quest]

I see

[native halo]

yea only pay for it if u really want those perfect images

[junior quest]

Perfect, I like that

[supple raptor]

If you didn't have hardware that could gen and had no interest in having a gaming pc, then I could see paying for it.

[native halo]

u can also just wait for the next pony / animage updates im sure ppl will improve them ( unless sd3 is better)

[vital raptor]

discord nitro gave me a free month of midjourney

never used it though

[junior quest]

https://tenor.com/view/mujikcboro-seriymujik-gif-24361533

[junior quest]

lets HOPE (fingers cromssed)

[supple raptor]

midjourney and nijijourney's models are pretty good. their prompt restrictions are obnoxious though.

[vital raptor]

yeah, i figured why bother playing with it when i can gen unlimited unrestricted locally for free and heat my room at the same time 😄

[junior quest]

I wonder if there are anyone who will hack sake of anime girls

[vital raptor]

given the success of pony, i doubt there's much need of a NAI leak at this point

with 1.5 it's pretty much the only reason SD exploded the way it did, but now it doesn't matter as much

[junior quest]

true(kinda)

[native halo]

there is,nothing out there compares to naiv3

[junior quest]

[vital raptor]

i dunno, I haven't seen anything from NAIv3 that's super impressive

it's good, but it's not better in every way

[junior quest]

isnt this impressive?

[native halo]

no

[vital raptor]

no

[native halo]

u can get that with animage

[junior quest]

eh? no?

I see

[junior quest]

hmm can I? sus

[native halo]

there are other styles in nai that cant get them right now,also the clean lineart and thousands of perfectly reproduced artists

[vital raptor]

i mean, if it's a style you like then that's great, but any model can do a generic headshot of a random anime girl

[junior quest]

nah Its not only headshots, I cropped, I cant send whole img

[supple raptor]

nai is absolutely better in every way.

[native halo]

i mean u right,anyone can do what nai did,u only need a few cs/ml engineers,a bunch of a100/ h100 clusters and like 7 million imgs with perfect tags and boom

[lyric zealot]

the crazy part about open source is that, it will almost always never be better than close source

since people from close source can just take stuff from open source, and make it closed source

kinda like how mid journey did and others

but eventually, there will be a time where using open source is already that much better, that you don't really need closed source stuff

[junior quest]

now thats the point

[supple raptor]

yep, as it turns out, having millions of dollars to run as many training runs as you need to figure out what works on state of the art hardware makes this stuff a lot easier.

[native halo]

i use supermerger that means i can out-do the engineers at nai/midjourney right

https://tenor.com/view/hacker-gif-19246062

[lyric zealot]

[junior quest]

[supple raptor]

totally. it's all the same shit. just mash them all together. don't worry what those numbers mean either. they're all nonsense

[native halo]

just steal it like the 1.5 one 👻

[junior quest]

coffee time

[native halo]

for animage it looks better if u use loras

[junior quest]

wut loras? I only saw character loras so far

[native halo]

just look for "style" in civitai and filter only the XL lora/lycoris

[junior quest]

got it

[native halo]

from nai

[vital raptor]

I guess it's like most other things that are 'better', the best part of open source is all the different models and merges plus loras and everything else, you can get what you want if you try

NAI is probably easier since it's a highly trained single model

[native halo]

yea,plus u can do videos with our current free models,cant really do that with nai

[vital raptor]

'better' seems to be subjective at best, at least I've never seen anything showing it being definitively better

[supple raptor]

reference for some of the nai styles https://rentry.org/NAIDv3artisttagtest

[vital raptor]

yeah, i have a list like that from @wispy canopy for pony

[supple raptor]

yeah, i have eface's pony list too

[vital raptor]

been playing with randomly mixing them together to find what i like

[wispy canopy]

that dude fell for some of the false positives

some tokens in the name cause a styel change, but it's just wildy different from what it should be

so there's some other activation going on

[wispy canopy]

oh wait, that's on nai

nah, on nai they could work

(ishikei being one -> doesn't work properly on pony)

[lyric zealot]

well the thing is that, user of nai can also do that with free source stuff

it's not just one or another

[vital raptor]

shiren \(monochrome structure\) on pony tends to make all the images predictably monochrome, didnt check to see if the style matches though

[supple raptor]

yeah, those are nai. some of them are definitely not just doing what they should though

[wispy canopy]

either bad tagging happened, or tag contamination

also a lot of that in pony

[supple raptor]

also...

