🏞｜general-with-images

[dense tapir]

they admitted to manipulating

[violet tulip]

it was duckduckgo

ye

i was wrong

[dense tapir]

I warned people about them years ago but I was the conspiracy nut

I wonder how well LoRA will handle a 1 image training?

[violet tulip]

i'll try to get this shark working

[dry crow]

So yea use firefox, with ublock origin, problem solved

[violet tulip]

quick q since it doesnt dsay

[dense tapir]

ublock origin ftw

[violet tulip]

do i clone shark inside of auto1111 folder?

[dry crow]

Shark is a webui for itself so no

[violet tulip]

or is it its own thing

[dense tapir]

I also use this for years too

[violet tulip]

ooo ok

only thing that i find that works is ublock

[dry crow]

Not needed anymore, since the training stoped and it raises your fingerprint

[violet tulip]

the zap tool is the shit

[dense tapir]

I still use it as it gets hits when ublock doesn't.

[violet tulip]

zap tool

works every time

[dry crow]

Show me one site where ublock fails xD

[violet tulip]

YIKES

I... I mean hey nice

if ublock is failing, it means you prob didn't even change what lists its using in options

[dense tapir]

I often wonder which is before the other in the chain as I will get hits on eff when I shouldn't if it is after ub origin (which is what I want).

[violet tulip]

ublock has many options

[dense tapir]

the order is important

[dry crow]

Ublock should be enough you can activate more blocklists in the settings, havent use pb and never saw ads

[violet tulip]

there is 1 other blocking ext i use

[dense tapir]

I don't care about ads as I don't see them either I care about other things.

[violet tulip]

adblock for yt

it actually works

[dry crow]

Dont needed with ublock

[violet tulip]

i do

[dry crow]

Dont use it xD

[dense tapir]

Check this out as I have been sitting here on google drive for a few hours

[violet tulip]

thats fine

i do

it works great

[dry crow]

Its completely unesseary

[violet tulip]

for you it is

not for me

and that's ok

[dry crow]

Then i would check my Browser or other extensions

[violet tulip]

i'm not you with your pc

i'm me

[dry crow]

Yea i just give Information

[violet tulip]

what works for you might not for me

and thats ok

[dry crow]

Can also recommend "I still dont care about cookies"

[dry crow]

Yea but in this case it should work for you xD

[violet tulip]

XD

[dry crow]

If not, then something is wrong

[violet tulip]

XD

XD

XD its so funny

XD

:/

this guy

[dry crow]

Other things? Trackers?

[violet tulip]

XD

lol ok i'll stop

[dry crow]

Thx

[dense tapir]

trackers, yeah. I haven't seen ads in over a decade.

[violet tulip]

it always sucks when there's a know it all in the room

just sayin

[dry crow]

Yea ublock filters trackers too, more then pb does

Depends on the settings too

[dense tapir]

Well, I am happy how mine is setup so we are good.

[violet tulip]

^

ok trying out shark, wish me luck

[dense tapir]

I wish there was an automatic caption for batch work that I didn't have to go in by hand. Maybe someday just not today.

luck

[dry crow]

Yea its fine just want to get facts right 👍

[dense tapir]

I use 5 caption models and always one will be the closest but never the same one.

[violet tulip]

Oh, AI is doing text in art now?

[dense tapir]

rare but sometimes they are so far gone none are even close

[violet tulip]

that's cool

[dense tapir]

yes

[violet tulip]

so like, eeeekk >.< we can make like... comics now?

[dense tapir]

I get studies/surveys lately for me to train AI on images

last one was a few days ago to tell it which images were buildings

paid well too

When this will have finally made it is when you can show the AI any image and it will describe it to you precisely as if you were blind with lethal accuracy.

we are getting there but still a ways off

[violet tulip]

so ahh Q, why is all the examples of shark telling you ti use cpu mode?

lol almost every example they use uses cpu lol

XD

but alright

every example is --device=cpu

😂

oh ok I found 1 example

Run any of the hundreds of SHARK tank models via the test framework

python -m  shark.examples.shark_inference.resnet50_script --device="cpu" # Use gpu | vulkan
# Or a pytest
pytest tank/test_models.py -k "MiniLM"

OR

lulz

just 1 example to use other than cpu

that's kinda sus

[dense tapir]

Well, I am just wow

It actually read it

[violet tulip]

cool

[dense tapir]

not bad for blip

[violet tulip]

make it say something stupid 🙂

i'm easily amused

[dense tapir]

naw, I am about to train something so I need to caption these

[violet tulip]

womp womp waaaaaaa

i have to edit shark already lol

just for the installer to work

easy fix

it's looking in the wrong directories for my python

[dense tapir]

GIT large is the most accurate

HOW does it know it is a woman?

It is, btw.

[violet tulip]

GIT large?

[dense tapir]

second one down

[violet tulip]

oh in the pic

still dont understand what it is but, i see it

[dense tapir]

you don't but it does which is amazing

[violet tulip]

well it's nice that the software is doing what it was programed to do, it's not like magic

[dense tapir]

You obviously have no idea

[violet tulip]

i'm a programmer

i kinda do

[dense tapir]

Of neural networks?

[violet tulip]

im not a very very good one

but i have a few projects under my belt and some code merge into official projects so, i'm like not a complete dumbass

[dense tapir]

If you never programmed a neural network then I stand by my statement that you have no idea what just happened and just how big it is. This isn't just connecting a for next loop together.

[violet tulip]

i just thought it was a silly thing to say, as if it's alive

[dense tapir]

Deep Floyd is coming too, btw

[violet tulip]

Floyd 🤣

oh no

[dense tapir]

Pink's brilliant brother

[violet tulip]

ooooh ok

I thought this was going in a totally diff direction

🤣

[dense tapir]

One smoked pot the other can AI image words.

something we presently can't do

[violet tulip]

i was thinking George

🤣

[dense tapir]

Oh FFS. Not

[violet tulip]

i thought it was a woke dev that named it that

i was gonna laugh so hard

[dense tapir]

When we can type text and it makes an image with the text we typed 100% legible that is a major advancement

SAI so who knows

[violet tulip]

SAI?

[dense tapir]

next up how about hands, ffs

Stability AI the makers of Stable Diffusion.

[violet tulip]

or retention of seeds

to make the same image

[dense tapir]

well, that takes some effort and the quality of the plant has to be primo. Errr, wait, what?

