🏞｜general-with-images

[wispy nest]

It works but as you said, it's super slow. I increased my virtual ram from settings but not a noticeable increase. It took almost 8 min to generate a single photo. Know any tricks to increase the speed?
@dense tapir

[smoky oak]

every time their heads move like that i hear the goofiest little sounds lol

[sonic vessel]

This looks like the model is fried

[smoky oak]

No u

:>

[fleet leaf]

hi

[glossy herald]

nah, I'm testing the styles, and some are really burning the pictures, but basic one is great

[wispy nest]

i may need to start trimming my model folder -_-

[sonic vessel]

rookie numbers

[smoky oak]

Mine was over 300 for a sec, and I trimmed it lol

was worth it

but now I have ~10 models I love

[wispy nest]

last time i trimmed it i was a bit over 200, but it just keeps going back up...it's a mystery lol

[smoky oak]

I just merged my favorite models together, and now I have a new favorite model

I now have realistic vision, but with noise offset mixed with more stylized generation capabilities in one model :>

[wispy nest]

i used to do a lot of merges, nowadays i just swap to something more specific. mixes can get some interesting results, but rarely do i feel like they are much better than what i've already got now

[smoky oak]

You can see how ell it worked here

now we have realism of realistic vision, and the noise offset and color composition of Fae

Fae is amazing, but too unrealistic for me, so I fixed it to what I wanted :p

[wispy nest]

eh, it's nice, but honestly just looks more animated...and already got stuff that can do that

the colors are great though

[smoky oak]

That was just one example

It shines more in more dynamic realistic scenes

you can see here how the different merge strengths affect it

What I asked for was a dark and underexposed image, which is where fae shines

v3 is my current favorite, as it can do the really dynamic images, but can also have some really detailed and realistic results

[wispy nest]

that's a great example, but i'd rather have either the left or the right there, everything in between just doesn't look as nice to me. i could see why people would want it, but it just seems like it's not worth what it loses from either one of the originals

i wll agree that v3 is the best of the 3 mixes there

[glossy herald]

FaeTasticerV5

[smoky oak]

loooool

[wispy nest]

took me a while lol

[smoky oak]

in its defense, I asked for a very underexposed image haha

[glossy herald]

it's the future of models

so much offset noise there is just black left

why bother training models Stability ! look what the people are asking for !

do a black picture generator !

[smoky oak]

Here I asked for a very dark and underexposed/sinister image of a black rabbit at night, and I think FaeTastic nailed my vision, whereas realistic vision 1.4 is just too damn bright (no noise offset things)

the Fae one captured that casually sinister look I wanted

its slightly unsettling, while still being really appealing tonally, IMO

[smoky oak]

priddy!

[glossy herald]

yeah, trying to get my PoW picture's base from SDXL

[smoky oak]

Also, I ADORE that offset noise can get these gorgeous high contrast and rich toned looks

[dry crow]

1girl, black

[smoky oak]

man, my Na'vi LoRA plays sooo good with this custom mix

[summer viper]

promt black color, dark

[glossy herald]

XL does really some OK stuff

[smoky oak]

Extremely satisfied with how well this LoRA is working

only issue is the second ear

My work is paying off

I've looked at that site, and its all just SD really lol

I have a feeling they are just running SD with a new coat of paint lol

[glossy herald]

thanks ! I'm removing the screenshot/message sorry

[wispy nest]

Understandable, thnx

[glossy herald]

and yes, feel free to contact us for this type of solicitations

[smoky oak]

yeah, they are just using stable diffusion haha

[glossy herald]

the best place to report this is in #1010934719455707218

it opens a private channel with the mod team

no need on that one though

already acted

[summer viper]

faetastic'

[wispy nest]

oh okay, ty

[smoky oak]

Oh yeah, that site is just a ton of microtransactions lol

they have a subscription model that would make adobe blush lol

5 different tiers of subs, with 11 different purchasable packages ontop of it. Sheesh

Anyways, blegh

[glossy herald]

plus they are mass contacting people here, we already got a ton of reports of exactly this unsolicited message. we try to protect people but it's a lost cause, we can only react here.
the ball is on discord's side

[dry crow]

noise offset is sooo good

[summer viper]

elemental flower. intense, dynamic, detailed, cartoon style, wide angled, overhead view, vibrant colors, whimsical, absurd, surreal, fun

[smoky oak]

flouer

[glossy herald]

https://www.youtube.com/watch?v=l5OR6k1fqdM

[smoky oak]

I love these haha

[smoky oak]

Oh my gosh haha. This is incredible!

my honest reaction

[glossy herald]

I'm still moving my head like the cat

[smoky oak]

https://www.youtube.com/watch?v=Lvxwj5204K4&t=0s

There is an hour long version lol

[glossy herald]

lol

just in case

nah I'm good x)

I got drugs instead /s

[smoky oak]

https://www.youtube.com/watch?v=fZhWA6F5LLE

Love this one too haha

I had a pact where I watched this video every time it ended up in my recommended, and youtube caught on and kept recommending it over and over and over

[smoky oak]

god, I probably watched it at least 50 times in a month

then one day, they just stopped recommending it lol

https://tenor.com/view/unspeakable-violence-unspeakable-violence-gif-25206571

I love how fast it is

[glossy herald]

so, what flower should I make...

[dry crow]

thats awesome haha

[smoky oak]

I could totally hear the g mod beating sound when I see that gif lol

https://www.youtube.com/watch?v=_GGfz-o5khc

this one lol

It fits that gif so well haha

[grand walrus]

if you update it should be fixed

[indigo heron]

Lineage 2

[summer viper]

sd does me a favor, he immediately gives me the right background

[smoky oak]

I just discovered the most incredible thing I have ever seen

I... Don't even know what to say

As an auditory to visual synesthetic, this is quite possibly the actual most incredible thing i have ever seen

https://www.youtube.com/watch?v=ywdRQ3zU6Uc

@glossy herald Please have a look at this. I think you would appreciate it

[rare jacinth]

ive seen something like this with actual compositions of music, and often you can notice interesting visual patterns, though some of them don't look very beautiful

[smoky oak]

That one is nothing but pure music made for 3D visuals

its incredible

https://youtu.be/ywdRQ3zU6Uc?t=229

this time is insane

I have no idea how you even do that, but I WANT to learn, oh my god

[grizzled sage]

this is funner than i thought. aiming for a 2000 image set of a smaller scope. going to try to make some items made of different stone types. here's a peak at the granite i'm generating right now. i'm genning way more granite boulders than i need just so i can curate a style. on them. i was thinking 50 of each material boulder but might do 20 instead

https://imgur.com/a/vmmBoZd too explicit to share. rocks too hard

[grizzled sage]

OOOO oscilloscope music. that stuff is really neat. i've fired some through a very old scope and got some ok results. its real.

[grizzled sage]

i love destin's take on this one https://youtu.be/4gibcRfp4zA?t=92

[shrewd carbon]

#1072220168534642768

[glossy herald]

ok that's a nice base

now into invokeAI for inpaint

[summer viper]

fking_scifi_v2

[glossy herald]

lots of variety !

[summer viper]

reminiscent of worms 3d

[glossy herald]

I think I got the fire part of the flower down

[glossy herald]

streaming the process of earth petal there https://discord.com/channels/1002292111942635562/1005535685375176875

[glossy herald]

earth seems ok, let' go to water now

[wispy nest]

football old

[glossy herald]

ocean done

air left

[wispy nest]

what model?

