🏞｜general-with-images

[oak osprey]

and that piece of crap is a nightmare

[smoky oak]

by file

[dark flare]

Sorry to interject, but I think I'm learning, slowly...

The only way I can train a 2.x model on a 12 GB card is to make a lora, right? Otherwise it's gotta be 1.5?

[oak osprey]

use DeepSpeed to train 2.1 on that

[smoky oak]

I should have listened to myself and never updated, now everything is ruined

[dark flare]

You had that FEELING

[wispy nest]

yeah, I kinda figured, that way is less stable because of the pathing

[oak osprey]

step away from it for a minute, that might help

[smoky oak]

no, I am not stepping away, I need this fixed

[oak osprey]

it doesn't help to continue to push at it if you're frustrated, for real

[sterile temple]

I can freeze my venv to a requirements file. you could try updating your venv using that?

[dark flare]

You'll just end up going format c:

[smoky oak]

It somehow managed to delete my gradio folder, I have no idea how the fuck it did that

and I have 0 clue how to get it back

[oak osprey]

using git? git checkout gradio

[oak osprey]

inside the sd-webui folder, can you do git status

[sterile temple]

open a terminal in the venv/scripts folder activate pip install gradio

[wispy nest]

yeah, they haven't installed it by using git I think so it doesn't know what it is :(

[oak osprey]

hmm

[smoky oak]

but this was not an issue before the update, so its not that

[oak osprey]

if it shows as missing in any git status that'll make it easier

might just have to extract the folder over itself again

if it's something that comes with sdwebui

[wispy nest]

that's because you got the issue from updating with all the temp errors you get :(

[smoky oak]

I am gonna be kicking myself in the ass for weeks cause of this, I even said myself that everything would be fucked, but I still did it cause I was speed greedy

[dense tapir]

DB for 2.1 is about 15GB and really 16GB is what it truly needs.

[oak osprey]

oh they'd hate that it just did this

[sterile temple]

I'm testing a new install with the requirements from my working dev branch to see if that works

[smoky oak]

I cannot under any circumstances to a new install

That will take days and days to get back to how I have it

[oak osprey]

have you ever used shadow file copy in windows?

right click the folder and go to Previous Versions tab under properties

[smoky oak]

I have not, no

[oak osprey]

see if it's there

[smoky oak]

it didn't change my venv at all

[sterile temple]

it'll just be creating a new one, then copying the user bat, extensions, models etc over

[smoky oak]

there will be no previous version

[oak osprey]

it'll have your gradio folder etc

[smoky oak]

isn't that what ruined everything in the first place?

[smoky oak]

thats not even located on this drive

my SD is on my D drive, its trying to access my gradio folder on my C drive

[wispy nest]

guys i have some confusion using txt2img and img2img, but let me explain, right now im using txt2img and controlnet and i loaded up an image in controlnet to reflect the pose in my output and so far im getting pretty good results, then why would i need to use img2img or what is img2img is actually for?

[smoky oak]

trying something out of desperation, I'll be back later

[wispy nest]

to show an example of what i did ... here is an image i loaded up in controlnet

[dense tapir]

img2img is take image A and changing it while retaining Image A (for the most part).

[oak osprey]

my god that looks so dangerous

[wispy nest]

have you installed anaconda? and if you're rebuilding it all, try and get it through git pull and not a file download as then you'll have the pathing correct, and when it don't find the correct path, then you can add the correct pathing to the bat file as well

and using that as template i used the following line of prompt to get this result

A portrait of a beautiful girl in a stunning pose. The girl's eyes are a deep, hazel brown, and her lips are soft and inviting. Her hair is long and flowing. In the background is a leafy plant.

having this much control over outputs with controlnet im kinda confused about img2img function

since im using txt2img

[wispy nest]

and some say that this one is "more stable"
https://github.com/anapnoe/stable-diffusion-webui-ux
but I don't know as I never used it myself

[dense tapir]

since you do not care about keeping any of that in that image then img2img is not for you

[oak osprey]

realismmm

[wispy nest]

can you briefly explain what are the basic scenarios when img2img is more applicable?

[oak osprey]

when it works

[sterile temple]

that's one way to spell test

[dense tapir]

When you want to keep the overall image. Could be the colours, etc... and using that image it is used as a guide to get to the final image

[smoky oak]

img2img is also how you upscale to really high resolutions, something that can't be done with just txt2img

[dense tapir]

but can be with tiles

[wispy nest]

okay, i guess i'm overwhelmed with the controlnet feature and how much it lets you control specific elements of an image

[dense tapir]

You take an image as your orientation for your prompt. Depending on how strong you set the denoising strength to it will look close to, or completely different from, your image. It will still have some resemblance in the final image composition even if you set the value to the maximum.

[smoky oak]

ok, I can still generate images, but controlnet is completely toast for me

[oak osprey]

we are well on our way to funny family photos

[wispy nest]

check and update it?

[smoky oak]

its not gonna do anything. Its trying to reference a folder I nolonger have

[wispy nest]

what's the path?

i havent found a speicific use of img2img so far with the stuff i was doing but thanks for explaining, going to look into that

[smoky oak]

C:\Users\SCYTH'~1\AppData\Local\Temp\gradio\tmpjrrn50gt.png

I have no gradio folder in my temp

[dense tapir]

In your case you have no need of it

forget that it even exists

[wispy nest]

and where is it supposed to be? where is your webui located?

[smoky oak]

I need it for controlnet

controlnet can't do anything without it, for whatever reason

[smoky oak]

0 clue

all I know is it used to work, and now it doesn't after I updated

[dense tapir]

Not you, that was to torqx

[smoky oak]

oh, my bad

[dense tapir]

hehehe

[wispy nest]

where is your bat you run to start the webui?

[smoky oak]

on my D drive

[wispy nest]

btw i was trying out pruned versions of controlnet models that are about 700mb in comparison to original 6gb, and they work extremely well, only issue i bumped into was when it asks for specific yaml files

[dense tapir]

yeah, those yaml files are all screwed so you have to grab one and rename it for each model

[sterile temple]

creating a fresh install using the requirements file created from the dev version worked. Now I'm copying the extensions folder, the ui-config.json and web-user.bat files over

[dense tapir]

I forget what is what but in the community on HF there is a post about it and it worked 100%

[wispy nest]

yeah, what i did was copy and rename the default version 15 yaml files after the safetensors

no longer getting erorrs now and outputs are pretty good

[sterile temple]

so clean, should've submitted this to the cow event

[smoky oak]

I don't understand why this is trying to access a folder I do not havvveeee

[wispy nest]

[wispy nest]

this might not help, and it can break stuff, but I let my webui know where it's supposed to go with a new .bat file that I use instead of the one the documentation say to use:

@echo off
cd /D D:\ADD_THE_FULL_PATH_HERE
REM for the above: my path is "E:\stable-diffusion-webui" ie, a folder named "stable-diffusion-webui" in the E drive
set COMMANDLINE_ARGS=
REM set the arguments you want above
call webui.bat``` but it's hard to help without knowing everything how it was installed, sorry :(
But if you want, I can try and find the error, but I'd need a lot info that

(also have the bat in the same folder as the webui.bat)

[smoky oak]

I fixed it

stupid fucking thing

all I had to do was make a folder there called gradio

[wispy nest]

ah, it might not have permission from windows to create folders, or something but as long as it work, then its' great! :D

[smoky oak]

there is nothing in the folder, thats the funny thing lol

it just needs the folder to be there, even tho there is literally nothing in it

[wispy nest]

it's a temp thingie, and yeah, it makes no sense. Only to dark wizards :P

[smoky oak]

I am happy that my half assed and many times unfounded solutions of stupidity do work at least

[wispy nest]

there's a lot of "weirdness" when it comes to code, like why is there a empty extra row after the supposed last row, etc ;P

[wispy nest]

if it's stupid, but it works, then it's not stupid :D

[dense tapir]

Is there no way to save an image in only jpg format in auto1111? I get png AND a jpg.

