✨｜sdxl

[polar epoch]

I was like. "still fuckin upscaling? Why the... oh..

[strong field]

does that message pop up when you run the automatic1111 setup?

[hardy cipher]

big memories

[shell pilot]

I run webui-user

[polar epoch]

time to acquire more ram to do way larger gens, as sure, takes pretty much 2x longer for larger gens, but ram acts as extra video memory

[terse pewter]

Kindly take some action against people generating bikini pictures in SDXL bots, its so uncomfortable generating images there, with my young brothers behind me

[strong field]

after following these steps right?
Automatic Installation on Windows

Install Python 3.10.6 (Newer version of Python does not support torch), checking "Add Python to PATH".
Install git.
Download the stable-diffusion-webui repository, for example by running git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git.
Run webui-user.bat from Windows Explorer as normal, non-administrator, user.

[shell pilot]

Yes

Not 3.10.6

[soft zealot]

Bikini Images are hardly NSFW, if you dont want your younger brothers to see them kick them out from behind you

[shell pilot]

I used 3.10.11

[hardy cipher]

are they wearing pasties?

[hardy cipher]

I mean, that's pretty cringe to do. but not sure if it breaks any rules

[terse pewter]

in bot 10

[soft zealot]

And yes @terse pewter I appreciate there may well be (based on yuor name) some cultural and relisgous iidfferences but as the olde rbrother you need to take some responsibility for what you are showing yuor siblings

[shell pilot]

lol

[terse pewter]

actually, even my parents are behind me 🤣 we have a small home so

[strong field]

Oh, I’m not sure of that works let me do some digging

[shell pilot]

thanks for helping

[soft zealot]

thats a "you" problem not a "me" problem

[hardy cipher]

it's like looking up crude images on a computer at the public library, lol

[terse pewter]

yup, definitely cultural/religious difference

i'll try another bot

[hardy cipher]

I don't know, not sure anyone here could do anything even if they wanted to

[strong field]

Hilarious that people aren’t ashamed in a public server

[polar epoch]

Hopefully this 60MB jpeg will embed

Shit

[hardy cipher]

unless someone here is a super secret mod

[terse pewter]

i think people would be able to soon, once sdxl gets open sourced, and run it on our machines

[strong field]

60mb?!

[vital ermine]

Ahem

[hardy cipher]

8k image?

[shell pilot]

That's why I download it

Jk

this is joke

[hardy cipher]

we know what chii gonna be doin

[shell pilot]

im a girl

[hardy cipher]

https://tenor.com/view/oh-my-looking-oh-yeah-gif-18221464

[shell pilot]

i said i was joked

[polar epoch]

Got an idea for it, pooh's jar floating down the river next to him

[soft zealot]

girls also have needs #justsaying

[strong field]

it should work with 3.10

[shell pilot]

sus

Idk why it wont work then

[strong field]

i guess you could try to uninstall python

and reinstall with 3.10.6

[shell pilot]

C:\AI\stable-diffusion-webui\scripts do i put copy of python exe in here?

[shell pilot]

i dont think it will work

[strong field]

shoot sorry

[strong field]

copy and paste it, see if it does anything lol

[shell pilot]

ok

[strong field]

ive had trouble creating venvs on windows before its always been a hit and miss thing for me

[shell pilot]

Dont work

[craggy ibex]

1024x1024

[strong field]

woa these are awesome

[wet nacelle]

Thanks

[shell pilot]

@strong field fixed it

I downloaded new phyton

[strong field]

which version?

[strong field]

glad to hear it

[shell pilot]

latest, since i think i will use fooocus instead

[uncut fiber]

WTP the hacker diary

@shell pilot not sure if stable diffusion isnt problem, not UIs

[shell pilot]

I think problem was python

maybe I didint click add to path but im sure i did

idk

[strong field]

sweet, now you learning the joys of python

[uncut fiber]

yes and i think it doesnt matter what UI, that python problem is with stable diffusion

[shell pilot]

I am going to study C++ and python in free time

[shell pilot]

ye but i download latest

and latest dont work with one of the one i have

[strong field]

python more important than c++ for this stuff

[shell pilot]

Nice mark zuckerburg

[wet nacelle]

https://tenor.com/view/this-is-elon-musk-elon-musk-meme-space-x-tesla-gif-25437348

[shell pilot]

ye for AI phython better, but I want to learn c++ too

I want to be software engineer or something

2 y

years until uni

[shell pilot]

ok

[strong field]

thatsthejoke.jpg

[wet nacelle]

That's the joke too igga

[uncut fiber]

still keyboard is oposite to programer, why!!! 😦

[strong field]

then we on the same page

[wet nacelle]

true true

I actually want to see what I can get from doing mark though. These colors are great!

I actually might use clip_vision as more of a art style type thing for a bit. Not great a making subjects yet.

[steady grove]

Looks like a StarCraft 1 portrait render

[steady grove]

Something very uncanny

[wet nacelle]

TRUE

[wet nacelle]

!!!!!!!!

[shell pilot]

😐

[wet nacelle]

https://tenor.com/view/ffs-baby-really-oh-god-just-stop-gif-20490509

Some people have found great ways of blending subjects. I have not found that yet.

naw

[steady grove]

That's the character Morgana from an old game

[wet nacelle]

this dunder head

[vital ermine]

there

[steady grove]

Prompt for Morgana

[wet nacelle]

prompt craft

https://tenor.com/view/huh-gif-23918002

[steady grove]

Write the word "Morgana"

[shell pilot]

Yes

[craggy ibex]

But when I try to generate other people oc

[wet nacelle]

@steady grovebroooooo

[craggy ibex]

Why is it like this?

[steady grove]

No. Even AI won't be your gf. 🫣

[craggy ibex]

Why.

[steady grove]

Someone's mother is neglecting then again. Log off @void zenith go live your best life. Forget your mother's neglect. She's not worth it

[vital ermine]

More the cushion the better the pushin

[steady grove]

Let's be honest. It's all because of mommy issues

[wet nacelle]

re tarded and or he's not American.

https://tenor.com/view/blm-gif-25815938

??

[wet nacelle]

k kool?

[noble shoal]

Is he trolling? 🤔

[strange mist]

seems like it 🧌

[wet nacelle]

You doing well this morning?

[strange mist]

I meant the other dude 😅
I'm fine. I gave fooocus a go. it was pretty cool.

[wet nacelle]

Naw I got that.

Yeah fooocus is really kool. I used a community model and built a small prompt that forced hyper realism throughout.

[craggy ibex]

Tbh

I would rather use 1000x1000

[craggy ibex]

It's perfect for me

ngl

I hate this

for fuck sake

generate only 1 person PLEASE

It's ugly

It's not cool

It's not perfect.

[hardy cipher]

use 23x45590 for all I care. just telling you what it should be

and you should use 1024x1024

unless you want extra arms

[craggy ibex]

But why am I so unlucky when I use SDXL?