[violet tulip]

but that would require a more robust database

[dense tapir]

seeds is a bit stupid because stable diffusion isn't really stable (makes me laugh).

xformers made that even worse

[violet tulip]

like i mean how animators animate using cells

they dont draw the entire image everytime

[dense tapir]

Oh, temporal

temporal encoding will be big when done

game changer

[violet tulip]

if SD could do that, that would also be amazing, cus then it could remember each charaecter

that would be balls

[dense tapir]

temporal encoding means no more flickering between frames and fluidity finally comes. Some person in their bedroom ripping out full featured films

Talk about fan films then

[violet tulip]

that would be sick, i tried the animation ext for auto1111 but all it did was make a gif of the images i made

so it was shit

[dense tapir]

yeah, the thing is I did deep fakes back in 2018-2019 and it had temporal shit in it (had to) so I am not sure why that isn't brought over here as the merging of the techs is inevitable.

[violet tulip]

maybe if i had turned on loop during creation

i do deep fakes with SD

🙃

its hit n miss

[dense tapir]

well, I saw a first attempt of a film of T2 as statues. It is coming

[violet tulip]

actually i learned how to inpaint from the MrDeepFakes forum

[dense tapir]

We just need them to merge with SD as SD must have the temporal stuff or it will always be hit or miss (mostly miss)

[violet tulip]

before i ever watched a tut on yt

[dense tapir]

Sweet, I remember there

GD that software I complained all the time was so damn primitive like 1990s kind of primitive when masking.

I hated it

took hours and hours

[violet tulip]

yea, ppl forget this stuff was being done back then

[dense tapir]

yep

[violet tulip]

was mostly a hollywood thing or say you went to college for that stuff

[dense tapir]

What I lol at that in came automatic1111 and inpainting. I lay a big black circle on the face and hit gen and sob there iti s. No more line, dot to dot bs.

that floored me the first couple of times

[violet tulip]

yeah, i can see this tool be more amazing to someone who did that stuff bk in the day the hardway

[dense tapir]

I would have killed for that back then, and I even asked for something like that back then, but no go

yeah, like the hand animators frame by frame cell animators when the big Disney machine came in. I know how they felt.

Draw every 8 frames etc.. and it filled it in. Just wow

You know I miss ublock origin in discord. I would love to have it in here for reasons

[violet tulip]

a few edits and now setup_venv.sh from SHARK works

[dense tapir]

Yeah, man, that's deep

[violet tulip]

alright that part is done

now to figure out these non linux instructions

and linux them

shark venv activated

😂

I already have everything installed shark needs

well that saved time

so we now know, it's using all the same shit SD uses

how do u start Shark

because it says double click the .exe

we don't use .exe's in linux

oh i see

(shark.venv) > cd apps/stable_diffusion/web
(shark.venv) > python index.py

alright

copying some models from other stuff to sharks models dir

so far so good

when i run it and open my GFX software, we'll see if it is running in cpu mode or gpu

i'm going to go ahead and guess cpu mode

i don't have hi expectations for this

File "SHARK/apps/stable_diffusion/web/index.py", line 23, in <module>
from apps.stable_diffusion.src import (
ModuleNotFoundError: No module named 'apps'
lol

who ever wrote this piece of shit web ui

can go fuck themsekves

they can't even write good python

[radiant orbit]

I received several messages and did not understand anything. Is there a new model that can also write text? Or is it not released yet?

[violet tulip]

who ever recommended Shark, that webui is fucking shit

and the python used to make it was also shit

if it can't do a simple from "this" import "that"

your fucking dumb

lol

i don't program in python, but i'v done enough tutorials in it to know how to use python's from/import

it's not rocket science

i'll just keep using invokeai

at least it works

sure it doesn't have all the fancy extra extensions like auto does but it works on AMD every time

[dense haven]

Trying to make this image less blurry. What would be the best approach?

[violet tulip]

negative prompt

"blurry"

positive prompt : idk "HD"

¯_(ツ)_/¯

nice lama farm or alpaka's

[dry crow]

Dude dont be mad at shark if your knowledge isnt enough to use it... These guys work daily on the webui to improve things and you insult the Software and dont even write python by yourself...

[violet tulip]

look you're defending software... cute

why

[dry crow]

You can ask for help on their Discord... They help you in minutes

Yea it works great

[violet tulip]

ok

fucking white knight for software

lmao

[dry crow]

Calm down

[violet tulip]

😂

am

[ocean moss]

Hello all, I would like to generate logos using an embedding found on Civitai https://civitai.com/models/4577/low-poly-hd-logos-and-icons

I found that these images are generated using the merge of two models** "ultrm_0.4-protogenInfinity_protogenX86_0.6-Weighted_sum-merged"**

Do you know what ultrm stands for?

[violet tulip]

the embedding uses those

[ocean moss]

I found protogenInfinity_protogen but not the other one

[violet tulip]

which means just use the emdeding

they are explaining to you how they made there emdeding

[ocean moss]

Oh ok, can I use it with SD 1.5 vanilla as a base model for the txt2img generation ?

Oh ok

thanks

[violet tulip]

yes

[ocean moss]

Thanks a lot! 🙂

[violet tulip]

if the emdeding uses a trigger word

use that word at the beggining of your prompt

(some use trigger words)

while you are on that site look at the other textual files to get an idea of what i mean

[dry crow]

You need a vae

[violet tulip]

or negative prompts "blurry"

don't mind this guy, he loves vae

loljk

@ocean moss

this is what i mean by some use trigger words

TO ACTIVATE THE EMBEDDING

some do, some don't

something to remeber to look at when you download these things

even some models themselves use trigger words

I make a list of these "trigger words" with the model name

and keep that note open when i create

[dry crow]

He merged Protogen Infinitiy and ProtogenX86 both on civitai, he then merged them with slider on 0,4

[violet tulip]

yea, that's what he did to create the embed

you can also click the image and see what was used to create the image

[dense haven]

Used one already

[violet tulip]

see

@ocean moss

also read the comments left by ppl

sometimes they can be helpfull

also not sure if anyone else missed that about trigger words

if you did miss that, now you know

and your images will now probably do what you want them to do

probably a good idea to make yourself a list of what trigger words go to what and to keep that list handy while you create

[smoky oak]

My friend an di just made an Andrew Tate LoRA lmao

[violet tulip]

lmao

[smoky oak]

Without the lora

With the LoRA

lmao

[violet tulip]

🤣

that's great

[smoky oak]

I'm pretty proud of how it came out haha

[violet tulip]

if you really want to troll the art world, i think you know who to make lora's on now

A+

lol

[smoky oak]

I made andrew tate arresting andrew tate lmao

[violet tulip]

lol

[smoky oak]

who should I train?