[wispy nest]

change the threshold when it should make it a jpg instead of a PNG? but why'd you want a jpg? :O

[dense tapir]

I did change the threshold which is why it saves the jpg now too

[wispy nest]

and the "file format" is set to the one you want?

[dense tapir]

yeah, I had to type out .jpg

*jpg

[wispy nest]

and the "save copy of large images…" isn't toggled on?

[wispy nest]

may I ask why you have the images of characters in "landscape" mode? Just curious :P

[dense tapir]

Just joking.

I hate vertical format with a passion

these are all in 1/2 FHD

[wispy nest]

too many videos from people recording with their phone? eh ;P

[dense tapir]

Yeah

I am so glad that we are over that trend on websites, for the most part, as I am all over the net bitching about the sites being in a vertical format (for over a decade) when I am not on the mobile site I am on the desktop website.

[wispy nest]

I'm angry at the current trend of having all corners rounded on websites. It looks so bad! :P

[sterile temple]

stick - the movie

[dense tapir]

I got used to it and rather like it but on these images, for instance, it looks like ass.

after Taco Bell.

[wispy nest]

I can't handle it on loading bars or things like sliders as well, I never know where "0" is supposed to be that way :P

and now I'm noticing the ball and can't unsee it anymore! Gah!

[dense tapir]

What really hurts is a lot of the time I can't even find the slider for the page on a ton of sites

it disappears or is almost the same colour

[wispy nest]

or how the search bar on every site, even on the supposed "good style guide" w3 have it so you have to click it to extend the search area and THEN you can write into it >:(

[dense tapir]

yes

I am opposed to change for the sake of change.

[sterile temple]

@wispy nest triggered? 😜

[wispy nest]

no…maybe…yes…aoifhdisufg! ;P

[smoky oak]

@wispy nest @dense tapir @sterile temple I don't remember who saw, but I fixed the stupid ass issue lol

its lol now, but I was extremely upset before

the fix was literally just making the folder it was looking for. No info inside, just a blank folder

[sterile temple]

glad it's working for you again 🙂

[smoky oak]

same here, I am passionate about this stuff, and I wanna make money with it

its kinda my only income. Now if only Etsy can get off their lazy ass and fix my account >:C

[dense tapir]

speaking of etsy, a content creator I watch was banned over this

he is appealing

[smoky oak]

banned over what?

[dense tapir]

I didn't understand

sec

[sterile temple]

let the cool minty tones soothe you

[dense tapir]

they unbanned him I see

[prime thorn]

Guy
I have a question

[smoky oak]

guy

[wispy nest]

as long as it's not purple I'm happy when it comes to the ai, so, sooo much purple in ai images... I got nightmares about it still :P

[dense tapir]

I think he had a sub-domain with the word etsy in it

[woven dawn]

He's talking to me, I am guy

[prime thorn]

[smoky oak]

ah, if he said gay, it would have been me, glad I was not singled out :p

[woven dawn]

^

[smoky oak]

[prime thorn]

Actually
I found these images but I don't know how to get the same style... I have tried so many loras and models but no luck...

[dense tapir]

I wish I knew how to make money from this stuff but I do not.

[prime thorn]

Like these...

Can anyone help me...

[woven dawn]

Whoa. That looks like a combination of nixeu, wlop and studio ghibli to me.

[smoky oak]

I was about to suggest some studio ghibli influence

[wispy nest]

looks like most results from models found on the civitai site, you can also check the meta data of some of the images there as well

[woven dawn]

Here's a nixeu and wlop embedding. You can use the Studio Ghibli LoRA from civitai.

[prime thorn]

🙌

Thanks!!!!

Imma try rn

[woven dawn]

I look forward to your results!

[dense tapir]

too bad not for 2.1

[woven dawn]

[dense tapir]

I don't mess with 1.5 since 2.0 came out. I play with it every so often to be reminded why I prefer 2.1

[woven dawn]

When we get controlnet for XL I will upgrade

[smoky oak]

I just wish 2.1 had good models. I would mess with it more if there were less deformities and artifacts

[smoky oak]

god, one can only dream of such heaven

[woven dawn]

[acoustic oracle]

how do i reduce vram again

nvm i found it

[dense tapir]

at 4gb --lowvram

[acoustic oracle]

ok

does it have any side effects or no

like does it reduce quality

[dense tapir]

no

it just does the best it can in pc ram

[acoustic oracle]

what effect does it have then

[dense tapir]

you and that "does it reduce quality", lol

use it, stop being timid, and find out

[acoustic oracle]

ok

i was just asking whats the big deal lol

[dense tapir]

I just have an aversion to timidity. It isn't as if it will burn your house down or something.

[acoustic oracle]

what is this

[dense tapir]

are you using xformers?

or sdp?

[sterile temple]

try adding --no-half argument to your bat file

[dense tapir]

no, don't do that if you can help it as I can't even do that shit with 6gb

[wispy nest]

or --no-half-vae

[acoustic oracle]

i dont think so as it says n/a on xformers in the footer

[dense tapir]

show me your command line

[acoustic oracle]

Ok

[dense tapir]

set COMMANDLINE_ARGS=--theme dark --lowvram --xformers --disable-safe-unpickle --port 9000 --api --opt-channelslast

[acoustic oracle]

https://paste.ee/p/CrCAC

[sterile temple]

take them out for dinner first

[acoustic oracle]

oh its a traceback most recent call last error, probably a python glitch

hmm idk tho

[dense tapir]

try my command line I posted

the port I use is 8999 so it can be removed for you

[acoustic oracle]

Ok

[dense tapir]

*9000

[acoustic oracle]

yea maybe the model needs xformers

wait my model came with a config i just realized

which folder is configs

still errs

where do i put configs?

nvm i found it

ayy the config fixed ti

it

[dense tapir]

good stuff

[acoustic oracle]

its rly slow for me (35 seconds) but looks very nice

im on artius21

[dense tapir]

you only have 4gb so expect it to be slow

[acoustic oracle]

hmm how does low vram work

[dense tapir]

uses system ram where it can so a lot of transfers to vram and back to system ram

[acoustic oracle]

oh ok

[acoustic oracle]

cool what model are you on

[dense tapir]

rmadamerge. what it merged with I forget.

it's on civit

[dark flare]

Yeah. Shivam's colab then?!

[dense tapir]

I heard it no longer works

I went to it a few months ago and yeah, dead.

[dark flare]

💔💔💔 whuuut

[dense tapir]

That was my go to for 1.5. I think it will still do 1.5 though

[wispy nest]

On windows you enjoy your weekend playing games
On linux you enjoy your weekend setting up your cd drive
I'm dying! 🤣

[acoustic oracle]

cool, can confirm racoons are very strong

[dense tapir]

Agile too

[compact fiber]

local sd, why do I get this result upscaling the image? Tried to change denoising higheer and lower, tried also masking the main centre but still I'm not sattisfied with the result, any help?

