#šŸ’¬ļ½œgeneral-chat

1 messages Ā· Page 92 of 1

twilit mural
#

i realized it was because of image resolution

#

mine were too low

north pelican
#

I need stable diffusion to come back! I can't create images. What option do I have?šŸ˜­

left gulch
#

Hi guys, I just recently launched my Youtube documrntary channel. I create some videos here on AI news and trends. Hope you can help me by subscribing, liking and sharing šŸ™‚ if you have suggestions also on content you want, feel free to let me know! https://youtu.be/Yt6YG6d2D54?si=etZdDSqjmozZjqDw

nova zodiac
golden gale
#

Anyone knows what is the problem with bots and it's for so long down? Thanks

north pelican
nova zodiac
golden gale
nova zodiac
#

The answer lied somewhere between "they were using the gpus from the bots to train new models" and "we don't have anything new that we need testing/want to show off so we aren't going to spend $$$ running the gpus"

#

I am not stability staff, so that is a completely unofficial answer

golden gale
nova zodiac
#

Based on past experience, they'll come back next time they either have a new model that they want us to provide feedback on, or they have a new model they want to show off!

golden gale
nova zodiac
#

No worries

twilit mural
#

@nova zodiac what model have you gotten best results with when using instantid?

fervent thunder
#

Lol

nova zodiac
left gulch
#

Hey guys. I just recently created my youtube channel. Part of the content I create is AI news and documentaries. Hope you can support me by subscribing, commenting, and liking. Please let me know also what topics you want me to create šŸ™‚ hereā€™s one of the videos I created so far. How do you find it? https://youtu.be/Yt6YG6d2D54?si=Z7V16DA2l3AYZmH7

clear oyster
#

lmfao nice analogy

fallow veldt
#

Inpanting is such an art. Now I notice that if you put a denoising stronger that 0.5 I noticed that the context messes up in the generation. Sometimes it does something, but it most likely it produce an aberration. So I keep trying to figure it out.

#

So I use fill if I want new stuff and original if I want to retouch something. But I'm paaying attention of the context I give to the inpainting-only masked

#

The context on inpainting only masked depends on the inpainting resolution, the area masked and "Only masked padding, pixels" (Automatic1111)

#

By getting the best context for SD to understand the part of the image I want to change and not going to high on denoising I'm being able to not be so frustrated while at it.

#

I got some trick like if I want to inpaint a square are (SD XL) I use 1024x1024, a landscape area 1344x768 and a portrait area 832x1152, then adjust the "Only masked padding, pixels" in some cases where I need more context. But I also sometimes masking certain things like 2 hands far away make it the inpainting zoom larger, I guess with less resolution, but handy in some situations.

#

Also to fix things that should be symmetrical or similar it is best to mask both, like both hands.

#

I would like to understand how it work as sometimes it works at higher denoising. It is like you promtp "hand" and you can see a giant hand in the whole generation instead of in the place you masked. And then someetimes it makes a hand. But I notice that if you don't go so high in denoise SD can see the context right rom the start and leads to a more intilligent inpainting.

#

So something practical, I'm currently working from 0.4 to 0.6 in original or fill, and got good results, I don't go higher than that.

fallow veldt
pliant raptor
#

What happened to the bot?

nova zodiac
wintry stirrup
#

dreamstudio not working for anyone else?

nova zodiac
wintry stirrup
nova zodiac
fervent thunder
#

whens the new model gonna be released ?

#

feels like we SD bros are falling behind

twilit mural
#

do xformers change quality or simply make it faster

nova zodiac
nova zodiac
twilit mural
#

Dope

#

Iā€™ve been messing w this all day lol

fervent thunder
charred thistle
#

have there actually been any newer UIs come out recently?.. I know comfy used much less resources than webui.. any new alternatives worth looking into?

fervent thunder
#

you can customize įµ‰įµ›įµ‰Ź³Źøįµ—Ź°į¶¦āæįµ

charred thistle
#

well comfy might be broken in recent updates with intel arc GPUs

twilit mural
#

is comfy faster?

fervent thunder
#

comfy is the fastest because it hooks into the model directly and what you see is the code

