🏞｜general-with-images

[reef root]

That's the only folder

I'm not

I have also shared screenshots of it's parent folder

[grizzled sage]

bro thats image browser not the controlnet folder

[reef root]

That folder is the only one

[grizzled sage]

install the extension. it'll be there. stop making things up

[reef root]

.

[reef root]

Yeah I'm here on this discord wasting my time for the funzies

[grizzled sage]

if the folder isn't there that literally means you didn't do the install step.

some people do that

[reef root]

I'm gonna do a screen record of my process

[grizzled sage]

k

[reef root]

@grizzled sage

I apologize for this 1 minute or retardation but it won't even connect unless I upload a model

[fast quest]

/settings

#settings

[grizzled sage]

stop wasting my time. you didn't even try to install it

you're just fucking around and trolling.

[reef root]

I'm doing that part now

As you can see it wouldnt connect to the ui

But I thought I'd show you the process anyways

I cant just keep the screen recorder on

[grizzled sage]

you may kindly fuck off. blocked so your pings won't get through. i suspect you're franktutor but gut feelings only get you so far.

[reef root]

Ok thanks bud

[green plover]

480 steps in on this TI final training for this one, the 2.1 768 version

[slender umbra]

@wispy nest I tried using keggle, I've got this far, now what?

[wispy nest]

Run the code.

[slender umbra]

did it

now what?

[wispy nest]

Click the link it gives you and you are free to use stable diffusion

[slender umbra]

it gives me no link though

Oh i was supposed to run all of them?

getting errors left and right

[wispy nest]

enable internet, somewhere on the right side and make sure you're using GPU

[slender umbra]

where? i'm not finding it

[wispy nest]

settings

[arctic geyser]

hi, can you share the notebook with me too

would appreciate it because google colab runs out of usage easily

thinking of switching

[wispy nest]

https://www.kaggle.com/code/miolovers1/stable-diffusion-automatic1111
You get 30h per week free

[arctic geyser]

thank you

[arctic geyser]

even i cannot find it in settings

[slender umbra]

we have to verify phone

[arctic geyser]

kk

[slender umbra]

and in the colab one it doesn't allow image uploading 😄

[wispy nest]

What do you mean? uploading where?

[fiery lichen]

for inpainting?

[arctic geyser]

which gpu should i use

P100 or t4 x2

[slender umbra]

for training, i have a folder called Kipe and it can't read it

[arctic geyser]

is the case smiliar

similar

like "Kipe" or "kipe"

[slender umbra]

the name in both cases is the same

[wispy nest]

doesnt really matter, sine you can't use 2 gpu's on a free plan anyway.
For higher resolutions p100 is better, since it has higher VRAM, for 512x512x8 t4 might be faster, not sure.

Either way, difference shouldn't be significant.

[arctic geyser]

ight

[wispy nest]

That's weird, cause it can read models from google drive

[slender umbra]

yeah is the path not correctly specified?

[arctic geyser]

what hsould i do here

ive enabled gpu and internet

it just stops running after this

[wispy nest]

try

/content/drive/MyDrive/kipe

[wispy nest]

run 2nd block, chose model and launch webui, if folders already exist - it's fine

[slender umbra]

nope still ntn @wispy nest

[fiery lichen]

I am at your "setting" stage cant find GPU either

I'm using a different kaggle notebook though.

I see comment on verify phone...

[arctic geyser]

youre supposed verify phone number

[wispy nest]

drive.mount('/content/drive')

but folder path is

/content/drive/MyDrive/kipe

I'm just assuming things looking at already existing and working collab code, no idea how paths work here.

[fiery lichen]

got it thanks

[arctic geyser]

ok i unloaded it

[slender umbra]

I'm not quite understanding should I do drive.mount('/content/drive/MyDrive/kipe')??

[arctic geyser]

i restarted it seems to be working now

[slender umbra]

@wispy nest I just added the images via link but now I'm running with this problem

[wispy nest]

Everywhere I check - everyone mounting just drive.mount('/content/drive')
and make paths relative to that.

/content/ - collab folder with contents
/content/drive/myDrive - seems to be folder with everything inside your drive

[slender umbra]

no worries I just did it with links instead it was easier

[wispy nest]

welp, no idea what's that 😄

[slender umbra]

told you it was rare it where as easy as it seemed 😦

is there no other way to run the training locally? I don't care if it takes longer but I don't have more than 4gb of vram

[grim swan]

#big cat

[wispy nest]

I don't do training , can't recommend anything 😦

Can't you download something to collab with wget?

if it's somewhere on the internet

[fiery lichen]

I kept thinking of kegels....exercises, training, so kaggle is natural for doing training on i thought

[arctic geyser]

lmao

[fiery lichen]

maybe kaggle is pronounced KA(car)ggle? kegel is pronounced ke(key)gel? where i am from they pronounce the same

[slender umbra]

I don't know what wget is

[wispy nest]

I was pronouncing it as kaggle (car way) . yea

[slender umbra]

nor what does it mean 'to collab with'

[wispy nest]

https://stackoverflow.com/questions/55755332/how-to-use-wget-to-download-in-certain-directory

[slender umbra]

I already solved the image download thing though

[wispy nest]

Let's go to #💬｜general-chat , we're making mess inside wrong channel

[arctic geyser]

what should i do here @wispy nest

i just get this

[slender umbra]

@wispy nest i tried restarting yet ntn

[wispy nest]

maybe try adding path manually ?

[slender umbra]

what path to what?

the images are downloaded already

[wispy nest]

I assume it wants model?
pretrained_model_name_or_path

[slender umbra]

great! i re-runed the model settings code and it worked

it will take time

thanks for the help! I'll keep writing you afterwards (when the process finishes

)

ok faster than I thought

@wispy nest how is it getting out of memory=???

[wispy nest]

lol, RAM?

You can check resources , it has tab

[slender umbra]

yeah it has no more ram i think

hpw to add?

i mean

Ok I give up this is practically impossible

I'm starting to think no one actually did it before

[wispy nest]

oh...vram...I don't think you can get more on a free plan

[slender umbra]

MORE?

why would it need more than 12gb?

[wispy nest]

yea , that's alot

[slender umbra]

I know people train with 10gb h0w is it it needs more than 10

[fiery lichen]

yea...i just ran out of memory on axelfar kaggle notebook and it forced me to restart

I'll go back to it another day, time to make images on collab

[slender umbra]

wait i'm in collab

and it says i'm out of memory

so I don't think you van on GC either

[fiery lichen]

i am only making images on GC

I was testing the extreme on kaggle

[slender umbra]

oooh making images

Oh welp, I'm making images locally but thought I could do training internationally since locally I can't

[fiery lichen]

i just got a kaggle notebook to work so I wanted to screw around to see what its limits are and it kicked me out

[slender umbra]

@wispy nest thank you anyway! it seems it is not currenly possible

[wispy nest]

I think it should be possible with this vram tho...