[glossy herald]

it's realbiter

but it's mostly about 200 inpainting per quarter

I'll stream the air quarter

https://discord.com/channels/1002292111942635562/1050795294113136710

[oblique nimbus]

here you go guys https://prompterguide.com/prompter/

[glossy herald]

air quarter done. now for the finishing touches and the center flower

[summer viper]

FaeTastic

[summer viper]

RealBiter

[floral flicker]

Multidiffusion upscaling is pretty awesome now that I understand it a little better

[floral flicker]

There is a Region Control setting, which allows you to add a specific prompt to a certain region during the upscaling proecess - so the tiles in those regions can have a specific prompt

[summer viper]

wow

new plugin

[floral flicker]

Whereas with other upscalers, the only option is for your prompt to be applied to each individual tile

[summer viper]

nice

[floral flicker]

I'm getting much better results now that I'm using the Region Control feature... I was getting poor results with many different attempts in both Multidiffusion and SD Upscaler when just using a "quality prompt" alone

[summer viper]

and

[floral flicker]

1. the face of a grinning determined latin male boxer
2. the abdomen of a latin male boxer, high detailed skin
3. the arm of a latin male boxer, high detailed skin
4. a boxing glove holding a card
5. a boxing punch with motion blur

Not perfect but I think I'll get a perfect one with a few more gens

[summer viper]

what does the original look like?

[floral flicker]

Input image

Here's the sort of results I was getting without Region Control

I've been getting excellent results going from 1024 > 2048, with or without this feature. But I've been having a very difficult time getting my image to 4096

[summer viper]

the result is stunning

[floral flicker]

The regions can each of their own separate negative prompt, too

This is pretty awesome for now, but hopefully someone comes up with a definitive perfect upscaling process that cohesively uses the original prompt

This is also upscaling from 1024 > 4096 in 1m 27s on a 4070ti, with a controlnet enabled

[summer viper]

my card 1050ti

it's probably going to take a lot longer for me

[floral flicker]

Ouch, yeah. I just upgraded from a build that couldn't possibly handle SD

I don't do much PC gaming so never had a good excuse, until recently

[summer viper]

hehe

I'm going to buy 2060 12gb on the cheap

[floral flicker]

With the advancements being made in video AI, I feel like I'll soon regret not going even higher than the 4070ti

But, yeah, budgets and shared bank accts, etc

[exotic lantern]

good evening)

[summer viper]

hi

[dry stone]

is an eta of 19 hours for 500+ images okay? also im just using the default settings is that fine

[dry crow]

@onyx yoke tried photorealistic images with skin details
Model: RealisticVisionV2
Steps: 30
CFG:7
Sampler: Euler a
Upscale: 4x UltraSharp

[last pawn]

Hey everybody! I have a question today!!! I want to fix something. I want this guy to look at the camera...I made an animation and I'm trying to fix the eyes with inpaint but I tried lot of stuff ta make this guy looking at me but nothing works...any advice? I have this in the prompt " looking right into the camera" and this in the negative prompt "look down, look upward, look to the side"

[last pawn]

very inteersting! where did you find this extension? Thank You in advance.

[floral flicker]

Its in the extensions tab, Multidiffusion

My input image was generated with CGF scale 5.
With Multidiffusion, I went from 1024 > 4096 with CFG scale 14 and denoise of 0.1

[last pawn]

that's amazing!

thank you

[floral flicker]

Just saying, the extension doesn't seem to have good results if you use a normal CFG and normal denoise.
Works better with high CFG and low denoise

[vague oasis]

try "eye contact"+

[last pawn]

juste "eye contact" in the prompt? or with the +?

I tried but I think it doesn't work! THANK YOU for the help 😉

[vague oasis]

the + was a mistake

shame it didnt work tho

[floral flicker]

Spend a minute or so in an image editor, trying to mock up the result you want. It can be crude

Your inpainting result should be much better

[wispy nest]

looking at viewer is prompt I was using, but looking at camera should work too

[floral flicker]

looking at viewer ⏫

[last pawn]

I got you but it's animated so I have like 200 images...hehe'

[last pawn]

I will try! thank you

[fiery carbon]

Some guy made a script for using controlnet to render multi frames in order to create consistent videos.
https://xanthius.itch.io/multi-frame-rendering-for-stablediffusion
Here is a great video that shows a workflow to use it:
https://www.youtube.com/watch?v=T2nw9BThSvo

My first attempt:

[glossy herald]

time for some invokeAi unified canvas over in https://discord.com/channels/1002292111942635562/1051238363421745182
if somebody's curious about trying something else than automatic, I'm working on my PoW ( #⭐｜pow-info ) in it

[fiery carbon]

Lower framerate version removes a bit of the flickering.
That can be also done in your video editor by reducing and blending frames.

[grizzled sage]

i've been paying attention to this stuff and it's good to see they've migrated it into an extension now!

[austere pilot]

how? did you used a video init?

[fiery carbon]

What do you mean by "how"? The gif is created from an mp4. The input for SD and controlnet are video frames from a 3D engine.

[austere pilot]

aha okay, almostI meant to ask how did you do it, thx for the answer

[strange jungle]

Keanu with blonde hair - see, the eyebrows are black.

So I add "blonde eyebrows" to prompt. They get blacker.

[fiery carbon]

Haha! Blondes do not usually have blonde eye brows!

[strange jungle]

I weight (blonde eyebrows:1.5) - now they get absurdly black

[fiery carbon]

It is simply not natural (existing in any training data)

[strange jungle]

https://imgs.search.brave.com/Y1eteKSKWIKKVWf9Z60fY-SD3BAfRgHkTWYwTUuX4fE/rs:fit:1080:1080:1/g:ce/aHR0cHM6Ly9pLnBp/bmltZy5jb20vb3Jp/Z2luYWxzLzNiLzAw/L2VhLzNiMDBlYTVk/NmZiMDU2ZTMzZjg1/MjQzMzBlYThlM2Ji/LmpwZw

[light totem]

try pale eyebrows instead?

[fiery carbon]

Even these are black

[strange jungle]

https://imgs.search.brave.com/p1SkBLvdt3HkrIADr-ibwgr2mvPnDez4JaLPyFOgQcg/rs:fit:675:778:1/g:ce/aHR0cHM6Ly93d3cu/cG91dGVkLmNvbS93/cC1jb250ZW50L3Vw/bG9hZHMvMjAxOC8w/MS9sb25nLWN1cmx5/LWhhaXJzdHlsZS1m/b3ItYmxvbmRlLW1l/bi0yLTY3NXg3Nzgu/anBn

[fiery carbon]

Try old

[strange jungle]

no, those are blonde

[fiery carbon]

Dark brown

[strange jungle]

not jet black like I get when I weight "blonde eyebrows"

[wispy nest]

That looks pretty natural

[strange jungle]

if I weight it, it's like they're drawn on black

[wispy nest]

I don't think anything lighter then that will look natural

[fiery carbon]

Like i said, blonde eye brows don't exist (in the data)

[strange jungle]

I agree, but look what I get with Keanu

black, blaker, blackest the more I try to weight blonde

[wispy nest]

maybe that's just Lora affecting it

[fiery carbon]

Error in the Matrix 🙂

[strange jungle]

those are prompts, not negative prompts

anyone with dark hair

Bruce Wayne

[fiery carbon]

Try "light" brows

[strange jungle]

Superman

Ok, I wil try light

Superman Light Brows

weighted 1.5

[fiery carbon]

Try "old brows"

[strange jungle]

(old brows 1.5)

[fiery carbon]

No brows?

[strange jungle]

and dont tell me SD doesn't do some learning in a session

[wispy nest]

🐧

[strange jungle]

notice Bruce Wayne and Superman both had hair like Keanu - that's because I did Keanu first

never mind that - I have "keanu reeves" in the prompt. That's why the hair is like Keanu hair

Superman alone gave blonde brows

[fiery carbon]

Maybe try "Brows like... (someone with blonde / grey / light brows)"

[strange jungle]

Bruce Wayne alone did not

[light totem]

sdxl did pale blond eyebrows on the first try blond man with pale blond eyebrows, handsome, detailed, sharp, neg: dark

[strange jungle]

let me try pale blonde

blonde Bruce Wayne with (pale blonde brows:1.5)

[fiery carbon]

Try another model!!