👁️ 🫦 👁️

[violet tulip]

here's a clue o/

[smoky oak]

NOOOOOOO

[violet tulip]

pussy

[smoky oak]

I'm not gonna do that lmao

[violet tulip]

🐔

[smoky oak]

I would be willing to do the rock tho

[coral mural]

God of war

[smoky oak]

put him on everything

[smoky oak]

Shut up, friend

(the friend who I did the LoRA with lol)

[violet tulip]

i'll give you $10 DODGE coin

[smoky oak]

what lmfao

[violet tulip]

lol

but it has to be troll worthy

[smoky oak]

this is pretty troll worthy lol

[violet tulip]

just for the lulz

i can't believe someone hasn't already

you would be epic for doing it

just sayin'

[smoky oak]

I am not gonna make a lora of one of the most horrific people in history 💀

[violet tulip]

pppfffttt, pussy

[smoky oak]

gonna do the rock

[violet tulip]

would you take commissions on real requests? @smoky oak

[smoky oak]

to train LoRA's?

[violet tulip]

not o/

[smoky oak]

I could try

[violet tulip]

Jeannie Lee from mxrplays or her twitch stream potasticp

she also does the voice for someone i think recorder but i forget in Skyrim

her bf Henry is known for making adult mods in skyrim

and let me tell you, for an old game like skyrim, there modding comminity is off the chain

looking at thee og game then looking at a moded game holy cow

its crazy

anyways, that's a whole new market for you guys to tap into

making comminsioned models, embeds, loras, ect....

and sell them

$$$$$$$$$$

you could probably sell'em on ArtStation too

with ease

i just bought some art that i like off deviantArt

so surly there's gonna be a market for stuff like this

something to think about anyways, if you're a broke artist, and instead of crying about AI, learn it, and profit ¯_(ツ)_/¯

start selling commissioned models

(Next Million Dollar Idea) Free

some of the models you find on civitai have patreons which have exclusive models (for sale)

might as well if you got the time

[wispy nest]

think I'll start doing some commissions myself

sign up for an artstation

[violet tulip]

i mean, right?

[wispy nest]

I'm giving away everything for free unless people ask for my time or effort

[violet tulip]

bingo

[wispy nest]

and i do get it i get pm asking me to help some one install something and sometimes its like yeah im sorry not doing private tuition for free

ive spent a grim amount of time i could have been outside to learn this

[violet tulip]

and if you are good, ppl will notice

then the ppl with big $$ hit u up one day

[wispy nest]

making small name for myself with the community on civitai and other creators I know my ai gens are good

;'D

[violet tulip]

I have faith in you bro

[wispy nest]

😄 ❤️

ty boss

[violet tulip]

https://www.youtube.com/watch?v=KL2T0XRzWUI

@wispy nest

np

listen to this poem

and you'll be in good hands

your own

if you're going to try, go all the way

[wispy nest]

there is a huge demand for this stuff rn sadly a lot of it is just seeming to be for low quality porn :L

[violet tulip]

#1 seller

sex

#2 seller weapons

[wispy nest]

in this case its starting to be anti progress beacuse people arent adopting newer better tech beacuse it's not able to make boobies as easily so people are staying behind the curve which again fine if thats what you want but I really hope it isnt where the entire thing dies

because AI is based

[violet tulip]

if you sell it, they will buy it

if you give it away for free, no one will care

funny how that works

but that's life

people associate good with cost

and i mean this figuratively or another word i can't think of

not literally, but i'm sure you smarter ppl understood what I meant

as long as what you sell, ppl like, it doesn't matter how it was made (in this case AI art)

there are ppl just selling jpg's at like 100 to 1000 jpgs just for use in art creation

which could be used in model or lora creation, ect...

and don't think that major studios don't keep an eye on artists there, cause they do

[wispy nest]

I'm just tryin to give somethin back with the public models I do because so many people have helped me to get to where I am at with the image gens I do personally!

[violet tulip]

sorry i kinda just rammbled on there lol

[wispy nest]

nah all good haha ive just got 200 things going on :L

[violet tulip]

i feel you on that

do you make the models as well?

[wispy nest]

yeah!

[violet tulip]

i'v been coming up with a way to automate the entire thing

[wispy nest]

https://civitai.com/user/captainzero

[violet tulip]

oh cool

[wispy nest]

o yeah?

[violet tulip]

mhm

i can give you a bullet point style of my thought on it

[wispy nest]

people have been talking about a tool to automate likelyhood of mixes being good by their outputs

i havent thought of how to do it exactly yet ''D

[violet tulip]

create a web scraper/spider
use a facial recognition to image quality control what the web scraper grabs
make a script to resize and convert images scrapped to png 512x512
train

[wispy nest]

retraining with good outputs is very good a lot of people are training data taht way to refine mixes, love that idea and want to do it next just need to make something so I can afford to rent gpu time beacuse im limited to 8gbvram ):

I'm going to see if I can train Loras to refine my mixes instead first!

[violet tulip]

hustle commisions

to buy a better gpu

[wispy nest]

trying mate I just dont grift I have to know im giving a good service and its worth some ones time/money

pro worker 😄

[violet tulip]

not saying your hustle has to be shady

just a suggestion

[wispy nest]

it is right tho , im just not very natured to sell anything haha that is the harder part for me than even making the ai

[violet tulip]

ahhh

[wispy nest]

people are making money off actual crap as far as im concerned lol

but fair play people need money

[violet tulip]

true

also true

[wispy nest]

i tend to be a bit all or nothing with things

if I was to lower my moral standard and start ~~scamming ~~ grifting, id probably make loads of money then hate myself

so gotta do things right

[violet tulip]

and it may be crap to a sophisticated eye that you have but if the ppl buying are happy, is it really crap?

[wispy nest]

nope! its not ! thats the thing , I just tend to not see it that way and want people to be like holy FUCK feel like they underpaid me

not because im subservient but just because that makes me satisfied

[violet tulip]

this question can have very diff answers if you are a consumer or a dev

[wispy nest]

and thats what gives me drive to do something

[violet tulip]

word

[wispy nest]

you have to be wary of people who will drop standard for cash cos its a slippery slope and show weak character

what else will people drop for money then?> ethic?

you know where I'm coming from ?

