✨｜sdxl

[timid garnet]

ZavyChromaXL on civitai

https://tenor.com/view/mr-bean-wink-wink-winking-rowan-atkinson-gif-15388297

[rugged crow]

thanks

[tepid surge]

what do you mean? model? xl base

[timid garnet]

still in beta tho. 1.0 soon

[rugged crow]

a lot happened since the last time i was here, and i never caught up to control net, trying to figure that thing out with my trt

[tepid surge]

oo control nets fun

[rugged crow]

yea

[timid garnet]

Man, the deforum discord is going wild, escalating so fast. Let me find you an example or two
(CN related)

Everything is going too fast at this point

[rugged crow]

a lot to digest tho

[timid garnet]

or this

[rugged crow]

idk how u do that yet

[timid garnet]

A last one

Isn't this cool !?!

not like all these tiktok bad videos

[rugged crow]

or how u do that thing they demoed with the lion where u move its face by dragging it with the mouse

[timid garnet]

Curious about this.
I'm aware of Deforumation, to pause/rewind your generated video, and change everything, like the direction to go during the video

Related to Deforum, in the a1111 webui ofc

Deforum is the way, for sure ! Love it

I'm more for pure txt2img tho, not edited videos like above

[tepid surge]

all i have is this atm though my basic animation graph is up and running

sadly there's no 3d transform nodes for Comfi yet

though FizzNode adds the Deforum style scheduling which i love

[untold crescent]

this about speed comparison between different webui's. of course the speed matters here. more images you push through, faster you iterate. and mind you this is on exactly same system. comfy literally generates 2 images in same time you gen 1 image on auto1111

so it's pretty dramatic difference

[rugged crow]

apparentl u can get a depth map from a 3d object and generate on it? i'm trying to do that but keep getting sidetracked catching up to stuff

[untold crescent]

controlnet depth yes

but we don't have controlnet for SDXL yet afaik?

[untold crescent]

oh wait you wanted to feed depth map yourself

[native knot]

Yes we do.

[upbeat summit]

it's evolving

[untold crescent]

ah, I see the repo now

[rugged crow]

yea

[untold crescent]

from week ago

can we do that on SDXL?

[rugged crow]

idk

[untold crescent]

we had that option on SD 2.0

[rugged crow]

didn't touch sdxl yet

[untold crescent]

but nobody seemed to use it

[timid garnet]

What's that ? 🤔

Can't find FizzNode

[timid garnet]

I'm using parseq for advanced s**t, or my 'deforum chatgpt prompter' for travel-like videos lol

[tepid surge]

beep boop, it's a custom node set for Comfi (seems i forgot the s)
https://github.com/FizzleDorf/ComfyUI_FizzNodes

[timid garnet]

Ho, okay

Never heeard about this, looks like deforum, same syntax

interesting !

Dang, i'll be way less dumb tomorrow

[halcyon tusk]

deforum under comfy

[tepid surge]

Wizz basking their wisdom

[timid garnet]

I'm mitigated on this one

https://tenor.com/view/velociraptor-push-cat-love-you-gif-12642577

[timid garnet]

https://tenor.com/view/obese-monkey-fat-monkey-summer-belly-eating-lettuce-summer-look-gif-13014350

Wizz be like :

[halcyon tusk]

https://tenor.com/view/supfkers-supfker-mfer-supfkers-moon-supfker-spiderman-gif-25529213

[timid garnet]

https://tenor.com/view/i-want-it-too-gif-15165305

[rugged crow]

btw it seems that i'm getting 29 its without xformers, ~24 with xformers on trt

[timid garnet]

Ho C'mon !!!!!!!!!

[untold crescent]

@rugged crowfound this https://www.reddit.com/r/StableDiffusion/comments/1130ljr/overriding_the_depth_estimation_in_controlnet/

[timid garnet]

I'm at... like.... like......... 1.5 it/s

%&£$# (insert insanity here 😭 )

[untold crescent]

no idea if that still works on SDXL but it just might

[rugged crow]

512x512 dog

[tepid surge]

poor monkee xnx

[halcyon tusk]

what are you rendering 16K ultrawide UHD++ ? if that's for 512 there's something wrong

[rugged crow]

fr

[timid garnet]

Man, i was at 14+ with my 'lil 3060, tensorrt and small prompts 😭

[rugged crow]

lol phat

[timid garnet]

768*1216 with sdxl 🥲

21sec per pic, at 30 steps

[scenic root]

Hi. I'm trying to make seamless pattern in SDXL installed locally. Without success 😦 I saw here that is possible to make such images.
I'm using "seamless pattern" or "seamless texture" or "seamless" alone. Even "style tile texture".
Can somebody help me with this issue?

[rugged crow]

it's the normal sd

[timid garnet]

But it's fine, i won't cry, i started with an AMD rx480

THAT was painful

[rugged crow]

are there any updates with trt, it's still super limiting, nevermind the retraining itself

[halcyon tusk]

OK since sdxl was trained at 1024 x 1024, try to get a pixel count that is exactly the same with your aspect ratio

[untold crescent]

auto111 and sd.next has "tiling" option for this.. I'm sure comfy has some setting for it too

[rugged crow]

isn't there a tile option?

[halcyon tusk]

it should total 1.048.576 pixels

[untold crescent]

you don't make it tile using prompt but setting

[timid garnet]

https://tenor.com/view/phrasing-archer-gif-8049838

*Rephrasing

Try again like i'm 5

😂

[rugged crow]

ah yes, decimals

[timid garnet]

Okay, i see what you mean. But what this ratio is standing for, related to 🤔

[timid garnet]

Hey, sorry to burry your message !
Here is what i get, can you elaborate ?

[untold crescent]

can we see timings in comfyui?

[scenic root]

@untold crescent @rugged crow The only options in ComfyUI fo SDXL i can see is "Batch count" 😄 How to use the option "tile" in this UI?
P.S. I've searched even for a node fo it, but not found.

[halcyon tusk]

well you are not making square images, but the pixel count is important, should produce better results and go a bit faster if you maintain the pixel count

what GPU do you have?