[strange jungle]

those were with eldreths lucid mix

[fiery carbon]

There's definitely something to that model that makes black brows

[strange jungle]

Dreamlike Hassan does blonde brows

[light totem]

i tried bruce wayne as a blond man and got young joe biden

[fiery carbon]

Sometimes it could be so easy 😅

[strange jungle]

So I am doing "man" in eldreth

Eldreth "man" - one was naked, so I had to get rid of it. Only 1 out of 10 blonde - with blonde eyebrows. Looks like 2 red heads, the rest black

I guess it didn't like picasso looking one with body parts

[wispy nest]

My best working and most stupid prompt at SDXL lol

That's what default human looks like

[grizzled sage]

many blondes don't have blonde eyebrows *

[wispy nest]

I think all of them, it's just albinos thing

[grizzled sage]

wait, why aren't i in there? i'm the master copy of everybody else obviously.

[wispy nest]

You aren't default, you bough cosmetics I'm pretty confident

[fading nymph]

#🏞｜general-with-images

[grizzled sage]

well i mean, i know a lot of people with VERY slavik gene sets and they absolutely have that cliche blonde eyebrow look. like its so common that it's a very known cliche of certain regions

[grizzled sage]

all those oils and moisturizers exist just to facilitate my default standard look

[wispy nest]

probably just isn't in training data or needs some weird prompting

[fading nymph]

#📝｜prompting-help

[grizzled sage]

ooo i know a reason why many blondes don't have it. cause thye box blondes. faaake

i was thinking albino keanu but i bet that just does him from the matrix

[wispy nest]

I guess I'm kinda slavik...I was blonde when I was young and I'm pretty confident my eyebrows wasn't blonde 😄

can somebody help me with something

no matter what it only uses my cpu 100% and not my gpu

[grizzled sage]

i have dark dark hair anddark eyes. i'm often not believed when i tell people i was bright green eyes and red hair for my first 3 yeras

[fiery carbon]

The color of the eye brows of that tough guy from Sweden? 🤪

[wispy nest]

Didn't know eyes change with age o_O
I had blue \ grey through my whole life (blue or gray depends on light)

[grizzled sage]

i think what it was was pigment cells came in later. i got green eyes still , like , under the brown

layers

like onions

[wispy nest]

idk tbh, the only thing I don't know out of this args is skip torch cuda test, but based on googling it shouldn't affect it

[wispy nest]

I have an amd gpu and it won't launch without it

ooooh amd...

That's better to ask in #🤝｜tech-support

[grizzled sage]

-skiptest allows it to launch without cuda, so in CPU mode

you don't want to skip that test usually

[wispy nest]

cuda is nvidia thing...

[grizzled sage]

you want it to fail

[wispy nest]

if I don't force it too skip it just refuses to launch

[grizzled sage]

well, CUDA is what pytorch uses. you want pytorch to interface with the GPU, whether Cuda, directML, or RocM

[grizzled sage]

yeah. you need to find a way to get pytorch to use your gpu. skipping that test just fails into cpu mode

[wispy nest]

K

[grizzled sage]

rocm is only on linux

[fiery carbon]

I find it more likely that Intel comes up with some amazing AI capable GPU than AMD for some reason.

[wispy nest]

didn't amd people had their own thingy?

[grizzled sage]

me, personally, i'm rooting for Jim Keller at tenstorrent. He basically designs a lot of what Intel and AMD use to be awesome

i bet he'll come up iwht something both those companies license

[wispy nest]

I mean not official a1111

[grizzled sage]

AMD has rocm, which matches the cuda api far as i know

[fiery carbon]

Sounds reasonable, thanks!

[grizzled sage]

amd64, the 64bit instruction set that allows interoperability with 32bit executions, him

[wispy nest]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs

[grizzled sage]

apple's machine learning silicon, him

tesla's machine learning silicon, him

[wispy nest]

Elon still freaking out about AI and want AI to stop developing further..so yea 😄

[fiery carbon]

The thing is, NVIDIA has the best GPUs and development in all regards of 3D and AI.

[wispy nest]

And I'm glad he won't be with us in this race, just feck this idiot

[grizzled sage]

https://www.youtube.com/watch?v=G4hL5Om4IJ4 150min conversation with the man. swooon

[fiery carbon]

NVIDIA has developed cloud processing power anyone else can only dream of.

[grizzled sage]

a 6 month pause is reasonable and signing a giant thing that Emad also signed, for the reasons that the academic group who wrote it stated, isn't such a bad idea. in 6 months, better tooling will be developed to guide the development better

[grizzled sage]

i'd say right now, apple has the most intense silicon on the market for consumers. i wonder what they got in their labs. they're all in

the silicon still needs software to use it though. that'll come in a deluge soon enough

[fiery carbon]

Agree. They have the best client-side architecture.

[grizzled sage]

i've never owned a macbook but i might consider buying one these days

same reason i moved from amd gpus to nvidia. machine learning capabilities

[wispy nest]

I'm not talking about wht he signed or whatever someone else is going, I just hate Elon as human being...pretty stupid human being in my opinion, based of what he's saying and doing...

Ukraine related propaganda, satellite nonsense , then twitter bans...
I don't want that kind of person to be nowhere near our future technologies, but, unfortunately, he's good businessman and I guess he will, nomatter how much I hate him.

[grizzled sage]

okay well, all that deflection aside, the "stopping ai for 6 months" thing was signed by emad too

what were you talking about?

[wispy nest]

I don't know anything about this what he signed

and that's not what I was talking about

[grizzled sage]

it's easy to hate something when elon puts his name on it. meh

[fiery carbon]

True. I have tried to open .usd on MacOS. It was not an enlightenment.

[wispy nest]

He has been freaking out about AI for long time now

[wispy nest]

I don't , you don't get the point

I didn't said anything about what he signed

[grizzled sage]

i think you're leaning into the anti elon hype cycle. he owns ai companies. i think he even was one of the initial openai investors

anti fanboy and fanboy are two sides of the same coin. you don't even know the guy irl

[wispy nest]

I have reasons to lean into that cycle

[grizzled sage]

there's a video of a monkey playing video games with his brain online. look it up

[light totem]

oh no, an AI chatbot that requires a server farm of highly specialized GPUs is too powerful, oh no, what are we gonna do

Other than notice, an unplug it, i guess

[grizzled sage]

elon is the ceo there

[grizzled sage]

heh. well, the server farm hosts the saas. these chat bots could potentially host on one machine in any facility anywhere

[fiery carbon]

Any one of those guys are only human. Don't forget greyzones.

[light totem]

i doubt chatgpt 4 could run on a single machine

unless it was the most bitchin amazing machine ever

[grizzled sage]

one beast of a machine, yup

[fiery carbon]

There is already something similar which you can run on your machine...