[violet tulip]

yup

[wispy nest]

😄 you seem like a nice dude I'll see you around, I kinda pop in and out of here cos i get too distracted lmao

[violet tulip]

i try to be, but more often then not i get a different responce from ppl lol

nice talking with you

Go All The Way

(that's pretty cool, I got to chat with one of the dudes on civitai) 😁

[wispy nest]

check the civitai discord if you want to meet way better ones than me

they like people who are coming up with idea and working and sharing for all benefit

you will be paid in knowledge by us as a community that way

(:

[violet tulip]

guess what

i just used chatgpt to make what i just said

a web scraper to use a start image for quality control

here's the code for just the simple scraper

import requests
from bs4 import BeautifulSoup

def search_images(query):
    search_url = 'https://www.google.com/search?q=' + query + '&tbm=isch'
    headers = {
        'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/58.0.3029.110 Safari/537.3'}
    response = requests.get(search_url, headers=headers)
    soup = BeautifulSoup(response.content, 'html.parser')
    images = soup.find_all('img')
    return images

query = input("Enter a search term: ")
images = search_images(query)
for image in images:
    print(image['src'])

now here's the code adding facial recognition to quality control the process

I'm fucking gangster huh? using AI to solve this Ai shit lol

after i got that answer here's what i asked it

How can I add facial recognition using a start image to quality control what the web scraper returns?

AI said: To add facial recognition using a target image for quality control of the web scraper's output, you can use a library such as OpenCV or Dlib in Python. Here's an example using OpenCV:

In this code, the detect_faces function uses OpenCV's CascadeClassifier to detect faces in an image. The match_faces function takes two images as inputs and first uses the detect_faces function to detect faces in both images. If there are no faces detected, the function returns False. If faces are detected, the function uses the compare_faces function to compare the faces in the two images. The compare_faces function uses a deep learning model trained on facial features to embed the faces in the two images and compare the embeddings using cosine similarity. If the cosine similarity is above a certain threshold, the function returns True, indicating that the faces match.

The download_image function takes an image URL as input and returns the image as an OpenCV ndarray. The script then loops through the images returned by the web scraper, uses the download_image function to download each image, and passes it to the match_faces function along with the target image. If the match_faces function returns True, the URL of the matching image is printed.

Not bad huh 😏

https://tenor.com/view/robot-bender-strut-futurama-gif-5665697

[wispy nest]

chatgpt is really good at writing code but be careful if you dont code much as it will tell you confident lies sometimes aha

[violet tulip]

true

[wispy nest]

I Only know csharp so i cant read it ;'D

[violet tulip]

you have to work at it, but i go back to gpt with the error and have it fix it

[wispy nest]

yeee!! did that for some now refuses to generate code which might or might not be some kind of encryption in memory and process hiding code 😄

[violet tulip]

if you can write a loop in FooBar lang, you can do it in any lang

[wispy nest]

they nerfed chatgpt badly sadly since the early beta

and its not as fun anymore

[violet tulip]

I just wanted to show you guys, that. idk. I thought it, and got it done

a web scraper that uses a start image to self quality control what it scrapes online

removing that manual labor for the end user

for finding all the images you need to create a model with

booyakasha

there are also some gpt groups that have jail breaks to use with the monkey extension

where u inject your script

allowing gpt to write naughty fan fictions

😉

any whodooly doodily just wanted to share that

maybe a few of you geeks out there can tweak that and make it better

hard part is done, there's the template to use

facial recognition web scrapper

with starter image

https://tenor.com/view/ali-g-sacha-baron-cohen-respect-gif-4606499

I hope this new year brings joy and wealth

to you all

simplified version using chatGPT

with search term and starter image

😏

(in only 45 lines of code)

that's impressive

making an new updated version:
Here's an updated version of the script that resizes the matched images to 512x512 and allows the user to specify the number of images to scrape:

getting all the points needed

i got you dudes

😁

it will now ask how many images to scrape for

convert the image size and extension

how you like me now

lol

version 3 refined

right there.

• What do you want to search for?
• Show me a pic of what you want.
• How many do you want?
• Resizes images to 512x512
• Converts images to .png

bam!

That's, how you automate the gathering process my frens

with motha fuckin facial recognition for quality control

so ya, it does it all

yw world

https://tenor.com/view/snoop-dogg-straight-gangsta-gangster-into-it-supportive-gif-10455620

only added 4 extra lines of code

[smoky oak]

I am just cranking out LoRA's

Dawyne The Rock in Protogen

After my Dwayne The Rock LoRA

[smoky oak]

So I have now trained an Andrew Tate LoRA, a Taylor Lautner LoRA, and a Dwayne The Rock Johnson LoRA, all of which produce fantastic results

[dense tapir]

I mainly do styles as above

LoRA isn't as good as DB but is way better than TI with them BUT is more flexible than DB too.

30m for 2k steps on Colab

[smoky oak]

I have been training all of mine you see in under 10 minutes, LoRA is dope

even at 3k steps for my shrimp one, it took only 8 minutes

[dense tapir]

I wish I could but a 3090 trains the same in 9.5m

[smoky oak]

What? I am on a 3060ti

[dense tapir]

768x768

[smoky oak]

ahhh, I am training 512

[dense tapir]

yep

[hasty nova]

New embedding soon! It's pirates

[dense tapir]

3090 is like 5m at 512

[smoky oak]

but I mean, mine look plenty good, even for the 1080p resolution I often generate at, I'd say

just wanna say really fast that xformers is a god send

[dense tapir]

curse too

[smoky oak]

without it, I can generate like 768 x 1024 max

with it, I can generate at almost ultrawide FHD

[dense tapir]

I wish Nvidia hadn't become greedy dicks or I would be doing so many on a new local card

[smoky oak]

1080x1920 no problem

[smoky oak]

Yeahhhh, they really held back the VRAM on their GPU's

[dense tapir]

yes, and the prices are OMG

[smoky oak]

the 3060ti would have been a great budjet AI card with 12GB VRAM

[dense tapir]

since when is a budget x060 600 USD? Since now. :/

[smoky oak]

xformers is the only reason i am able to do high res generating right now, which is at least a benefit

[dense tapir]

@smoky oakOriginal leak was 4060 16gb at 379 USD MSRP. I saw the 4080 and did an ut oh. Saw the 4070ti and knew it was over. I was saving for it too.

[hasty nova]

https://discord.com/channels/1002292111942635562/1073240057038782514 Tis released!

[dense tapir]

8gb of ram for 600 USD

[smoky oak]

was able to generate this in SD (Color graded a bit and upscaled in Gigapixel, but its still almost fully native 2560x1080 on an 8GB card)

[dense tapir]

Yeah, 16gb for me minimum or no dice of buying it.