[arctic geyser]

@fiery lichen

man ill just google oclab at this point atleast its hassle free

@wispy nest do you know how to fix this

[wispy nest]

Either url doesn't work or you have internet turned off

[green plover]

Training test image for my next synthetic artist

[arctic geyser]

both the url and internet work

[wispy nest]

idk , I'm downloading everything just fine, double check your url

[smoky oak]

Currently using AI to upscale images of a character to train an AI on lol

[arctic geyser]

triefd opening the url on google and it works

[wispy nest]

I just reinstalled completely clean version and it works , idk what can go wrong.
just launched 3 code blocks in a row and opened webui

The only thing I did differently - commented out anything v3 model and it's vae, since I'm not gonna use them anyway

[stoic pivot]

Why do I get so bad colors with SD v1.5? I tested with "SD VAE" settings on Automatic and with "vae-ft-mes-840000-ems-pruned.safetensor" that some recommended, but the colors still are bad. Prompt "modern (gray loveseat) against a white wall, cute warm decoration".

[wispy nest]

vae name looks a bit off o_O

https://huggingface.co/stabilityai/sd-vae-ft-mse-original/blob/main/vae-ft-mse-840000-ema-pruned.ckpt

[stoic pivot]

I have that, but I used .safetensors rather than .ckpt, but I will thest the .ckpt now.

Ty.

[candid palm]

/

[smoky oak]

I can believe my AI assisted AI upscaled images were able to train a LoRA AI to decently recreate a character from a show with less than 5 scenes lmao

[smoky oak]

gotta love when the denoising breaks lol

[smoky oak]

Am I crazy or does she have like a huge forehead lol

[wispy nest]

after some measurements , I'm proud to announce - I think within norm, eyes are kinda weird tho...

[open sun]

i love this

[potent hornet]

I asked ChatGPT (in a long conversation) what it would look like in real life

do you guys want to take this description and see what Stable Diffusion has to offer? haha

Ah, also, ChatGPT said that it would like to be called "Sage".

Or maybe.. ChatGPT as an RPG character?

[wispy nest]

here's a little preview of a Lora I am working on, it's only v1, final result will be more detailed ❤️

[silk torrent]

Mhm, something tells me my sd2.1 is broken? Not sure what though if anyone is able to help?

[lilac pilot]

Creativity isnt creating something out of nothing
Creativity is creating something with everything.

Water looks decent on first one. But really like the feathers on the second one trying to mix and get them on the first image

[grizzled sage]

you're using 1.5 models. 2.1 is the model that was released

[lilac pilot]

Flow what model do you prefer?

[silk torrent]

No way did I not download the right one?

[lilac pilot]

is there a guide on here that I can reference when to use what model when?

[grizzled sage]

i have a lot of fun with dreamshaper its a nice generalized style and is 1.5. protogen photorealism is good too. I really love embeddings in the 2.x sphere too. it's hard to pick just one favorite.

lately i've been using more 1.5 models though since controlnet doe a lot there

[lilac pilot]

Yes. got a taste of using controlnet. its on another level. devs are aliens haha

[grizzled sage]

Wait.. these images are for different VAE's. Are you trying to create images on 2.1 with a 1.5 vae?

[grizzled sage]

1.5 embeds and vae's should only work with 1.5 models

[silk torrent]

Yes I've got the 2.1 sd model and I am trying to figure out why my images look so weird on it. At the very least the last vae should work because thats what everyone uses with 2.1 isn't it?

Not using embeddings

[grizzled sage]

everyone uses orangemix vae on 1.5 anime models. all 3 of those vae's are meant for 1.5 models

[silk torrent]

Ok then what is the correct 2.1 vae? Please send me a link

[grizzled sage]

i think the default one should be fine for 2.1 but i dont know. i've never tried using others

[silk torrent]

The "default" is what I thought I downloaded

[grizzled sage]

when i do, i get ones for a specific model that might need one. i never really had much need for a vae

[silk torrent]

vae-ft-mse-840000-ema-pruned.ckpt

This was even linked under the 2.1 stable diffusion site on hugging face

but I must have gotten something confused somewhere

[grizzled sage]

honestly vae's are at the edge of my understandings. i know they're the final encoder. maybe you can use old ones for 2.1 models. i shouldn't speak so confidently there

this might not be a vae issue. those images do look pretty cooked and i dont know why

[silk torrent]

it looks even worse if I don't use a vae, is my model just broken?

[grizzled sage]

i dont think its anything like that. just a software setting somewhere i'm sure

vae-ft-mse etc etc does appear to be the correct one used so its porbably not the vae issue. those images dont look like low steps or low cfg or any settings like that.. hmm.. do you have a clipskip set?

[silk torrent]

Clipskip is 1 now but is also like this on 2. My steps aren't too high and my prompt is very minimal right now but high steps do the same

[grizzled sage]

yeah 1 or 2 shouldn't be an issue. i thougth maybe it had gotten knocked to something like 4 or 5 by accident. trying to speculate on what could cause these sort of results

[silk torrent]

Yeah same, trying a much more detailed prompt with higher steps quick to see if that matters

[thick flint]

😹

[hallow sparrow]

I'm, running the new ContolNet, but encounter 2 problems. With mostly the generated image being blown out. No colour contrast? And some models Model as base, creates this error? Hope someone can enlighten me... "RuntimeError: mat1 and mat2 shapes cannot be multiplied (154x1024 and 768x320)"

[sterile kiln]

hahaha, français? Je vais entrainer EM avec LoRA, ça va chier

[thick flint]

Oui français, a la base je voulais qui fasse une pub pour des produits de la marque Old Spice mais voilà le résultat

[wispy nest]

Jamerais bien use mon visage comme ca mais jsais pas comment faire.

[sterile kiln]

Tu as quoi comme GPU?

[delicate wharf]

have you solved it? me too having this problem

[grizzled sage]

just posting for discussion sake in the other chan, thanks

[wispy nest]

I had this on KDE I have this here

but on KDE it was glitchy and fucky

so fuck linux

[grizzled sage]

nvidia?