[grizzled sage]

many bad actors coudl buy a $100k machine and run an instance of gpt4 for nefarious reasons. there are reasons to be concerned

[light totem]

alpaca runs on consumer hardware. you can run the 30b model and all you need is 32+gb RAM

[grizzled sage]

not like "cancel everything" concern, just, "eassssy" concern

yeah. and i haven't even mentioned alpaca yet

[light totem]

I dunno, I'm just not worried at all. remember when we embargoed selling the PS2 to Iraq because the 'advanced graphics chip' could pilot a missile

[fiery carbon]

Please a link for "alcapa"

[light totem]

https://github.com/antimatter15/alpaca.cpp

[grizzled sage]

the first dual core cpu's in apple power pc's were embargoed because the USA didn't export super computers

[light totem]

super easy install too

[grizzled sage]

apple of course leaned into it stating they were selling super computers

[fiery carbon]

Thanks 🙂

[light totem]

you cna dl the models here https://huggingface.co/Pi3141

[grizzled sage]

people wer elike "soon computers will have 10 cores!! why does it matter!" and otheres would be like "PREPOSTEROUS!!" ahh. the old timey years

[light totem]

turns out the real civilization-destroying technology development wasnt' supercomputers, it was kids using social media

who woulda thought

[fiery carbon]

Ok, well do i find the instruction where to put the files in the github?

[grizzled sage]

zuckerberg would've been coming up during those days

[light totem]

yeah, i hvae windows and there's an easy download. unzip, drop it in a folder, put the model in that same folder, go

you need a command line flag to run the 13b or 30b models tho

default is the 7b

[fiery carbon]

Thanks, will go for it the other day.

[light totem]

my pleasure, have fun

[grizzled sage]

Thanks for the link! i was looking at the cpp files and their instructions on how to build it and was just so confused. maybe might be able to get things running on my own now with that help.

[light totem]

i'm sure you will. I found 2 wrong models with the same name before I found those but they all work

there used to be a torrent and dl for them on the main github page but he removed them due to legal ambiguitiy

[wispy nest]

SDXL

I have a weird feelings - it seems like it doesn't know "Celeste" game, but also kinda trying to make same style

[light totem]

maybe there's a redhead woman named celeste that everyone draws on deviantart

[wispy nest]

I think it has Celeste original art pictures , but transferring only style

Red hair is my addition to prompt, if I don't add girl with red hair - it doesn't do red hair

[fiery carbon]

The blonde guy, using the 4x-Ultrasharp Upscaler 😉

[wispy nest]

That's just Madeline, Celeste game, on SDXL
Style is still there...well , except 1st image

[fiery carbon]

🙂

[ripe cedar]

What extensions are the ones most commonly used?

[dense tapir]

depends as I use the tagger a shit ton since I train so much.

[wispy nest]

controlnet

[dense tapir]

I personally never use that. Eventually, I hope, it will be vestigial and the AI can do it to my cues.

[wispy nest]

A1111 updated again...completely breaking it...

[dense tapir]

yeah, I rolled back and not updating again but I still have his memory leak issue so might be the version before this one that didnt have the issue.

[wispy nest]

It's recent update

webui doesn't even load anymore

instant crash

[fiery carbon]

Oh, the update mess. I opt out.

[dense tapir]

my commit is 64b7e838 and it has the mem leak but everything else works fine

not really sure how far back I need to go to get to before they created it

[wispy nest]

eh, I'll just pass on it for today, maybe will do some collab gens later

[dense tapir]

colab updated pytorch to 2.0 and broke a lot of shit

[wispy nest]

mine woks still I think

[dense tapir]

broke my trainer

I forked it and went ahead and went to 2.0 and it worked, and the dev the next day went backwards. Later I will check and see if colab is slower on 2.0 pytorch as I was or others locally. I didn't notice a difference

on colab for training

I think the issue, or at least was, is xformers

it isn't working with pytorch2.0 so you slow way down.

[ripe cedar]

I haven't tried any extensions yet because I heard they don't work well with the recent update. I'll have to see how to use an older version (since I used the exe installer so it might be different but shouldn't be) and then. I heard a9fed7c364061ae6efb37f797b6b522cb3cf7aa2 is a good one

[dense tapir]

how far back is that one?

[ripe cedar]

I think it was 1-2 weeks ago

[dense tapir]

cause I think that is the one I was on then updated and everything broke then I went to 64b7e838

all work in it but it has the mem leak

[ripe cedar]

For mem leak it needs restarting?

[dense tapir]

yep

shut down then load it back and all work

until the next time

this is what I hate about github because I can't just view all branches and immediately see their commit hashes.

[long trench]

I still need midjouney to work

and it just not

can someone help me generate?

[dense tapir]

I see, my commit is 1 week ago so yeah, 2 weeks ago is a good one

D:\stable-diffusion-webui>git reset --hard a9fed7c
HEAD is now at a9fed7c3 Merge pull request #8503 from mcmonkey4eva/filename-length-limit-fix

[long trench]

the exact style i need

[fiery carbon]

Needs some controlnet work, but i'm tired.

[stoic pivot]

A dust bunny under a sofa.

[stoic pivot]

According to a negative prompt, this is something folk do not want in a image.

[dense tapir]

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 678.00 MiB (GPU 0; 6.00 GiB total capacity; 3.76 GiB already allocated; 0 bytes free; 3.83 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

[stoic pivot]

I like the Result of Bing Image AI, according to Bing AI it use DALLE-E neural networks, in Stable Diffusion, what modell is closes to that or it it like compare apples with golfballs?

[stoic pivot]

Bing and DALL-E compared to Stable Diffusion and FaeTastic.

[long trench]

im fucking done, the model im using isn't even working

made this and thats it

[dry stone]

Looks fine

[long trench]

i ment it stoped working

[dry stone]

But I'm scared now. Spent 8 hours training my model hope it turns out good

[dry stone]

Wdym

[long trench]

it wasn't generating anymore

when i tried again

it gave me errors

[ripe cedar]

I didn't make this but I saw it on the internet and was made with SD but the poster never said what model; anyone know what model it might be?

[stoic pivot]

Inspired by POW "¤ "Elemental Flowers", but out with just one flower and not in the competition.

[young solar]

You could still submit it. You don't need a flower for each element, just one.

[grizzled sage]

lots of people misunderstood the concept i think. rules say 1 image per person an challenge calls for images of each of the flower. so do you do four images with one of each element? or one image only with 4 flowers? or what i like is the prohack people came up with, and just stitched 4 imagesinto 1

i believe its you choose your favorite element, and submit that

also, all submissions welcome

honestly, has potential to win. it's simple and beautiful. looks like a candle flame. quite nice

[crystal bloom]

Made this renders in Marmoset Toolbag. Textures created with Runway and Stable Diffusion.

[crystal delta]

FAQ: How do I use the Stable Diffusion website & what do all the settings do?

Check out these videos for a great overview of the website and how to use it! https://youtu.be/014J2Yo1aGI ｜ https://youtu.be/M3jCa6qTHpQ

[light totem]

looks to me like you need to turn down your cfg scale a little

[outer lotus]

A beautiful rose

#🏞｜general-with-images a beautiful rose

[dense tapir]

it's at 5

[light totem]

huh. weird. it looks oversaturated but i guess it's me

[dense tapir]

I am having a bit of an issue that I am unsure how to fix for training.

[light totem]

what exactly? maybe I can help out

[dense tapir]

well, I trained that loha last night/early this morning. You good with loras?

[light totem]

i am not 😦

all I know about those is turning the strength up or down

[dense tapir]

ahh. That is the problem I have no one I know to ask

[light totem]

ive seen a couple guides but I haven't tried it yet myself

[dense tapir]

guides are for anime so I found they are worthless

well, for what I do

trigger word 1.5 and the loha at 1.2 for the above and I tried all kinds of weights and that is the best combo. What does that mean I need to do?

That is what I need to find out

@smoky oak might know but they do subjects while I do styles so not sure if the same would apply.

[smoky oak]

ping

👁️

Ok, whats the question?

[dense tapir]

trigger word 1.5 and the loha at 1.2 weights for the above pic, and I tried all kinds of weights and that is the best combo. What does that mean I need to do?