[wispy canopy]

HOWEVER

that list also has names i don't have already

[supple raptor]

it's just CLIP being dumb.

[wispy canopy]

always worth ceckign out

[lyric zealot]

as @nova remnant would say "grabby"

[wispy canopy]

ah, sora 72-iro with your beautiful art: thanks to the leak and AOM i'm totally sick of your style xD

[vital raptor]

ultimately, I think quality wise, pretty much anything works these days if you bother to figure out the right prompting and settings, my wishlist is stuff like proper proportion and depth (without controlnet or similar)

[native halo]

another stolen one from nai

someone with a nai acc should try the sexy prompts and send them to me (just for reference ofc)

[supple raptor]

apparently that artist's name can't be sent in here so i'll explain without using the words. that artist's name and the phrase that could be made by splitting it into 2 words are tokenized exactly the same so there is no way to reliably train a model that they don't mean the same thing without just using a different tag for the artist.

[vital raptor]

I thought I read somewhere that the tokenizer checks single words like bigthings and only if it doesn't find a match it would break them apart to find matching tokens big and things, making it possible to train the combined version assuming it's not already trained as both

maybe i misunderstood, or the article was wrong

[supple raptor]

there is no token for a blank space. so thing1 thing2 is tokenized exactly the same as thing1thing2. swarm has an easy to use tool to show you how your prompt is tokenized. I've seen an a1111 extension that does it to, but don't know the extension off the top of my head.

[junior quest]

pony (anime style lora)

[vital raptor]

it's called tokenizer

apparently it doesn't work with XL models

[supple raptor]

kek

[vital raptor]

seems to me like a pretty bad oversight on the functionality of the tokenizer, there are a lot of things that are different depending on if it's 1 or 2 (or more) words

moonlight is something completely different from 'moon' and 'light'

although that explains why it always puts a moon directly in the image

[supple raptor]

that's just how CLIP functions. CLIP was never meant for this. it was just found to work reasonably well.

it's small, fast, and usually does a reasonably decent job.

[nova remnant]

https://media.discordapp.net/attachments/857955668829011988/961381752805425222/kok.gif

[supple raptor]

this is a stickup! hand over the money and nobody gets hurt

[nova remnant]

[native halo]

https://tenor.com/view/bye-go-away-get-out-good-bye-dog-gif-1315026558192484972

[nova remnant]

She just came to say hi

[supple raptor]

robbery successful

[nova remnant]

Where is the balaclava

[vital raptor]

she doesnt need one, noone is gonna be like "yeah, the robber was a cute young girl.."

[supple raptor]

needed money to buy one. but then she got a gold plated gun instead.

life of crime has paid off

[vital raptor]

took awhile to get an acceptable image.. kept giving me nsfw views, and putting nsfw in the negatives just put socks on her 😄

[supple raptor]

yeah... that's a tough angle to not get pantsu

[vital raptor]

if only it had that much..

[lyric zealot]

||is SD3 here yet? ||

[supple raptor]

nyo

theSaddest

[lyric zealot]

[vital raptor]

i guess shorts are the answer

[junior quest]

pokemon?

[nova remnant]

Definitely "lopunny" from Pokemon

[junior quest]

https://civitai.com/models/28169?modelVersionId=265102
found a one that similar to NAI

but its chibi

[magic wave]

Trying mixing styles

[bitter latch]

not sure where the glass came from

[magic wave]

I think I like this one more

[native halo]

https://tenor.com/view/oh-my-goodness-gracious-gif-25726284

[nova remnant]

Just put chibi in negative

[junior quest]

I dun have the model, need to try

[nova remnant]

It's a femb..

Definitely from Pokemon

[bitter latch]

nano

[nova remnant]

Won't fit in Mario tunnels

[magic wave]

Some more pics

[magic wave]

The pipes, yes

[nova remnant]

I couldn't remember the name

[magic wave]

fAIR Xd

[wispy canopy]

dont'care about the hands, the expression is pure gold

[lyric zealot]

any forge user here now?