[wet nacelle]

He's actually just using 1000x1000

[hardy cipher]

you're not

[wet nacelle]

odd

[hardy cipher]

do what brings you joy

[wet nacelle]

I WANT TO EAT RIBS!!! I FUCKING LOVE RIBS!!!! OH MY GOD!!!!!!!!

https://tenor.com/view/bbq-ribs-ribs-meat-pork-ribs-gif-26281071

[strange mist]

@wet nacelle do you think ksampler approach implemented in fooocus is the proper way to do it? 🤔

[noble shoal]

Ah, good old 23x45590. That's the best resolution.

[deft coral]

i keep getting black sqaures when training

[wet nacelle]

I really have zero idea. It could be though.

It might just be me or does he have a different method of denoising? It just seemed different.

[craggy ibex]

1024x1024

Welp

I tried my best

[wet nacelle]

looks betteer right?

[shell pilot]

😐

[noble shoal]

Dude, we know now that you hit puberty. Congrats

[craggy ibex]

I guess yes

[wet nacelle]

@shell pilotDid you click a link or were you actually hacked?

[craggy ibex]

Maybe change something in here might do

[shell pilot]

link lol

Because i was retard

[craggy ibex]

What are these sampler names???

Euler?

dpm?

[strange mist]

he does something different with the sampling

[wet nacelle]

Unlucky. I two times did that. I actually joined a server and then accepted a friend request by a user I didn't know.

I now which server have the bots that spam you with groups.

[wet nacelle]

Oh yeah I remember reading that now. really interesting stuff.

[shell pilot]

wow

[noble shoal]

What is that foul language you use? Are you talking about woman like that?

[craggy ibex]

A.I. sometime being stupid

[craggy ibex]

even tho my prompts explained everything

[hardy cipher]

sometimes it just happens. use another seed number

[hardy cipher]

wow

[strange mist]

I don't really understand all of this... as long as it gives me pretty pictures 🤷‍♂️

[wet nacelle]

Same here

[uncut fiber]

He looks here as real toy.

[craggy ibex]

maybe seed 0

[wet nacelle]

Can we ping a mod to let them know of this guy? He is adding nothing to the table.

[noble shoal]

I am up for that idea

[hardy cipher]

there are lots of seeds, bud. just try any of them

[hardy cipher]

strange vibes in here today, lol

[craggy ibex]

Are there any seed that can force A.I. to generate only 1 person?

jeez

[hardy cipher]

yeah, 123456

[wet nacelle]

Here. send in the prompt.

[craggy ibex]

Tragic.

[wet nacelle]

WICKED

[shell pilot]

Why does this take forever?

[wet nacelle]

It won't

It's weird

[craggy ibex]

When I click "Queue Prompt"

the seed randomly generated a number

idk why

[shell pilot]

I waited like 10 min now

[wet nacelle]

I swear that it will just work

something may be wrong hen

[craggy ibex]

maybe it's becuz the randomizing seed cause the problem for me

[shell pilot]

Do i wait or reload

[wet nacelle]

Close the cmd and reopen it.

[craggy ibex]

Disappointment

is that a fucking dog or a delusional dragon?

also

[shell pilot]

It's a furry

[craggy ibex]

you are not a mother fuckin centuar

[wet nacelle]

https://tenor.com/view/i-know-what-you-are-boykisser-ominous-stare-gif-1854943194524266546

[shell pilot]

I am not

I am a furry hunter

[wet nacelle]

I am a degenerate furry.

😇

[shell pilot]

😕

[wet nacelle]

https://tenor.com/view/jerma-gif-14663292704966997212

[zinc cargo]

anyone here runs swarm?

[steady grove]

Bragging about being degenerate 🫣

[vital ermine]

I love taking images and running them through my lora

Never know what is going to be spit out

[craggy ibex]

FINALLY SOME GOOD SHIET

URGH

IT TOOK A FUCKING FOREVER

[vital ermine]

Seems like it makes yours flowers

[craggy ibex]

FUCK the "allboymix_v20" this piece of flumery of a.i. generating image industry

such a mother fucking disappointment

[steel mist]

I hadn't and would never in a million years figured that out. Thanks!

[craggy ibex]

I'm so fucking done

[craggy ibex]

time to send this image to my friend

😡

[vital ermine]

Well, at least this one didn't make flowers for me.

[shell pilot]

still dont work

[uncut fiber]

it is stuck on processing text encoding, it seems it doesnt generate image at this stage

[craggy ibex]

I'm so sorry to hear this

[uncut fiber]

@shell pilot try post what cmd error it show

[vital ermine]

No, that is a good thing

[shell pilot]

it just wont load fast

[wet nacelle]

Can you check the advanced tab to see if the base and refiner are selected properly?

[shell pilot]

ok

[strong field]

What gpu did you say you had?

[hardy cipher]

yo

[shell pilot]

2060 rtx

[shell pilot]

its stuck

[hardy cipher]

@polar epoch Made this abomination for you

[wet nacelle]

That does look right. odd

[shell pilot]

the thing wont load

thats the problem

[wet nacelle]

Have you tried to just uninstall it all and reinstall it?

I hate doing that with this tech but it might help.

[shell pilot]

idk how i just ran bat file

how do i unsintall that?

[wet nacelle]

Just all the files. I would just delete the folder.

[uncut fiber]

or you can try delete only venv folder

[wet nacelle]

AH

yeah that too

[craggy ibex]

How to not make A.I. become stupid even tho I applied so many prompts

Such a impossible dream

[hardy cipher]

yeah

[craggy ibex]

I did everything you told me

[soft zealot]

Interesting take on The BatGirl using the ReVisioned technique @strange mist @native knot

[craggy ibex]

but a.i. still not being cool with me

[hardy cipher]

that's how it works out sometimes

[elfin flare]

Where is this addon and where can I get it.

I want to be able to control LORA weights.

[wet nacelle]

I really like this style

[elfin flare]

Is that a LoRA?

Btw is the thing that guy is using up there

LoraCTL?

What's the name of that extension?

[wet nacelle]

No. I'm using clip_vision

[hardy cipher]

you just need this

[elfin flare]

I'm not using Comfy.

Lol

[noble shoal]

Hey quick question. Does anyone know what class prompt to use for a style in Kohya? For my last analg photo style i have used "style". I am talking about LoRa training btw.

[hardy cipher]

well, you could be using a 10 stack, bud

[wet nacelle]

Oh wait should I use this too?

oh wait. rad bro.

[elfin flare]

Well no, I couldn't.

ComfyUI is slower on Intel Arc than it is on Vladmantic

And half-broken as well

I don't want a 10-stack.

[hardy cipher]

I just made it a while ago because I could. might actually use it though

[elfin flare]

What is this extension?

[hardy cipher]

that's the extra models extension or something along those lines

[hardy cipher]

sd-webui-additional-networks

[elfin flare]

I see it.

Question now

is if it works on ARC.

We'll see.

[hardy cipher]

apparently I can't loop clip vision conditioning back onto itself

lame

[elfin flare]

Clip vision is indeed interesting

[hardy cipher]

unclipXL

[craggy ibex]

How to stop A.I. switch colors?