[smoky oak]

I have no worked with LoHa, I don't think I understand

I know LoRA's for subjects though, obviously

I am not too sure how different LoRA and LoHa are

[dense tapir]

well, does that mean more steps or lower LR or no way to help it?

[smoky oak]

I also have to go really soon, but I will try to help

[smoky oak]

this image is the problem?

Problem is too oversaturated?

[dense tapir]

no

[smoky oak]

ok, sorry, I am trying to figure out what you are trying here

ohhh

[dense tapir]

a woman with blonde hair wearing a tie and a hat with a blue and yellow tie on her neck

[smoky oak]

you are asking what it means if you are at weight of 1.2

That means you are good, but you need an additional epoch. How many epochs are you at?

[icy matrix]

Hey sorry if im being a nude but is InvokeAI good/safe?

[dense tapir]

I can't make the lora work without heavy weights for trigger word and the lora

[smoky oak]

needing to over weight means that it knows what it is doing, but not enough that it can do it at low strength

[dense tapir]

only happens when I tell it the tie

[smoky oak]

ok, yeah, you need more training time. Steps should be good, if the result looks as you want, you just need more time

[smoky oak]

hehe a nude

[dense tapir]

more time?

[smoky oak]

yeah, its fine

[icy matrix]

;-; noob

[smoky oak]

how many epochs are you training?

[icy matrix]

https://media.discordapp.net/attachments/992578863244771380/1086741376928071710/tSoyMtM-1.gif

[smoky oak]

hahaha

This is amazing, I am saving lol

[dense tapir]

1 because epochs are the devil's work for styles and you can see in the loss graph. 11 epoch 800-1200 steps (I forget now)

[smoky oak]

ah ok, very different for subjects

I recommend 2 always

the difference between 1 and 2 is massive. I have found that lower steps with an equal total, but across 2 epochs produces far better results

[dense tapir]

you can actually see in the loss graph where epoch 2 hit as it jumps up in loss

[smoky oak]

yeah, I am not sure for styles, but I know for my various subject LoRA's, that 2 epochs have always been massively better. Any more and I get overtraining

[dense tapir]

how many total steps?

[smoky oak]

how many images are you training on, and are they of the same person?

[dense tapir]

no, not the same. I strive for 800 steps so repeats/image count

or 1200 or w/e

[smoky oak]

ok, can you show me the data set in DMs?

[dense tapir]

all 100?

[smoky oak]

jesus, you are using 100 images in a LoRA?

[dense tapir]

yep

[smoky oak]

what are you even training? lmao

this image is from only a 9 image data set LoRA 😅

[dense tapir]

works beautifully and the anime people use about 250-500 avg 350

styles are vastly different in this area

takes more steps and more diverse data

[smoky oak]

yes, as they typically have to make tons of sub set tags for facial expressions, as they do not translate with basic understanding of human expressions in the model

(Ie, if the character looks human, then it can pull from images of people smiling or angry in order to translate that on the subject. Whereas for anime, they need individual images of every emotion as the model doesn't know what crying looks like in the style)

[dense tapir]

real easy for the AI to lock onto something with styles

[light totem]

how many images do you think it'd take to get a comprehensive hands lora

[smoky oak]

can you just explain to me what you are training, and what it is for?

[dense tapir]

cyborgs

[smoky oak]

depends on how in depth you need to go

[smoky oak]

cyborgs? And those are the results you got above from it?

[dense tapir]

sure, and better

problem is I am weighing heavily due to "wearing a tie blah blah"

some reason wearing a tie makes it do nothing

[smoky oak]

I am very confused. I am not sure what wearing a tie has to do with anything

can you screenshot your whole data set in DMs?

no need to send individual images

I think you may be tagging incorrectly got a LoRA like this

[dense tapir]

Naw, tagging is fine it is probably I need more steps

[smoky oak]

If you need to tag for a tie, then something is off

do all of the images have a picture of somebody with a tie?

[dense tapir]

no, no ties. My point is I can put something under a tree, milking a cow, riding a bike, or anything but I tried wearing a tie and my lora ceased to work

[smoky oak]

I think we are having a severe miscommunication, cause I still am not sure what you are doing

I highly do not recommend 100 images, personally

I recommend 10-20 diverse images with more steps, ran at 2 epochs

[dense tapir]

which works for subjects, yes

[smoky oak]

Was that not what you were asking for? I am so confused

you literally pinged me saying my knowledge of subject LoRA's could help

[dense tapir]

no, I was asking does that mean I need to lower my LR, or raise it, or add more steps, or what?

[smoky oak]

In that case, I have no idea

I was misunderstanding everything you were asking cause I thought you were trying to train a specific person

unfortunately, I have to go actually

[dense tapir]

alright, well thank you

[smoky oak]

no worries

I'll let you know if I find out, as I have plans to try out a style LoRA soon that is a little diverse

[strange jungle]

I have to start trying to make LORAs

[summer viper]

delib+(tastic-biter) * 1

[strange jungle]

Robo Cat Logic: "I fits: 1. I sits: 1"

[summer viper]

illuminatiDiffusionV1_v11

[dense tapir]

locally loras are dead easy on colab it was pure hell

[grizzled sage]

when i train loras of people, i've seen better results when i lower the lora:tag:0.8 weight and then add emphasis or not on the subject token.

like when i go to use them

maybe that's what the anime guides mean

a 1.0 will often see results that seem more like the captioning and the images i used. turning it down allows the weights to play in the new attention better. i think training the unet for lower than the default 75% of steps might work here. total speculation.

[summer viper]

delib05+tastic05

[flat pike]

very nice one...meow!

[spark hinge]

Blackbeard for One piece

[smoky hull]

gozilla eat fried rice 3d logo - with - imagine

[summer viper]

create?

[smoky oak]

Gonna be using some SD generated images I made myself in conjunction with my uncle in collaborative VR sculpting in order to finally (hopefully) bring my VRchat character to fruition

Hes a really pretty tiger guy, but I don't think I can send the images I made of him here

[cunning gazelle]

How can I fix this mirror reflecting problem?

The hair reflection is always wrong

[grizzled sage]

its a ghost!

ummmm, i would maybe train a reflection lora? where the hair is always right? or inpaint

hard to say how to solve that one

arms i notice too are wrong

[cunning gazelle]

Yes

Mirror shots is not as easy as I thought

[hollow heath]

Student pose

[smoky oak]

Man, I really just wanna work on my VR chat model with my uncle in VR, but I swear, Meta's Meta Link wireless connection is unusably bad

[summer viper]

deliberate 2

[grizzled sage]

wireless tech for that level aof bandwidth is all shit. for vr, i'm all about the tether

as much as i hate wires

[smoky oak]

virtual desktop works damn good. Its very reliable, and can even pass higher bandwidth than my USB can, bit the issue is I can't run the apps I need outside of anything but Meta's trash meta link

I just wanna sculpt in VR man

at this rate, my VR chat model is never gonna be made

I just wanna use Adobe Medium, but its an absolute pain in the ass now

[grizzled sage]

i still have a vive and haven't upgraded that yet. it's a very narrow fov compared to modern sets now

[smoky oak]

A vive should still be plenty good to have a fun time

[grizzled sage]

oh it is

[smoky oak]

My uncle just got a meta quest pro

[grizzled sage]

has done me well

is there any benefit to the pro? i heard it was a flop. i was looking at a q2 for real though. maybe a new htc thing

quest pro has eyetracking that facebook feeds on don't it?

maybe i was just reading hater reviews

vive xr elite doo look good tho. now that you got me looking at the qp i'm considering the options

[smoky oak]

The pro is really just a lot of cool features at a premium cost

[smoky oak]

IDK, sounds like some stupid conspiracy to me lol

[grizzled sage]

often is

[smoky oak]