[wispy nest]

AMD

back then

[grizzled sage]

ah back then

[wispy nest]

I used RX 470

[grizzled sage]

linux is fun. i go back an forth. a lot of "fuck microsoft" over the years too

windows 11 feels fresh and i like what satya has put together here

[wispy nest]

open source is like communism. people keep going back to it but do realize it can't build a proper usable system

user friendly system

[grizzled sage]

lol i think it can, it's just the culture has a ux denial problem

[wispy nest]

I left linux with these thoughts

[grizzled sage]

there's tons of good open source successes

[white birch]

@grizzled sage

[grizzled sage]

there's a ton of closed source failures too

[wispy nest]

last time I complained in InkScape discord about it not having a crop canvas feature people told me I'm weird and told me about workarounds that take 30 secs and a lot of nerve wrecking

so this is exactly why it doesn't have a crop canvas feature

[grizzled sage]

lol mentioned this in the other too.. i don't want tiny 11. i want fully bloated with a dual turbo blower and giant piped exhaust, bells, whistles, everything. i got the machine to do it so why would i limit software from using it?

ALSO, those tiny builds are a neat experiment but not right for primetime. big thing they do is disable security updates. boo

[wispy nest]

the community is forgiving

[grizzled sage]

maybe neat for embedded systems but i think microsoft caters there now a days too

[wispy nest]

nobody gives a shit about UX

[grizzled sage]

feel that

[white birch]

this one doesnt disable any updates.
and there are much better alternatives to native windows apps usually

to each their own 😛

[grizzled sage]

like people do, but on a level they're not thinking about

oh i know alternatives. i'm a chameleon completely. i mostly hate terminal alternatives on windows over the years though. but now we're in some new space. windows 11 has reinvented the terminal

it's got fugin tabs now!

also notepad is getting tabs soon too but who cares honestly

[white birch]

when I eventually get a P+E core chip ill do 11. until-then; I refuse to change my right clicking habits 🙂

[grizzled sage]

a big thing i swapped back to windows for was gaming and all the new dx12 features coming. i dont wanna hobble windows update to save 12gb of system space

i'll buy another ssd instead

[white birch]

for me its the random drop is CPU perf.
Defender mostly at fault, but all the phone-home nannys are partially to blame as well

[grizzled sage]

yeah supposedly 11 does better thread scheduling with the e cores? i heard that. i had lots of glitches on linux that would turn off if i pressed scroll lock to dsiable ecores

[white birch]

I havent always had good CPUs. Optimization matters to me.

yeah Win10 is hopeless with an Ecore chip

[wispy nest]

my point as well

even with good CPUs I still care about effectiveness

[white birch]

dropping defender litterally doubled my FPS back in the day

[grizzled sage]

real time scanning will always be a huge performance overhhead no matter what software you use to do it. that's the compromise there. performance or a seatbelt? Personally, i'm comfortable enough with my usage habits that i run without real time scanning at all, and only load a virus scanner for scheduled scans

[white birch]

same.

[wispy nest]

I want to be in charge of all my resources

[white birch]

No defender, no AntiVirus. nothing.
Same Dreaming 🙂

[grizzled sage]

me too but usualy i'm like "CHARGE!!!" and launch all the things

[white birch]

if I messed up bad enough to need em. Ill just wipe it.

[grizzled sage]

i want every transistor tickin

[white birch]

Yerp

[grizzled sage]

gets harder as i upgrade

[white birch]

For me uses 😛

[wispy nest]

ah we have regular blackouts so I have to work on battery

[white birch]

Start CPU mining when your idle 🙂

[white birch]

u need an alderlake 😛

[grizzled sage]

hear that. storm season is good hear. been calm this year for some weird reason (as the rest of the world storms on)

[wispy nest]

I choose the most efficient software

[white birch]

Girlys home, peace holmes

[wispy nest]

yeah the eastern storm coming from Russia here with rocket rains

taking down power plants

never been calm in the region

first WW1, then WW2, now here we go again

on some days we have like 6 hours of electricity a day

I am lucky to have friends abroad who brought me this bulky portable power station

so I can work

my work is usually on idle, eating about 10W

unless some windows service starts motherfucking eating 40% CPU and RAM

or another mofo starts using my discrete GPU for no reason

then I run out of charge pretty quickly

one of the reasons to hate windows is I have no control over it. it does what it wants

ideally I would like to have a terminal OS which can run windows 11 apps

and doesn't have background processes that I do not schedule myself

[grizzled sage]

honestly, i found that withh linux too. i had to get into the guts constantly to find out what it was doing. And i do that on windows too. The guts are closed source on windows, but i'm going into a blackbox on eithher side. It's not like i'm reading the manpages for half the shit i do in linux

[wispy nest]

yeah it depends on the distro tho

[grizzled sage]

i liked plasma while i used it. i miss teh windows that wibble wobble when they're dragged

[wispy nest]

plasma is not a distro it's a DE

[grizzled sage]

i know

[wispy nest]

if you use linux from a terminal only it's pretty much perfect

no DE

it's what it was designed for

[grizzled sage]

another thing it did that i miss, windows has so many awesome start menu button hotkeys now in 11, but plasma does a thing with it thats abbbbsolutely genius

[wispy nest]

server side

[grizzled sage]

holding the start key lets you click a window anywhere and drag it

[grizzled sage]

linux yeahh. not gnu/linux though

[wispy nest]

have you seen the new KDE interface? it looks pretty cool tbh

[grizzled sage]

i have not

[wispy nest]

but it's still glitchy af

[grizzled sage]

as is tradition

[wispy nest]

I've seen a dev presentation where the guy demos the new design

and it started flickering in the presentation

he's like

"I dunno why it's doing that"

"but let me move on"

that's the fucking essence of KDE

and then there's gnome which is pretty stable in comparison but has lots of not so obvious bugs

[wispy nest]

the spaghetti guy!

macaroni or something

ah yes, Macron

[sterile kiln]

in his thoughts, it is a wild dragonfly, which meditates on the harmony between living beings. He loves seasonal honey and the smell of dandelions in the spring.

YOu just donnu him dude

[fiery lichen]

you people who train the a.i. ....

i hope this building image does not get flagged

all i wanted was a skyscraper with some details, and the ai drifted its mind away

[wispy nest]

can you reconstruct the howl's moving castle with SD infinity?

[fiery lichen]

oh it gets worse somehow with nsfw and penis on neg prompt

[modern willow]

magic

[wispy nest]

nice guitar

not put on some strings

[grizzled sage]

controlnet! lol

[wispy nest]

@grizzled sage installed and ran the default setup

[grizzled sage]

very hot

this gives me hope if it was that easy. i've been meaning to dive in but ngl, was a little intimidated. also distracted by all the milliion other things

[wispy nest]

it was stupid easy

just needed to git pull the webui repo

it has the extension manager in the latest version

then you go to the extensions tab

available

search for "deforum" and click install

then once it's installed restart the ui

go to the deforum tab

switch to 3D mode in the Keyframes tab

[mighty phoenix]

Sup?