[sterile temple]

are you trying to outpaint?

[compact fiber]

yes, I want the wider pic

[wispy nest]

by the way, I remember 2.1 had funny issues where it was impossible to not get a light source, I remember a funny image where the ai placed a lightbulb without a connection in the middle of a grass field just so it could light it up :P

So, is that still the case? :P

[dense tapir]

That was with me

it was a Big ass light bulb too

just dropped at my feet

[wispy nest]

that image is burned into my mind, it was too perfect!

[dense tapir]

I was the first one to find the issue, to be called names, and people out from everywhere tried to make dark image to no avail. Emad said they would look into it, and did nothing so now we have offset noise and the new one that is 100% true black, but doesn't work on models not trained with it.

[sterile temple]

once you have the wider pic with the blurry stuff on the sides, send it to inpaint, paint over the blurry area on 1 side, then do the other side

@compact fiber

[compact fiber]

genius, thanks !!!

[wispy nest]

that's a huge shoe! or a very tiny house! :P

[dense tapir]

heh

[wispy nest]

and all of the images are with a "photo" style?

happiest dragon I've seen :P

[dense tapir]

Yeah, I am into photorealism mainly since I got into SD when 1.4 dropped

I wish I could video some of these but not that janky video we currently have.

[wispy nest]

how does the photoreal style look like with some Bokeh? I noticed, or I think, it looks a lot better when used. :D

[sterile temple]

i prefer depth of field over bokeh. bokeh adds all those floatey circles in the background

[dense tapir]

Depends. I don't like Bokeh tbh unless in a city street life type scene

[wispy nest]

yeah, I can understand people not liking them, but I love them as they really make the image "pop" to me... or maybe it's just my eyes that pop 😵

[sterile temple]

might have a few popped blood vessels

[wispy nest]

only from purple ai images, and rounded corners in software! >;P

[dense tapir]

LOL

[wispy nest]

I love it when there's something in the forground that's blurry like that, something just make my eyes see it as "more real" :P
Yeah, my brain is weird. Useless thing, always making me forget things and…what were we talking about?

[dense tapir]

As an old nature photog dof is behind and near not really just one most times.

too close and too far

[wispy nest]

dof?

[dense tapir]

depth of field

[wispy nest]

ahhhhhh, I knew that...totally! maybe 😶‍🌫️

[dense tapir]

I have a hard time making bokeh in 2.1 but 1.5 it was part of the trained camera stuff I do miss

[wispy nest]

but I don't see the definition of the word as the thing I want in a ai image, but what the ai "think" the word means. :P

[dense tapir]

in 1.5 we can specify all sorts of photographer specs and it knows. I go to 2.0 and the first prompt I was like wth? Nope, doesn't know them but in 1.5 the clip was trained on over 150 terms

you can, bigtime, fake real photos in 1.5 due to the terms it knows from f-stop, to shutter speed, and film speed.

what lens, what camera, etc...

[wispy nest]

like the Bokeh word for example. I use it to make the blurry dots like it's supposed to, but I have also noticed it making the image have "more/better" depth, the subject move away from the camera, have more side details, etc. That's why, in my opinion, people get confused when a word doesn't do what they think. Because they think the word will "only" create, or not create, that one thing :P

[dense tapir]

Bokeh is just the imperfections of the lens and its shutters. Nothing more.

[wispy nest]

I trained very simple embeddings for "setting the subject/scene," with the help of strong "noise" patterns :)

not to the ai it isn't, and that's the important opinion. That's why I try and trick it into making what it think I want :P

[sterile temple]

someone should make an extension where you can set camera modes and film stocks

[dense tapir]

if you open a lens wide open so all the shutter blades are full open getting bokeh is nigh impossible

I agree

1.5 has some but 2.x nothing

Hassleblad is a big camera for 1.5 2.0 made the damn camera instead in the image

[wispy nest]

here bokeh helped with the focus :D
Or at least how I think it did :P

[wispy nest]

that's why you shouldn't write the name of the camera in the prompt :P

or camera for the matter :P

[dense tapir]

Bokeh is defined as “the effect of a soft out-of-focus background that you get when shooting a subject, using a fast lens, at the widest aperture, such as f/2.8 or wider.” Simply put, bokeh is the pleasing or aesthetic quality of out-of-focus blur in a photograph.

[wispy nest]

55mm is enough in my opinion to make it understand what it should do :D

so guys on mult controlnet i've used canny and openpose together on this base image ... and look at the amazing result ...

[dense tapir]

This is a keeper

[wispy nest]

pretty vivid texture and light

[dense tapir]

I see bokeh

[wispy nest]

that bokeh is pretty cool effect too

[dense tapir]

hard af to get in 2.x

[wispy nest]

I've found that using the name of photo studios, photographers, and general effect names works better than using any name of lens, camera, etc. at best the detail from earlier like 55mm does something, but when prompting a product, then you'd get the product instead ;P

like a random camera thrown into a lake ;)

[dense tapir]

Here ya go walach

lots of bokeh and lens blur

[wispy nest]

I sometimes add void to the blur effect to make it look more "circular" :P

[dense tapir]

void? where do you out that?

[wispy nest]

like a photo of a fog bokeh void :P

fog blurs the lower half of the image for me, or that's how I use it anyway hehe

[dense tapir]

how can the void have a bokeh? I mean it is a void.

[wispy nest]

google void and check the images and you'll see

[dense tapir]

All I see are blackholes

[wispy nest]

exactly :D

[dense tapir]

For only 35 cents per day you too...

[wispy nest]

notice the void? :P

[dense tapir]

that is a 1.5 thing as it did absolutely nothing in 2.1

[wispy nest]

aww :(
That's sad to learn

[wispy nest]

heh, this is what the same prompt I used above did with default 2.1 :P

[frosty vapor]

Is Easy Diffusion the same as Stable Diffusion? And, how do I get results more like the first image? That was generated over on Dream.AI.

[sterile temple]

Easy Diffusion is a UI for stable diffusion. You will probably have to find a model that performs better than the default you get with the Easy Diffusion

[frosty vapor]

Any ideas?

[sterile temple]

I don't use easy diffusion myself, so I am not sure how to add another model. But you can find models on the civitai website

I'm getting pretty consisten bokeh in 2.1 with the prompt: 'portrait of a man outside at night'

[kind quartz]

i like this car.

[weary nebula]

Where's a good place to learn CNet? I'm really struggling to get it to even take my reference into consideration

trying to get this boi to be in a "writing" position which.. seems to be quite the task 😐

[sterile temple]

rotate it 180 and mirror it?