I know it is way lighter than all the other headsets

Cause it uses the super flat lenses

[ebon zephyr]

Hai

[smoky oak]

And it has full built in face teacking

[grizzled sage]

same look for the new vive xr too. lighter looking

like, i just wont even consider the index anymore

[grizzled sage]

yeah lol not just eye tracking is right. face

[ebon zephyr]

@lapis flame a young teenager boy profile picture

[grizzled sage]

https://tenor.com/view/zuck-zuckerberg-meta-metaverse-gaming-in-the-metaverse-gif-23714674

[smoky oak]

I don't think anybody should have ever considered the index, it's not a very good headset lol

[high tundra]

Avatar

[smoky oak]

It's riddled with a lot of problems, it's lower resolution than even the very affordable quest 2, and the quest 2 can almost match it's refresh rate lol

[grizzled sage]

the tracking index has is still by far more accurate. it's just cumbersome. its the reason the vive is still viable

but yeah. it's also yhuge now

[smoky oak]

And it's considered low resolution IMO

Not only is it lower red, but it has a larger FOV, so the pixels are much bigger

[high tundra]

@high tundra avatar

[smoky oak]

Whudd

[grizzled sage]

imo resolution isn't always the magic sauce in these spaces. throwing higher resoolution screens at it was something a few headsets did early and flopped

pixel shape and subpixel arrangement bigger deals

[smoky oak]

I'm not talking like 4K resolution or anything, I just think that a $1,000 incredibly overbuilt headset that can't function on its own should at least have the same resolution as a $250 standalone VR headset

[grizzled sage]

i'm not sure where the index lands there tbh

[smoky oak]

Short

[high tundra]

/avatar

[smoky oak]

If you're trying to generate an image in the server, you don't do that here

[grizzled sage]

"function on it's own" i don't think thats a feature set i need. i'm not impressed with the arm graphcis the meta headsets do at all

any headset i get is gonna be workstationed at my pc

[smoky oak]

It's pretty good for me at least. I mean I can play beat saber at 120 FPS with no problems, so I'm pretty happy

[grizzled sage]

or gaming from my pc

[smoky oak]

But the thing that's really attractive to me, is running virtual desktop wirelessly from my computer to my headset. Because I get 500 megabits direct connection on 5 GHz, and it runs even better than my headset on cable, because my computer has poor USB bus

And then I counterbalance it with a battery on the back, and I get 10 hours of continuous play with no problems

And then it also has built-in tracking, so you don't have to deal with that BS

[grizzled sage]

trade off really. the steam vr system is a superior tracking system, but only in the defined space between the light houses. while yeah, issues, i've managed

i got mine mounted on drywaller posts. they piston up like a jack and press against the floor and ceiling. move them easy

compromises

[smoky oak]

My uncle just told me that he is giving me a full kit of SlimeVR trackers as a gift

[grizzled sage]

put them all on your hips to get the most accurate vr chat gyrations

https://tenor.com/view/elvis-elvispresley-elvis-dance-elvis-dancing-gif-23242630

[smoky oak]

I am gonna be doing a standard 5 point system

[grizzled sage]

on the knees, hips, andnnnnddd

[smoky oak]

They aren't quite as good as lighthouse trackers, but they also require 0 external sensors, and function behind barriers with about 80% the accuracy of full light house

[grizzled sage]

new vive trackers look sauce but they're unnamed prototypes rn

self contained. no externals needed

[smoky oak]

yeah, the ones with the camera

[smoky oak]

thats how the slimes are

but for a fraction of the cost 😅

I am sure the slimes aren't quite as good, but they are damn good for the price

[grizzled sage]

lighthouses uses cameras too. just IR cameras looking for polses, not HD cameras looking for optical landmarks

[smoky oak]

I know how they work 😅

[grizzled sage]

i just love technically moments

[smoky oak]

but I was meaning the new self contained vive trackers have the cameras in them

[grizzled sage]

no i know

<< jerk

[smoky oak]

lol

[smoky oak]

My uncle has an insane VR setup

But he makes money off of VR

[grizzled sage]

maybe it needs some xanex

level it out

[smoky oak]

lol

even when he was on a quest 2, he had full mouth tracking

now I believe he has 9 point vive tracking along with full mouth and eye tracking

[grizzled sage]

i was more into it before. but then i lost wind there. was saving up for an azure kinect, but got a car instead

[smoky oak]

lol, good call

I am just really grateful that he is just gonna give me a full kit of slimes 😅

[grizzled sage]

i think there's a million% potential with the kinect still but i lke the car

[smoky oak]

he bought them so he could tinker with them, but he said when he gets back to America, he will let me have them cause he has his full vive tracking

[grizzled sage]

noice

[smoky oak]

He tried out the quest pros face tracking on one of his friends models, and I still can't believe how well it works

https://cdn.discordapp.com/attachments/427011414851321866/1071297439778226237/face2.mp4

the bottom jaw sliding is crazy to me, I didn't realize they were tracking depth and offset as well

I thought it was more like pattern recognition, but nope

I just need to get my shit together and make my VRChat model, but its gonna be soooo hard cause hes so muscular lol

I already trained some LoRA's to help me generate references to use for sculpting of him, which worked shockingly good

[grizzled sage]

yeah if the model is made good, it should line up ez

standard face marker setup

speaking of https://huggingface.co/spaces/georgefen/Face-Landmark-ControlNet

[fleet goblet]

I just realized I had access to SDXL in DS.

bruh

And I also have 173 credits. Nice

Okay.

[high tundra]

avatar

[fleet goblet]

I don't think that's generated.

I do not trust this man.

[high tundra]

/I don't think that's generated.

[grizzled sage]

was conceptualizing some discord profile pics earlier an this was a fun one

[ebon zephyr]

@lapis flame a perfect oil painting of a handsome guy

[ebon zephyr]

@lapis flame a animated photo of an Young teenager boy

[tame trout]

[summer viper]

@lapis flame a perfect oil painting of a handsome guy

[lapis flame]

#🏞｜general-with-images message

[summer viper]

@lapis flame a bad old painting of a handsome gay

[lapis flame]

[lapis flame]

💩 🤣

sorry I am not a bot I'm just someone in the chat

[summer viper]

hehe

[grizzled sage]

i am using chatgpt to code. i'm just telling it what i want and it goes. simple tool and i keep getting errors so i go back and we're having a neat back and forth. it's even comenting the code it makes to describe fixes. shits neat

[grizzled sage]

told it to write a python script to look in a folder where all my cropped images and captioned txt files would be, load the two side by side so i can easily edit the text files with the pic right there

[glossy herald]

if you go step by step, and describe correctly each feature, it's really quite powerful.

[grizzled sage]

yeah seriously. i gave it a list of features i wanted like save and undo. was more just a trial but it worked really well

[summer viper]

Classic Negative

[grizzled sage]

i'm actually going to use that tool in my work i think. later for now. goodnight generators

[glossy herald]

Have a good night Flow 🙂 I'm just getting up here :=)

[summer viper]

Illuminati Diffusion

the most interesting turned out

[wispy nest]

why all 2.1 models look like Loras, trained on 1 very little thing

[oblique nimbus]

is this what you were after or is it just a bit of randomness?

[covert portal]

The smallest seed carries the Whole within itself!

[crystal delta]

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

[deep bay]

#1072220168534642768

pank

[vast pebble]

Moonlight, Mirana, DOTA 2

[glossy herald]

We don't have a bot for making pictures on the server, sorry : #🏞｜general-with-images message

[chilly lance]

using random pictures of RBG PC cases as a img2img basis for some old master style fantasy art gives pretty cool results ^^

[wispy nest]

This is new, been running for months (Easy Diffusion), did something change?