[wispy nest]

and hit generate

it will download everything it needs

about 3GB

and then it will run the default setup

it even generates the mp4 file for me

it has ffmpeg with it

[hidden dew]

https://github.com/ddPn08/Lsmith
A new UI for Stable Diffusion that apparently has significant speedup (~3.5x faster than xformers+diffuser)

[dense tapir]

Leaves me out.

[static tusk]

8.58s vs 30s..damn!

[dense tapir]

I was reading about this when VortaML was the thing and if your card has tensors this will rock.

For tensors you need a 2,3,4k card.

[static tusk]

ahh so it rebuilds the models to tensorrt

interesting

surely that could be implemented into A1111

[dense tapir]

yes, it could

tensors are so damn good that, did you know, they spend over half their time doing nothing waiting for more data even in large data training.

[static tusk]

yeah the vram is the bottleneck

[dense tapir]

well, they have started giving them their own data but they even wait on it, lol

[static tusk]

I imagine the tensorrt method would not work with any embeds etc

tempted to have fiddle but I will wait

[dense tapir]

I am not sure but if embeddings die then a lot of people would never use it

[static tusk]

pretty much

[dense tapir]

I have some embeddings I made that I tried to LoRA and no dice.

[static tusk]

Ive not dipped my toes in any Ti or lora dev yet. still busy learning to gen images first

[dense tapir]

it did not like my nir I made into an embedding at all

I tried my nirphoto emb today and nothing popped up. Not sure what lora failed at with it while the ti loved it.

Lora hates this

fantastic loss though

Interesting to see how fine SD can go. It isn't as fine as I originally thought

probably because base is 512x512

[smoky oak]

I found somebody who is trying to cope so hard with the fact that stable diffusion is so much better than wombo dream lmao

My stable diffusion result

vs the wombo dream result

his response was "they look equally as good"

like bro, keep lying to yourself lmao

[grizzled sage]

i mean sure but... does the differences really warrant your reaction?

[smoky oak]

It does when he sent me 42 messages in DM's for the last 4 hours

about how wombo dream is just as good as SD, and I just don't know how to use wombo dream

[grizzled sage]

i think maybe he's turning your crank for amusement

[smoky oak]

I barely even responded to him in DM's at least

[smoky oak]

The part that really got me was him telling me hes used stable diffusion v1.4 and knows that the results I am sending couldn't have come out of stable diffusion... Thats what really got me

[dense tapir]

wombo dream? wth is that?

[smoky oak]

its an app I used before running stable diffusion locally

[dense tapir]

I never heard of it.

[smoky oak]

its a very very fast image generator that can do pretty good honestly, but this guy is trying to act like its the best

it can make some good results, but its still behind midjourney and RAW stable diffusion

[dense tapir]

each one has its strengths and weaknesses.

[smoky oak]

yeah, dreams strength is its speed

it makes 4 images in like 6 seconds flat, and for that I say its dope

[dense tapir]

Shit

in SD, on my 1060 it is 45-65s

768sq

[glossy reef]

damn, when i started using local webui with loras and stuff i really given up on midjourney for now, dream isnt the best for sure

[smoky oak]

yeah, its really fast, and the results are great, but I was just saying I was cancelling my dream susbcription to run local SD, and he came up in my DM's to say how I was switching for no reason when dream is just as good

[dense tapir]

SD's strength is its flexibility. What doesn't it do that you need? Train it.

[glossy reef]

good luck squeezing that much quality on dream or midjourney xD

[smoky oak]

right

again, I present this image lmao

the detail in the hair alone is already leagues above MJ or dream

[dense tapir]

I still think MJ has the quality edge tbh but it has that I made this with MJ look to everything so no thanks

[smoky oak]

MJ has good quality, yeah

bit its still no RAW SD

[glossy reef]

MJ could be more amazing, they're just limiting resolution for now, and censonrship is killing me sometimes

[dense tapir]

yep

[smoky oak]

censorship is why I left MJ for dream

[glossy reef]

or ok, a maybe more than sometimes

[smoky oak]

you mean to tell me I am paying $30/m for limited generations and a censorship filter so strong I can't generate "blood red lips"

thats when I noped out for dream

[dense tapir]

censorship is in RAW sd too so go on civitai and grab sucking dick models, etc...

[smoky oak]

yeah haha

[glossy reef]

lol yeah, you want something? search for it, use it local and voila

ofc if you have decent gpu

[smoky oak]

the censorship in midjourney was insane dude

tube skirt? censored

bruises? censored

[dense tapir]

damn

[glossy reef]

i got term

In a sauna censored

in midjourney

[smoky oak]

a shirtless guy at the gym lifting weights? censored

[dense tapir]

The stuff I posted in here yesterday MJ would have censored?

[smoky oak]

oh for sure

you can't have ANYTHING remotely NSFW

Like I wanted to make a fighter who was like bruised up and stuff

censored

[smoky oak]

oh yeah, for sure too sexy for MJ

[dense tapir]

LOL

[smoky oak]

I remember even lace dresses were getting flagged

[dense tapir]

they will lose then

[smoky oak]

like it was so bad, I stopped using it as a whole

I have heard its not as bad anymore, but come on

[glossy reef]

its sometimes more like capacity show of model than true model for people to use, its so constricted

[smoky oak]

wombo dream does allow NSFW gen on their discord bot at least

[smoky oak]

for real

[dense tapir]

MJ eat those

[smoky oak]

lmfao

[glossy reef]

and for anime really, civitai to the rescue

mj cant handle this

xD

[smoky oak]

for real

I just made this one as well

compared to dreams Anime

like it looks good, sure, especially for 6 seconds for 4 images, but like

[dense tapir]

CivitAI needs to have SFW as the default cause I signed up and the first page was all porn. I am mean porn worse than I would see on pornhubs's front page. Eeeeeeeeeeeeek, thank goodness I wasn't at work.

[glossy reef]

yeah, dreams looks nice to, but kinda simple

[smoky oak]

YES

Big agree, OMG

[dense tapir]

The first one was literally an erect penis and a female sucking it. Ahem

[glossy reef]

"Look he's still using midjourney"

😂

[white rain]

https://imgur.com/l9dU9PE highly detailed, digital painting, artstation, concept art, smooth, sharp focus, illustration, art by wlop

[smoky oak]

yeah

[smoky oak]

lol

midjourney could never

[white rain]

https://imgur.com/l9dU9PE highly detailed, digital painting, artstation, concept art, smooth, sharp focus, illustration, art by wlop

[smoky oak]

oh, i also trained a LoRA on an extremely under represented character from Arcane, and it did pretty damn well lol

I trained one on scar from Arcane

had to AI upscale all of the images

[dense tapir]

I don't mind that shit is up there but don't have that shit as the default. They will eventually get taken down I think.