[timid garnet]

This finally explain why i see people talking about precise ressolution here and there

Thank you Sir ! 😘

I got +4 IQ today

Amazing

[timid garnet]

RTX 3060 12 gb

Bought it just to run a discord bot, dedicated

Was the best choice to me, like 5 months ago

[untold crescent]

@scenic root https://www.reddit.com/r/comfyui/comments/150ktna/creating_tiling_seamless_textures/ apparently there is no built-in option.. but you can modify the source code to force it to tile... but this isn't ideal unless you only ever gen seamless textures with this 😄

[halcyon tusk]

that should be fine.. it/s may not be the best but I think 1-2 is too low

[timid garnet]

Curious about your recommendation, was like 300-340€ atm, don't remember

[timid garnet]

This is the kind of card being r***d every driver update

The gain is for the new generation, pretty crazy, as i saw

[rugged crow]

i never touched sdxl, i just figured it has to have one, i was surprised you tried to get it with prompting

[scenic root]

I don't understand what to elaborate on this...

[scenic root]

Thank You. I'll check this out.

[untold crescent]

I'd just use sd.next or auto1111 for seamless tiling unless you need a ton of them

[timid garnet]

Like a screen of what you get, for example, lol

[untold crescent]

they both have checkbox

I'm frankly surprised comfy doesn't have it

someone could add that option and PR it I suppose

[rugged crow]

when do these with the entirety of shrek saga

[scenic root]

heh... i'm using prompt to tile because it's no MJ (it has option --tile) and it's no bot on this Discord. Here I see option to tile, and it work. So, i thought than must be a possibility to achive it on local instalation of SDXL.

[tepid surge]

yees

[timid garnet]

Look at this one, made with deforum by a dev, experimenting.

Like, 0 noise

[rugged crow]

there is some noise

[timid garnet]

https://tenor.com/view/meh-mediocre-so-so-kinda-maybe-gif-13935570

[rugged crow]

ok

[rugged crow]

this dude replied to me lol

[timid garnet]

These guys are mathematicians at this point, complexe stuff

[uncut fiber]

so with --xformers you got performance down

[timid garnet]

lol, so true

[rugged crow]

i did, yes lol

[timid garnet]

Here is the settings, if you wanna try. For the video above

[uncut fiber]

it is weird, did you delete --lowvram?

[timid garnet]

🤔

What sorcery is this

[soft zealot]

Yes I fully understand the matths but why do so many people make such a big deal about how many images they can generate per day? The only people benefitting from this are SSD manufacturers lol

[untold crescent]

huh? to get good results with SD, you iterate... alot

[timid garnet]

a1111 new version will use optimization automatically, but tranformers still need an argument in the start batch.
So without, it should use the first defaut one

[untold crescent]

of course the speed matters

[rugged crow]

not sure if u remember, u helped me get trt a while back, i was getting 24ish with xfoemrs. with this trouble of not being able to use trt with control net i found this out by accident, 20 minutes ago

[timid garnet]

All my knowledge is here lol

[timid garnet]

omg, no way !

I doo now

[rugged crow]

i was always getting 24ish w xformers, lowvram seems to have absolutely zero impact on performance

[uncut fiber]

@rugged crow you can try instead of --xformers, which i believe not working because not installed properly, using this
--opt-sdp-attention

[timid garnet]

I did a PR to the tensorrt for the webui lol, to batch files, this repo was deadborn

[uncut fiber]

it has bad impact on performance

[timid garnet]

Buy a real card, stingy you !

[rugged crow]

yes, does it work?

[timid garnet]

Sure

[rugged crow]

wdym lol

[timid garnet]

But deprecated now lol

[rugged crow]

rip

[timid garnet]

MOOOOAAARRR vram

https://tenor.com/view/star-wars-more-gif-21856591

[untold crescent]

it was all fun and games when SD still generated image in few seconds using gaming capable GPU 😄

now you need 3090 or 4090

[timid garnet]

😭

[rugged crow]

really never heard any details about trt except from u. heard about them but it seems to be a black hole. why does it break my controlnet, why does it break other models if i pick the old ones, why is it so limiting in general. does it work on sdxl? no answers

[rugged crow]

i've a 44070

[untold crescent]

and nvidia is still cheaping out on their GPU VRAM

[timid garnet]

If Stable Diffusion could win the """"model war""""

[untold crescent]

I guess to sell those ridiculously priced models

[timid garnet]

What model is this ?

Your optimization 'issue' is new to me

[untold crescent]

4070 would be nice card had they put more VRAM on it

but still only 12GB

[tepid maple]

how to avoid "RuntimeError: CUDA error: misaligned address " ?

[rugged crow]

thinking of manually changing the ram but i'm not that smart and idk anyone who did that yet

[timid garnet]

I'm pretty sure this model will exist at some point, tho

[rugged crow]

44070

[timid garnet]

Year 2036

[rugged crow]

this means in 88 years

release cycle is every 2 years, on a new node

[timid garnet]

Don't underestimate the marketing team !

[rugged crow]

true lol

[timid garnet]

https://tenor.com/view/william-shatner-james-kirk-captain-kirk-star-trek-maximum-speed-gif-16487705

[rugged crow]

what does this do

[timid garnet]

Well, nothing in the new versions, cause using automatically the best one. Also available in settings/optimizations

[uncut fiber]

it is for gpu 3xxx or 4xxx similar to xformers, but cant be used both, so one or another

[timid garnet]

But before, yeah

[rugged crow]

so it's pointless now?

gona try anyway

[timid garnet]

transformers is the only last one needing an argument in the start batch, at this point, yes

[tepid surge]

Check your clip dimensions, least something like that happened to me though I on comfy xnx

[timid garnet]

Use the settings/optimization combobox, and see in the console

[tepid maple]

I use sdxl recommended sizes

dont give any clip change all default

dont use any lora too. very plain prompt and negative prompts only

[tepid surge]

yee it was long shot

but i did shoot!

[tepid maple]

is there log file for 1111

it happens after like 10 generation

[rugged crow]

🏹

[tepid maple]

I am automating with the sd-webui-agent-scheduler

[tepid surge]

チーズ!