[summer viper]

typical win 10

[wispy nest]

Just weird that it popped up now... like 2 months after I started using SD. Guess I hit 'public network' and move on?

[merry vigil]

° °

#1072220168534642768

[summer viper]

this is the norm for windows to suspect everyone of everything

[wispy nest]

I should probably upgrade to 11, but I've been lazy. 😛

[summer viper]

why?

[grizzled sage]

that's a great new way of photo bashing. diffusion is legend

[grizzled sage]

i switched to 11 after leaving 10 for linux for the last 2 years. 11 isn't actually bad at all. lots of great new features. biggest one is by default there are tabs in the command prompt now

and tabs in explorer!

[sterile kiln]

and with Stable.art, it's even easier...
An exemple with my Arthas generation
https://github.com/isekaidev/stable.art

[grizzled sage]

thats a cool plugin. ive been considering getting a home copy of photoshop again

[sterile kiln]

yes, a copy under the coat? Want a link?

[dense tapir]

Yes, and no. That was done with 2k steps (took amost 50 mins) vs 800. It was only marginally better in my tie prompt. I simply do not have the time on colab needed to train the amount of steps I may need.

[grizzled sage]

lol no i live somewhere where they actually enforce license use. seen afew friends who thought special copies would help them until an adobe lawyer got ahold of them one day

thanks though

[sterile kiln]

North Korea?
otherwise I don't really see which country does that

[grizzled sage]

Canada west coast. It's not the country doing it. It's adobe's operating policy here. While they don't go after all, i just don't sail any commercial software that way

[sterile kiln]

You're worrying for nothing x). No country analyzes network traffic this way, especially if you know how to use a VPN (ipleak.net).
They are helpless, like everyone else. And even if you are afraid of the behavior of the software itself, all you have to do is make firewall rules to block network traffic, even if it's completely useless.
These are the kind of comm they give each other to scare, nothing more.
Anyway, that was just for explanation.

[grizzled sage]

it's not the countyr doing it. Look bro, i just don't pirate commercial software and i have reasons that are very leveled.

thanks but no thanks. Don't get all "everybody is doing it" on me here. Grade 8 is about 30 years back that way

[sterile kiln]

will understand who can

https://www.southparkstudios.com/video-clips/pxg0qb/south-park-the-perfect-storm-of-self-satisfaction

[grizzled sage]

i understood the reference right away tyvm. your link don't work as i'm canadian and it's blocked here.

you think it's smug to avoid legal troubles and thats where i ended the convo.

[sterile kiln]

I believe that a large group, which passes from licenses everything that is more normal, to subscriptions at all costs, goes beyond understanding. It's simply hitting the consumer in the wallet.
So when someone can't afford it, I think that in any case he won't pay for it, and can afford to have a "pirated" version.
I bought PS 3 and 6. I'm not going to pay for it 50 times.

If what you want is to procrastinate on the economy, free to you. But it still makes for long speeches, all that instead of directly saying "no thank you,
-I'm lazy
-I am scared
-I don't hack"

Hence this totally deserved smug.

[glossy herald]

hey guys 🙂

working on pow

[sterile kiln]

saw that

GJ

[glossy herald]

playing with comfy UI and some similar thing to composable diffusion

just takes an hour to set up in comfy instead of 10 minutes

[sterile kiln]

Did you try again with Multidiffusion?

[wispy nest]

guess I should try comfy too, since my a1111 is broken anyway

[glossy herald]

it's working but nope, I was focussed on comfy today

if you drop those pictures (open in browser first) in comfy, it should load the full workflow

[wispy nest]

Do you have link for comfy official docs?

[grizzled sage]

i'm not saying anything about you bro. i don't pirate because i'm not a consumer in that field, i'm a professional. im here and on other public servers talking about my work. i'll only talk about myself.

i'm not personally attacking you at all, in fact, i don't really care about you. less and less the more and you try to make it all about you. i'm glad the plugin works great for you. .. how bout we move on.

[wispy nest]

now when I think about it, I probably won't be able to do anything with comfy on my 1050ti babe

[inner summit]

Hi, can anyone help me understand why my images are generated like they are "painted" even tho i set negatives..
Tried different models, kinda the same.

[wispy nest]

vae?

[grizzled sage]

thats a very cool composition

[coarse bridge]

Hi guys !

[wispy nest]

yo

[inner summit]

using default one unless model specify

[coarse bridge]

How I can use the bot for make art?

[wispy nest]

usually color problems comes from wrong vae or no vae at all, try re-selecting it , resstarting , idk

[inner summit]

do i need to reload or just simply apply settings? will try that

[wispy nest]

apply is enough, but who knows, a1111 is broken lately

@glossy herald why can't I use /faq anymore

[sterile kiln]

Okok, its me this time.
1st it was canadian FIB
2nd you're not a stealer
3rd I blame you x)

It's cool man, don't panic
We hurt no one and it's the internet, enjoy

[grizzled sage]

wow

stop making things up

[sterile kiln]

it was not sarcastic, I really wished you to chill, have a good afternoon

[wispy nest]

I don't even udnerstand what's happening

[inner summit]

oh, apparently it did not took the right .vae
thats huge difference

[grizzled sage]

i said i didn't want a pirated copy of photoshop and thanked him. kind of went haywire after that

[glossy herald]

hum

[wispy nest]

[glossy herald]

get an error or not even seeing the bot in the / command list ?

[wispy nest]

no commands, I thought bot is down or smth first, but people used it some time ago

[glossy herald]

I got it working, but I gotta say, bots have been a pain on some task recently, and restarting discord has had to solve it for me multiple times too

[wispy nest]

oh yea resstarting helped, weird

[glossy herald]

like report bot just disapears on me every day until I restart

I think the problem is on discord's side on big servers

[summer viper]

every time you face a choice to pirate or not, ask yourself the question, do they deserve my money or not

[crystal delta]

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

[stoic pivot]

The future according to Open-Ai.

[wispy nest]

eh forgot user =.=
#🏞｜general-with-images message

[marble sail]

Guys I need your help😭😭

[wispy nest]

future looks pretty bad for AI 😄

[sterile kiln]

Thank you, Finally someone who understands what I'm saying

[marble sail]

I’m having issues installing stable diffusion

[wispy nest]

What issues?

[stoic pivot]

Yes, as fast the humans find anything that may be smarter than them they will kill it with fire 😄

[marble sail]

What can I do

I’m a new user 🥺

[wispy nest]

oh...you might need to install older version...a1111 is broken rn

[grizzled sage]

read the error. they got no gpu.

[marble sail]

How do I do that?

[summer viper]

Need install drivers Nvidia

[wispy nest]

yea and that too

[grizzled sage]

generic grab bag advice is often bad advice. especially for new users.

don't just throw that stuff around like it's confetti

[sterile kiln]

Apparently this commit would be the last stable one (thanks @glossy herald )

git checkout a9eab236d7e8afa4d6205127904a385b2c43bb24

[wispy nest]

I thought he got same error I have, didn't read whole thing

[glossy herald]

not sure what is causing this. #🤝｜tech-support would be of better help though

[wispy nest]

after update, UI doesn't work at all 🐧

[glossy herald]

you remade the venv from 0 to be sure ?

[wispy nest]

I didn't do anything

[glossy herald]

I have the version from 2 or 3 days ago, not sure

but it was post the big update

[wispy nest]

but your checkout works, yea

[wispy nest]

memory leak one?

[glossy herald]

and it works currently for me.
The checkout Seb shared is correct too

[grizzled sage]

--reinstall-pytorch and --reinstall-xformers often helps too

[glossy herald]

yes... thus comfy

[stoic pivot]

This is why AI is scary, I asked for "Elon Musk as an AI robot" and it gave me just Elon Musk.