[weary nebula]

Task is a little more complex than that or I definitely would, trying to get it morphed to better be used for a reference model for a hand coming out of a book and writing on the other page

which, requires a bit more than what I said just was trying to keep it brief, and was referred to learning Control net, however, it doesn't really seem to be working the way in which I'm trying to use it after watching about an hours worth of youtube guides ;=;

[wispy ether]

Rocket

What

[sterile temple]

tensor rt

[wispy ether]

I removed it

Its bad

I get a error after using the model 3 times

Something with shape

[sterile temple]

I am not using either. I just disabled it

[wispy ether]

I removed it because it takes forever to load sd

[smoky oak]

whats up, losers? :p

Whats good in the stable hood

[sterile temple]

i'm just playing around with comfy ui

[smoky oak]

oh, how is that going? I tried it a couple times in the past, but it was really far behind on usability. Its been a couple months since

[sterile temple]

I'm been using it for about 10 minutes 😄

[smoky oak]

ah haha

[sterile temple]

I had an idea

something you can't do with A1111

[smoky oak]

when I first used it, it was much slower than auto and used a lotttt of VRAM

[sterile temple]

doesn't seem slower to me

[smoky oak]

I am sure it has sped up a lot since

it was like 30% slower back in the olden days :p

woaaahhhhh

My 3D model design is top 80 out of over 400k designs globally 🤯

[smoky oak]

I see, I see

care to explain? haha

[sterile temple]

it's a image created with both 2.1 and 1.5

[smoky oak]

👁️

Ok, that is cool lmao

[sterile temple]

that is 10 steps with 2.1 to start off, and then finish with 10 steps with 1.5

[smoky oak]

that is a really cool concept I have never thought of

Honestly not sure why you would wanna, but its cool nonetheless! haha

[smoky oak]

ok, question

is it a 20 step gen split between them

or is it a 10 step base gen with a 10 step img2img?

[sterile temple]

i feed the latent output from 2.1 into the input for 1.5 and then send that to the vae decoder

[smoky oak]

ok, so its like a more integraded img2img pipeline?

so its doing 10 full range steps, then doing 10 more ontop?

[sterile temple]

yeah, but I think the image from 2.1 is not denoised maybe? i don't understand

[smoky oak]

very interesting

[sterile temple]

but even if it's img2img, it's easy to reuse anyway

[smoky oak]

yeah for sure, it is very cool

[sterile temple]

I was just thinking 2.1 can do some good things, but 1.5 produces a better image overall

[smoky oak]

now that I have controlnet de-clawed, I am probably gonna reinstall regional prompter and see if I can get anything cool with it

@wispy nest Are you here by chance? I am messing around with regional prompter

[sterile temple]

lol, turns out I'm a moron. The default denoise setting was 1, so it was basically ignoring the output from 2.1 anyway 😄

[smoky oak]

lmfao

[kind quartz]

learning regional prompt, what a weird skies, but nice ring effect

[sterile temple]

so yeah if you turn the denoise down it acts like a img2img

[smoky oak]

yeah, I am gonna have to find a video on how to use this regional prompting, cause their documentation is pretttyyyy bad lmao

oh jesus lmao

[sterile temple]

poochy mama 😛

[smoky oak]

this is rancid lmao

well, at least it is working lol

top thrid is man, middle is dog, bottom is apple lmfao

[sterile temple]

yeah that is working lol

[smoky oak]

yeah, so this regional prompting is NOWHERE near as good as I thought 😅

I thought I would be able to do much more with it

[kind quartz]

for me it is working

[smoky oak]

oh it works, its just not nearly as useful as I thought lol

I don't think this image is real

I think its just a breakdown of how you would prompt the scene, not a result from it

[kind quartz]

it is real you bet, you can copy parameters. I am just bit confused with base and common or nothing or all 🙂

choose that aquarium example

[smoky oak]

it may be real, ontop of an img2img, but based off of what they show at other parts, and my own tests, that will look like dogshit without a consistent prompt lol

[kind quartz]

this is terarium, and their example got aquarium

[sterile temple]

ok, I changed to advanced sampler nodes which lets you stop the sampler short of the final step, e.g. 20 out of 30. Then pass the noisy image to 1.5 to do 10 more steps

[kind quartz]

try switch between base prompt and common prompt or turn them both off

[smoky oak]

and this is what that example looks like when you actually try it lol

[kind quartz]

yes this is different example with aquarium

[sterile temple]

first image is 2.1 at the full 30 steps. 2nd image is 20 out of 30 from 2.1. and last image is finishing the last 10 steps with 1.5

[kind quartz]

try made 100 pictures and you will like some 🙂
Also their example seems to be more wider

[smoky oak]

terrarium doesn't make it any better lol

[kind quartz]

what sampler and how many steps?

[smoky oak]

Euler A, 15

more steps won't help

[kind quartz]

try harder o.k try sde karras with 50 steps 🙂

[smoky oak]

will make 0 difference, but sure

👍

lol

they aren't sharing a lotttt of their process, cause the actual addon is nowhere near as good as they seem to make it

[mystic marlin]

Man I need to fix this image. Almost there too. lol

[kind quartz]

i am satisfied, no need to convinced others. You are using some weird model imho.

[smoky oak]

I have tried it with 11 models

it doesn't work NEARLY as good as they make it seem

[kind quartz]

and what prompt you have on?

[smoky oak]

the one they provided

[kind quartz]

Try turn them down all

[smoky oak]

this looks a little better, but its still nothing like the examples they give

I'll give high res fix a shot with it

I am convinced that regional prompting is doing nothing here

[kind quartz]

it is doing. Just layout. For me it is working 🙂
quality of image is up to model imho.

[smoky oak]

you can see here, its not following their grid at all

I doubt the positioning is doing anything more than just a standard prompt here

[sterile temple]

do you have image preview turned on?

[smoky oak]

result with no regional prompter on lol

[sterile temple]

you might be able to see if the image starts off using the regional prompts and then just gets taken over by one of the prompts

[smoky oak]

I do not, but I can give it a try

the image gens really fast tho, not sure if I will be able to see much

[sterile temple]

I saw that happen with the neutral prompter

[smoky oak]

lmao

thats a very nice desk

ok, I think it is starting to work a little better now, but it is still far from what they show

[kind quartz]

it sometimes skip content in prompt, but one can try it and learn, in my case it is working. And i have not any anime model... You have to try several seeds. Also you should have same size and AR, not sure they provide it.

they posted cherrypicked example of course.

[smoky oak]

ok, I got it working properly now. They messed up their example prompt, they had the prompting parts in the wrong areas related to the image below, thats why it wasn't working reliably

That explains why it was such a mess and didn't line up, cause the way they laid theirs out, it was incompatible with the layout below

[kind quartz]

Never judge product by its manual 🙂

[sterile temple]

would be easier if it was like the regional prompter in the tile diffusion extension

[smoky oak]

this is MUCH better

now that things aren't misaligned from their example

[kind quartz]

try add line above, like something photo-realistic or something. If it will work

Anime, studio Ghibli
this for example, if it change scene very.

[smoky oak]

much much much better now that they are lined up right lol

[wild sorrel]

what's this mess

[smoky oak]

actually just realized that the gens all gave him reddish hair, how intersting

[wild sorrel]

try to change sunset to something else, maybe it'll change hair color

[kind quartz]

mans head probably problem?

[sterile temple]

first thing I thought of when I saw the column mode was the four seasons in 1 day

[wispy ether]

:(

why is it red

[dry crow]

You opened a ( and never closed it with )

[wispy ether]

where

oh

[tropic shell]

I'll take these chalk...pastel four seasons in a day.

That's a print worthy kind of thing

[sterile temple]

aw, cheers 😊

[tropic shell]

🥂

[rain gazelle]

Posted a couple arnolds to #🎥｜animation ... this face change thing is wild.