#

the nodes are the code blocks

twilit mural
#

i started today and as iā€™ve started doing more advanced things itā€™s gotten quite a bit slower

#

on webui

fervent thunder
#

do comfyui, its so much faster

#

i switched to comfy fully because of speed advantage

twilit mural
#

was it worth

#

i only have 8gb vram

fervent thunder
#

i got a 4090 and it got bottlenecked by a1111

twilit mural
#

i have a gtx 1080 lol

#

so

fervent thunder
#

thats what i had before and it still was better to use comfy

twilit mural
#

oh really

fervent thunder
#

you get faster generations

#

more control

#

less bloat

#

more options to do what you want

#

right now i got a workflow that generates wallpapers 2560x1440p in like 1 minute

#

all automated

twilit mural
#

thatā€™s crazy

#

itā€™s taking me 6-7 minutes to do instantid on 512x512 deadass

fervent thunder
#

with face detailing, SDXL refiner, wildcard prompt, postprocessing like adding noise

#

and upscaling

twilit mural
#

iā€™m using realvisxlV30Turbo

fervent thunder
#

thats a good model

#

juggernaut is also good and can do more styles

fervent thunder
twilit mural
#

i got my 1080 for free

#

founder edition

#

my friend had it unopened and didnā€™t need it so he just gave it to me lol

fervent thunder
#

i had to pay $800 for mine and got only $200 back šŸ˜­

#

when 5090 releases imma sell my 4090 and get the new one

nova zodiac
#

1080 still holds up all these years later

twilit mural
#

i can run games and ai models so id say im chillin

fervent thunder
#

i hope blackwell architecture has some neuromorphic cores

twilit mural
#

iā€™m due for an upgrade at some point though

#

cpu not the best tho

fervent thunder
#

just stretch it as long as you can, right now we dont have good AI hardware

#

AI currently is nothing but high dimension function approximators, we dont have dedicated consumer hardware for neuromorphic computing

#

so its not real artificial intelligence

#

its just simple machine learning

#

the real AI stuff is only possible with neuromorphic hardware that can do model fitting and eval super quick as an operation like a single tensor operation

twilit mural
#

Oh i also tried realisticVisionV60 and got insane results

#

Not with instantid obv just from prompt generations

#

but itā€™s crazy

#

pics look real

fervent thunder
#

you can make it more real in comfy by adding some noise lmao

twilit mural
#

im gonna try comfy tmrw

#

kept seeing it today when i was doing research

nova zodiac
fervent thunder
#

wait come to general with images @twilit mural

twilit mural
nova zodiac
#

Just dont forget to wind the cfg down on ito

warm junco
twilit mural
#

yea i went in and added those

#

earlier

#

itā€™s a bit faster still slower than earlier though which is odd

fervent thunder
#

check out the images

#

they are generated with a single prompt & workflow in comfy

warm junco
#

You shouldn't get 5 minutes for instant ID

twilit mural
#

iā€™m not sure if itā€™s because i have more models installed now

#

but all of a sudden it got a lot slower

twilit mural
#

Painfully slow

fervent thunder
twilit mural
#

maybe my pc needs a break iā€™ve been doing this nonstop for hours lol

warm junco
#

For instant id faceswap

fervent thunder
#

ah

#

i gave up on controlnet sadcat

twilit mural
#

itā€™s pretty amazing but i canā€™t even use it now because my images are taking 15 minutes lmao

#

i give up

#

for now

warm junco
twilit mural
#

i just turned my pc off šŸ˜­

fervent thunder
#

i tried using photomaker to get instant ID but it was so bad

warm junco
#

Oh ok

twilit mural
#

Omg i literally spent all day doing that

#

iā€™m addicted

warm junco
#

As external tool

fervent thunder
#

also gave up getting controlnet to work in comfy

#

always giving out errors

twilit mural
#

Yeah anyway iā€™m not sure whatā€™s going on, most images were taking a minute or two if i had like higher res settings and now theyā€™re taking 10 on low