[glossy reef]

first of all, mj is not a model, its a system, you can't really make what you really want because there is back-end of this model filled with systems like "starters and shape defaults" and front end like coloring and aesthethic

[smoky oak]

thats from the show

and this is after I trained it

[glossy reef]

even if you put simple stuff like "i need a square" it will give you some crazy scenery

[smoky oak]

like for only having 6 images, it did pretty damn good lmao

[glossy reef]

so you cant really use it like you want

[dense tapir]

you can train SD with 1 image. Insane.

[smoky oak]

yeah

[glossy reef]

and thats good and bad at the same time

good for newbie "players"

and bad for more experienced users who want to have control

[smoky oak]

thats how I ended up here

I found out how hot the men stable diffusion could make are, and I bounced lol

this is another person LoRA I did

[dense tapir]

I train styles mostly and 1 image is what I use up to 10 but mostly 4-6

for people 1-30

[glossy reef]

never trained lora, but i must dig that stuff up, i heard i could do it on 8gb vram

[smoky oak]

oh yeah, no problem

I do it on 8, and my LoRA's come out great

[dense tapir]

lora is nice but I ran into its limitations today

[smoky oak]

I made a markiplier LoRA

Came out pretty great IMO

[dense tapir]

I used my ti embedding data to train lora and the loss was fantastic but forget it when I called it nothing was there

[glossy reef]

huh strange

[dense tapir]

I played with resolution tonight in lora and I can see we need more than base of 512x512

my lines were too fine for it

I had to quadruple the lines and finally lora acted right

I have a nirphoto embedding that simulates that near infrared photography. works great as an embedding as a lora nothing, zilch

[glossy reef]

yeah, so it looks it can be tricky sometimes on more complicated subjects

[dense tapir]

yep. if lora fails go to ti

[smoky oak]

I made a capybara LoRa lol

No Lora

LoRA :>

[dense tapir]

I wish HN had better optimized tools cause I really like the ones I made but 3 days of 2.5h each day to get 600 steps is insane

[glossy reef]

ah

ahahaha

capybaras are cool

[smoky oak]

The capybara LoRA also works alarmingly well on furry models as well lmao

[dense tapir]

See, perfect example. base had an idea of what it was then Lora will work fantastic but if you can't get anything close from the raw SD bas then dreambooth it

[smoky oak]

That's not how it worked for my shrimp, but if you say so

[dense tapir]

I did a lot of film noir stars in 1.5 and base had no idea so DB it was

[smoky oak]

Stable diffusion has 0 idea what a shrimp looks like

[dense tapir]

it had a general idea

[smoky oak]

Did it? Lmao

[dense tapir]

you would be amazed

[smoky oak]

Looks about right lmfao

[dense tapir]

it had an idea of my nephew and my nephew is not on the net

[smoky oak]

I guess that's a "general idea" of a shrimp

[glossy reef]

haha damn yeah, stuff i've seen people generated still really sometimes surprised me even after more than 1 year in ai generating

[dense tapir]

why? because my nephew had features from all those images so it could make almost him

[smoky oak]

My LoRA just helped a bit guess lol

So yeah, LoRA's can help a ton of things the AI doesn't know at all lol

[glossy reef]

i was starting on Vgan and later disco diffusion stuff

[dense tapir]

*can't

[glossy reef]

so its amazing to see how its now advanced

[dense tapir]

DB,LoRA,HN, TI in that order from the most training into the AI to the least

not sure why HN was left out there to rot

emad said new HN was coming but yeah, emad

[glossy reef]

yeaah

well

[dense tapir]

that lora failed me too 😦

[glossy reef]

i dont know if stuff with raw stable models will be all so bueno right now with all that copyrights drama

even 2.0 and 2.1 are ehm

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

[dense tapir]

If they will fix base it can be

[glossy reef]

yeah i wish that will be the case

[dense tapir]

I wish auto would do diffusers

[dense tapir]

Yeah, 1x1 too fine. 2x2 better, 3x3 too course so didn't work

that was lora style

3x3

[slender umbra]

what's the difference between a lora and a ti?

[dense tapir]

Interesting. I might have a use for this

[smoky oak]

I'm not sure I can properly explain it

Main thing is TI's are way smaller size wise, but take a lot longer to train, whereas LoRA's are bigger size wise, but use less memory and time to train

[slender umbra]

OMG how much vram is needed for lora's?

[smoky oak]

~6-8GB

[slender umbra]

😦

i've only 4

[smoky oak]

Oh, I'm sorry. You're not gonna be able to train anything on that sadly

[dense tapir]

I can't train anything due to speed but I can train TI with 6gb. VERY slowly so I use colab

[slender umbra]

tried using collab and kaggle for TI's but it's waaay to complicated

[slender umbra]

can you explain me how you used colab? I always get stuck and it says I need more vram?

[dense tapir]

if that is too complicated then stop now as they only get more complicated

[slender umbra]

it's mainly the fact that there are no cleear tutorials, and every single step is an error

[dense tapir]

not kidding lora colab was the worst to learn for me

[slender umbra]

do you know how to use the ti one?

[dense tapir]

To train ti I use automatic1111 and train it, or Hypernetwork, in it with the extension.

[slender umbra]

huh ok, unfortunately I always run out of memory hehe

[dense tapir]

I like this style

[slender umbra]

@dense tapir any tutorial for lora colab? or maybe a link?

[dense tapir]

no tutorials at all and you just have to try and try again until it works.

[slender umbra]

😄

can you send me the colab link to try and work it?

[dense tapir]

https://github.com/Linaqruf/kohya-trainer

[slender umbra]

thanks!

[dense tapir]

you're welcome

[slender umbra]

wish me luck 😉

[dense tapir]

you'll need it

run into any issues ticket and he will get back to you

[slender umbra]

he?

[dense tapir]

yes

I said morgan freeman and it made him look like Nelson Mandela

[grizzled sage]

i got bored with my powershell and spiced it a bit. maybe i can make it a hud feature like i had in kde.

wish i could make it do colors on the output though

[grizzled sage]

omg. it's an easter egg. name a terminal window anything_quake and it is drop down

[smoky oak]

Oh that one is awesome haha

[grizzled sage]

not working oh well.

yo what? it woks somehow! i push start ` and get a quake style term

🌠

[slender umbra]

have you tried adding "colours" to the tag?