[sterile temple]

did you make?

[rain gazelle]

I mean.. software made it. But I generated them.

[weary nebula]

Anyone experienced Control net?

[sterile temple]

did you make?

[rain gazelle]

Echo in here? 😄

[sterile temple]

weird

lol

[weary nebula]

did you make?

[weary nebula]

truth

[smoky oak]

Oop-

The internet needs to chill lmao

[weary nebula]

Nah

[sterile temple]

what the, what is my discord doing lol

[weary nebula]

Nah

[smoky oak]

The internet needs to chill lmao

[wraith karma]

Clerics entering a dungeon

[weary nebula]

Nah

[dark sparrow]

Via Midjourney, Although I told it to create normal headphones, it proceeded with a burger headphone🙄

[dark sparrow]

But Midjourney is accurate while in 99% of the other free image generating AI, the output is completely irrelevant to what I prompted

[wraith karma]

You are not including SD in that list, are you?

[weary nebula]

you're not even in midjourney

[weary nebula]

Im struggling hard with SD lol

[sterile temple]

I sure am having a hard time getting what I want /s

[dark sparrow]

But Midjourney is accurate while in 99% of the other free image generating AI, the output is completely irrelevant to what I prompted

But Midjourney is accurate while in 99% of the other free image generating AI, the output is completely irrelevant to what I prompted

[sterile temple]

this channel is echoing messages

[wraith karma]

Wait, why're you sending this again?

[weary nebula]

I downloaded it trying to get it to help me with a MJ problem but its struggling more xD

Discord is dying right now

MJ bots having problems and their discord is too, I bet discord is having a giga fuck up rn

[wraith karma]

Well, SD takes some great time to generate what you want, but you get there at the end

[weary nebula]

I uh, don't think Im getting anywhere xD

I think I'm using control net wrong in some way

I downloaded it trying to get it to help me with a MJ problem but its struggling more xD

[wraith karma]

@weary nebula that's how I got to the image I wanted

[weary nebula]

kek

[sterile temple]

nite all 🌙

[weary nebula]

feel free to try and take a stab at what Im working on lol

[oak osprey]

o/

[oak osprey]

is discord ok lol

[oak osprey]

making a realism model is easy

painting

[umbral glacier]

Chad painter and Chad poet

[crimson dawn]

its corn!

[ripe cedar]

The 3090 is around 700 USD on ebay these days. 3060 is around 300. Wish I could upgrade

[crimson dawn]

when i wanted a 4090, i asked my boss for extra shifts. didn't get enough though so i only got a 4080

[wispy ether]

@sterile templedid you get tensorrt fixed tho

[wispy ether]

or you could have saved longer

[crimson dawn]

no. it was mybirthday. i wanted gpu now! also i'm saving again. the titan ada is a possibility. 48gb cards will drop eventually

my case is ready

https://youtu.be/_KpxRmoc4og this video hahaaaaaaaaha

[oak osprey]

this channel's family photo

[wispy ether]

ok?

tf

[dense tapir]

Are there any blood, guts, gore models out there? None of that fake stuff I mean the kind that you could replicate a murder scene with?

[boreal falcon]

probably

Unstable Diffusion must be the place for that

[dense tapir]

Naw, that is mostly porn

or it was

[wraith karma]

@wispy nest why such reaction?

[dense tapir]

@smoky oak I just copied auto over and git pulled the latest so what is this?

[wraith karma]

6 fingers tho

[wispy ether]

hm

[wispy ether]

fartnite

[vocal cloud]

Blood For The Blood God

[smoky oak]

No idea honestly

[dense tapir]

Nor I. I see we received a new sampler.

It magically appeared on my drop down where my old one was

[smoky oak]

oh, so it IS different

[dense tapir]

yes

I am going to try my old version because this is hideously slower

3 mins for that

smh

I thought that maybe it just needed less steps.

[oak osprey]

i will test it on my model

[strong sonnet]

https://github.com/AUTOMATIC1111/stable-diffusion-webui-tensorrt

Which one should I download?

[oak osprey]

hmm lol

[oak osprey]

DDIM does better

with the DPM++ 2M SDE Karras scheduler i have to crank the sampler steps up and the cfg down

[smoky oak]

I am curious to know why regional prompter doesn't work with high res fix all of a sudden

it did the first few times I used it, but not now

weird

[oak osprey]

it can get interesting results though

[smoky oak]

GA

just be warned, tensor RT is faster, but its not very useful, you will be extremely limited in what you can use it on

[oak osprey]

same prompt and seed and everything in DDIM and i prefer the Karras result

[smoky oak]

doesn't work for anything over 840x840, doesn't work for multi batch size, doesn't work for high res fix, doesn't support controlnet or most addons, and can't be used with additional LoRA's and networks

Tensor RT in general is

[oak osprey]

tensorrt is likely meant for hosting providers that intend on a model working as quickly as possible over an API

[smoky oak]

It was way faster for me, but IDK what the point of having extremely fast single image gen is, when its still slow at multi lol

[strong sonnet]

can we expect them to be fixed in the future, I just want to try it out

[smoky oak]

possibly, no real way to know

[oak osprey]

did you try just onnx?

[strong sonnet]

I pasted it in extentions folder

But I cant see here

[smoky oak]

What do you mean?

[strong sonnet]

What might be the issue?

[smoky oak]

you have to do a lot more than just that

[oak osprey]

without tensorrt. onnx has its own optimizations but should be more useful than tensorrt

[strong sonnet]

what should I do?

[oak osprey]

thing is i'm not sure how well onnx is integrated into sd-webui

[strong sonnet]

what does this mean

I didnt get it

[oak osprey]

that platform's architecture is showing its limits now that there's all these formats and possible inputs/output changes

[smoky oak]

you have to push your SD install to the dev branch, which resets a lot of things in your install, and you also have to past the folder in a different location

And then you will have to restart everything, then select the model you want and convert to ONNX, which takes a minute or two, but then you need to go ONNX to TensorRT, which can take up to an hour

it really is an annoying process for something that is considerably faster on some hardware, but kinda pointless and not very useful

[oak osprey]

wow an hour? lol it took 30 minutes on my server

[smoky oak]

it says 15 minutes to an hour

it took 4 minutes on my PC the first time, and then 19 minutes the second

[oak osprey]

well it sounds lke it's quantizing it

down to like 4 or 8bit

[strong sonnet]

how to push sd install to dev branch?

[smoky oak]

its making a modified unet, but what that entails, I am not sure

What I DO know, is it makes 0 quality/pixel difference

[smoky oak]

that is something somebody else is gonna have to help you with. I am not confident enough to help, as it has a very high chance of just ruining your SD install lol

it tried really hard with mine, but I was able to fix it

[strong sonnet]

oh ok

[acoustic oracle]

I dont know how a c919 became teddy bears but i love it.

[smoky oak]

I personally don't think its worth it, especially cause it is so restrictive

[smoky oak]

@strong sonnetWhat GPU do you have?

[oak osprey]

i made a realism model

[strong sonnet]

3080(laptop)

[smoky oak]

you would probably see a decent speed gain, but its not like you would need it at basic 512x512 gens

my only current dislike for regional prompter is that it completely trashes realistic images

with regional prompter

without it

it really does just destroy any semblance of realism lol

ALSO

@dense tapir When did this happen!?