[hardy cipher]

ask politely

[craggy ibex]

I am having a headache fr

seriously with this

[steel mist]

ok the filename prefix there almost made me spit out my coffee 😂

[hardy cipher]

LOL

just noticed that

[steel mist]

I have many figma files with layers named something along those lines

[hardy cipher]

https://tenor.com/view/me-explaining-how-to-get-free-robux-me-explaining-how-free-robux-robux-bobux-gif-20114366

[soft zealot]

a few more "ReVisioned" takes on BatGirl @strange mist & @native knot & @crisp owl

[hardy cipher]

that's some clip vision there

[soft zealot]

Any South Africans in the room ?

Damn they can play some good Rugby (and yes I'm Welsh & currently depressed lol)

[delicate kelp]

Hi guys, good morning! I'm happy to announce the release of ProtoVison XL 0.5.2.0!

This version of ProtoVision has had some additions to add more variety and complexity to scene backgrounds, more facial and skin detail to subjects, better coherence. Eye and hand improvements. PVXL should also be more responsive to cinematic prompts. Facial variety has been improved, and racial biases are less prominent. - See listing for full changelog details and sample images. Thanks so much for all your downloads, likes, stars, and shares, I LOVE seeing all your art filling the galleries for my listings, it really is amazing!!! 🫶

https://civitai.com/models/125703?modelVersionId=144229

[upbeat summit]

this looks amazing!

[delicate kelp]

Got updates for Dynavision and NightVision in the pipeline as well, testers have been creating some pretty amazing stuff with them, can't wait to show you all what we've got coming next!

[craggy ibex]

I wonder how do I check what's my version of SD hmmm

in ComfyUI

[uncut fiber]

what SD? It is only model. Comfui, you can update with manager

[craggy ibex]

Stable Diffusion

[uncut fiber]

i am telling you. SD is just and only model. You know what model are you using, comfui or a1111 UI version you are looking for, and you can install update of any of them

[empty orbit]

any good textual inversions like easynegative that work with sdxl?

[craggy ibex]

Oh oke.

What's yall GPU btw?

[uncut fiber]

3070 only

[craggy ibex]

damn

mine is gtx 1650 ti

[uncut fiber]

i used to have 1070ti, and switched to 3070 because not needed changing PSU 550w

[craggy ibex]

very cool for learning how stable diffusion work

[craggy ibex]

daaaamnn

that's drip

very cool drip.

[soft zealot]

GTX1080ti +5600X +64Gb system RAM

[zinc cargo]

what's the prompt?

[craggy ibex]

Get iced

[steel mist]

Is there a comfy node that'll pull metadata out of a png

I recognize that there would be quite a few issues with it, including the sheer amount of MD in a large workflow

But it'd be rad if there was a node that would let you grep/search for an output by name and then use it

[soft zealot]

a tower unit with green lighting sat on a desk that has 3 monitors and a Mandalorian Helmet

[zinc cargo]

and a tardis!

[soft zealot]

and a Dalek

[terse pewter]

This is my second attempt at creating a sad story through ai images of stable diffusion

I am trying to create stories, that would make us cry, for tiktok

You need to max out the video for full view

[uncut steeple]

Why is it so wide

[terse pewter]

It's tiktok resolution, 1080x1920

[zinc cargo]

if you aim for social, try to do 16:9 resolution (unless you want to put captions above and under - then never mind)

this is how we see it in discord
that's why i've commented

[terse pewter]

yup, if you press on the rectangular button below, it will show correctly

to expand it

[half ivy]

genuine imma bout ta leve comfy

how tf is this wrong

@visual glade soz bt dafk

[hardy cipher]

oh, okay

[half ivy]

dont

[hardy cipher]

I see

[half ivy]

show me

[half ivy]

modules dnt count

[hardy cipher]

dang, what will you do now?

[half ivy]

move back to 1.5?

[strong field]

That’s not the problem

[half ivy]

i am

[strong field]

What modules are you missing

[half ivy]

i get it

tx

[half ivy]

0 appraz

[strong field]

Manual install.

[half ivy]

@strong field gimme yr best CN,

[strong field]

CN?

[half ivy]

need a prompt and a preprocessed img?

[hardy cipher]

if you look at the red nodes you don't have installed you can see what they're called, then just find the node packs with those nodes and install

[half ivy]

yea right click on node

get it

[hardy cipher]

https://tenor.com/view/die-laughing-funny-kids-dies-gif-24902870

[half ivy]

ever tried making a barcode\

same vibes

[cedar tusk]

For Comfy nodes, is there a way to pre-define the names of the outputs of a custom node in the code? So instead of saying "INT" it might say "height". I know you can rename them after in the UI

[hardy cipher]

you could do it in the code for the node

[half ivy]

right click / rename

jesus

guis

ffs

[hardy cipher]

shush

[half ivy]

thats what she said\

anyway

will still be here i support

[cedar tusk]

Can you point me to how? If I do a custom RETURN_TYPES() entry then it'll limit what that can connect to

[half ivy]

need to get bk to production

[half ivy]

cant be so hard, surely

[cedar tusk]

Problem is that only works on my local saved workflow. Not if people load in the node on their own workflow

[strong field]

Bruh what is happening

[half ivy]

bro

[hardy cipher]

RETURN_NAMES (tuple):
Optional: The name of each output in the output tulple.

[half ivy]

same q's

[hardy cipher]

he spelled tuple wrong, lol

[strong field]

Gotta hard code

[cedar tusk]

ah, thanks! I had no idea I could tuple that

[half ivy]

serious, avoided blender nodes ast all costs, dove into sdxl open hearted, bt...

[azure oxide]

god why cant comfy be easy to use for people like us

[half ivy]

amen

[hardy cipher]

why can't this free thing that I choose to use cater to me?

[half ivy]

whats wrong here?

[cedar tusk]

Using Comfy and developing for Comfy are two different things. I think using comfy is pretty easy if you just drag in a workflow

[half ivy]

bro, im bitching for all normies,

[hardy cipher]

this is how I node

[mossy canopy]

creating nodes are simple too.. but it's so time consuming

[cedar tusk]

aight, so it looks like that doesn't work. If I have something like:

RETURN_TYPES = ("INT", "INT")```
the node has outputs of "INT" and "INT"

[half ivy]

honetly, b4 you devs make yr workflow, send it to me, i haz teh dumbs, we can simplify

[cedar tusk]

oh I am dumb, disregard

It's RETURN_NAMES

[hardy cipher]

indeed

[half ivy]

pcs showing off again

[half ivy]

from a non node creator, this is very confusing, well done, you missed the point, you kept it selfishly

is GA a bot?

[vital ermine]

Last time I check no, but who knows.

[half ivy]

going bk to auto at 1.5
sad story bt fun'er

AI is a thing, so mebe the response is not = (will delete this)

[vital ermine]

At least I am not running around in here asking for anime chicks to have sex with like that other dude did this morning, lol

[queen adder]

if you want your Lora training to really affect your base model, how high on the network rank would you go before considering it to be too much? i tried using 128 and felt my Lora didn't make too much difference on the underlying model when i called the instance.