I installed several models so idk if thatā€™s why, or if itā€™s bc of how long i was running my gpu or what

fervent thunder
#

most likely because of dimensions

#

the controlnet image needs to have the same dimensions as the image its applied to

#

and i dont get why its not in the node itself, why do i have to fiddle around with that

warm junco
twilit mural
#

I had tried it with and without hires fix and basically the same

#

which is also a bit weird

warm junco
#

Strange. I have a 1080 too and not such slowdowns

twilit mural
#

Itā€™s like itā€™s getting stuck, itā€™s not throwing any errors but the ETA will pause for like 30 seconds and then start going UP

#

and then itā€™ll stick again and repeat

twilit mural
#

Do you have 8gb or the higher

warm junco
#

Or figure out why its taking so long

twilit mural
#

Aight iā€™ll revisit it tmrw

warm junco
twilit mural
warm junco
#

Yea we can look at it tomorrow then

twilit mural
#

comfyui looks like an unreal engine blueprint

pearl elbow
#

How can generate it's totally change

sudden ruin
dry epoch
#

I have a question regarding automatic1111 v1.7...

#

How can I tell it in which directory to look for models?

warm junco
dry epoch
#

cheers ...

#

should I add this directly in the .bat file?

#

huh.. so that used to be a dialog some versions ago... now it is straight into bat?

warm junco
dry epoch
#

ok, cool, gotcha

#

I jumped a few version numbers...

warm junco
dry epoch
#

ah ok, I will head over there then ;D

#

cheers$

lucid marsh
#

Just patiently waiting for ROCm on auto1111

#

My XTX so trash with directml

nova zodiac
lucid marsh
#

Like whoever makes auto1111 also has to support it

nova zodiac
lucid marsh
#

I know we're lacking windows ROCm and that's cringe

lucid marsh
#

So it's a direct translation layer then

nova zodiac
#

the rocm just adds a layer that converts the cuda call into something the amd gpu can understand

warm junco
#

But for now I can do anything on my XTX

nova zodiac
warm junco
lucid marsh
#

Like higher quality with less fuckups

#

Not sure how much truth there is to that

warm junco
#

Nah quality isnt defined by the Backend

#

You can get 4k images on CPU

nova zodiac
lucid marsh
#

Speaking of CPUs

#

Has anyone tried the ryzen 8000 gen 2 AI tiles for something like this?

nova zodiac
#

Theoretically one should be able to get rocm going through wsl

proven mason
#

I haven't been able to generate images for sixteen hours, since last night. I keep getting the message "something went wrong on our side, please try again later". I wrote to support, no one responds. can i fix it myself?

warm junco
nova zodiac
lucid marsh
warm junco
nova zodiac
warm junco
proven mason
#

when will it be available for use? is this the end of the website? do I need to buy services from another AI?

nova zodiac
lucid marsh
#

Amd kinda lame for advertising the shit out of RDNA 3 ai capabilities and then being lazy as fuck and not actually developing it

nova zodiac
lucid marsh
#

It's been like a year and a bit and there's still no rocm

#

What a joke

nova zodiac
#

to be fair, cuda has been around for well over a decade..

#

2007 release

lucid marsh
#

Rocm has been out for ages now as well

#

RDNA 2 had it no?

#

AMD shouldn't have advertised the XTX's "AI" capabilities if they didn't actually want to implement them

nova zodiac
#

not sure.. haven't followed it that closely

lucid marsh
#

It's been out since 2016

nova zodiac
lucid marsh
#

As is the use case for rocm

#

Rocm 5.6 was supposed to have RDNA 3 support and that's been out for a while

nova zodiac
#

I agree.. but not all AI applications use it which is why they can tout their AI prowess while still be totally shit at AI art

#

the moment they release rocm that works and a 48gb top end card, that shit will sell like hotcakes

warm junco
#

There are a few library's on github that need adjustments

lucid marsh
#

I'm reading the documentation now and it looks like rocm on windows is part of the pro driver?