[smoky oak]

let me see if I can find the actual res I generated that at

2432 x 1024

[dense tapir]

damn

[smoky oak]

completely unprocessed image

what can I say, xformers is the only reason I am using SD

[dense tapir]

nice

[smoky oak]

768x768 is too low res for me, and that was about what I was locked to on big data sets

[dense tapir]

I find with LoRA neg prompts mostly are unneeded and/or break them.

[smoky oak]

I also just got public SD A1111 working, so I can lay in bed and gen

[smoky oak]

for mine, they have been very helpful personally

[dense tapir]

I never need them tbh, and another I know says the same thing though we do 2.1 exclusively

things are different with 2.x

[smoky oak]

my LoRA's are trained on 1.5 (not what I wanted) cause they were giving me errors on 2.1 base for whatever reason

[dense tapir]

I just wish Automatic1111 worked in 2.1 with LoRA

We have to use the extension

[lofty trout]

Is this fine tuned? What promp did you use to get that style?

[smoky oak]

It was a lot of fine tuning with 7th layer Anime v3-a

I don't have the prompt I used, I'm sorry

[sterile kiln]

LoRA is unbelievable

[lofty trout]

Np, it doesnt really matter since its fine tuned. Whats v3-a? Im kinda new to this.

[smoky oak]

It's a specific model

It's anime v3, model a

Let me see if I can grab the link

[lofty trout]

Alright

[smoky oak]

https://huggingface.co/syaimu/7th_Layer

You want 7th Layer Anime v3a

That's what I used at least

[dense tapir]

Yes, it really is AND would be even better if Auto will fix his 2.1 implementation of LoRA so it actually works.

[lofty trout]

How do i use this? Do you have a tutorial? I only know how to add my own images to dreambooth.

[dense tapir]

@sterile kilnThe colour vibrancy is just WOW as everything just pops.

[sterile kiln]

Bro, it just work if you add the right extension

[dense tapir]

I have the extension I meant native in auto as 1.5 does. There is a reason we need it fixed.

[hazy viper]

Did you follow a video or tutorial or how do you create your own custom model?

[wispy nest]

why dont you go ahead and contribute to the github then ?>

it's a fucking free software that is in heavy development we need more coders

the fact these UI even exist is immense

sorry if im coming across harsh but you are sounding entitled as fuck

[hazy viper]

how did you train your model? cant find any decent tutorials or info for it

[wispy nest]

yes you can

go onto youtube type in and search by date

[fervent rune]

https://www.youtube.com/watch?v=70H03cv57-o

Used this one and some modifications of my own, I will be releasing a video for doing LoRA on styles though since all the ones out there seem to be for subjects

[wispy nest]

yeah literally Aitrepreneur has so many good videos

it takes 10 seconds to find

[hazy viper]

hes never shown up on youtube for me

good to know thanks!

[wispy nest]

i get that sorry if I sound attacking

[hazy viper]

all good lol

[wispy nest]

its just the amount of time you see these questions when you work in it

[hazy viper]

😄

[wispy nest]

it gets immense

[hazy viper]

rip

[dense tapir]

No thanks. If I were to program it wouldn't be on his swiss army knife throw in the kitchen sink scripts. Free or not everyone is entitled to mention when something isn't working and wishing that it were. You sound like an arrogant fuck, but you have been on the defensive with it before. It will get fixed when it does and maybe, just maybe, another tool that is less of a catch all will come out eventually.

[wispy nest]

how is it arrogant to want people to contribute for the benefit of others

you sound like an entitled fuck yourself

go make that tool

dont be a leech

[dense tapir]

Because you got butt hurt, and defensive then attacked me. Don't try and play the victim here because you jumped down my throat for merely saying I hope it gets fixed.

[wispy nest]

"if i was to program it"

PROGRAM IT THEN

Im not a victim

i just think you sound like a prick

and disagree

deal with it

what are you giving back to the community beside shitty opinions ?

if the answer is nothing then well done

[dense tapir]

I am dealing with it it seems you can't.

[wispy spindle]

let's not be vulgar

squash it and let's move on

[wispy nest]

yep will do

[wispy spindle]

[haughty crane]

I'm a little surprised how well it can do the grimdark of warhammer 40k

[wispy nest]

I released a free SD 2.1 based model that you might really like @haughty crane it can do that style

[haughty crane]

Ill check it out, thanks :D

[wispy nest]

https://civitai.com/models/7615/czeros-sd-21-space-fi-mix

[haughty crane]

A bit of SCP inspired exploration

[wispy nest]

I love this kinda style

[haughty crane]

The polaroid tag works magic

[serene juniper]

Minecraft

[glass sequoia]

How do I get the system to stop producing two people when I just want one?

[serene juniper]

where to generate images and how?

[glass sequoia]

Is the prompt gonna be a negative one like cloned?

[dry crow]

you have to lower the resolution or better make a Resolution as a Portrait Ratio like 512x768 or 768x1024
The tag 1girl, often helps too xD

[glass sequoia]

Understood. Thanks

Can you make the image lower rez and then step it up to higher rez?

[dry crow]

yes then you need the HighresFix, you can upscale an image with it

[glass sequoia]

Got it

[dry crow]

Most people create 512x768 and then upscale it, because the models are trained on 512x512 so with higher resolution you will get more artifacts and double heads, persons etc,

[glass sequoia]

Ahhhh that makes sense

[haughty crane]

I used the description from SCP-610 'The Flesh That Hates' and generated a couple hundred images, I think it did a decent job, there is nothing overly gore-y but it is a tad nsfw.
https://imgur.com/a/24cDAIX

[left sparrow]

there's a character on a webserial i read that has a witch's hat made of fire, i've been trying to get it for like two weeks now,but today playing with playground i used their playdoh style

created this for me and i was like YES that hat but better

[haughty crane]

thats adorable

[left sparrow]

img2img is great

[coral mural]

Nice style

[haughty crane]

love the cartoon flames

[limber flower]

skate

[sterile kiln]

Trained with LoRA

[shell breach]

is there a way that i can run Stable Diffusion on an AMD card? Windows if possible

[ornate wagon]

Did you use the playdough image as input?

[dry crow]

With Shark from Nod.ai

[left sparrow]

Yep

[floral horizon]

Next Dead Space is going to be at an ocean xD

[zealous nest]

Yes.