[junior quest]

yups

[lyric zealot]

do you use adetailer?

[junior quest]

nope, but I have it

not using it

[lyric zealot]

ahhh, i need someone that actively uses it

[junior quest]

then its not me

[lyric zealot]

cos i found out, if with adetailer, my gen goes up by like 20 seconds

which is weird

[junior quest]

is it weird?

[lyric zealot]

no idea

but it's a lot higher

[junior quest]

I mean it need sm extra time to change so not weird I guess

[junior quest]

ouh I see

[lyric zealot]

well

[junior quest]

for me its like 4 - 6 seconds

[lyric zealot]

the total generation time from 30+ seconds to 50+ seconds

which is kinda crazy, no?

[junior quest]

yah yah

[supple raptor]

I use it and yes, this happens to me too. VAE pass is slow AF in forge.

[lyric zealot]

is it such a big difference?

like 20 seconds big?

[supple raptor]

Yep. Every vae pass adds about 15 seconds for me.

[lyric zealot]

well, so basically saying, turn off adetailer, unless it's needed?

[supple raptor]

I've just been using swarm or comfy unless I'm away from my PC and in that case I just suffer.

[lyric zealot]

hmmmm

[supple raptor]

I'll probably abandon WebUI completely if Swarm ever gets a mobile layout.

[vestal crypt]

https://media.discordapp.net/attachments/1041682785418629125/1215229286458331196/00097-1724088460.png?ex=65fbfd77&is=65e98877&hm=0442f6df80b7591e89855babf78d1c45593a8456ae0d3e06a36a1bb6e1bfaaf5&=&format=webp&quality=lossless&width=291&height=437

[bitter latch]

hmmm

random but interesting

[bitter latch]

goes back to bed in hopes of sleeping, tho out of my adhd meds so yeah my brain hates me

thuds

[bitter latch]

ok now I goto bed nite

[vital raptor]

pony kinda really sucks at making dark images, but man, the lighting and shadows are pretty damn consistent and accurate

[supple raptor]

Yeah pony seems worse at dark images than even old 1.5 models from before noise offset was discovered. That's by far it's weakest point.

[vital raptor]

everyone wants their hentai to be nicely lit 😄

[supple raptor]

It almost makes me wonder if they tried training with noise offset but set the value wrong and ended up clamping the brightness range more than base XL instead of expanding it. IDK if it would actually work that way though.

[steady grail]

you can't change the color interacting with the noise?

i even remember some nodes in comfy that can influence that aspect

[nova remnant]

where are the custom workflows stored in comfyui?

can those be loaded with a click?

[steady grail]

https://discord.com/channels/1075337975497498665/1086059095590715472

move the file inside the gui of comfy, or one of the buttons on the right

load, if i'm not wrong

[vital raptor]

the workflows are stored in the images it generates, i think it requires an extra plugin or something to have stored one-click workflows

[nova remnant]

totally new to this, just intalled comfyui out of curiosity and trying it out

i think when saving workflows they are saved as templates which can be loaded with Load button but im bit lost where to look for those templates

[vital raptor]

it's in the images you generated

[nova remnant]

oddly i dont have image saved but i saw a template loaded when i started comfyui with all the advanced feature, after clicking Clear I can't seem to load that back up

but before clicking Clear I did click Save just to be sure

[steady grail]

every image generated contain the workflow

if you prefer to start from scratch, double click with the left button on a free space and select a model loader

than connect everything else starting from it

a straight forward use a loader, conditioner+ and conditioner- for the prompt, a ksampler, a vae-decoder and image saver

[nova remnant]

no way to save workflow w/o saving them to images?