[timid garnet]

meh focus, but awesome. I lost one eye on it

[timid garnet]

ah

[timid garnet]

🤩

What model is this ?

[tepid maple]

download it it is raw png

[timid garnet]

I need to test some other models, but only if recommended

[rose abyss]

Hi guys, please I need help refine a prompt request, which channel should I join?

[tepid maple]

#📝｜prompting-help

[tepid surge]

i need to set up a face fix pipe

casue melty face

[timid garnet]

https://tenor.com/view/phrasing-archer-gif-8049838

[rose abyss]

Thanks a lot

[rugged crow]

discord strips metadata

[tepid maple]

oh

[timid garnet]

lol, thinking about something. Let me find it.
I named this masterpiece "Life of a Woman"

[tepid surge]

counter argument: goo

[uncut fiber]

@rugged crow not at all

discord doesnt strip metadata

[timid garnet]

https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/d1c49a24-a03f-421e-ab96-6926b838b09c/transcode=true,width=512,optimized=true/20230508012316.webm

[uncut fiber]

from comfui png at least 🙂

[timid garnet]

SD 1.5, months ago

[tepid surge]

the spoook

[timid garnet]

'Life of a normal woman'

[tepid maple]

is there way to lower the sdxl vram usage so it doesnt memory leak to crash 🙂

[timid garnet]

--medvram, but slower

[tepid maple]

tried still uses 11gig again

[tepid surge]

or --lowram

[tepid maple]

didnt tried lowram testing

[tepid surge]

what 11 gig?

[tepid maple]

I have 3060 12gig

[tepid maple]

sdxl using like 11,5 like stupid

I bet it is crashing because of this

[rugged crow]

multi-dollar company making 70 series cards with 12 gigs

[uncut fiber]

can anybody dragdrop it in comfui?

[tepid maple]

I bought it soooo cheap

[rugged crow]

what card

[tepid surge]

awe, i even had a graph for back up:

[tepid maple]

3060 12gig

[rugged crow]

4070?

ah

ye

mine is 4070 and also has 12 gigs 🙂 multidollar company

[rugged crow]

idk what that is

[timid garnet]

Recent pic

[tepid maple]

more than 12 is overkill tbh

[rugged crow]

for sd? no lol

[rugged crow]

u look stunning

[tepid surge]

ah ha! as epected @uncut fiber

[tepid maple]

you have no idea how good iamges I can produce

no
idea

[uncut fiber]

so you see 🙂 @tepid surge thank you!

[timid garnet]

https://tenor.com/view/seth-meyers-right-am-i-right-totally-same-gif-9191853

did you see the prompt

[rugged crow]

does it have to do with ram tho?

[uncut fiber]

that red thing is performance booster for 3xxx and 4xxx

[rugged crow]

do we still need 24gb for training?

[tepid maple]

nvidia making us blevie we all need 32gig vram for future but no. models will be optimized in time .no need to rush

[timid garnet]

https://tenor.com/view/showme-show-me-morpheus-gif-5144220

[tepid maple]

***except training

[timid garnet]

🤩

FEED ME

[tepid surge]

i havent seen any meta data wiping nodes but i did want to try something so i'll do it rn since we bot here (will ping when done)

[rugged crow]

8 gigs of gddrx6 costs $30 lol, no excuse

[uncut fiber]

o.k.

[timid garnet]

Are you on civitai ? Show me your profile image gallery ((:
https://civitai.com/user/wizz13150

[tepid maple]

agree

[timid garnet]

Single batch of 4 images.
Only one word as positive prompt

But i'm not a creative guy, badly need inspiration 😭

[rugged crow]

u actually look french in that profile

[tepid maple]

well just visit photograpyh sites randomly you will find some inspirations 100%

[timid garnet]

https://tenor.com/view/enojado-rooster-cockerel-gif-8701701

[tepid maple]

specialy those very artistic photography sites

[timid garnet]

Of couuuurse 🥖

[tepid maple]

not like cheap stock sites

like 500px

[rugged crow]

le coq

[tepid maple]

try with that

[timid garnet]

et Allez l'OM ! ((:

[rugged crow]

what's om

[timid garnet]

Lately, i'm producing many many piscture, with words like 'Creativity', or using the model's name only.
You should try, it's like unleashing the model, doing what he really should do

Probably an illusion, but here are the pic .. pretty realistic for a realistic model

[tepid maple]

less is more works I know

I dont use lora or embeddings. just very clean prompts and negatives

[timid garnet]

Man, this thing is alive !

[tepid maple]

medvram works lookslike fingers crossed

lets see if it will crash with 100 iamge on the Q

[timid garnet]

https://tenor.com/view/excited-so-drums-gif-11982656

[tepid maple]

only thing I didnt optmize is good ssd

gotta buy those 1tb ssd

[ionic gulch]

txt prompt:
https://i.imgur.com/TocE7Er.jpg
txt prompt + img2img:
https://i.imgur.com/B0uHj5b.jpg
txt prompt + candy mode:
https://i.imgur.com/AWajTcf.jpg

revision prompt + txt2img:
https://i.imgur.com/wt76vef.jpg
revision prompt + img2img:
https://i.imgur.com/UzoWZNT.jpg
revision prompt + candy mode:
https://i.imgur.com/fCZ0llE.jpg

[tepid maple]

nice

voodoo makeup 😄

love it

[rugged crow]

Not sure how I installed it the first time 8 did, but any image larger than 512 would make it spillover into system ram and cause it to spend 2 minutes on an image

[tepid maple]

my guess is that too. I am one shotting high res in first generation. probably strugling in some prompts. randomly

now people will say why not use scaling. well it is slow.
this is better
1344-768

[rugged crow]

It was also I think doing some weird caching or just not clear ram and spend even more on some prompts, maybe just reinstall

[timid garnet]

Yeah, got one for like 80e, pretty awesome (wasn't up to date at all regarding my disks)

⚡

[uncut fiber]

I have SDXL with 3070 at 3it/s, you should have at least 5

[tepid surge]

Light but no cigar @uncut fiber , i'll put a feature request after digging around a bit :3

[tepid maple]

who ?