[violet tulip]

cv2 keeps failing for that script i came up with trying to use PIL instead

i havn't givin up on this

in this updated version, the cv2 functions have been replaced with equivalent functions from Pillow, such as Image.open for reading images and Image.fromarray for converting a NumPy array to an image. The Image.resize method is used to resize the image to the desired size, and the image.save method is used to save the image in the specified format. The code should work the same way as the previous code using cv2, but using Pillow instead.

quick edit, and I think it fucking works!!!

woohoo

well.... shit it runs, with no errors

maybe someone smarter than me

can figure it out

it's probably something simple

i don't usually program in python

[violet tulip]

ok i got it working but without the facial recognition so far

and I'm using a custom google search (which is easy to get)

man, redeeming this is a pain in the ass lol

[smoky oak]

I can't believe that xformers took me from generating max 768 x 1024, to now allowing me to generate full 1080 x 2312

Was able to generate this natively on 8GB VRAM

[violet tulip]

cool

what were you using to create it, prompt model ect..

i'd like to try

i give up on that fucking automation script

i'm not smart enough to solve simple shit i guess 😒

oh wait you use automatic

nevermind

i don't even know why i try sometimes

see u guys around

[cunning geode]

https://imgur.com/a/tkWLRs7
Since yesterday generating a high-resolution image using high-res fix always generates some additional bodies or heads in the image as can be seen in the imgur link.

This didnt used to happen. I am using a new custom trained model of mine that I merged with another custom trained model of mine but it is happening with the unmerged model too, as well as in an older Automatic1111 version, and with older models where this definitely did not happen too.

This seems to have started when I updated to the Automatic1111 version where the high-res fix was changed, as in it now includes highres steps and an upscaler that you can choose and such. Any idea how to fix that?

[cunning geode]

is there a way for me to get an old version from automatic1111= like from december or something?

[onyx forum]

May I ask?

[honest dune]

Aonde eu crio as minhas imagens? não tenho o dreambot
Algum BR aqui pode me ajudar?
Kkkk

[cunning geode]

Okay yeah seems to be an issue with my models. Not sure why. I didnt encounter these issues before. Weird.

[graceful hawk]

https://www.youtube.com/watch?v=CWoWRunbiWs&ab_channel=PythonInOffice I used this tutorial

[smoky oak]

@violet tulipI sent you a friend request to discuss me training that LoRA you wanted

[dry crow]

Denois on 0,5 should work

[cunning geode]

nah i always do 0.5
this time i even tried as low as 0.2

always same issue

seems like a model issue, but idk why

[onyx forum]

what is the name of this model? I search but nothing

[green plover]

My firewall is blocking the site, but reverse image search shows it to be a model being developed in S. Korea

[onyx forum]

is this datasets not available in huggingface?

[green plover]

I don't know. I wasn't able to get much information on it. Just that the site had a .kr TLD and what google translate was able to tell me which wasn't much. It was featured in an online magazine a couple days ago

[wispy yarrow]

maybe https://civitai.com/models/7183/dalcefopaintinglora300

[onyx forum]

I search in LoRA but nothing, but everyone says it's available on Huggingface/similar web

[onyx forum]

no, the photo I sent was before the upload date

[wispy yarrow]

tbh, that could be any realistic model merged with anything

[onyx forum]

well, looks like it doesn't exist

[strange cliff]

Maybe the model wasn’t shared publicly who know

🤷

[onyx forum]

yea

but I'm disappointed, why are many people using those datasets, but I don't get it

[olive thunder]

Eating at the buffet

[wispy nest]

https://civitai.com/models/7797/czeromix-20-hybrid-sfwnsfw-anime-mix new 1.5 mix dropped

[smoky oak]

does anybody here have any info on how to create a style LoRA instead of a subject LoRA?

I am trying to make a full LoRA for a specific art style

[dense tapir]

What do you need to know as that is all I work in now

2.1 but for 1.5 no differences

[smoky oak]

If I am trying to train a style that encompasses many different objects, the more images and the more diverse, the better, right?

[dense tapir]

wrong

30+ will degrade fast

it confuses the AI

[smoky oak]

alright, so the somehow training it on almost nothing allows it to transfer that style to... anything?

not sure how that works, but if you say so

[dense tapir]

I have a 1 image style training, yes

[smoky oak]

interesting

[dense tapir]

mostly try to keep styles to 6-10

I like 8-10

20 it began to get wonky and hard to handle unless a 24 gb card too

[smoky oak]

24GB card, what?

are you using dreambooth?

[dense tapir]

yep, because LoRA has a bug with Gradient Accumulation

[smoky oak]

I am so confused

I trained my shrimp model on 30 with just 6GB VRAM used

I am not sure what you are referring to

[dense tapir]

I am deep into the woods with LoRA so yeah, a lot of what I say you might not understand and a lot I still do not understand about LoRA like the maths it uses.

Confuses the hell out of me

[smoky oak]

all I know is I did 30 just fine with no more than 6GB VRAM used, and it came out great

IDK what else to say

[dense tapir]

Then do it like that. If something works stick with it

[smoky oak]

alright, maybe it will work for me just fine like that shrimp one did

here's hoping

[dense tapir]

fingers crossed

I will give you a weird thing that happened tonight

I did a 51 epoch training that took over 90m

[smoky oak]

if this works, this is gonna be an awesome style to upload for others haha

[smoky oak]

so, what exactly are Epochs, if you don't mind me asking. That stuff I for sure have not touched on

[dense tapir]

I saved every 10 epochs and every even epoch (20,40) the style was not there. nothing, nada just not there

[green plover]

1 epoch is one time through your training data set

[dense tapir]

yes

[smoky oak]

how do you do more Epochs?

[dense tapir]

set it

[smoky oak]

oh, thats a setting in LoRA trainer?

[dense tapir]

yes

[smoky oak]

so it just does the same training over and over effectively? Or am I missing something?

I am trying to learn about this

[dense tapir]

it learns

refines

then if done too much blows up

lol

[smoky oak]

That sounds like an issue I had were I made 5 loras that didn't work at all, turns out the base model I was training on was incompatible, and it kept throwing a background size error in command prompt or whatever

so I had to retrain them on a different model, and they worked just fine after

[dense tapir]

Naw, this has something to do with the LR and the scheduler as it was bouncing over the data. epoch 60 would have been zip as well as I can almost guarantee it

Next time Iwill go to polynomial as I used in DreamBooth

I am also unsure if 51 epochs (1020 steps) was really needed

[smoky oak]

@dense tapirSo you are telling me fewer high quality images, and more Epochs is better for styles than more high quality images and less epochs?

[dense tapir]

I am saying what I was told and what I have learned. Throwing too many images at it for a style is bad. Now for a subject (like your shrimp) it is good.

[smoky oak]

alright, interesting

I did train it on several angels and colors

which offered a very visible benefit

[dense tapir]

I know it does for subjuects

*subjects

[smoky oak]

Interesting

[dense tapir]

for dreambooth I kept my subject images around 26-30 with 30 being the target. Styles I don't think anyone has pulled off well using DB yet as I asked around. Not really made for that anyway.