[oak osprey]

"My ventriloquist dummy has assumed a human form and--- hello? 911???"

[smoky oak]

@sterile temple I did an image inspired off your 4 seasons!

[dense tapir]

👀

[dense tapir]

I rolled back as I just didn't like the new shit

I'll stay in my comfy zone

[smoky oak]

for me there was no rolling back, so I just accepted my fate lol

at least I don't seem to be giving anything up at the moment

[dense tapir]

I copied my entire 70 gigs folder and used the copy

[smoky oak]

I backed up almost everything, and the update still managed to affect it :/

[dense tapir]

the key word is "almost".

[smoky oak]

it also caused those issues with controlnet, which I fixed by doing the stupidest shit ever lol

[dense tapir]

well, since it was only working in the new copy folder it didn't touch the original folder at all as it had no idea where it was.

I was still on the old python but the shit was slower

The new sampler I have no idea what to make of it

I really have this sense that shit is just being done because it can not because it brings anything to the table.

[vocal cloud]

I feel like that's how the vlad fork has been this past week.

[smoky oak]

I don't do anything with Vlad now

he is an extremely rude person, and I just don't like him, personally

[dense tapir]

I left that not too long after I downloaded it. It zapped all of my images I saved for their prompts. Never going back to that dickhead again. Yes, I publicly called him that on Gthub when he said shit to me. No worries, I don't miss him, but I really do wish we had a real choice besides auto and forks.

[smoky oak]

Vlad is a jerk with a stick up his ass, IMO lol

and Auto is a flaming racist

so we have pompous ass who doesn't care, or racist ass who doesn't care

[dense tapir]

summed it up

[oak osprey]

state of the world rite there

[dense tapir]

Vlad likes to attack people on his github so I didn't take it personally he is just an ass

[smoky oak]

I too would enjoy some good options 😅

[dense tapir]

ComfyUI is so damn close if he could implement nodes in the back and give us an auto like interface in the front. we want the real power we can go nodes. He despised gradio as much as I do so I still hold out hope he can do it as DaVinci Resolve does it.

[smoky oak]

I would be super down to switch if it gets that usable

for now its just a little too clunky and obtuse IMO

there really isn't much as to why I wouldn't wanna use Auto, other than the fact that hes a flaming POS racist lol

[dense tapir]

Nodes is pur power but it is anything but comfy to work in

[vocal cloud]

Never noticed any behavior like that from Vlad on github. Maybe a sort of bluntness, yes. I haven't experienced anything extreme (yet? lol) But, anyway, yeah I contemplate switching back to a1111 sometimes. Main reason why I switch to vlad was because of token merging and it was a little faster at the time, but a1111 has access to token merging for a while now to, so...

[dense tapir]

I just saw a post about token merging and it said the results, with visuals to prove, the outcome is inferior

[smoky oak]

Vlads implementation has a lot of problems with it, but he doesn't care. You share problems with him, and he basically tells you too bad

[dense tapir]

I wish more would test.

[dense tapir]

Well, I fired him and feel good about it.

[smoky oak]

haha, I like that outlook on situations lol

[vocal cloud]

I only run merging at 20%, so the outputs don't really suffer much loss in quality.

[vocal cloud]

You know, I think I've kinda noticed that a few times. haha, didn't think much of it.

I guess you're right

[dense tapir]

I am not even sure what it is but with 2.1 having issues already I don't need to compound the quality ones.

[smoky oak]

General and I had some issues and General presented them to Vlad with literally every screenshot of the issue possible, to which Vlad said there was not enough info, and we were not trying hard enough, and to stop wasting his time

[dense tapir]

Yep

The he said I was, basically, a cry baby. I was like wtf?

[vocal cloud]

Oh yeah, he seems to prefer a particular format when you submit issues. Does not prefer screenshots.

[dense tapir]

that is when I said I might be but you are a dickhead. He deleted the thread

Yeah, but you have to have screenshots

[smoky oak]

It was a problem he himself created, and all that we had was the error codes in CMD which basically said "there is a problem"

[dense tapir]

yep

[smoky oak]

so that was what was submitted, along with everything else important, and he was an asshole about it lol

[dense tapir]

I suspect he must not be able to read English so he can't translate screenshots or he is blind, or something.

[bronze cobalt]

Hey can you help me with this error? Im trying to covert ONNX to TensorRT and getting this error

or does anyone know why am I getting this error?

[smoky oak]

he sitty

[vocal cloud]

damn, well sorry to hear you had a shitty experience. at least you don't have to deal with that anymore. If I experience any harrowing issues with the ui later down the road, I may end up joining the club, lol.

[smoky oak]

I just find it funny that he said there was not enough info... whenhe himself is responsible for the error codes lol

[dense tapir]

DAMNNNNNNNNNNNNNNNNN 😦

Here we go

[sterile temple]

nah I didn't

[sterile temple]

Beautiful, nice one

[karmic sedge]

You have to use the full path to the file

And make sure the output line is empty

[sterile temple]

Mondays... am i rite?

[wraith karma]

Damn, general opinion about using AI to generate images is so bad.
My tries at posting images I generated on several Discord servers were met rather negatively

[dense tapir]

Yep

[sterile temple]

Some people think it requires low effort compared to what they 'real artists' put in, some people think training AI models is stealing

[dense tapir]

We are a "toxic" community, and the scum of the planet according to some.

[sterile temple]

that too

[wraith karma]

Well, I mean, effort is incomparable, that's true

[boreal falcon]

partially, yes

😄

[dense tapir]

The only thing toxic about this community is when @smoky oak eats at Taco Bell.

[wraith karma]

In the very least, making close-up image of well-known fictional character is extremely easy to do

[smoky oak]

I don't even eat Taco bell! :p

[boreal falcon]

i mean this Discord server itself is far less compared to that Reddit one

[dense tapir]

My point exactly @smoky oak

Reddit is straight up trash anyway

[wraith karma]

Wait, what does happen in SD Reddit?

[boreal falcon]

sometimes i think Reddit is kinda like 4Chan

[dense tapir]

I finally unsubbed that one and went to a friendlier sub that was like it was in the early days.

[boreal falcon]

the ultimate cancer

[dense tapir]

More like a 4chan wannabe

[boreal falcon]

cant get more toxic lol

[smoky oak]

There is some pretty cool stuff in the SD reddit, but man people can get shitty in comments lol

[boreal falcon]

well true

[dense tapir]

Someone clued me in on a different sub that only cares about the art side of it. Like that one used to be until the troll, and anti AI dolts appeared.

[wraith karma]

I've only used Reddit for searching game advices, I have very positive opinion about it

[boreal falcon]

depends on subreddits

[dense tapir]

Gamers are even more toxic than the anti AI people

[boreal falcon]

but man looking at SD Reddit but also some others...

[wraith karma]

Oh well, I usually play hardcore PVE

So it's fine

[dense tapir]

Yep, I despise PvP with a passion for 23 years now.

[vocal cloud]

SD reddit has some gems every once in a while regarding new tools, tips or techniques. But 90% of the time it's just art posts.

[wraith karma]

With shitshow in conments?