[keen hollow]

Ive tried comfy like 4 or 5 times and always leave.

cant stand how everything seems to take 9 steps to do

[steady grail]

it's pretty nerdy comfy

it's more helpful to understand how ai works

[keen hollow]

I guess im more interested in image results than how many nodes i can string together

[steady grail]

using the nodes, you assemble every single step, from the model, to the generated image

i'm more obsessed on how to gen the best image possible

i never use controlnet or region, but i have it there, assembled

just because i was curious of how it worked

[keen hollow]

at a point it's diminishing returns and "best" is subjective

[vital raptor]

same, i just want consistent results, I don't need to feel like i'm inventing something every time

[steady grail]

and instead i made a totally chaotic workflow, where every gen is totally imprevedible and creative

i love to see flowers fall, sparks of dust here and there, birds in di sky, flash of light, or shadows from the trees

i love to see the gens take life

[keen hollow]

exactly...
that's a big issue in comfy really... there is no standard for anything

everyone and their mothers have released their own "workflows" online, 5000 ways to do the same thing

[vital raptor]

yep

[keen hollow]

many of them also don't really do it properly

[steady grail]

but every workflow do it slightly differently

[vital raptor]

and most of them have no idea what any of it does

[keen hollow]

many others are just broken

[steady grail]

i know what every node in my workflow does

... kinda

[keen hollow]

then you have everyone using different extensions and plugins from all over the place, and now you have 5000 node extensions that do the same thing, spread across 5000 workflows that do the same thing

then this extension updates, or breaks, or merges with that one... it's a compete mess

[vital raptor]

"but it's so much faster!"

[keen hollow]

Oh, and it was slower for me. 😄

[steady grail]

yeah, whenthey break, i break with them XD

[keen hollow]

at least at higher res

[steady grail]

going back to the default nodes makes me cry XD

if i'm not wrong, comfy is capped, in how much ram can use

[vital raptor]

😄 it's like 0.2s faster on average for me, doesn't even save enough time to make up for all the time i wasted on the workflow

[steady grail]

now with forge, the difference in speed is not so much any more

[vital raptor]

forge on the other hand, that has some significant improvements, for the most part

i still get better speed using a1111 and TensorRT

[steady grail]

but with comfy, you can do customizations that on a fixed gui can't be done

[vital raptor]

but, forge isn't meant for me

[steady grail]

like editing the light of the gen

[vital raptor]

that's a node, not comfy, and there's extensions for that

[steady grail]

nodes that can only be supported from comfy

and his freely customizable workflow

[vital raptor]

i mean, you're free to make an extension for a1111/forge too

it's basically the same thing

[steady grail]

what makes unique comfy is that you can automatize every step of a generativer workflow.

literally every step

nothing is absolute

[vital raptor]

that's the one part i do like about comfy, is stringing multiple workflows together to do an entire process

[steady grail]

without scripting codes

that's the point and the only point of comfy

[vital raptor]

but, i find i rarely actually want that since I only upscale a chosen few images, and i don't want to change workflows to make that happen

[steady grail]

i have an upscaler, but most of the time, it's disconnected

[vital raptor]

forge, and now a1111 1.8.0 have this handy little button, i love it

[supple raptor]

There isn't a specific place for them. You can put them wherever you want. The save button in the queue widget will save a .json and you can either load it by using the load button which opens a file explorer window, or you can just drag and drop either the .json or an image with the metadata anywhere into the canvas space.

[supple raptor]

CTRL+B is your friend. select nodes you don't want to fire and press CTRL+B to bypass them. As long as they weren't critical to other parts of the workflow, the rest of the workflow just does it's thing and those ones are ignored.

[vital raptor]

yeah, that didnt exist last time i used comfy

[supple raptor]

That's existed since like a month after comfy was released.

[vital raptor]

i tried trying it out again recently but after a few minutes of building a workflow i got bored

hmm

tried extracting a lora to just get the adjustments, but it doesn't seem to work

[supple raptor]

this makes building most workflows much less of a pain. https://github.com/jags111/efficiency-nodes-comfyui/tree/main

[vital raptor]

i don't think you understand, i'm lazy and impatient (for the dumbest things) I bought a 4090 because my 3080ti was generating too slow

[supple raptor]

hmm. the darkness effect is pretty good though it did produce an undesireable pattern, especially in the dark areas.

[vital raptor]

yeah, it was more a proof of concept, and as a lora might not be as bad, but something in the extraction didnt work

currently it's just teh autism confetti model with some tweaks

[supple raptor]

this is the entire workflow for sdxl txt2img+hires fix in comfy with efficiency nodes. also you can plop down any one of those colored nodes and just drag the noodle out to an empy spot to get suggested nodes to add so you only need to search for one of them.