[half ivy]

k wait, ya'll got anime girls inhere. i might stay then

jk

super upset though, thought more devs wld congregate

[vital ermine]

SD is in a state atm, ngl

[half ivy]

gimme decent rsrcs and i will learn

[vital ermine]

lol

there is the fly in the ointment since forever

[half ivy]

every 2nd person has a generation that has red blocks that isnt fixable

[vital ermine]

??

[half ivy]

red

[vital ermine]

Yeah, well I am not doing comfyui for reasons.

[hardy cipher]

I guess you probably shouldn't use it

[half ivy]

i envisage openness, not selectiveness

[cedar tusk]

Alright, so another question: how can I get Comfy to load an image preview on a custom node like LoadImage does? It looks like it might be hard-coded

[hardy cipher]

very strange people coming in here today

[vital ermine]

well, I can hear SAI now. "Then for you we have the simple web interface for xx per month."

[half ivy]

dont you dare

[hardy cipher]

hmm, that's a good question. I should figure that out

[cedar tusk]

name appropriate, haha

[steady grove]

all this apprehension about comfyui reminds me of when governemnts made leaded gas illegal and many people were pissed about having to put less toxic gas in their cars

its what they've always known and they don't want to change

[half ivy]

hmm

imagine changing and it doesnt work

no manual to guide]

just code bro

do it

[hardy cipher]

then don't use it

[steady grove]

there are manuals and guides

[half ivy]

or dont

[half ivy]

classic

called it

[steady grove]

you don't have to use comfy if you dont' like it. it's not like it's legally mandated. it existing doesn't harm you

[half ivy]

i want it

[hardy cipher]

it existing doesn't make a1111 suck right now either

[steady grove]

this conversation is very predictable. I agree

no a1 just sucks on it's own usually

[half ivy]

no

it responds to instructions

[eternal fog]

You could look at something like StableSwarm. That's got a comfy backend, but an easier to use front end.

[hardy cipher]

people are acting like comfy stole it's a1111's soul or something. like they're outlanders and there can only be 1 stable diffusion UI

[steady grove]

that's different too though. so much change

[half ivy]

comfy is like, can you guess the formula

then can you guess the rigfht ingredients\

[steady grove]

some people like to make their own problems and manufacture their own roadblocks. it's weird

[hardy cipher]

people that complain about things that are free and that they aren't obligated to even think about make me sad

[half ivy]

wish i cld code

[steady grove]

complainign about no manuals and docs is just lazy too

[half ivy]

,(excuse AGAIN amirit)

[cedar tusk]

complaining about free things can be valid, just needs to be actionable complaints

[half ivy]

wots yr gpu?

[hardy cipher]

that's fair

not saying it should be immune from criticism

but vague nonsense

[steady grove]

google is free and i got a LOT of complaints about them as a company

facebook too

[hardy cipher]

"I'm going back to 1.5"

evidently thinks comfy is sdxl?

[half ivy]

bro

[steady grove]

lol yeh. That's just nonsense

[half ivy]

send me yr best img2img CN

[half ivy]

lets talk

[steady grove]

i think a lot of it is fanboy logic. Because stability is paying comfy, that means auto isn't getting paid 😦

[half ivy]

its here

[hardy cipher]

well, I wouldn't mind if they both got paid

[strong field]

Do you no like maths

[half ivy]

make it

ill load yr result

[half ivy]

no

[hardy cipher]

a1111 taught me a lot, but then it kind of sputtered out

[strong field]

“Load default”

[half ivy]

im a non coder, non math'er

[steady grove]

you just linked someone trying to helpyou figure out comfy, and you sarcastically treating them with an attitude similar to "Yeah okay. sure buddy!"

Not sure what the point was

[half ivy]

entertain me

[hardy cipher]

it's weird tribalism nonsense as well

people have to choose a side

[half ivy]

help

[steady grove]

pikachu, i choose you!

[cedar tusk]

Well, for now I'm just using a preview image node. And with that I now have a node that can load in an Auto1111 image and let you grab the metadata to deliver to other nodes!

[half ivy]

how so

[hardy cipher]

ooh, that's pretty awesome

[half ivy]

yusss the angled hate to join in

[hardy cipher]

I've been looking at the code for nodes that have some of the features I want. and then looking at it again and again since I barely know python, lol

but figuring it out

it's pretty straightforward compared the C and C++ or java

[strong field]

Hey that’s a great idea

[steady grove]

python is a scripting language. it's leagues easier than c++ or java

[hardy cipher]

thinks he's trolling. I guess he is

[strong field]

Python gg ez level compared to those

[hardy cipher]

frustrated he can't noodle so he's lashing out

[half ivy]

no

[steady grove]

again, you're focusing on drama instead of actually using the UI.

this isn't looking for assistance. it's looking for attention and dragging things out.

[strong field]

Hey I get it, I’d be sad if I couldn’t noodle

[steady grove]

comfy got them noods everyone wants

[half ivy]

sj\how me yr assistance, you literally went to dramas bro lol, read bk, you pushed it

[hardy cipher]

people's insecurities often manifest into aggression

[half ivy]

is true

[steady grove]

thats a great way to ask for help after you've spat on a lot of it being offered

[half ivy]

bt

should try understand

[steady grove]

maybe it'll work

try asking that way again and doublel check

[half ivy]

hoooowwww

JC

[half ivy]

kkkk

[hardy cipher]

I'll help pretty much any doofus that seems to sincerely want to know about things

[half ivy]

you right

[craggy ibex]

I did my best

[half ivy]

my bad

i see

[hardy cipher]

but I think some people don't want answers

[steady grove]

yup

[steady grove]

yuuuup

[half ivy]

anyway, doofus here

[hardy cipher]

nah, I don't know what you are

[half ivy]

wot is top model?

[steady grove]

they're all different tools for different purposes. there is no best

[hardy cipher]

those are all 1.5 controlnet models

[half ivy]

it worksw

top odel/

[steady grove]

there doesn't need to be a best either

[half ivy]

fk my "m" is going bust

[half ivy]

shhh

[steady grove]

tried to offer advice and then this. okay

[half ivy]

the top 1.6 works with sdxl base and refiner, so it works

[steady grove]

"i've changed guys. help me!" /acts the exact same way

[half ivy]

bt

argh

how?

[eternal fog]

If you mean what is the top model, as in what is the one on the top of the list. It's hard to tell because it's not named very well.

[craggy ibex]

This checkpoint is really meant for me

[half ivy]

i need a workflow

[craggy ibex]

I love male anime character

[hardy cipher]

allboymix

nice

[steady grove]

if i explain then you'll spit on my answer

[hardy cipher]

I did not know that was a model

[half ivy]

true

dont worry

[craggy ibex]

yeah

[half ivy]

ill wait for some1 elser

[craggy ibex]

I'm searching for stuff like this

[half ivy]

lol

[steady grove]

i'm not worried about other people failling. i think they're the most worried about that one

[half ivy]

\llloooooollll

[craggy ibex]

but not 3d anime tho

[half ivy]

sensitive sally

can i nrename him

yessss

dont

[steady grove]

a lot of people when failing will blame anyone but themselves

[half ivy]

i teruly blame me

serious

[craggy ibex]

I LOVE HIM

🫶

[half ivy]

like i sai, avoided nodes in blender,

avoiding them now \

just think it could be easier

[cedar tusk]

Here's another question: what's the trick to getting a node's input to default as an input port and not be in the node. For example, having a STRING type be an input point and not a text box?