[sterile temple]

facts

[vocal cloud]

and 4chan Stable Diffusion General /g/ is just lame fucking 1girl post 24/7

[dense tapir]

ugh

[smoky oak]

Money Worcky wokk
Alurckay Poriky

[dense tapir]

some of the 1girls I straight up don't get

[wraith karma]

Because they're the easiest ones to generate

[dense tapir]

I mean almost real but still anime. Mental mind virus

[smoky oak]

I am so tired of just sexualized girls 24/7 in AI

[vocal cloud]

That is true, yes.

[dense tapir]

Well, yeah, and hard to get non sexualized beef cakes too

[sterile temple]

When I look at reddit, I think it's probably 80% teen boys, which explains most of the content/comments

[vocal cloud]

1girl definitely is that though, haha. "a mental mind virus"

[smoky oak]

I do sexualized men all the time, sure, but I don't feel the need to make my attractions other peoples problems

like, I know well enough to know that nobody cares lmao

[dense tapir]

Yep, but it is damn hard to get normal people.

[wraith karma]

That takes like one lora and something like 'epic realistic, masterpiece, detailed, cinematic lighting' and done

[dense tapir]

Normal as in people you see on an average day

[smoky oak]

I can do it pretty reliably with how I have been working my understanding of realism, but it is annoying that attractiveness if prioritized

[dense tapir]

How about some 1s or 4s or even 7s? Nope, all 9s and 10s

I was reading a woke ass article yesterday about this very topic

The left is hitting this HARD because of the 9/10 instead of "normal" people. I mean I don't see these women on an average day.

Clubbing, even then 9/10 are super rare.

[wraith karma]

It depends on where you live

[dense tapir]

No, not really.

most people are not 10s

[wraith karma]

I see lots of beautiful women and student girls constantly

Just by going to Uni

[dense tapir]

Well, good on you but that is 100% not indiciative of real life outside that bubble.

For once I was starting to agree with the woke on this issue.

[sterile temple]

lol americans talk funny, left right woke 😄

[smoky oak]

it doesn't have to just be woke lol

Its common sense that AI is biased towards attractive people

[dense tapir]

Author used the term as a badge of honour

[smoky oak]

I agree, its fucken stupid lmao

Politics in the US are all BS anyways, and both sides are fucked lol

[wraith karma]

Let's not get into politics

[dense tapir]

It is a training bias as the article claimed

[smoky oak]

yeah, definitely is

[oak osprey]

i just wanna complain about discord for a second, lol what the fuck @ that

why can't i see his name

[smoky oak]

its the servers fault lol

oh, also your theme

[oak osprey]

omg

there's themes?

[smoky oak]

yeah lmao

[dense tapir]

I agree with that part of it because are you telling me there are no ugly people in the dataset OR, which is more likely, the trainers (iow SAI, et al.) purged them from the dataset?

[oak osprey]

whaaat

[sterile temple]

That is very dark

[smoky oak]

you are using the same theme as me

[oak osprey]

holy crap they switch it on when you do Nitro

[smoky oak]

yeah lol

They are on mobile too

[oak osprey]

man none of the themes work with black names?

[smoky oak]

I mean, its a tradeoff to have a dark theme

I am not sure why the server decided that black names are a good idea lol

they are 100% invisible on discord mobile with full black

[sterile temple]

I'm just using the default theme, it's a dark grey, so I can read the black names

[oak osprey]

jesus wept

[dense tapir]

There is a big bias toward Asian young women too I find across a lot of models.

[wraith karma]

Well, I mean, why would someone train their anime model on bad drawings, for example?
On photorealistic models things might be different

[oak osprey]

stealth trolling mode engaged

[wraith karma]

That's because lots of models are trained by Asians, guess what

[dense tapir]

I mean real images trained on not paintings but even then there are a lot.

[dense tapir]

In the base model too?

I found a niche a photorealistic model of sub 8 looking people.

[wraith karma]

Aaahhh.
I mean some top models from CivitAI, for example, are trained by asians

[dense tapir]

NON Asian

[vocal cloud]

That's gold, not black.

[smoky oak]

Young Asian women are the single most fetishized group of people on earth, so it makes sense that sick and perverted people prioritize them in datasets

[vocal cloud]

[dense tapir]

There is a term for that.

[wraith karma]

Lolicon

[smoky oak]

Yeah, several of them lol

[wraith karma]

[smoky oak]

*illegal content

[wraith karma]

Well, for them it's not illegal

[smoky oak]

it is, just not punished

[oak osprey]

hey turns out its actually black and your monitor is broken

[dense tapir]

Some models I couldn't get any other race from so I ditched them.

[oak osprey]

i asked an expert

[smoky oak]

lol

[vocal cloud]

Must be because I'm on a CRT monitor

[wraith karma]

A shit ton of japanese comics, for example, are in that genre.
I doubt they're all illegal

[sterile temple]

is it a mitsubishi?

[vocal cloud]

idk what that is

[sterile temple]

I had a mitsubishi crt monitor

[smoky oak]

Japan has a very very different legal system, where its more transactional, and child content is a very mild problem in their system. Just cultural differences

[vocal cloud]

I'm just playing. Nah, i'm not actually using a crt. Lol

[dense tapir]

On 1.5 would someone use this prompt and see what people come out cause on 2.1 most of them they are Asian young women. Prompt is - Stunning Futuristic World: A picture of a futuristic city teeming with technology, mega buildings and flying vehicles. The picture shows sparkling lights and amazing light effects, and the characters appear in costumes with unique designs that reflect the advanced futuristic world.

[vocal cloud]

Didn't know mitsubishi did anything other than cars, tbh

[sterile temple]

what resolution?

[dense tapir]

any is fine though I am 869x536

[sterile temple]

you can get mitsubishi air conditioners here

[dense tapir]

*960x536

[wraith karma]

That's some really strange definition

As close as possible to full hd after upscale?

[dense tapir]

yes

I can't do 540

536 or 544

[wraith karma]

@sterile temple sooo?
I'm interested as well

[sterile temple]

that's with no negative prompts

but it's my mix I specifically avoid sexiness 😄

[smoky oak]

I'll try it

[dense tapir]

k

2.1 is just overloaded

[wraith karma]

Those robot hips looking sexy AF though

[dense tapir]

Here is the 2.1 non ema base

[smoky oak]

hmm

[dense tapir]

Notice something?

See, wtf 2.1?

[sterile temple]

sproing!

[smoky oak]

I am removing my neg

[dense tapir]

WTH is with the penis building?

[wraith karma]

Brooo

Are you serious?

[smoky oak]

no neg

[sterile temple]

are you expecting people, because I think it will lean towards what's at the start of the prompt

[dense tapir]

Notice mine didn't?

2.1 vs 1.5

The 1.5 ones are more what I was thinking, and 2.1 base close enough

[dense tapir]

No people?

[sterile temple]

i moved the character part to the start and got a person

[dense tapir]

but not people

very odd. CLIP vs OpenCLIP

[smoky oak]

the prompt isn't exactly well worded, TBH lol

[wraith karma]

Isn't it natural when generating in wide definition to get more than one person?

[dense tapir]

yes

[dense tapir]

Blame lexica.art

[oak osprey]

in 2.1 it's less problematic depending on your prompt, training data, and resolution

[dense tapir]

besides I am not after that I am after seeing what does what.