[supple raptor]

that one looks even better than the last image. could have been something in the last prompt just acting funny perhaps. It does make the natural graininess of the model a bit more noticeable, but I don't think it's terrible.

[supple raptor]

I forgor the screenshot

[vital raptor]

same prompt, different seed

[supple raptor]

ah, maybe just an a1111 bug then perhaps

[nova remnant]

my very own actual first workflow

and it took me several mins just to set everything up, opposed to max 3 mins to come up with the image idea on a1111

@keen hollow

[supple raptor]

do you want a workflow that's ready to go with hires fix and a better version of adetailer than you can get in a1111?

[nova remnant]

absolutely anything to save some sanity haha

[keen hollow]

I hear ya... I get it. It's why I only go to comfy when absolutely necessary.

[nova remnant]

i can relate, im kinda waiting for SD3 so fiddling with comfy in the meantime

Thighs

whoa

i got a cool workflow from loading one of my better images done in a1111

Why comfy for sd3?

[vital raptor]

it'll be the first to support it

[nova remnant]

[supple raptor]

ok, first thing you're going to want to do is install comfy manager assuming you don't already have that. https://github.com/ltdrdata/ComfyUI-Manager

This will help you to install any nodes from my workflow or anyone else that you don't already have without having to go searching for them.

Next open the attached image in browser and save the file somewhere that will be convenient to find it. Then drag the image anywhere onto the canvas. This will load the embedded workflow and you will likely get a prompt saying that you are missing some nodes. More on that in the next message.

[nova remnant]

Cat typing gif

[supple raptor]

To install the missing nodes, click the manager button in your queue widget. Then the Install Missing Custom Nodes in the new menu that opens.
Wait for it to finish and it will prompt you to restart comfy, do that with the button it gives.
Wait some more while it installs dependencies and finall reload the webpage when that finishes.

[nova remnant]

awesome thanks, yes i have manager already installed, going to load that image of yours in a bit

[nova remnant]

i guess this is the image you refered to, its in webp extension btw

[supple raptor]

Yeah, that's the image. Make sure to click "open in browser" so you get the original version. If you save it from discord it won't have metadata

[nova remnant]

dont think that file format is supported

ahhok got it now

[supple raptor]

Here's a similar one for SDXL if you or anyone else wants it. This one has a few extra nodes that you can delete if you want.

I like exporting them as PNG like that instead of just a .json so that you can easily just look at the image to see what the workflow is.

[nova remnant]

great thanks!

updating all the missing nodes

[supple raptor]

btw, you can replace the top (positive) Prompt node in either workflow with the ImpactWildcardProcessor node if you want to use wildcards.

[nova remnant]

ok thanks, i rarely use wildcard btw

[supple raptor]

👍

[nova remnant]

getting errors with face detailer

checked the manager, nothing missing currently

The message suggests versions 0.10.1 for torchvision and 1.10.1 for torch might be compatible.

tried device mode to cpu and gpu .. still throws errors so im guessing its the torch version that's incompatible?

[supple raptor]

yeah, that is probably the problem. those are quite old versions. I would recommend activating venv and running a pip install -U -r requirements.txt

[nova remnant]

did pip install -U -r requirements.txt when installing comfy couple of hours ago

ddnt use the switch -U tho

[supple raptor]

activate venv and run pip freeze it will show current versions of everything you're running

-U indicates upgrade so it's needed to ensure existing packages are updated, but since you were installing it when you ran it, that wouldn't have been needed at the time.

[nova remnant]

torch==2.2.1+cu121
torchaudio==2.2.1+cu121
torchsde==0.2.6
torchvision==0.17.1

Installing collected packages: Pillow
Attempting uninstall: Pillow
Found existing installation: Pillow 9.5.0
Uninstalling Pillow-9.5.0:
Successfully uninstalled Pillow-9.5.0
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
img2texture 1.0.6 requires pillow<10,>=9.2, but you have pillow 10.2.0 which is incompatible.
Successfully installed Pillow-10.2.0

Is there a reason sd3 will work only on comfy at first?

[supple raptor]

For one, comfyanonymous works for stability. and for two A1111 is a lot slower at implementing new features. I suspect SD3 will probably work on the dev branch of A1111 at least within the first week. But comfy likely already has the code to run it ready and just needs to push to his repository.