"colorized"

and negative "blackandwhite

[grizzled sage]

well like you see there is some color. its enabled. but the rest of the script is plain

its only the server process that outputs some terminal color goodies

[slender umbra]

again, have you tried using those prompts?

sometimes getting more emphasis on what you want works

[grizzled sage]

ohh you think i mean for the terminal background?? haha no that i want to be a neutral image. i want the terminal letters to be colorized. not that big of a deal. just i got into reconfiguring my term for a minute

[slender umbra]

Aaaah I thought you wanted the prompted image to be colorized sorry

[grizzled sage]

thanks anyways

[hallow sparrow]

No not yet will post resolve if I find it.

[green plover]

@strange jungle pick a couple celebrities

[strange jungle]

for what?

[green plover]

For me to make a SD character to take into iClone

[strange jungle]

Bing Crosby

Frank Sinatra

I would say Ptolemy Shizo - but nobody knows him

he was popular in Egypt around the first century AD

SD does a horrible job making Bing Crosby

looks like a cross between bing and stan laurel

[green plover]

Here's our starting headshot bingxfrank

[magic forge]

Logo of Traditional women with saree in blue color with white background in indian style

[green plover]

Obviously I could have done more with that prompt to make him more photo realistic, but I'm being quick

[strange jungle]

Not bad...Frank Crosby, Bing Sinatra

looks like Roy Rogers

[green plover]

Out of character creator

And that is a posable character

[strange jungle]

so far so good

[strange jungle]

so lets see it in action

[green plover]

you mean I have to put clothes on him?

[strange jungle]

Who cares about clothes. I want to see the head speak

throw a suit on him

Bing Crosby would be all over AI and Daz and iClone stuff, motion capture

he was always looking for ways to perform without having to be live, to work

avatar Bing Crosby - he'd love it

[green plover]

Not doing lip sync. That's a time investment, but I'll make him dance

[strange jungle]

by shooting at his feet?

[green plover]

By dropping a mocap on him 😛

[strange jungle]

Have you tried cascadeur?

[green plover]

Looking at it now

[strange jungle]

the issue with mocap files is they're canned motions

many generic motions...nothing custom

mocap rig is too expensive for me

[green plover]

I'll have to look deeper into it later

[strange jungle]

free version has limits for vertices and length

[green plover]

Exports as FBX can always transfer

[strange jungle]

right, but it still has limits

[green plover]

300 frames is 5-10 seconds so that is plenty for a single motion

[strange jungle]

what about the joints?

also, it's some type of hosted - I would expect in order to impose those limits

[dense haven]

How do I remove the 'no preview' pictures? It's taking up a lot of space

[strange jungle]

I don't like cloud

cloud services can delete stuff in an instant

[green plover]

ok told you I wasn't gonna go crazy with it

[strange jungle]

just something simple like singing while riding a camel

[green plover]

o.O

[strange jungle]

Thanks for showing it. Looks good.

[static tusk]

settings > extra networks. change to thumbs

[slender umbra]

@dense tapir hey you up? How did you manage to get an output with the colab?

[smoky oak]

thank you so much OMG

[lament wyvern]

hacking

[delicate wharf]

you need to use 1.5, not 2.1 for controlNet

[hallow sparrow]

Do you mean the Checkpoint Base model? I followed the instructions to a tee. This only happens with my own sketch that I'm trying to develop. Is the problem with Hand sketch? Just black & White... Trued al scenarios... see thumbs at bottom.

[lilac pilot]

Did you do in painting for this?

[hard wagon]

No. It’s ControlNet.

[plain pier]

I don’t know if I did this correctly but I typed in Messi World Cup and got this

[lilac pilot]

My gen

Ex

How come the example pops out way more

I was using the same seed and prompt I added restore face didnt do much. Ex looks way more polished... cherrypicked, but did they used an upscaler?

Upscaled.. alright seeing a little adjustment

[hexed swift]

@wispy nest These were some out comes. Issue with it is it doesnt get text based logos correct. so they would probably have to be added through PS

[wispy nest]

Awesome!. Did you use any specific prompt or model? Just to get an idea on what steps to follow

[hexed swift]

There's no specific model just sd2.1.

With the prompt I played around with the likes of:

Hyper realistic editorial photograph of an epic tiffany x Bose speakers campaign.
An award winning campaign photograph of .....

then added at the end a bunch of awards like promax, webby, clio, D&AD etc.

[wispy nest]

nice

[hexed swift]

negative prompts being stuff that i didnt like, like wires, or deformities etc.

[wispy nest]

Amazing.. Cool I'll try it out

thanks for the reply

[hexed swift]

np

[smoky oak]

Anybody know any good photorealistic animal models?

I haven't been able to find any that do animals particularly well

I have one that does decent actually, I wasn't prompting right for it before

[vague oasis]

have you tried this one? https://civitai.com/models/4488/classic-negative-sd-21-768px

[smoky oak]

I haven't, as it didn't look like what I needed, but I suppose there is no harm in a free download lol

[steel yarrow]

ControlNet on SD is mind bending tech, just the value for previz alone... amazing

[vestal needle]

Any idea why my ControlNet images look great in previews when generating, but once it finishes, it ruins the color and stuff? Left is the last preview image before finishing, right is the finished image

[smoky oak]

Ov that's very weird, I've never had anything like that

It's almost like a gamut curve

[vestal needle]

Yeah I'm not getting any errors in the console, and each model type is doing it. No problems with normal img2img, etc.

Maybe I'll do a fresh a111 install

[vague oasis]

ong, im never using normal img2img ever again

controlnet is too powerful

[steel yarrow]

that shot is finished with img2img for faces, after ControNet for the ensemble shot

[tidal anvil]

send me a picture of a dragon driving a car

[lilac pilot]

anyone have a portfolio online?

[pallid ruin]

Epic

[lilac pilot]

Anyone try animation with deforum?

[steel yarrow]

yeah

pretty awesome

[jovial rune]

Comando ?

[grizzled sage]

Serious. Why add noise when you can add control

[civic crag]

so i was talking about ControlNet that i discovered today, i really like that

[fiery zinc]

nice one

[lilac pilot]

So far using it you have to hit the sweet spot in the denoise or it comes out way off

[civic crag]

i didn't have time to tweak it so far, it seems really good with defaults

i only adjusted the prompt and the controlNet weight

[fiery zinc]

also depends on the source. Your source has a clean background to work with.

a busier background would add complications

and, of course, the AI doesn't understand intent

so sometimes something in the scene doesn't quite translate

but I'm rambling. It's still really cool.

[civic crag]

i think i didn't have problem with busy background in my tests

[lilac pilot]

nah appreciate that so make sure your backgrounds isnt too busy

[lilac pilot]

it didnt smudge or blur?