[cedar lark]

gen speed depends a lot on sampling method and vram load with loras/controlnet

[uncut fiber]

4070 owner

and resolution 1024x1024

[rugged crow]

me? yea i think i'll just skip xl lol

[tepid maple]

I wish 1920x1088 was stable

tried alot but it is hallucinating alot

almsot crashed again

mfr....

[timid garnet]

[tepid maple]

when generating it is using 8gig. there is like temporary state I dont know saving stage ? vae stage ? that makes it jump

[cedar lark]

I can gen in that res just fine, you just need to work on prompt

[uncut fiber]

i dont understand?
that node is 20%-80% boost

[cedar lark]

example

not upscaled at all

[tepid surge]

i talking about the meta data thing cx

[tepid maple]

tried lots of simple prompts it is definitly halucinating mroe than other recommended sdxl sizes

[uncut fiber]

o.k. they are there, discord doesnt deleting them

[tepid maple]

recommended sizes are the best.

[rugged crow]

it's doing it again, not releasing memory lol

tried a 1024x1024

[tepid maple]

scary

[tepid surge]

when you're not allowed to forget cx

[rugged crow]

what's that

oh it's an emoji

[tepid maple]

[tepid surge]

:3

[rugged crow]

why did u reply and then unreplied

[uncut fiber]

what realy drive me nuts, is water in water 😄
Wrong picture

[rugged crow]

i made this w midjourney and flowframes

i don't need controlnet

[tepid maple]

for this yes you dont need

wtf I didnt expected this

[ionic gulch]

JPS-4in1Rev-v2.0 workflow (examples see above) - TXT or IMG Prompt (Revision), TXT2TXT, IMG2IMG, Candy, Depth, Lora Support, 70%/80%/90%100% Base Refiner
https://github.com/JPS-GER/JPS-ComfyUI-Workflows/raw/main/JPS-4in1Rev-v2.0.json
(all required nodes are available through ComfyUI Manager)

[waxen berry]

hope yall noticed the new models in the Bot are crazy good right now

[tepid surge]

oh wow what soft lighting

[rugged crow]

https://github.com/hnsywangxin/controlnet_stable_tensorrt

ok

[tepid surge]

the only ones i have from testing one of the pixel art models is nsfw cx

[tepid surge]

yooo ps1/ dk graphics cx

[shy kelp]

how do you made this, any flow?

[wet nacelle]

I'm just blending images.

[tepid surge]

mack wizard, you're on another lvl then me cx

[wet nacelle]

It's actually really simple. I'm adjusting the image weights.

Also with a prompt

[tepid surge]

'm still wrapping my head around face restoration cx

[shy kelp]

can someone help me, this is my workflow, any tips to get better photo

[shy kelp]

im using two ksampler

[tepid surge]

average space agers cx

[tepid maple]

comfy still too complicated for me

dunno

and I dont know why stabiltiy choice that path instaed of forking 1111 similar ui

user experience 10x better.

at one point in time someone in the lead said "fuk it lets go with nodes"

I wanna know who is that person

[shy kelp]

yeaa brooo

frfr

i would beat em

[tepid maple]

nah I woudl jsut ask. do you regret ?

😄

[wet nacelle]

they are the smartest one in the group then. You sound like a lazy mothafucka that chooses not to learn new things.

[native knot]

Because nodes are more versatile, easily changed, capable of more complex things with a simple UI.

It's more difficult than A1111, sure, but it's not that difficult if you spend time starting with the basics. Besides, if you want to just jump into creation with ComfyUI, get someone's image that they've posted, drag it onto the ComfyUI screen, and then boom...you have a workflow that works (assuming you have all the nodes). And if you don't have all the nodes, download ComfyUI Manager to have it help you identify what to download and just click Install next to the ones it tells you.

[shy kelp]

also true

[tepid maple]

me is the last preson you cna say that

haha

[wet nacelle]

why the fuck would they regret using a better system?

[tepid maple]

I look at the tools as UX

ux = user experience

and as a daily web developer who makes 20-30 projects a year. trust me I know couple thing

[tepid surge]

Yeah tbh i'm drinking up this node system, it teaching me a lot, kinda actually forgetting to other work i so in-it rn cx

[shy kelp]

i only have 4 gb vram, and i can still use to generate images on comfy, but not on a1111

[native knot]

If you like ComfyUI, you should also check out NodeRED for home automation, among other things. I sometimes use NodeRED in tandem with HomeAssistant...it's wonderful.

[tepid maple]

how comfy and that gets connected ?

[shy kelp]

thanks for the info

[tepid maple]

any easy way? docs ?

[ionic gulch]

nodes are fine, but the manager should be default and not require new users to install manually, also many math nodes, converters, etc. should be default and not require addons. for better user experience there should also be logic nodes like switches included in the core.

[tepid maple]

simple math nodes not default? didnt know that

[native knot]

I agree with you on this. But it's a work in progress, right? So like any other actively developed project, those things will probably come in time and they're available now with simple add-ons.

[ionic gulch]

at least not for all cases.

[waxen berry]

Prompt: film still, mexican girl, imperfect skin, in the middle of a town, Hasselblad X1D film, 50mm lens, realistic, sharp focus, 32k UHD
Negative: anime, paintin, cartoon, drawing, digital art, open mouth, blurry, Bokeh, bad eyes, artifacts, z-clipping, typo
Style: Photographic

[shy kelp]

agree

[half ivy]

bro

[stone fossil]

😉

[native knot]

I don't have NodeRED connected with ComfyUI. I'm making the comparison because they're both Node UI/UX.

[tepid maple]

if there is a good solution in the comfy for automation I would siwtch in a second 😄

11 is pain in the but

[wet nacelle]

I dunno what to warn you of here.

[tepid maple]

for now I use sd-webui-agent-scheduler
works fine but laggy ıdont know why. and UI bad. cant ready all details in window

[tepid surge]

that's a cool logo

[tepid maple]

yep

can I get that .png file pls 🙂

[tepid maple]

haha

[wet nacelle]

I thought so too!