[craggy ibex]

DID HE JUST GIVE ME A MIDDLE FINGER?

[hardy cipher]

that one is really good

[craggy ibex]

HELLO?

[hardy cipher]

this is my favorite model right now https://civitai.com/models/128456/pe-shitty-fanart

that lora is also potentially super useful if you're going for photorealism

just send it negative

[steady grove]

make one then copy paste it

ctrl c ctrl v

or alternatively, ctrl shift v to paste with connections

[cedar tusk]

Ah, I mean in the code. So that the first time a user loads the node it's showing ports not boxes

[hardy cipher]

hmm

[steady grove]

Not sure about that. I think you'd have to recode each node.

setting it up once and pasting it seems a lot easier for new users

[hardy cipher]

I don't know about that

[steady grove]

maybe we'll get an option when importing new nodes. "all values as inputs"

[strange mist]

I see they're more on realistic side 😄
top middle is my fav. amazing detail on the costume.

[polar epoch]

You mad fuckin lad, i love it Now to tandem up 2 or 3

[craggy ibex]

He is so cool so me

I will call him...

[hardy cipher]

still haven't gotten around to uploading it, lol. but I will shortly

[craggy ibex]

Strad Grandgaze

[half ivy]

@steady grove a challenge?

i can dm input img

[cedar tusk]

Is there a server rule about at'ing the comfy themselves?

[hardy cipher]

wait, wallish, what exactly were you wanting to do?

[polar epoch]

Aye, ping me when it's ready lol

[hardy cipher]

will do, keep getting distracted by various things

[cedar tusk]

I have "STRING" inputs to a node, but I would like them to default to 'input' type (node inputs) and not to widgets (like multiline)

On first load via the node definition in code

[hardy cipher]

oh, well that's easy I think. unless I'm misunderstanding you

[eternal fog]

Why don't you look at another node like the Save Image node that only has an input and look at how it's coded.

Should be able to get an idea what way

[steady grove]

you admitted to not respecting what I say. Why would i engage that?

[half ivy]

i do though

[cedar tusk]

So it looks like some types default to having widgets made

[polar epoch]

That's me as well every damn day from everything lol

[steady grove]

this feels like troll bait and you'll just string me along saying my generation isn't good enough

[half ivy]

you tend to be negative, is okei, a challenge if you will

[cedar tusk]

But also I've accidentally had it where I turned an input into a non-widget on load, but I'm not sure what the syntax was that did that and it isn't consistent

[steady grove]

its just an opportunity for you to act as an arbiter towards me. no thanks

[hardy cipher]

can't you set it up as input?

[steady grove]

i'll challenge myself

[hardy cipher]

well one second, I'll look at the code

[steady grove]

okay well, no thanks

[half ivy]

just keep the shape

CN vibes

[hardy cipher]

        return {"required": {
                    "model": ("MODEL",),
                    "clip": ("CLIP",),```

then it shows up as inputs, that format

I realized a while ago I actually don't care about anything this guy is saying

[cedar tusk]

here's an example:

def INPUT_TYPES(s):
  return {
    "required": {
        "prompt": ("STRING",{"default": '', "multiline": True}, ),
    },
  }```
This will start with a widget for the 'prompt' value, but you can turn that into an input port in the UI.  If I remove the definition parts (so it's just `"prompt": ("STRING",),` for example, it may not even load the node

As the "STRING" entry is needing to know the type, I think

[hardy cipher]

use this as an example

    def INPUT_TYPES(cls):
        return {"required": {
                    "model": ("MODEL",),
                    "clip": ("CLIP",),
                    "switch_1": (["Off", "On"],),
                    "lora_name_1": (cls.loras,),
                    "strength_model_1": ("FLOAT", {"default": 0.0, "min": -10.0, "max": 10.0, "step": 0.01}),
                    "strength_clip_1": ("FLOAT", {"default": 0.0, "min": -10.0, "max": 10.0, "step": 0.01}),
                    "switch_2": (["Off", "On"],),
                    "lora_name_2": (cls.loras,),
                    "strength_model_2": ("FLOAT", {"default": 0.0, "min": -10.0, "max": 10.0, "step": 0.01}),
                    "strength_clip_2": ("FLOAT", {"default": 0.0, "min": -10.0, "max": 10.0, "step": 0.01}),
                    "switch_3": (["Off", "On"],),
                    "lora_name_3": (cls.loras,),
                    "strength_model_3": ("FLOAT", {"default": 0.0, "min": -10.0, "max": 10.0, "step": 0.01}),
                    "strength_clip_3": ("FLOAT", {"default": 0.0, "min": -10.0, "max": 10.0, "step": 0.01}),
                }}
    RETURN_TYPES = ("MODEL", "CLIP")
    FUNCTION = "apply_loras"
    CATEGORY = "loaders"```

model and clip are inputs

[cedar tusk]

The problem is with the "STRING" type, mostly

[hardy cipher]

ahh

[half ivy]

dang

night then 💚

[hardy cipher]

I always just look for other nodes that have whatever it is I don't quite know how to do

see how they do it

[soft zealot]

probably my favourite as well.

On the realism point its intersting because as you can see I wasnt particulalry prompting for it

[steady grove]

this is the way

[cedar tusk]

yeah, unfortunately I'm not seeing anything in the default nodes or in the sets i have installed

[strange mist]

it's probably nightvisionXL, must have a higher bias towards realism.
what's this Revision technique btw? are you using two images as inputs alongside your prompt? 😮

[soft zealot]

something SAI announced yetserday. Yes uses one or more images that get clip encoded and added to the prompt

See the lower part of this link:

https://huggingface.co/stabilityai/control-lora

[craggy ibex]

HE HAVE MUSCLES

[eternal fog]

Thats a big eye

[soft zealot]

Ooh just reran that top middle one back thru but this time using ProtoVision

[eternal fog]

Interesting

[hardy cipher]

I'll look around and see if I can find anything in mine. curious now

[soft zealot]

Which is intersting when you see what he inputs were lol

[strange mist]

still realistic but in a different way. the mask needs inpainting. other than that def a good one.
looked up the revision thingy, mindblowing stuff.

[strange mist]

those are nice eyes 👁️
which model are you using?

[eternal fog]

Newest ProtovisionXL model for those

[cedar tusk]

https://github.com/wallish77/wlsh_nodes Feel free to check out the Auto1111 importer and mess aroudn with my nodes to see if you can figure out that STRING input thing. I gave up for now.

[strange mist]

wow, no prompting or after detailer?