[smoky oak]

Very interesting

[dense tapir]

LoRA, HN/TI

I would do more with HN but the tools for it are horrible

I trained a few that I still use today but OUCH, they were never optimized so BS1 on 15 GB cards and BS2 on 24gb with hours and hours to train one.

took me 3 days on Colab at 2.5h per day (max allowed for free) to get to 600 epochs on it and I said enough.

[smoky oak]

jesus

[dense tapir]

tbh it should have been 3k

if I can find the dataset I used I will try and LoRA it since LoRA leaks even if called or not

[smoky oak]

I am doing 2 epochs on 130 samples of 12 images. I wanna see how they differ from one epoch to 2

[dense tapir]

130 samples?

[smoky oak]

130 iterations, my bad

so a total of 3120

or wait, no

1560

or wait, yes

3120, its 1560 per epoch

[dense tapir]

steps you meant. Yeah, I get so mad as I never know wth to call things as I am used to DB, TI/HN then up pops LoRA to do it its own way

[smoky oak]

not bad, 3120 steps in just over 10 minutes

[dense tapir]

we need a standard and all things use the standard or be shunned. a SI of the AI world

[smoky oak]

yeah, I agree

[dense tapir]

yeah, 1.5 is super fast

I just can't bring myself to go back to it

[smoky oak]

Yeaahhh, the style didn't seem to take very well at all

:/

[smoky oak]

@dense tapirwhen you do style trainings, do you describe everything in the images in detail? Cause this style only sticks when generating one of the characters shown

actually, it doesn't even take then

[dense tapir]

I used to not caption but with LoRA you have to

I hate it as I spent a few hours captioning 20 images so I am not going to do 20 images again. Was a test anyway.

It worked VERY well though.

[smoky oak]

so you did painstakingly caption things in detail?

that does help?

[dense tapir]

not super detail

but in detail, yes

[smoky oak]

interesting, lets see if I can do that for these 13 images

[dense tapir]

if you don't caption with LoRA it doesn't work well

for TI I never caption

[smoky oak]

damn, I was researching how to train style loras, and the website went down

[dense tapir]

not a lot of good info out there on that tbh

what is out there are mainly models/subjects

[fleet sluice]

pt logo type, yellow background, minimalism

[wispy nest]

https://civitai.com/models/7615/czeros-sd-21-space-fi-mix-fixed-v2 fixed error with my 2.1 mix

[violet junco]

Snake ciberpunk

[shrewd delta]

doctor preparing for the future

[radiant abyss]

what is the difference ?

[dry crow]

1.5 got released with the compvis account, 2.1 with Stability and all new models too

[last path]

o ok, thx

[green plover]

"by Yoji Shinkawa" Seed 258651764 CPKT: 1-5-pruned

[wispy nest]

Nice

[ebon dust]

a girl of 20 years old with red hair in full growth, slashes with katan robots, night city, neon lights :: widescreen, film effect, HD, anime mod

#1072220168534642768

[green plover]

The channel doesn't have a render bot. I think the auto censor might let that one through

[sharp beacon]

Note sure, how much of these prompts are legit, and actually effecting the result.

insanely detailed cinematic lighting, sticker Rebelle water color, ice-cream soft-serve cone, Pinterest design iridescent ,melting art professional, 8K,

Can anyone give advice how I can improve this? Some prompt suggestions ?

[near turret]

#1071918339850502277 message

[sharp beacon]

I think , if im not mistaken called " in painting" and or img2img

tons of youtube videos, of high quality and explanation on that. Can even train models based on type of art, or on your own art

i have the option of img2img in the Automatic1111 webui setup of stable diffusion

i just never used it

i heard photoshop get throw around with stable diffusion plugins. blender i dont know.

im new so just stuff i recall from a ton of random vids

right now im just using the web ui

not exactly sure. until i look into it more. but just at random

https://youtu.be/aYia7Jvbjno

Blender im totally clueless. always found blender super complicated.

so i just run from it.

Best thing you can do, is enter Stable Diffusion inpainting, into a search YT, and just digest one vid at a time. and if it dont apply just try another

Im sure if you ask enough here, someone will know more about it . Sorry

[sterile kiln]

GUys, could someone explain me the difference between theses versions?

[sharp beacon]

from what i remember, safe tensors is like a more safe thing for security for your pc. so people typically use that one.,

not sure what the others mean.

but i typically just use safetensors. that beginner video i posted above, talks a little about that

[sterile kiln]

they are all safetensor files, what is supposed be "baked" and "vae"?

[sharp beacon]

i dont know.

@dry crow may know.

[sharp beacon]

can i use this for ice cream cones XD or is it just going to make landscapes and faces ?

[static tusk]

VAE files improve the lighting and colours of your gens. Baked means the VAE is included in the model, so you dont need to use your own

[cunning geode]

look at the "feet"

[sharp beacon]

tell me.

[sharp beacon]

she got monkey feet

[dry crow]

@sharp beacon @sterile kiln Hey,
pruned is like compressed, (unnesseary parts are removed from a model) so it uses less space.
Safetensor is the model variant (.ckpt / .Safetensor)
VAE is an Addition to a model that can increase the colors and Details of a model.
Baked in means that the vae (which is normaly a seperate file) is already included in the model itself.
Clip fix is for clip (goes to deep to explain now)

Choose the newest model of those. Pruned, vae inbaked and safetensor are preferable

[sharp beacon]

is there a page for all the prompt variables and options? like a glossary

[tawdry wraith]

Pruned isn't compressed, insignificant weights are pruned

[sharp beacon]

im getting some issue. not sure why but all it makes is this

[icy marlin]

https://youtu.be/dheRfUrLPsg

[ripe cedar]

Where do I click to see what version of stable diffusion I'm using? In automatic1111

[sharp beacon]

i think in the webui u can select it? not sure if the folders show it. unsure what file it would be

[ripe cedar]

Script of batch file?

*or

[sharp beacon]

i meant just in the actual ui on the browser

theres a option top left

if there is a file u can set it , not sure

[ripe cedar]

I think those are just the checkpoints it shows, not the model it's running on

[sharp beacon]

i think its the same thing?

checkpoint is the model. Or maybe im wrong? thats what i always thought.

[sharp beacon]

are the checkpoints top left, the actual model its running on web ui?