[tepid surge]

that'll be -$0.40

here you go

cx

[tepid maple]

maybe I shoudl make similar logo for my new site

still thinking about it 😄

[half ivy]

tattoos look slightly better

[uncut fiber]

E.A. Poe by Gustav Klimt
i like how figurines vary by artist

[tepid surge]

me and my 4060 so sloow cx

[uncut fiber]

@tepid surge you should benefit from that AITemplate

[half ivy]

brrrooooo

how??

[tepid surge]

which one?

[vocal rapids]

how can I find the latent switch ? thanks

[uncut fiber]

That node, that was red, when you tried if it has metadata

[stone fossil]

I made a PCB lora for the lolz.

[ionic gulch]

if you use comfy manager you can select "install missing custom nodes"

[tepid surge]

to be in nature again

[half ivy]

a pcb lora is also another thing lolz

[vocal rapids]

I installed those missing custom node once, after restart the comfyUI, it still prompt me that latent switch is missing, is that the latent switch related to ComfyUI Impact Pack?

[stone fossil]

Circuit board styled.

[half ivy]

bumping a share?

[stone fossil]

I made 4 so far unsure or I should share. 🙂

Probably when time is there.

[tepid surge]

woof had to get coffee and pee! Mack the real zoomer cx

my slow ass xD cx

[half ivy]

living...

[tepid surge]

awe cute cheeks!

[ionic gulch]

you mean the 4in1 switch?

[half ivy]

ser pls, muh family

[wet nacelle]

https://tenor.com/view/wont-somebody-please-think-of-the-children-think-of-the-children-think-about-the-children-children-the-simpsons-gif-20004023

[vocal rapids]

yes yes , it's a great work indeed but I am figure out how to install the missing latent switch

[tepid surge]

the face structure kinda reminds me of Taliesin Jaffe

[wet nacelle]

This is just funny

It's really just a good way to make styles.

[half ivy]

tease me with a portrait of a cyberpunk ninja

[silent sleet]

/

[half ivy]

i need this lora bra

[stone fossil]

I did so far: hellraiser / borg / ice / pcb.

🙂

[quartz sequoia]

anyone having an issue with stableswarm it aint using the comfy workflow

[stone fossil]

Crazy consepts never die.

[ionic gulch]

should be this pack:

[tepid surge]

if anyone listening to some music too:
https://www.youtube.com/watch?v=ty0EezB5qkI

[tepid surge]

yo you wanna throw some promps or ideas our way? I'd be happy to gen stuff and explore at your intrest! xwx sounds fun

[vocal rapids]

yup I installed that already but it still fail to load "latent switch", nevermind I am gonna to re-install all plugin again to try, thanks.

[ionic gulch]

can you select it manually?
https://i.imgur.com/tAJmkqy.png

[tepid surge]

you peep following the uap hearing's and stuff? i am, excited

[half ivy]

im worried abt more fake pictures though

[wet nacelle]

That stuff is re tarded honestly. It's a show for the public. The actual information is being talked about in private.

[tepid surge]

ai generated radar metrics! xwx

[half ivy]

can we agree as a community to never make ufo stuff until its proven

[tepid surge]

fr fr ive been following Dr Greer for a while

[tepid surge]

plus... other things

[tepid surge]

xDDD

the ufo studders like an old car and just shits on to the street

[half ivy]

what if, teh aliens are in the diffusion models and we're slowly getting made used to their real presence

[tepid surge]

the reflections 👌

[uncut fiber]

hearse driver grave.

[native knot]

Made a clean nice quick text version of the resolution table in case anyone wants:

║                      RESOLUTION RATIOS                      ║ 
╠══════════════════════════════╦══════════════════════════════╣ 
║           Portrait           ║          Landscape           ║
╟────────────┬─────────┬───────╫────────────┬─────────┬───────╢
║ Resolution │ Decimal │ Ratio ║ Resolution │ Decimal │ Ratio ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  512x2048  │  0.25   │ 1:4   ║  2048x512  │  4.0    │ 4:1   ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  512x1984  │  0.26   │ 8:31  ║  1984x512  │  3.88   │ 31:8  ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  512x1920  │  0.27   │ 4:15  ║  1920x512  │  3.75   │ 15:4  ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  512x1856  │  0.28   │ 8:29  ║  1856x512  │  3.62   │ 29:8  ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  576x1792  │  0.32   │ 9:28  ║  1792x576  │  3.11   │ 28:9  ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  576x1728  │  0.33   │ 1:3   ║  1728x576  │  3.0    │ 3:1   ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  576x1664  │  0.35   │ 9:26  ║  1664x576  │  2.89   │ 26:9  ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  640x1600  │  0.4    │ 2:5   ║  1600x640  │  2.5    │ 5:2   ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  640x1536  │  0.42   │ 10:24 ║  1536x640  │  2.4    │ 24:10 ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  704x1472  │  0.48   │ 11:23 ║  1472x704  │  2.09   │ 23:11 ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  704x1408  │  0.5    │ 1:2   ║  1408x704  │  2.0    │ 2:1   ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  768x1344  │  0.56   │ 9:16  ║  1344x768  │  1.75   │ 16:9  ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  768x1280  │  0.6    │ 3:5   ║  1280x768  │  1.67   │ 5:3   ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  832x1216  │  0.68   │ 13:19 ║  1216x832  │  1.46   │ 19:13 ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  832x1152  │  0.72   │ 13:18 ║  1152x832  │  1.38   │ 18:13 ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  896x1152  │  0.78   │ 7:9   ║  1152x896  │  1.29   │ 9:7   ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  896x1088  │  0.82   │ 14:17 ║  1088x896  │  1.21   │ 17:14 ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  960x1088  │  0.88   │ 15:17 ║  1088x960  │  1.13   │ 17:15 ║
╟────────────┼─────────┼───────╫────────────┼─────────┼───────╢
║  960x1024  │  0.94   │ 15:16 ║  1024x960  │  1.07   │ 16:15 ║
╠════════════╧═════════╧═══════╩════════════╧═════════╧═══════╣
║                           Square                            ║
╟─────────────────────┬───────────────────┬───────────────────╢
║      Resolution     │     Decimal       │       Ratio       ║
╟─────────────────────┼───────────────────┼───────────────────╢
║      1024x1024      │       1.0         │        1:1        ║
╚═════════════════════╧═══════════════════╧═══════════════════╝```

[half ivy]

just gonna have to use 1.7 weights on pcb (kicks stone, sulks)

[wet nacelle]

https://tenor.com/view/yes-sweet-hell-yes-pump-it-gif-22136641

[wet nacelle]

Guy could be on Big Brother

[tepid surge]

(slides meme in)

[uncut fiber]

Those are big brothers

[uncut fiber]

And those are confidents 🙂

[native knot]

I adjusted the portrait decimals to follow the ratio instead of being just the flipped decimal, so now they're more "true".