[hardy cipher]

hmm, I think I might have them installed actually

[eternal fog]

No just that Revision technique

[hardy cipher]

lol, didn't even make that connection

[eternal fog]

@strange mist These as the input images

Wasn't what I was expecting, but looks cool and if you push the weight more towards the earth image, it starts to make eye based space scenes

[soft zealot]

this is one from when I was playing with the original SDXL0.9 release

[eternal fog]

It's not always a win though

[strange mist]

cool stuff. I'll give revision a shot tomorrow 😄

[hardy cipher]

once I get a few put together and am confident they'll be useful to people I might look into having them put in the manager list. haven't really looked into how that's done though

[cedar tusk]

They are already in the manager list

Have been for a while

[hardy cipher]

yeah, I think I use them actually

yup, got em

I tend to delete the node packs when I keep getting dependency issues and realize I don't really need them

but seems yours passed the test, lol

[tepid surge]

Anyone else working on getting a setup for ComfyUI to do animations? :3

[hardy cipher]

I wish

[soft zealot]

FYI you can do this into an OOTB Image Save node using built in S&R fuunctionality.

Eg I use

SDXL/%date:yyyy-MM-dd%/%date:yyyyMMddhhmmss%%SWT.seed%%BML.ckpt_name%%RML.ckpt_name%%STYLE0.style%%STYLE1.style%%STYLE2.style%_STD

WHich piut it in a ime dated Subfolder of Outputs/SDXL with a name like

20230819203614_3347873819_protovisionXL.safetensors_protovisionXL.safetensors_base_base_base_HRF_00001_.png

Co BML.ckpt_name is Base Model Loader (yes I know Im using a WAS NODE witha Name_String on it, just not using it, OOTB S& R & WAS Tokens dont plauy nicely, reasons) you just need to edit the Node Name in Properties

[cedar tusk]

oh dang, I had no idea

[soft zealot]

Its all about learningthough so its never a waste 🙂

[cedar tusk]

Well it's too late to undo the dev time, haha

Plus i wrote a PyQT tool that lets you drag in images and get prompt info from Auto or my nodes, which is super useful and wouldn't work with the default save, so not a waste

https://github.com/wallish77/sd-prompt-inspector to try it

[hardy cipher]

was wondering why this upscale was taking forever. turns out I had somehow stacked two upscalers on top of each other and they were both in the chain

[tepid surge]

FizzNides, WAS Node Suite & MTB Nodes seems to be a good starting point so far but once I figure out something more full I'll throw the set up in here

[hardy cipher]

256?

lol, well big image

16^2

[indigo carbon]

hey, I'm trying to make an image fusing workflow, final image has nothing to do with the 2 images provided. any idea?

[cedar tusk]

Maybe let them get to know each other first? Get them some movie tickets and dinner reservations? (Sorry I have no idea)

[tepid surge]

there might be something like Interpolating latents

I might be making a technical blunder saying that but like vae encode the images and then some lerping thing

[strong field]

See if this works

[indigo carbon]

currently messing with the image weights

[strong field]

I hope so he image didn’t get stripped of meta

[indigo carbon]

no metadata

[strong field]

Darn on mobile so I don’t have my goodies

I got u tho 2 seconds

#✨｜sdxl message

Did you stay here, he’s got the weights preset

Prompt also matters here, empty prompts can get random

[hardy cipher]

love it when I inadvertantly srun a 1.5 model in the xl workflow

[strong field]

Two heads are better than one

[indigo carbon]

huh, I'm trying to make it not prompt dependent, gonna look into this

[hardy cipher]

man, I spent about 20 minutes adjusting clip vision and various other things. then realized I hadn't even thought about the model

ugh

[strong field]

Ah I misunderstood

[soft zealot]

ok not a traditional look but I am likingthe way the suit has been rendered

[cedar tusk]

Decided to make some ComfyUI app icons

[hardy cipher]

thought you were cracking joes at my expense. which would be fair

[hardy cipher]

joes

jokes

[hardy cipher]

I should just stop typing for now

[hardy cipher]

those are pretty cool

[strong field]

Isn’t clip dependent on text anyway, not sure how to give revision direction without a guiding prompt

[indigo carbon]

my other approach is to maybe use interrogate and merge tokens, I know that's possible

[strange mist]

a grittier re-imagining of batgirl. the color scheme reminds me of the awful titans show.
I like the expression on her.

[soft zealot]

Revision can be used without any textual prompt what soever and can generate a LIP encoded COnditioned output 100% from input images

[soft zealot]

oh god that was dire

[strange mist]

oh I like the image 😅
the show is bad, but the costumes are kinda good. gritty and grounded.

[soft zealot]

I watched season 1 then stopped

[indigo carbon]

yeah, but when you don't give it any prompt it makes SDXL's incoherent instinctual image

[strong field]

Without clip text encode you get random tho, could be unrelated to source images because of the way it works

[indigo carbon]

just a tile of fruits and leaf

[strange mist]

I think I watched a bit of season 2 as well.

[strong field]

This sounds like the right direction

[eternal fog]

Use the ConditioningZeroOut node

[soft zealot]

a few more variations onthe grittier looks. I like th eidea of a painted on Domino mask and no cape, its mor e"functional"

[indigo carbon]

is it a built in node? or a custom one?

[eternal fog]

Pretty sure it's built in, I didn't download anything to get it.

[indigo carbon]

huh, that's also dependent on text though

it wants a tokenized prompt and outputs conditioning

[eternal fog]

It zero's out the text conditioning, so it sends zeros, rather than an "empty" token

[indigo carbon]

ohhh, nice! I'll try this rn

[strange mist]

are you prompting something for the facial expression? 🤔

[soft zealot]

there was a response last night from , damn it forgotton who now from the SAI team) about the Zeroing node and waht it does

[eternal fog]

I'm trying to figure out the best way to use a negative prompt as well as using Clip Vision, without the negative massively overtaking the prompt, as the text prompts have way more strength than the image prompts

[soft zealot]

depends on if you think its "cute & sexy" lol

NB dont forget that im using the ReVision Technique so its also influenced by the 2 input images

a cute and sexy picture of The BatGirl in aan aggresive fighting pose (clenched fists) in a dirty, dritty street,she is wearing a head covering mask that has pointed ears on top of her head,she is wearing a black cape that is blowing in the wind, she is wearing a dark blue one piece leather suit with black fabric inserts , she is wearing with flat soled boots
,art style by (frank frazetta:1.3)

[soft zealot]

FWIW & IMHO with SDXL you can keeep or completely omit the negative prompts most of the time

[strong field]

True

[eternal fog]

Yeah it's for things like, it always seems to make women with earrings, so I want to remove them with a negative. But at the moment it completely changes everything in the image.

[strange mist]

yeah no, that is a fierce expression. I wouldn't want anyone to give me that look 😁
image 2 has a high strength and it has a sort of determined expression. that has to be the source.

[strong field]

Does a node exist to retokenize clip ?

[soft zealot]

meanwhile a ReVisioned generation with 0 textual input

[strong field]

Lol I guess that’s kinda related

[soft zealot]

same seed but with added style info

[strong field]

Better

Maybe it’s best to use weights @indigo carbon

[soft zealot]

Notem the way I have it confgure in my workflow is that the ClipVISION ddta is to BasePrompt G Positive only

Dont care if that the right way its the way I'm using it 🙂

[hardy cipher]

I wing it and do what I want

[strong field]

That’s clever

Lol is there a right way? Only comfy way.