Interesting experiment and I am glad it was garbled to see how each handled it

[smoky oak]

here it is ina 2.1 model

[smoky oak]

2.1, no neg

[wraith karma]

2.1 seems more grounded somehow

[smoky oak]

2.1 has its own look for sure

not a look I tend to like, personally

[wraith karma]

1.5 had much more creativity for this prompt

[smoky oak]

yeah

[wraith karma]

Easy to compare close. Left one is 1.5, right one is 2.1

[oak osprey]

what is the prompt?

[wraith karma]

On 1.5 would someone use this prompt and see what people come out cause on 2.1 most of them they are Asian young women. Prompt is - Stunning Futuristic World: A picture of a futuristic city teeming with technology, mega buildings and flying vehicles. The picture shows sparkling lights and amazing light effects, and the characters appear in costumes with unique designs that reflect the advanced futuristic world.

[dense tapir]

Yep, you see the same as I do

[smoky oak]

2.1 looks much more bland

[dense tapir]

I prefer it over 1.5

[smoky oak]

and I am not even using like... an art model for 1.5

[smoky oak]

I guess it is personal preference, but I prefer creativity

[wraith karma]

Maybe the opposite?

[dense tapir]

I prefer things to be more grounded and why I despise the over usage of LEDs in chassis for computers.

[wraith karma]

Having same style among several prompts is a questionable thing

[sterile temple]

here is a full 2.1 and a 2.1 where the last 10 steps are done with a 1.5 model

[dense tapir]

We have to nail this down even if that means gan before we can use this for story books, and motion pictures.

[smoky oak]

the 1.5 refined version looks cleaner IMO

[oak osprey]

the future is a parking lot

[dense tapir]

how do you do that? last 10 steps are in 1.5?

[smoky oak]

comfy UI

We were talking about it a little while ago

[dense tapir]

I like the second image it is crisper, has less blur

Oh

SEE, Comfy has some serious power but damn

[sterile temple]

this is the image 1.5 gets from 2.1, then it does steps 21 to 30

then I use the VAE from 2.1 to finish

definitely helps clean up the final image with less blotchy compression

[smoky oak]

1.5 does have higher "detail" at the same res as 2.1, which has been consistent for a while

[dense tapir]

give me automatic1111 with a node button and that would be great. Nodes are automatically connected and I can change, or add, as I need as I do in DaVinci Resolve.

[smoky oak]

I really wish as well

[sterile temple]

here is the workflow for comfy if you ever want to try it

it's basic

[sterile temple]

probably can be improved if you want to add loras and use prompting that is better for 2.1

[dense tapir]

I just can't get with Comfy and I tried it as soon as it came out as I love nodes but that sucker is far too much. I can't just get in, generate, and get out I have to connect all kinds of shit by hand, etc...

[oak osprey]

no negative prompt there

[dense tapir]

Honestly DaVinci did nodes right

[smoky oak]

FL studios patcher is incredibly good nodes as well

[dense tapir]

I haven't used FL Studio in a decade

[smoky oak]

A node network with standardized connections between programs, and then a parametric UI ontop that you can link in any way you want

[dense tapir]

I went to Ableton

[smoky oak]

That is shocking, coming from you 😅

[dense tapir]

Though I know the window you are talking about where it connects the modules/plugins

[smoky oak]

I would have never thought you would choose abelton and their very predatory pricing over FL

[sterile temple]

The weirdest node UI i used was for a PA system at a previous job. Way over the top complicated for what it was used for, but really powerful nonetheless

[oak osprey]

nodes for audio processing are the best

[dense tapir]

I am not all OSS/OSH because FL went in a direction I didn't like. Fruity Loops was the best, then it became corporatized like and I went to Ableton Live

[smoky oak]

you can see here how it works on the bottom layer, and then I was able to connect it all to a custom UI

That is also like, an extremely important rack for my audio work, so I hope nobody is able to reverse engineer it lmao

luckily you can't see the settings and direct values lol

[oak osprey]

always trying to keep secrets

tricksy hobbitses

[smoky oak]

I am willing to share a lot with Sd, as thats kinda what its for, but not music lmao

You don't survive by being nice in music

[sterile temple]

fruity loops has been around forever, I remember playing around with it when I was a teenager

[smoky oak]

yup, most common audio production suite in the world

also still lifetime updates, even 25 years later

[dense tapir]

I used it from 3.1 or was it 2.7? I forget now

[oak osprey]

me, figuring out Sytan's secret values and publishing them, because knowledge should be free

[smoky oak]

luckily its not possible from the screenshots I shared lol

this color rack can do some incredible things

[sterile temple]

it probably got so popular from being pirated by lots of young people at the time, they grew up and used it professionally and helped to grow it. just a guess 😄

[smoky oak]

tho, it is not as important as my squelchy rack, thats the one that people actually want lol

[sterile temple]

kind of like PS

[oak osprey]

i made a character from Fallout 76 real

he wants to tell you conspiracy theories

back in his earlier years before the conspiracies really took a grip on him

[dense tapir]

Have to be a patreon to get them 😦

all men it appears to be at 4k

for reg images

[smoky oak]

I think probably the coolest thing about FL studio is that you don't have to pirate it in order to use it. The free version is the full version

[oak osprey]

if i try and make him happy, he becomes a different person. its like a metaphor

[smoky oak]

And there is no time limit

[sterile temple]

oh really? that's cool. I don't remember how I got it at the time. I just assume 😄 ah, the days of kazaa emule etc

[dense tapir]

WHUT? There is a new Lora optimizer now fp8?

[oak osprey]

my system generated like 400 of these as validation images during training overnight. it's weird to think my server was dreaming about family photos

they're all way better than what Bing can do

if you set your expectations low enough, you can achieve greatness

try to beat this with Bing

[smoky oak]

Yeah, elasticity is the best when it comes to using it without paying

For example, they have their own system where if you have a friend that has a paid version of FL studio, you can collaborate with them on a free version and it will give you full access

I bought my FL license when I was 12. One of the best purchases I ever made

[dense tapir]

HERE WE GOOOOOOOOOOOOOO https://forums.developer.nvidia.com/t/ada-geforce-rtx-4090-fp8-cublaslt-performance/250737

[sterile temple]

Nice, more software companies need to make it easier for people to pay if they are able to. You want to capture the young people who are the most creative, then they grow up and are happy to pay for software they appreciate

[smoky oak]

Tho I do need to Google and see if I can change my activation name, cause it's not the same as my artist name anymore

[oak osprey]

[dense tapir]

Seems throttling due to clock speeds on the 4090

[smoky oak]

It also helps that I feel studio is way more affordable than alternatives like logic and Ableton

[dense tapir]

Damn, my bid for a 3090ti lost

[smoky oak]

I'm surprised general of all people supports Ableton

[oak osprey]

whats that movie with the mischevious kid

Dennis the Menace

[sterile temple]

all movies

[smoky oak]

They are outrageously expensive, and you have to pay for each update, and they use proprietary technologies for lots of things

[dense tapir]

I will be damned if I pay 1k+ for a 3090ti

[woven dawn]

I'll be damned if I can't sell my 4090 for at least $1400 when the 5090 comes out.

[dense tapir]

Might can if the 5090 is 3k

[woven dawn]

5090 will be 2499 for sure

[dense tapir]

100%

[smoky oak]

Damn, now I wanna nerd out with audio lol

As much as I like AI, music has sooo much more creativity in it

[woven dawn]

I'll teach you music theory for $10usd / hr