[fiery zinc]

not really, for me anyway. Generally it's just misrepresenting stuff

[civic crag]

i think lowering the weight solved most of the problems

[fiery zinc]

like a rack in the background becoming something the subject holds in her hand, or something odd like that

[civic crag]

oh i see

[civic crag]

about that, is there a way to edit the mid render to remove anoying things ?

[fiery zinc]

it couldn't separate the background from the bow, so it looks like she's holding a... rackbow?

[lilac pilot]

devs are godly so in few weeks we will see this evolve to having minimal issues

[civic crag]

xD

[fiery zinc]

yeah, exactly

this is moving so fast it's making my head spin

my to-do projects have already exceeded my capacity to even look into them

[lilac pilot]

Im trying to keep up with all the items i can do haha

[civic crag]

oh they are working on fingers too ?

[lilac pilot]

just saw 3d now we can do

[lilac pilot]

I hope so

[civic crag]

😂

another controlNet (with no background again)

this is magic, i love it

[wispy nest]

How did you prompt that bow?
I was trying to make horizon zero dawn style art with a bow some time ago and all my bow attempts went...pretty bad

[lilac pilot]

Is there a eraser for inpainting?

[fiery zinc]

I used controlnet. I literally used a HZD image as the control

let me see if I still have the source image somewhere

[wispy nest]

yea, controlnet wasn't there yet when I tried =\

[fiery zinc]

controlnet is a game changer

in this case, literally 😆

[wispy nest]

hey bro can u pls help

?

with stable diff

[fiery zinc]

damnit, I think I deleted the source image. Let me see if I can find it online

[wispy nest]

do u know how to run it?

anyone pls

[wispy nest]

dw, there are plenty on the internet I can use...like this one

[vague oasis]

which gpu do you have?

[wispy nest]

grip is weird...but whatever

[wispy nest]

idk

[fiery zinc]

https://www.youtube.com/watch?v=VXEyhM3Djqg

assuming you have Windows

[wispy nest]

i have windows 64 bit

if thats what u asking

[fiery zinc]

if you have windows and a decent gpu, then follow that video guide

[wispy nest]

yea but i did everything

[vague oasis]

do you have a gamer pc?

[wispy nest]

yes

i downloaded that last thing in cmd

with 2.3 gb

then i closed the window

whenever i open the batch file, its downloading again

2.3 gb

what should i do?

[vague oasis]

invokeAI is easier to install than auto1111

[wispy nest]

eh, kinda same thing...maybe just less dependencies

any solution pls

[vague oasis]

without specs its hard to know if it runs or not

[fiery zinc]

Oh here we go, for those who were curious, this was the control image for the above archer render:

[wispy nest]

but is it supposed to download again 2.3gb?

everytime i wanna open stable diff, does it need to download 2.3 gb?

[fiery zinc]

the fence in the back got merged into the bow in the controlnet process

[wispy nest]

yea, makes sense

[wispy nest]

nope

[fiery zinc]

Sounds like it didn't complete? Or did you add something into the .bat file?

[smoky oak]

My web UI keeps crashing, and I have no idea why

[wispy nest]

i added nothing

[fiery zinc]

yeah, fingers 😆

[vague oasis]

are you using the sd 2.1 model?

[fiery zinc]

the bane of AI

[smoky oak]

There are plenty of AI's that do good hands most of the time luckily

There are definitely things that AI's do worse

[wispy nest]

sd-v1-4 this one

[smoky oak]

oh 1.4?

[wispy nest]

is it bad?

[vague oasis]

use the sd 1.5

[vague oasis]

its just old

[smoky oak]

SD 1.5 is the oldest you should use IMO

[wispy nest]

I'd suggest following official gh page insteead of random youtube videos...
https://github.com/AUTOMATIC1111/stable-diffusion-webui#installation-and-running

ohhh

[smoky oak]

1.5 an beyond have a ton of support

[wispy nest]

thx!

[smoky oak]

Ok, i am so confused as to what is going wrong with my stable diffusion right now

it keeps freezing when it gets to the upscale part, and its never done this before, WTF

[dry crow]

for technical help there is #🤝｜tech-support

[smoky oak]

I am so confused

it breaks every single time now

[wispy nest]

Look what console is saying

[smoky oak]

its saying nothing lmao

no errors, no anything

this just started happening today

it just freezes when it gets to upscale, and then breaks and doesn't respond

[wispy nest]

if webui crashed it might be still fine , it happens sometime , console will still process what you're trying to do

[smoky oak]

no, the console freezes

like it stops the generation and just freezes

I reload and try to generate again and it just says "in queue" and then after a couple seconds the webui starts to throw errors

[wispy nest]

might be cpu bottleneck

[smoky oak]

so weird

[smoky oak]

its not

this is an actual problem

[wispy nest]

yea if it is saying "in que" - it means it is still processing something currently

[smoky oak]

I have generated 1000's of images before, this isn't even high res generation

[smoky oak]

it freezes

[lilac pilot]

it wont let me post my example

[smoky oak]

its not processing anything

[lilac pilot]

it does have anything explicit

[smoky oak]

this is so weird, I have never had this issue before

trying it again, lets see

[tawdry geode]

It happened to me, its faster to restart

[wispy nest]

maybe disable console progress bar temporary to see if it helps , but doubt it spams enough to kill cmd

[smoky oak]

restarting is doing nothing

it just freezes when I restart

[tawdry geode]

disable live preview helps also

[smoky oak]

yup, it just did it again

[wispy nest]

That's good way to increase gen speed too lol

[smoky oak]

what in the world man, these aren't even high res gens

just keeps freezing here

the UI is still responsive

but the generation keeps completely breaking

[wispy nest]

give it some time maybe?

[smoky oak]

its not even using my GPU

[lilac pilot]

is this artifact on her nose?
Is my weight on something to high?