[untold totem]

is there an xy plot that works with SDXL in ComfyUI?

[hardy cipher]

wouldn't any xy plot work with sdxl?

[indigo carbon]

ok wtf, it got way more coherent when weight is 3

[hardy cipher]

I think it depends on what your input images are

[soft zealot]

Overview of the data flow in my workflow

Orange is Text
Yellow is COnditioned
Blue Are Switches to siwtch between Normal, NormAl + COntrolNet, Normal + ControlNet+Revision or Normal +ReVision depending on where they are thrownsiwtch

[steady grove]

https://civitai.com/models/31126/imagesgrid-comfy-plugin-xy-plot i used this once. worked ok

[strong field]

Damn nicely done, you always have great workflows

[soft zealot]

Cheeers, theres other stuff in there as well to do with textfile generation but thats a simplified view

[untold totem]

I saw that one, but i didnt quiiiite see how to integrate it with Searge

[soft zealot]

Scrolls out firther and turns lines back on lol

[indigo carbon]

nah, it looks like clip_vision isn't the way to fuse images- even with 2 identical images and CLIPZEROOUT the final image had nothing to do with the first 2

[steel mist]

Why do all these mega workflows have options to load a separate vae? There aren't any available on civit and it seems like it's always just best to use what's in the checkpoint

[steady grove]

used to use the straight lines but i've gone back to splines because i canmore easily tell where lines are coming and going. the straight ones all blend together. makes it really hard to follow with an eye

[eternal fog]

It doesn't really fuse the images, it's supposed to take the concept of them. I don't really understand how it works, but it can get interesting results. But I don't think it can "fuse" images.

[steady grove]

what about revision?

[eternal fog]

Clip_Vision and Revision are the same thing

[soft zealot]

remind me, how many different Seeds can SDXL potentially use?

[indigo carbon]

2.1 trillion (my educated guess, that number is the max on most computing stuff)

[indigo carbon]

and trust me, tested multiple times

[soft zealot]

and you're basing your statement on a sample siize of? 'o)

[steel mist]

It's the size a FP32 variable can hold

[soft zealot]

Anyways from my sample ssize of 1 where I have 2 identical images it does appear to have generated an image related to them with Zeroed out Prompts

[soft zealot]

my question was rhetorical

[steel mist]

/shrug

[eternal fog]

For some reason people seem to be expecting that they can put 2 images in and it's going to sort of merge them together. Not that it's going to use that concept.

[indigo carbon]

did over 20 tests, none had anything to do with image.. if it will after the ENTIRE 2.1 trillion it will make a few good images I don't give a shit- I'm not going to wait for an infinite amount of time -_-

[steady grove]

less about the seed and more about initial noise. so possible combinations of all pixels as a starting point equals xres*yres*8*8*8 possibilities. 2^24 * the pixel count essentially.

seed just is used to generate initial noise

[eternal fog]

Seems to work fine for me

[steel mist]

I'm wrong anyways a 32 bit is 2 billion, a 64 bit is... a lot more than that 😆

[soft zealot]

if people want t merge / blend inages theres a node for that 🙂

[indigo carbon]

that overlays them without even using AI. it can be a base for redirecting to sampling though

[steady grove]

17,592,186,000,000 at 1024x1024

[soft zealot]

yes and I never said it was AI influenced did I ? @eternal fog merely pointed out (correctly) that some people ar ecionfused about even at a basic level ReVision works.

[soft zealot]

bottom line its a shit load

[steady grove]

when i looked at joe's revision post it showed a lot of multiple images being used

but these 20 tries shows it can't do it i guess

joe must've done it wrong

[indigo carbon]

what does "ecionfused" mean?

[steady grove]

he doesn't understand it doesn't work

[soft zealot]

TBH I dont really care how it works, all I know is based on my BatGirl images its prodicing sme rather nice images that I can see are tied to the prompt * source images

[steady grove]

my internal typo recognition figured it out. maybe you're just ecionfused

[soft zealot]

it means I have sausage fingers and the girls love me ;o)

[steady grove]

daddy, would, you, like,

some sausage?

[soft zealot]

I cant write the word I want to write as its banned so I'll settle for the one that sounds like a snack bar instead.

snicker

[steady grove]

big ol dongus fingers

[soft zealot]

and yet I can happily wirt Scunthorpe

[steady grove]

wirt

[indigo carbon]

oh damn, didn't notice I guess I am ecionfused, and maybe even and maybe even ar

[soft zealot]

which for those that dont know is a seaside town in the North East of England

https://www.google.com/maps/place/Scunthorpe/@51.6812787,1.5814547,7.25z/data=!4m6!3m5!1s0x4878e14c66603897:0xf5db46553ac8b350!8m2!3d53.588646!4d-0.654413!16zL20vMDFwcjZu?entry=ttu

[eternal fog]

Revision does do odd things sometimes though. And the weights seem to matter a lot.

Like? lol. Not sure how these concepts exactly merge.

I guess maybe, dirt on the ground, soldier(armor/space), but not sure where the tiger comes in

[soft zealot]

note some people would argue that as its South of the Humber its isnt the North East

[eternal fog]

No one cares about Scunthorpe anyway

[indigo carbon]

I think the better approach is to interrogate an X amount of images and then fuse the tokens- seems more reliable

[soft zealot]

well this is true, it is a dump

[eternal fog]

This isn't using tokens from what Dustin said yesterday.

As someone was asking how to get the tokens out that it uses.

It does something on a conceptual level, rather than just generating tokens from the provided image.

[untold totem]

I guess I'm also curious, what is everyone's favorite SDXL workflow with comfyui? I've tried Searge and Sytan

[indigo carbon]

yeah, but for consecutive image fusing the best way is likely interrogate->token merge->conditioning->sampler

just one piece is missing for that

[eternal fog]

No idea, once you start getting into the details and all the maths that goes on for generating the images my understanding stops.

[soft zealot]

well ask a silly question, of coutse I'm going to say my own (which starte dlife as being based on @high skiff s and then "eveolved a little lol

[high skiff]

me

[tepid surge]

nice lighting

[untold totem]

What's it called? And can I integrate an xy plot with it?

[untold totem]

same question, i didnt quite see how to do an xy plot w/ yours

[untold totem]

for ref, i'm just trying to have a side by side of several loras i've trained

[high skiff]

I am not sure I am in the right headspace to help right now, I am really anxious right now. My mother and I are just 3 miles away from the direct path of Huricane Hillary right now

[soft zealot]

I dont actively publish it cos I get irrated by inane questions espeically when people ask where do I get such and such a node from even though Ive included a credit & notes file with all the information in.