[wet nacelle]

sweet

[uncut fiber]

@native knot you adjusted, still div by 16?

because if not /16 then it is wrong

[half ivy]

is this the maths i hear i need to learn?

[native knot]

What do you mean, exactly? All entries on the table are values divisible by 64 since that's what's required for SDXL.

[copper spire]

Hello sorry to bother you again hehe 😅 i just wondering can we run controlnet with collab because i have to put a lign in the terminal but i don't find how in colab

[uncut fiber]

o.k. just as you said you adjusted, if not only f.e. divisible by 8

[native knot]

I adjusted the decimals, not the ratios themselves.

[uncut fiber]

o.k. 👍

[native knot]

Do you think @visual glade would pin the post so others could see it easily? 😉 (Pretty please, comfy...)
#✨｜sdxl message

[uncut fiber]

weird they didnt train it on 576x1024

[shy kelp]

any prompt? suggest me some

[wet nacelle]

wha?

why bud

blud

[tepid surge]

lace

[native knot]

lattice

[wet nacelle]

jerma

jerma

[native knot]

some of my favorites:
lattice, fractal, bismuth, gunmetal, kaleidoscopic

[tepid surge]

jerma in lace playing tf2 with a bowl of salad xwx

[tepid surge]

He is scout cx

[shy kelp]

😂

[tepid surge]

@uncut fiber

xD cx

[uncut fiber]

😄

[tepid surge]

i want this as shader code

[uncut fiber]

This is upscaled and converted.

[tepid surge]

flower's a nice touch

[copper spire]

Anyone run sdxl 1.0 comfyui with control net on Google colab ?

[tepid surge]

isnt it just installing the webui and uploading the model?

like git pull, execute type something?

[copper spire]

I don't think so i will double check but I can't see the control net on my interface (sorry If I answer completly wrong I don't know much yet)

[tepid surge]

ditto, I dont really know how to collab work or how to run stuff from a command line

you mihgt be able to get it up if you look at how the other models get up and running, looking at the code, seeing what it mighg tb e trying to do and if it works for the xl model

[copper spire]

I put the canny and the depth in the control net but after I don't know how to finish the install

Okay I m gonna check that thank's for your time 🙂

[native knot]

Nope rope.

[tepid surge]

of course being! >u< hope it works out!

danger noodle

[hardy cipher]

looks too good in that one

[upbeat summit]

looks like your got inspired by that lora to do your own fan art drawings

[hardy cipher]

well I figured out how to make it plug and play, lol

[tepid surge]

the face cx i'm dying

[hardy cipher]

Astra, they look like they're on the wrong kinds of drugs

well maybe fun for them but I'd want to avoid them

[hardy cipher]

(((meth))), (((bath salts)))

but as long as they're having fun

[steady grove]

dont know if sd knows that slang very well.

[tepid surge]

i add the bath salts

[tepid surge]

more. bath. salts.

jesse

[spring fulcrum]

@soft zealot @visual glade @high skiff @indigo carbon Check this out... Upscale Resolution Factor of 10 in 100 seconds

[zealous fable]

Hi, everyone! I'm working on getting SDXL. Which should I be grabbing here?

[steady grove]

0.9

[tepid surge]

it has the 0.9 vae embedded in it

[zealous fable]

Okay, thank you! Was a tad confused with the two versions.

[tepid surge]

other one doesnt

[steady grove]

the original released vae was not liked by people so they released the beta vae

[upbeat summit]

is that a mountain dew crypto server rack in the wild?

[tepid surge]

is that was that is xD i jsut typed server rack on mountain/in nature or something

they grow in the wild and mingrate every winter

[steady grove]

the mountain dew is how it cools for the morning mining rush

i'm so happy that crypto mining with gpus is obsolete

[tepid surge]

yea that was crazy

[sage sand]

Sorry guys i'm a noob, what is SDXL?

[steady grove]

i also do love disruptive technologies

[steady grove]

a larger upgraded version of stable diffusion with two prompt inputs. two! also, TWO models for one pic. WHaaaat?

[soft zealot]

guess I need some extra searge nodes then to take a look ;o)

[sage sand]

ok so should i install as usual in automacit1111, in the models folder?

or need something different?

[steady grove]

automatic1111 works but have you heard of comfyui?

[sage sand]

Yes, but all say me it's work like automatic1111 so i never interest in that

[steady grove]

an updated version of automatic1111 will use sdxl. just copy the safetensor into the Stable-Diffusion model folder

[ionic dragon]

whats the res of the img?

its the highest res, SD i have seen someone posting

[spring fulcrum]

8960 x 11520

[steady grove]

using topaz. i can tell

[ionic dragon]

oh damn, 8690x11520

[ionic dragon]

how many mins did it take to render?

[spring fulcrum]

2 mins... (119 seconds)

[ionic dragon]

how is that possible?
which gpu are using?

8xa100?

[spring fulcrum]

4090

[steady grove]

topaz gan is fast but .. a gan

[ionic dragon]

which workflow are you using?

[spring fulcrum]

searge 3.4

[ionic dragon]

i cant drop into my UI, as I dont have nodes

[ionic dragon]

oh cool

[spring fulcrum]

I'm going to keep upping the factor till it crashes or reaches a maximum pixel which is pretty high

these files are huge though so I'm not going to do that for long

[ionic gulch]

is it just high-res or does it add details and improve textures? i'm looking for a solution to at least keep the details and textures, as most upscalers remove them instead.