[sterile kiln]

Just copy/past the error on their Git page

[wispy nest]

let's not talk about this guy or I'll start swearing and telling bad things about him

[glossy herald]

that's just the truth coming out. not the best of AIs though, just ElonAI

same

[sterile kiln]

Elon AI, hahaha. That's how you see that when a successful software is open source, paid copies quickly appear. (special dedication to Adobe)

[grizzled sage]

if i decide not to pirate, that's not me ruining any economy or making any personal attacks against those who do. all i said was i didn't want a pirated copy of photoshop, especially one linked from a rando on discord.

i also have this great security policy where i never trust any cracked software because why would i in 2023?

[marble sail]

what do I do

[grizzled sage]

"Whatever are you talking about" sigh. passive aggressiveness. One of those ways people make themselves feel like the smartest in the room.

[sterile kiln]

Guys just scroll and see by yourself

[marble sail]

I’m new!

[glossy herald]

let's maybe stop before we start attacking each other and going against the here.

[marble sail]

I have no idea of what you said

[grizzled sage]

heh. i'm glad. this is one of the fastest growing new digital hobbies

[sterile kiln]

there were no attacks, just explanations

[glossy herald]

please, go check in the #🤝｜tech-support channel, you'll have people taking care of you in a more tech environment, this channel here is quite active

[marble sail]

Okay Thanks

[sterile kiln]

pm me if you want @marble sail

[glossy herald]

Yes I see it, and I'm ok with where it's at currently. i'm just saying, you guys seem to go a little higher each time you respond currently, so I'm just preventing before one word goes definitively too high

[glossy herald]

that seems the best solution for more open discourse for sure 😉

[grizzled sage]

relevant starting position for those who don't want to scroll. since it's being reframed like i am the one starting shit now. like, why even is this a problem ithat i don't want pirated copies of professional software? bro. it should've never became some sort of "you're so smug" situation

[sterile kiln]

I don't intend to insult him, but as soon as I explain something to him, he goes on to the other chosen one. And then he got mugged
👀

[sterile kiln]

Man, they can scroll to 2 years if they want, they can see everything

[grizzled sage]

i actually linked it though

[wispy nest]

I hate the fact that lots of services just making money of out community or open-source work
People committed lots of time just for someone to take it and start making money out if it....

Most of them doesn't even give credit to SD or creators

[grizzled sage]

SaaS is problematic, but also, professional software licenses are not a joke to test if it's part of your livelihood

[sterile kiln]

I could spend long minutes answering you but I'm too lazy. Think what you want

[sterile kiln]

depth layer with local prompts?

[glossy herald]

still comfy UI

same as the last one with the flower, but with an infinite loop

it's a scribble mode

using higres fix + upscale, and 5 different prompts targeted

[shadow iron]

Frodo Baggins

[shadow iron]

made it better

ok now it's even better

[wispy nest]

[stoic pivot]

The magical words, "snuggly snuggle, fluffy fluff, manly man, cutie cat", to summon a beta chad.

[steep frost]

Stable Diffusion XL - beta is out now

[light totem]

sad mine still says preview

[glossy herald]

mine too

[rustic barn]

@wispy nest

These were all created by me using leonardo

It's good

Right?

[wispy nest]

well, if it works for you - that's all that matters

I just didn't find there models I wanted to use

[rustic barn]

Everyone has a different taste

But
I can not generate image 2 prompt in leo

It doesn’t generate right
I have find another Ai🙂

[brave mauve]

magical man on the background of mountains

[glossy herald]

Flower of 4 elements

[glossy herald]

we don't have a bot making pics, sorry

[potent sparrow]

#Due to extreme demand we can't provide a free trial right now. Please </subscribe:987795925764280351> or try again tomorrow.

Due to extreme demand we can't provide a free trial right now. Please </subscribe:987795925764280351> or try again tomorrow.

[glossy herald]

hello and welcome here but what are you talking about ? and please don't start here by spamming ...

[wispy nest]

bot looks good and actually looks like 1 picture

and last one

[glossy herald]

yeah, I've tried improving different ways but the first one was better

it melted too much after

[wispy nest]

previous looked like different pics merged

[glossy herald]

I messed up conditionning for a few

[glossy herald]

I meant to ask Soul, what do you mean usually by this emote ? you use it a lot, responding to lots of things, and I'm not sure ^^

[wispy nest]

I'm not sure too

[glossy herald]

hehe

[wispy nest]

I keep forgeting to change my seed , when I import prompt from another image

generated same things like 3 times now

[glossy herald]

for sure happens ^^

same here

[wispy nest]

Does this counts as nsfw yet?

[glossy herald]

and by the way, the images tend to melt as one because they are made as one.
I use 5 conditioning areas. one per petal/element, and one global for the flower, but the picture is made at once after that, with some controlnet to help on top

[wispy nest]

yea I understand that

it's just...transition doesn't look natural

[wispy nest]

wanted moon halo tho

[wispy nest]

🥺

[crimson flower]

[wispy nest]

can also do things like this 😄 not perfect, but pretty dang good considering no inpainting

[empty notch]

can someone make a rainbow salamander?

@cursive lance i am you

[empty notch]

use this promt

[glossy herald]

nope, I'm not in an editor with a single prompt

go on dreamstudio if you want to run prompts ^^

[empty notch]

rainbow salamander, render 4k, ominous shadows, drawing style

[glossy herald]

still nope sorry

[empty notch]

I don't have permissions or I don't know how to use it

[glossy herald]

we don't have bots

[crystal delta]

FAQ: Where can I access the DreamStudio beta website?

https://beta.dreamstudio.ai

[glossy herald]

go there, you can make pictures

[empty notch]

thx

[glossy herald]

#1072220168534642768 in case, there are lots of resources like this one

[fickle gorge]

My friend told me I shouldn't let this image die on my hard drive, so I'm posting it here
Small details still seem be troublesome, but I think it's not that bad this time

[hardy tulip]

"Woah there! The owner of Stable Foundation has requested that Discord block any messages our mostly-accurate robots deem to be explicit, so your message has not been sent."

there's nothing

[glossy herald]

clyde can be quite the bother sometimes yeah..

[fickle gorge]

This is why it's mostly, I guess

[hardy tulip]

added lines

now it doesnt flag

amazing

anyways, i went to shower and left it to generate

no prompt, only negative

[fickle gorge]

I never tried only with negatives, but no prompt at all generated mostly landscape-like outputs

[coarse jetty]

The moonlit forest, with its gnarled trees and dark shadows, seems to shift and sway in an otherworldly manner. The ground underfoot is covered in a blanket of snow, crisp and sparkling against the black, starless sky. The silence around the small, rickety bridge is absolute, as if all the world's sounds have been muffled by the thick blanket of fog. The dark water of the dark, still pond reflects the moon's glow in an ethereal, shimmering manner, while the only sound is the drip-drip-drip of water onto the ice. (Image with SD and Text are generated with my AI!)

[stoic pivot]

I asked MS Bing if it could create a prompt for Stable Diffusion, it do not understand the prompt format, but it worked when I gave the words -fantasy with butterfly in a jar; the suggestion was "Create an image of a butterfly in a jar that is surrounded by a magical aura. The butterfly should be colorful and have intricate patterns on its wings. The jar should be made of glass and have a cork stopper. The background should be dark with stars shining in the sky. The butterfly should be illuminated by the light coming from the magical aura surrounding it".

[coarse jetty]

It's so artistic and beautiful!

[stoic pivot]

And me just typing "glowing red|yellow intricate-detailed glass flower surrounded by blue ice, seen from close to the ground, snow particles, ice crystal, backlight light bloom, bokeh, midnight star sky"

And no /%¤& stars 😄

[coarse jetty]

here's some multiple others images that I created!