I do howvwever leave it in all images I post here (that have been generated in Comfy)

[untold totem]

ooof, no worries at all. we're in LA, so not as bad. stay safe

[high skiff]

we live 120 miles away from the ocean, at a mile in elevation in a desert, and we are on expected emergency evacuation watch right now

[soft zealot]

tie to batten down the hatches

[high skiff]

we live in the high desert, but we are only 3 miles from the projected center of its path

[high skiff]

we are all hunkered down right now, hoping that our elevation and our mountain bowl shields us, but the local mountain communities were just emergency evacuated

This is the first time in the history of California this has happened

[soft zealot]

@high skiff just as a point of reference the furthest any point inthe UK is away from the sea is 84 miles (or it might be 70 depending on the source)

[high skiff]

Hurricane Hilary is expected to reach all the way past Las Vegas, its insane

[eternal fog]

We also don't really get hurricaines

[untold totem]

neither does socal.... but here we are

[high skiff]

Its not that socal "doesn't really get hurricanes", its that we never have

[indigo carbon]

idk what y'all use, the workflow I made does stuff like this in ~14 seconds on average

[high skiff]

only thing we have ever had was a tropical storm in 1931

[soft zealot]

tell Michael Fish in 19827that lol

[untold totem]

yeah, its wild times

[high skiff]

Hilary just got downgraded to Cat 3, so I hope so much for San diego that it keeps going down, cause if not, there are going to be unprecedented casualties

[eternal fog]

Before I was born, doesn't exist clearly.

[soft zealot]

Point of interest , whats the difference berween a Typhoon & a Hurricane?

[untold totem]

location i believe

typhoons are near asia?

[soft zealot]

I would have been 23

[glad grove]

why was he a fish

[soft zealot]

probably about the same age (or older) than many of yo u here are now 🙂

[soft zealot]

because his dad was a Fish, and his Dads dad before that

[eternal fog]

You are older than my Parents. And I'm not "young" lol I'm 32.

[indigo carbon]

also, @high skiff, @strong field successfully compiled 3090 AIT modules on windows, they can probably give you the module

[soft zealot]

hey I have 2 grand kids and 3rd on the way

[high skiff]

OH?

Yippie!

[indigo carbon]

yeah, we figured it out

it gave them an extra 1.2-ish it/s (from precompiled)

[high skiff]

thats dope! I would really appreciate it

[soft zealot]

now all we need is an AIT module forth e1080ti and oither cards that would benefit form a little boost more than a 23090 does :p

[high skiff]

I am trying to distract from the huricane fear ATM

[strong field]

Not at comps but I’m definitely send when I can

[high skiff]

not sure non tensor cards will benefit

isn't AIT based off Tensor RT?

[soft zealot]

it was mostly tongue in cheek 🙂

[indigo carbon]

you wish, TensorRT is SO unfriendly and less beneficial for performance

[high skiff]

I thought AIT was just a better optimized and made version of Tensor RT

[indigo carbon]

plus, TRT modules can't move from CPU RAM to VRAM and back while AIT modules can

plus bigger boost

[soft zealot]

so theoretically it may be possible at some point to compile it for older cards such as the 1080ti ?

[untold totem]

@high skiff Hurricane Hilary has weakened to a Category 2 storm, so that's good news

[high skiff]

much love <3

[steady grove]

https://huggingface.co/stabilityai/control-lora/resolve/main/samples/revision-sample.jpeg maybe the documentation is wrong

[high skiff]

just 40 minutes ago it was demoted to 3, not 2

[upbeat summit]

Clip(Re)Vision AIT when?

[high skiff]

It was 4 yesterday

[untold totem]

10 min ago it was down to 2 according to cnn

https://www.cnn.com/us/live-news/hurricane-hilary-path-08-19-23/index.html

[glad grove]

maybe someone will do it in the future (soon)

[steady grove]

i thought with storms, cat 4 was highest

[high skiff]

it is

[glad grove]

there was a little rascal who was cat 5

[high skiff]

for hurricanes

oh right, there was 1 cat 5

[soft zealot]

serioulsy it would be good if older cards could be given a boost

[steady grove]

oh we've got into cat 5 storms now. nice

[indigo carbon]

also AIT is universal and doesn't need some stupid conversion like ONNX for TRT, you don't have to rebuild the code if converting different architectures. AIT learns the specified model's architecture and makes a ~100mb MSVC module, making it not checkpoint specific

[high skiff]

Yeah, I get that its much better, but I swear I was told it was the same base idea implemented in a much better way

[indigo carbon]

nah, it's just inspired by TRT- but executed better and relies on much more universal languages

@high skiff if you want I can guide you how to compile yourself in DM. it's somewhat simple after figuring it out

[high skiff]

I am too scatter brained for that at the moment, but thank you

[steady grove]

the thing with the pacific coast is it has a lot shorter of a shore till the continent's edge. Deep ocean is really near land. Atlantic has huuuge shores. the sudden cutoff near the pacific coast really changes weather patterns. systems make landfall like it's a 45 degree incline suddenly.

its the main reason we don't get a lot of massive storms hitting the west coast and systems are a lot more localized

[indigo carbon]

you can do it right now infact, but doubtful it will profile any better than without optimization, but you can try it

AIT is very new tech, so it's just common sense it performs best in newer cards such as 3000 series and 4000 series

but if TRT does anything for a 1080ti, I'm confident AIT will do better when compiling on your own

heh, I like SDXL's waterguns

[high skiff]

moist

wet, even

[indigo carbon]

is water wet?

[native knot]

😉

[high skiff]

there is no wet without water

[indigo carbon]

yeah, that's true. but the definition of wet is "x with water on it"

does water have water on it?

very philosophical question

[high skiff]

[soft zealot]

I may have to look into that then

[marble kindle]

why is my revision not working properly can anyone help??

[autumn forum]

whats it doing?

[marble kindle]

i give the 2 images and the generated image isnt related at all

[autumn forum]

https://civitai.com/models/130726/revisionxl-with-additive-prompting

use my workflow

the workflow comfy has doesnt really explain how to us it well. this workflow solves the issue.

[unreal plover]

Issou

[marble kindle]

ill try yours this is mine, using stabilities workflow weird output

[eternal fog]

Not seen that name used before, but yes

[autumn forum]

you are using zerouts so it should work. i mean try mine if you want. clipvision is interesting some concepts dont blend well. its not really going to add the 2 together but it takes elements from both and creates a brand new image out of it.

[eternal fog]

yeah it can do odd things sometimes

[marble kindle]

oh ok

do you normally use strength of 0.5 for both images

[eternal fog]

Depends

[autumn forum]

ehhhh no. lol

put whatever you want

[indigo carbon]

also, am I the only one that gets weird fruit drawing tile when doing no prompt on SDXL?

[eternal fog]

I do wonder though when you have a zeroed out conditioning, why there are differences between 0.5 on both and 1.0 on both, what is that strength used against

[autumn forum]

its a multiplier of sorts as comfy mentioned.

[eternal fog]

that isn't safe

[boreal gate]

SD cannot use http links as reference in a prompt. ?

[autumn forum]

sd doesnt use the internet

[eternal fog]

The random car mashups it makes when you use Clip Vision are interesting

[ionic gulch]

who wants to test my 4in1 workflow (txt2img, img2img, controlnet-candy, controlnet-depth)?
https://i.imgur.com/D9Yz1DS.jpg

[boreal gate]

hand🖐️

[boreal gate]

wow