[spring fulcrum]

It seems to keep details and add pixels... but I'm not really the best observer of that.... If someone wants to really look into it and weigh in on that I welcome it... I will upload the next image and then the stock generated one next to it. I'm not doing this to try to be better than anyone I am just a nerd that wants to push these limits and have some fun... 🙂

[tepid surge]

what is this cx this is like absolute pedo territory:

[spring fulcrum]

the one you can see is the original... the one you have to download is obviously the upscaled version... that one is at an upscale factor of 13

[native knot]

What was the prompt?

[tepid surge]

holy . s#t x0x

[tepid surge]

3 girls, 2 guys, reaching to hug the camera, painting, scifi armor

[native knot]

Try women instead of girls.

[tepid surge]

non of the renders are looking at the cemrea like this one cx

ahh right

[steady grove]

the details are too vague to say those are children for sure. calling it pedo territory seems a little bit elony

[tepid surge]

yee it'd just fs pull my social heads up cx

[spring fulcrum]

Ya I used some different nodes with the searge 3.4 workflow.. two different models and a LoRA in a pair tree....

[tepid surge]

i keep saying it, but this sd xl kicks butt its so great

[ionic dragon]

is it latent upscale or gan based upscale?

[spring fulcrum]

I honestly don't know

@soft zealot Is this logic correct or am I mistaken here? This is what I think I know... Latent means it comes from empty space right? gan is working with the pixels that are already there right? If that is true then I think its gan... For the upscale... but I am not sure.

to say the least the images do look identical

[tepid surge]

Noone: What have you been doing all day?
Me: Generating, lots of generating

[timid garnet]

[ionic dragon]

i am not really sure

[tepid surge]

average stable diffusion user:

[soft zealot]

at a qucik squizz it looks like what I called a "crude upscale" in earlier version of my workflows (and yes is in pixel space not latent space)

[ionic dragon]

@spring fulcrum when i generate I dont find any massive diff b/w 2x upscale and 3x upscale and barely anything when going higher, then why did you upscale 9x-10x?

[tepid surge]

Words:

[copper spire]

@tepid surge you were right I didn't see I can just take off few # and it install it with controlnet:) thank's again

[tepid surge]

oh cool! mm brain growth

Say cheeese

[spring fulcrum]

I just wanna see where it breaks thats all... just wanna see where the upper limits are

[shy kelp]

should i turn on hardware acceleration?

[tepid surge]

yes (i dont actually know)

[soft zealot]

too many primary colours around here, ltes tone it down a little ;o)

[sinful falcon]

This thing definitely sees into the future.

[tepid surge]

you got it boss @soft zealot (this last batch, was pixar "wet floors")

[native knot]

1920x1088* native render

[steel mist]

That is surprisingly good rigging compared to any time I've asked for a ship. I mean still pretty unrealistic but less obviously so

[soft zealot]

on board the good ship Venus...........

[native knot]

Check this one out

[tepid surge]

gb childhood

[native knot]

That 2nd one..."We're here to talk to you about your car's extended warranty."

[tepid surge]

5 "ambassadors" show up cx the mormon church there to convert your life insurance

[void zenith]

Sorry bro

[peak dove]

That was a fun two weeks at the Edinburgh Festival ...

[tepid surge]

A what a what!?

---

(not generated)
my kind a festival

can someone pick a color please i gonna see what i can do with it

[shy kelp]

red

[upbeat summit]

purple

[tepid surge]

how dare you both pick seperate colors

just gimme a sec i got this

[shy kelp]

HAHAHA

[lilac wren]

They are made juste beside my town, Haribo is at "Uzes" in the Gard (France)

[native knot]

lol...which one of you did this?
https://civitai.com/models/131063/virtual-insanity-sdxl

[upbeat summit]

not me but that's great 😄

[tepid surge]

there it is @shy kelp @upbeat summit purple and red

the only thing i could ask more for would be a fat stream of fire breath coming out cx

[hardy cipher]

when is the Robert Z'Dar lora coming out?

[hardy cipher]

when is the Robert Z'Dar lora coming out?

[wet nacelle]

Mr leno is that you?

[shy kelp]

🔥

[hardy cipher]

I wish. not sure who Mr. Lenno is though

[hardy cipher]

ahh, you accidentally an extra n

[wet nacelle]

too true

damn

[tepid maple]

anyone know automated way of replicating custom sdxl fork prompts
like a cloud api something

[hardy cipher]

yeah, they have strong chins

[tepid maple]

I want to make a tool for my site dont know how yet

[tepid maple]

damn gpu server stupidly expensive 😦

[hardy cipher]

beauty

[waxen rock]

have you tried runpod or vastai ?

[tepid maple]

nope looking now

expensive

[tepid maple]

cant make money with this prices

[waxen rock]

I'm working with runpod when I need a big batch, the lowest I found
otherwise im working local

[hardy cipher]

if it was that easy everyone would do it

[tepid maple]

for ex ?

[waxen rock]

ex ?

[vital wraith]

Is LoRA training on SDXL insanely slow? Or is it just me?

[tepid maple]

lets say you generated 1000 HD image ?

with 30 step or 50 ?

how mcuh you pay

[waxen rock]

should take around 40 min I guess on RTX 3090

and 0.44 is hourly

do the math

[tepid maple]

hmmm

thats actualy not bad

[waxen rock]

as long as you dont run it 24/7 it should be fine

[tepid maple]

does keeping models cost too?

or just gpu run times ?

[waxen rock]

the files ?

[tepid maple]

sdxl forks

models

I am planning like using 10 fork makes 60 gb cant upload everytime

[waxen rock]

hmm runpod has network storage feature which you pay 0.07 per GB

im using it for my customs

[tepid maple]

ok than kyou man

will go in dee pwith runpod

[tepid surge]

what exactly is money

[tepid maple]

make something people need and sell it= money

🙂

[waxen rock]

anyone knows if its cool to mention comfy ?

[void zenith]

Hot

[tepid surge]

it is very cool

[steady grove]

s / need / want

[void zenith]

I want her to bee my ai gf

[tepid surge]

mood i got do money stuf

[waxen rock]

I meant if it's ok with him lol

[tepid maple]

: )

[void zenith]

Type s/e/x after posting gif from trending

Do it bro

[waxen rock]

ai chicks onlyfans 🤭

[tepid surge]

there's a money

cx

[void zenith]

Yo What Discord robot did you make?