[ripe cedar]

It just shows the models being used like depth2img, NAI, etc

[spring steeple]

/giphy query:https://s.bw.run/yck48n7w A cover for a poetry book, watercolor illustration

[smoky oak]

Man, I keep trying to train a LoRA on a style, but it just will not work man

I don't know what to do better honestly

For whatever reason, the full color dataset produces only black and white results

[sharp beacon]

check points top left

[ripe cedar]

Checkpoints are one thing but I meant the Stable Diffusion version. Like 1.5 vs 2.1, how can I check that?

[sharp beacon]

unsure. i just assume the Automatic1111 is updated upon start up. i guess you can check the github and just see when was the date verse there. to when installed.

[sharp beacon]

i always thought the 1.5 or 2.1 is the stable diffusion. its the model itself.

can get it from hugging face or something

https://huggingface.co/spaces/stabilityai/stable-diffusion

at least, that is the language used in that beginner video i saw.

gave that impression. like that.

[ripe cedar]

Oh, so the different models don't actually run on specific stable diffusion versions? So I can just basically use any checkpoint I want as long as automatic 1111 is updated?

[sharp beacon]

just use automatic1111, and the models should run assuming the model files are in place

maybe some variables to how the files work. maybe how they named and where they go. some are stable diffusion some run with dependences. best bet is look for youtube videos that talk about it for finer points of info. the 2.1 should be just drag drop into the stable diffusions models folder

https://www.youtube.com/@Aitrepreneur/videos

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

[smoky oak]

Does anybody have any resources on how to train Style LoRAs instead of subject LoRAs? I have been trying (5 different variations for a single style) but they all keep coming out really bad

[sharp beacon]

Olivio Sarikas channel has some good info too

[smoky oak]

This video helped me a ton. Had a1111 running in under 20 minutes

https://youtu.be/cFgmXLLnHp0

[dusty hemlock]

Can anybody tell me what this means man

I’ve been sitting here for about 6 hours now troubleshooting

[sharp beacon]

try adding that variable like it says to the bat

[dusty hemlock]

I did it just goes on to download more and then say runtime error

[sharp beacon]

im curious what it does. i added it to my bat

set COMMANDLINE_ARGS=--xformers --medvram --autolaunch --skip-torch-cuda-test

seems to work not sure it even applied it even though its in the file

[static tusk]

the line in green. --upgrade pip. add that argument line after autolaunch, remove "skip ...."

once its upgrade remove upgrade pip

[sharp beacon]

i had to uninstall and install like two times before i did it right. What gpu you have? What vid tutorial you used?

[dusty hemlock]

I’ve uninstalled 4 times now and I used Scott detweiler

I’m about to just give up

[sharp beacon]

be sure, you using python 3.10. not the latest version 11.

[dusty hemlock]

I did

[sharp beacon]

hmm im new like you so unsure. im watching his video right now see what he said

[static tusk]

also do you have an nvidia GPU? xformers is for nvidia

[dusty hemlock]

I do not

I haven’t put money into this thing lol I just know it has a gpu so whatever comes with it

[sharp beacon]

i got a 1050 ti nvidia gpu, i7 non k, and 16 gb ram, Samsung ssd. and pc struggles to run this.

my pc rather old.

[static tusk]

better to move to #🤝｜tech-support to get you setup as xformers wont work, so you need to use the CPU for generation I think

[spring steeple]

/https://cdn.discordapp.com/attachments/1053101349908271114/1073786360059658371/aa18a243-9e53-4ee8-9cf9-89b8b889b10d.jpg Uma capa para um livro de poesia nostálgico, Pintura à Lápis de cor, HQ, 105k

[green plover]

Today's poster

[smoky oak]

can somebody PLEASE FOR THE LOVE OF GOD help me with this?

I have only 1 model that does this, and its the best model I have

about 26/30 gens fail

when doing img to img

and about 6/30 when doing text to image

and it can't do batches at all without breaking

I have no idea what to do, and it sucks cause this is hands down my favorite model

[green plover]

I get that when I run a safetensor model without running a normal checkpoint first

[smoky oak]

Oh?

so I should run a ckpt before this model?

I wonder if I can get this model as a ckpt

[green plover]

What I used to do. I just added the model converter extension and converted all my models

[smoky oak]

do you have info on how I can do that?

this is the only safetensor model that gives me issues

[green plover]

yeah. go to the extensions tab and there is a model convertor extension

[smoky oak]

just installed it, I PRAY this works

[green plover]

bon chance!

[smoky oak]

if it does, I am forever in your debt haha

[green plover]

hehe, we're here to help each other

[smoky oak]

its stuck on installing, weird

[green plover]

sometimes it will take some time. just be patient.

been there

[smoky oak]

I was just mentioning cause its not using any CPU, disk or network

[green plover]

Yeah. it's probably network lag. it has to pull from git

Just because it's not using network doesn't mean it's not waiting for a response

[smoky oak]

ummm

let me restart SD

I am very new to all of this haha. I just installed SD 4 days ago, and started training my own LoRA's 3 days ago

[green plover]

you have --shared without --enable-insecure-extension-access

[smoky oak]

I have a ton of experience with AI's like Dalle 2, dream, and midjourney, but nothing so local haha

[smoky oak]

Is that an issue? I am using the local URL, not the public one

[green plover]

if you are accessing from your the same machine, you don't need to use --listen or --shared

[smoky oak]

I use --shared cause I usually gen on my phone, but for now I al local

I restarted it and its stuck on installing again, lets see if it ends up working

some of these extensions sound amazing, like stereo images

[green plover]

then temp turn it off so you can install extensions and you should be fine

[smoky oak]

I take stereograph images in person haha

[smoky oak]

alright, let me try that

[green plover]

I just use --listen because I punched a hole in my firewall and set a password

but since I quit my job to become an artist, I'm pretty much at home anyways (really I quit for health reasons)

[pearl cradle]

super realista. 8k. super resolucion

[green plover]

nice!

not sure if the one I did earlier will get past the filters, no nudity, so I'm going to try

Multiple explicitly defined materials

Made my own call after rereading the rules and pulled that one.

[smoky oak]

ok, it installed first try now

time to try it

fingers crosssedddd

[green plover]

Good luck

[smoky oak]

I LOVE this model, and I reallyyy want it to work

I used it to generate my character for my DND campaign with friends, but its so hit or miss that it took nearly an hour of hoping the inpainting wouldn't crash

My final result

[green plover]

Nice!

[smoky oak]

7 foot tall Tiger Tabaxi sorcerer

he big boi

[green plover]

which model is it?

[smoky oak]

Anything furry v1

I would link, but its incredibly NSFW lol

[strange cliff]

[green plover]

LOL I know. I could get it if I wanted it 😛