[smoky oak]

could be too high weight, yeah

looks like an attempt at like a glossy nose or something

[smoky oak]

still frozen

here, let me restart my PC and see if that helps. this is all so weird

These are very low res gens compared to my normal

[civic crag]

i have a similar problem (it happened today, i did nothing new, SD takes 9GB of vram when idle, just started)

[wispy nest]

¯_(ツ)_/¯

[smoky oak]

I only have 8GB VRAM, but with xformers, I have been able to do really high resolution generations. This is the first time I have had this problem

gonna restart my PC, please let this fix the problem

[civic crag]

lol

[smoky oak]

It's updating right now, I don't know what would be causing problems like this

I'm just trying to do a 1.5x upscale on a low res image. It's nowhere near as high res as I usually go

I can usually do native 1080x1920 gens no problem

[civic crag]

wow ... i can do 880x880 max

[smoky oak]

Xformers is a massive help if you have a 3000 or 4000 gen GPU

Max I could do before was 768x1024

[dry crow]

its a massive help for every nvidia gpu

[smoky oak]

Now I have been able to go all the way up to 2560x1080

[civic crag]

mine is pretty old it is a 1000 i guess

[dry crow]

you should use it

i have a 1080

[smoky oak]

X formers is the only thing that allowed me to get into SD how I wanted with no resolution limitations

[civic crag]

yes i have a 1080ti with 11GB

(was quite good at the time)

[smoky oak]

To get images this high res and quality straight out of SD is just gorgeous

And that's not even as high res as I canngo

[civic crag]

😍

[smoky oak]

That's only 1024x1536

[dry crow]

yea then edit your webui-user.bat and add --xformers --autolaunch behind Commandline_ARGS=

[smoky oak]

This one is 2316 x 1080

[crystal delta]

FAQ: How do I generate variations?

There’s no explicit way to generate image variations, but you can “DIY” it a bit, by re-running the same seed with either slightly different prompt text or slightly different settings. Try shifting the CFG scale value by a few points, reordering the words in your prompt text, or adding another new keyword or two.

[civic crag]

what autolaunch does ?

[dry crow]

starts the webui without you copying the ip in browser

[civic crag]

damn 😮

[smoky oak]

Oh for real?

I didn't know that

Cool

Let me add that lol

[dry crow]

after xformers its my fav argument 😄

[smoky oak]

I just bookmarked the address, so it hasn't been that big of a deal

Please please please let the restart have fixed this issue

PLEASE

anndddd

its working now

so weird

[civic crag]

hmm what does that mean ?

[smoky oak]

I guess I just needed to restart

@wispy nestRestarting PC helped. No idea what happened lmao

[dry crow]

it starts the webui automaticly in browser, you dont need to copy the ip or click on your bookmark

[smoky oak]

waoh-

I just got a massive error

what in the world

ok weird, it seems to spit that out now when I run out of VRAM and it does backend loading

oh well, still seems to work

[dry crow]

nice

[smoky oak]

I love how Xformers just goes to background generation when you run out of VRAM\

its the only thing that lets me go so high res

[dry crow]

yes same, i can go 1920x1080 only with highres fix

[smoky oak]

what GPU?

[dry crow]

even then it sometimes stuck with vram

GTX1080

[civic crag]

oh i didnt try to go higher res since i added xformers

[smoky oak]

I went from a max of 768x1024 all the way up to native 2560x1080 if I push it to the max

before and after xformers

[civic crag]

oh wow let me try

[dry crow]

native? what gpu?

[smoky oak]

it will say it ran out of VRAM, but then continue in the background

3060ti, 8GB

this one is native 1024x2432

same res for this one

[civic crag]

oh it actually works, but result is a bit ... weird

[smoky oak]

you need to use high res fix

high res fix gets rid of those problems

it does a low res gen to get composition, then redoes it at high res

[lilac pilot]

Not sure if I asked but you said xformers is good for 3x+ gpus? It wont hinder potential?

[smoky oak]

if you go too high res on initial generations, you get some really weird generations

[civic crag]

hires fix with no upscale ?

[smoky oak]

no, you have to upscale it

that is the fix

[dry crow]

its good for every gpu 😄

[smoky oak]

so for example, I usually use 2x upscale

[civic crag]

so i gen in lower res and use hiresfix to make it higher ?

[smoky oak]

yes

thats how you get the super clean generations like I sent

like this one

[civic crag]

what is your base res ?

before hires

[smoky oak]

I try to stick with the smallest dimension at 512, as most datasets are trained on 512

so for those images, 512x768 and then 2x upscale

[civic crag]

ok let me try that, thx

[dry crow]

with highres fix

[lilac pilot]

we cant do 1440p resolutions or higher yet?

[smoky oak]

high res fix is insane

[smoky oak]

you can if you manually enter them or use high res fix

high res fix can go all the way up toooo

[lilac pilot]

Oh nice

[smoky oak]

8192x8192

good luck doing that tho lmao

[dry crow]

but sd upscale can get you even higher res xD

making 6000x4000 images or 8-12k

[civic crag]

i usually use "ultimate SD upscale" for upscale afterwards

[smoky oak]

I haven't even heard of that, is that an extension?

I just natively generate at 1080+

[dry crow]

yes and no

under img2img you can select script sd upscale

it will tile the image and render each tile itself

[smoky oak]

ohhhhh

right

I used that once or twice

Before high res fix

after high res fix

[lilac pilot]

is rentry good source for in-depth reading?

[smoky oak]

no idea, but best of luck haha

[dry crow]

depends on the date of the blogs

[lilac pilot]

You too

[smoky oak]

also, I just discovered how dope batch size processing is

8 images at the same time all at once, and its faster than if you did them all individually? And it doesn't use that much VRAM? Sign me up lol

fantastic way to do seed searching

all 8 of these in 50 seconds with full res max speed live preview

cannot complain

and only 7GB VRAM used

[lilac pilot]

Batch count and size should matchup?

[civic crag]

damn i need a new gpu

[smoky oak]

no

[smoky oak]

no

count is how many gens it does back to back

size is how many it does at the same time

so a batch size of 8 does what I just did

all at the same time

final result

so it allows you to see 8 generations at the same time, so you can get an idea for all 8 seeds if you want

[civic crag]

😍

[smoky oak]

I didn't know it worked that way until I hit it on accident lol

and it generates faster than 8 individual images

[lilac pilot]

just did it thanks

[smoky oak]

tada

no problem :p

lets see how fast I can go with no live preview, cause I use max res max speed preview cause I am a pixel peeper like that lol

HOLY SHIT

No preview is WAY faster

OMG

8 images in 32 seconds

thats a 20 second saving 👁️

[lilac pilot]

Sweet, in settings?

[smoky oak]

yeah

you can't see as they generate tho

[civic crag]

oh damn, thx for the tip

[smoky oak]

no problem, its very useful

I am gonna scout 64 seeds

[lilac pilot]

off topic: Can we bookmark text in discord?

[smoky oak]

lets see how long this takes to do 64 512x768 images

[lilac pilot]

Im taking progressbar off and seeing with that too

[smoky oak]

I did that as well

I don't even have my GPU OC on

actually

let me try with the full OC as well

lets see just how fast we can go lol

[dry crow]

i think my batch size is broken xD

getting 3 times the same picture

seed is -1

[smoky oak]

weird, never had that happen before

[lilac pilot]

Reload ui?

[smoky oak]

alright 64 images are almost done

WOW

64 images took...

252 seconds