[steady grove]

umad?

[waxen rock]

😶

[void zenith]

No bro its a trick

[tepid maple]

holy it accepts bitcoin

hell yeah

[void zenith]

Plz tell

[void zenith]

What server do you own?

[tepid maple]

just checked runpod

[waxen rock]

still working on it lol, but how do you im making a bot ? 🤔

[void zenith]

Fiscord DEv badge

D*

[dusk mica]

does clipdrop do any additional prompting comapred to local comfyui?

[steady grove]

it's someone's alt troll account. maybe somene that got banned recently. they've been full throttle tilted for a couple days now

[void zenith]

Dude im not trolling

I am just curious

My pfp isnt me

[waxen rock]

can't share it yet, but it will be private anyway tho

[void zenith]

No problem bro

[tepid maple]

sadge

[void zenith]

What server do you own LIL bro

[tepid maple]

no server

I hav e3060 12gig machine atm

[void zenith]

Your badge bro

[tepid maple]

oooh the server

[void zenith]

Yee

Discord

[tepid maple]

it is not english forget about it

it is a gamer server

[void zenith]

Ok How many members?

[tepid maple]

we see 1,5k online sometimes

[void zenith]

🥶

[tepid maple]

maybe 2 cant remember

that happens when you have 600k yotuube channel stuff

[void zenith]

Wtf

[tepid maple]

pretty normal

[void zenith]

Oh i see your yt

W bro 600k subs so much

[tepid surge]

the adventuuure

[native knot]

*mutant adventure.

[tepid surge]

how.. dare you

[tepid surge]

i'll have you know I mopped the entire floor

[high skiff]

Thanosketti

[tepid surge]

thanos could so do a food channel

[native knot]

On the surface he looks calm and ready to drop bombs, but he keeps on forgetting what linguistic prompt he wrote down.

[hardy cipher]

such beauty

[void zenith]

hot mommies

[void zenith]

So good you have good thanos pictures

[fair stag]

whoever that is, he needs a donut, stat

[tepid surge]

why you always sexualizing

[glad grove]

bros look like it needs a donut

[tepid surge]

did you say free donuts (and I go by they/them)

[glad grove]

thats cool and all but i was talking to the ai pictures

this is peak male performance

[tepid surge]

https://tenor.com/view/anime-head-pat-anime-pat-pat-anime-blush-gif-23411369

all the same

[glad grove]

https://tenor.com/view/neco-arc-gif-25039455

[tepid surge]

Okay, donuts are cool, time 4 zzz

[hardy cipher]

deep fried

[glad grove]

new movie confirmed

[steady grove]

Even ai beautification filters can't make it not look like a toupe

[eternal fog]

lmao

[steady grove]

ai doesn't know what to do with the thing and just gently plants it down upon his liver spotted scalp

[glad grove]

now make them kiss

[eternal fog]

Terrifying images

[glad grove]

real

[tender timber]

Is that still slower with the openVino update from three days ago? https://www.tomshardware.com/news/stable-diffusion-for-intel-optimizations

Are you able to do 1024x1024 in SDXL for Comfy or AUTO before upscaling? Some people are saying there's a 4GB VRAM limit but for Intel, but other places say there's a workaround. Here in Canada an A770 is ~$430 for a 16GB VRAM version

[eternal fog]

Why does it do him so well lmao

[vital sail]

lol

[vital wraith]

I am making my first SDXL LoRA. Does anyone know what's going on here? I am used to seeing the epoch progress bar, but noticed at some point this morning it looks like this:

Don't want to interupt it, as it hasn't actually thrown any errors. >.>

[glad grove]

that means it workin 🙏

[soft bone]

unless you have thousands of images, this is too many steps.

total steps is calculated like this:
img count * repeats * epochs / batch size.

"repeats" is set in the name of your training folder ("100_man" would be 100 repeats). it looks like you have epochs set to 1 and repeats set way high. You can trade repeats for epochs and get the same steps. so change repeat to 1 and epochs to 100 or whatever, to get multi-epoch training that can be saved incrementally.
Also increase batch size to divide the total step count and train faster, if you have vram headroom.

[vital wraith]

I am using 5 epochs at 10 steps

[soft bone]

img count?

[vital wraith]

About 900

[soft bone]

"steps" is not a number you manually set. only repeats and epochs. do you mean repeats?

[vital wraith]

Epoch progress kicked off like normal, saw it hit 2 or 3 prior to this happening. Yes, repeats.

[soft bone]

no idea then. I've never seen epoch count disappear mid training. maybe telling it to save or sample every N epochs will keep it in line, idk

you should open an issue on github

[vital wraith]

Weird. I wish the console screen kept more of the text so I could see when it occurred

Happened about 1/3 of the way into the 3rd epoch from what I can tell

[soft bone]

the whole process should still be there if you scroll up, but sometimes it wont let you scroll while in progress

[vital wraith]

Looks like the text cleared, I thought the same, likely a buffer limit or something on the console

[soft bone]

weird. I would recommend regular CMD in that case.

[vital wraith]

I am using a standard CMD line, I just think due to the progress bar spam, it hit the text limit heh

2 lines for every image for thousands of steps lol

[soft bone]

when did standard cmd get multi-tabs? I've only seen that in Powershell. my kohya training has no tabs at the top

[vital wraith]

Windows 11

Added some nice features to your usual cmd box

[soft bone]

ohhh, well that could be it, they probably introduced a limit. I've done training with that amount of steps before and it kept the whole thing even after tens of thousands of lines

still whatevers doing that shouldnt ignore your epoch count. worth making an issue on Kohya's git

[glad grove]

new movie

[vital wraith]

I am going to check and see here how it progresses, perhaps the progress bar for the epoch got split somehow...or some other asthetic difference. Will know soon enough I suppose.

Welp, just accidentally closed it. lol

Hit settings to check buffer size and accidentally closed multiple tabs, ending the script 😛