[nova remnant]

:oh:

[near elbow]

now you made them angry (don't ask why 🙂 )

oh all happy again

[nova remnant]

that reminds me of a really funny image

cat girls protest xD

[near elbow]

"catgrls riot"

[steady grail]

should be proposed in real life XD

[distant remnant]

SD3 already works on ComfyUI by the way

[supple raptor]

POGGIES

[nova remnant]

@lyric zealot :oh:

[steady grail]

sd3 is already out or you are on the white list?

oh, you are one of the devs

the dev of comfy i suppose lol

[supple raptor]

yesh

[nova remnant]

the comfy ui himself

[steady grail]

ok, i must be dumb

comfy became a sentient and probably lewd...

we should gen less lewd <.<

[supple raptor]

the man who was so determined to make bigger fennec ears that he wrote his own ui just to do it

[steady grail]

well, that makes sense

i would do the same

but we all know what will decide if sd3 is a success or a flop

[nova remnant]

dragon girls!

👍

[steady grail]

if you mean milf dragon girls <.<

[nova remnant]

[nova remnant]

[distant remnant]

it can generate cute fennec girls so it's going to be a success

[steady grail]

you know that ai it's just for wifus

[supple raptor]

the real question is.... when can we download weights?

[steady grail]

everything else is optional, let's be honest

t5 was for sd3 or i'm wrong?

[supple raptor]

yesh, it uses 2 clips like SDXL as well as t5. the paper indicted t5 may also be optional, though maybe that was just from earlier testing.

[native jasper]

thank you sir... the recent changes for StableCasscade and SD3, does this require a complete reinstall OR are they supported through the standard update process at startup?

appreciate your essential contributions to the community.
(i realize some of these posts are regarding improved / updated workflows)

[steady grail]

i hope i can run it on 4gvram

[distant remnant]

obviously the code isn't public yet but it will be once the weights are officially released

you will just have to update comfyui normally

[nova remnant]

big ears @supple raptor

[steady grail]

is impressive the interaction with the floor

[safe raven]

can i run sd3 on 8gb with 3070

[supple raptor]

i would assume you'll at least be able to run one of the smaller versions. almost certainly not the 8b model at least without a healthy dose of model switching trickery.

[steady grail]

yeah, the 8b is out of question

but with some tricker, i can run sdxl, so i should be able to run the middle one

sweating a little

[distant remnant]

you might be able to run the big one slowly

[steady grail]

oh, that's cool

maybe i'll play with the workflow as usual to trick things

the black magic i have done on comfy is simply mental lol

[safe raven]

can you try "Hatsune miku walking on top of construction site"

[nova remnant]

[steady grail]

i wonder if the model can gen an image of a comfy workflow XD

[native jasper]

I've had good luck running most Stable Diffusion versions on Colab. Takes some finessing.

[steady grail]

inception lol

[nova remnant]

goth girls best girls

those ears starting to look like a bat haha

[steady grail]

well, cool, now we have a big model trained to do big fennekin ears

exactly what we where expecting 👍

[distant remnant]

here's your miku about to fall

[steady grail]

a flying tower crane

that's so creative!

[steady grail]

why did removed the flying tower crane from the prompt? lol

[stable crane]

cool!

[steady grail]

cool

[nova remnant]

big bow

[steady grail]

how about giving here a big sword and a rifle?

maybe while sitting on a chair

[nova remnant]

too violent for my family friendly discord!

[steady grail]

how about a big lollypop and big water rifle?

while sitting on a baby chair? XD

[steady grail]

cool

[bitter latch]

wha wha , horror setting

ok not really 😄

kinda

I'm like nope. silently go back the way I came from

[near elbow]

just SDXL 🙂

[near elbow]

she's mocking you 😸

but joking aside, great job!

she totally approves

[distant remnant]

cute big ears ^

[bitter latch]

those are from bing dall-e 😄

BING

[distant remnant]

yeah it's pretty obvious that it's inferior to SD3

[keen hollow]

I like the ol' 6-legged horse.

[keen hollow]

kinda magazine-y

reminds me a little of jamie tydall's art

that would be a cool lora/model....