#

I may be stupid

#

Or is HIP just a part of ROCm and we don't get the whole thing

warm junco
nova zodiac
#

just that lol

lucid marsh
warm junco
nova zodiac
#

you'd think they'd just get the AI to whip up some code for that and she'd be done in a week lol šŸ˜›

lucid marsh
#

I'm just gonna tell chatgpt to finish it for me /s

warm junco
lucid marsh
#

"chatgpt write me a functional ROCm package for windows"

Go to install it and your GPU fucking explodes

#

Ok gaymers I am landing in Toronto rn

#

Pilot be doin some wild shit today

#

Last time I tried this in MSFS it did not go well

warm junco
#

MSFS?

lucid marsh
#

Microsoft flight simulator

warm junco
#

Ahh

lucid marsh
#

That game low-key trash ngl

#

Want me to pay 25 bucks for a single plane

#

Eat my ass

#

I'll just go play ace combat

warm junco
#

Lol did they include micro transactions?

lucid marsh
#

So many

#

You get a handful of shitty lame planes for free that just die when you pull up too hard or that are abysmally slow

warm junco
#

Okay, goes on the not buying list

lucid marsh
#

And you gotta fork out a lot of money for each plane that isn't trash individually

warm junco
#

So back to Google Earth's F17 jet xD

lucid marsh
#

Sometimes like 90 euros

#

Whatever that equals in usd or cad

#

For a single plane

warm junco
#

Thats more than the game itself wtf

nova zodiac
#

jeesus f

lucid marsh
#

I played it once

#

Saw I couldn't use an f22 without forking out like 25 bucks

#

Checked out the rest of the store, flew a commercial airliner once cuz it was the fastest coolest plane I had (everything else being a shitty prop) and never touched it again

#

There is no progression and no way to get these planes for free

#

No in game currency

#

Just buy with your credit card

warm junco
#

Wow

#

Didn't thought they f this game up that far

fervent thunder
#

Server is pretty much useless, now that they removed the bot

#

Theme is too narrow

#

Will die soon

warm junco
plush night
#

has anyone experience with the ai influencer trend rn?

gleaming bane
#

what upscaler do u guys use?

long wigeon
#

Does anybody have any experience with Kohya_ss?

nova zodiac
gleaming bane
#

does anyone use deoldify in webui?

novel echo
#

What's up everyone?

lunar nebula
#

Hello, I have a beginner question šŸ˜‰
I follow this picture:
https://civitai.com/images/4232027
I tried the prompts and other. What wonders me is that they only make 7 steps. Does that means sampling steps? I have tried 20 to 30 usually. When I try 7 it is way to undetailed ...

long wigeon
#

Have you used the exact same generation data? Then it might just be that he mistyped it and it originally said 27 or something

#

https://civitai.com/models/4201?modelVersionId=245598

On this models page it says the following:

`Realistic Vision V6.0 (B2) Status (Updated: Jan 16, 2024):

  • Training Images: +380 (B1: 3000)
  • Training Steps: +76k (B1: 664k)
  • Approximate percentage of completion: ~12%`

Does that mean that model is only 12% done? If so, why is it available already?

still glacier
# lunar nebula Hello, I have a beginner question šŸ˜‰ I follow this picture: https://civitai.com/...

Why do I get different results from civit.ai examples :

1/ --xformers (and some other flags but mainly this one) will make stable-diffusion non deterministic. Meaning for the same prompt+settings running on different hardwares you'll get different results. The output should still correspond to whatever the prompt is.... It might just be slightly to very different if you're unlucky. So yeah, some command line arguments can alter the outputs of your prompts and your hardware might also do the difference.
2/ Do you have all the loras, embeddings mentioned in the prompt ?
3/ Are you really using the same settings ? Copy pasting manually what's inside the "Generation Data" boxes is not enough. Civitai is hiding a bunch of data there. If you want to get the exact same settings you'll have to click the "Copy Generation Data" button at the bottom of the image page, paste its content into Auto1111's prompt field and then click the blue arrow under generate to automatically apply each value to the correct field. Hidden values can be; clip_skip, ENSD, token merging ratio, etc
4/ Did you have any overrides set beforehand ? They should show up at the very bottom of the page (some will probably show up too if you use data from civitai like I mentioned in 3/)
5/ Maybe it's using some extensions that does not record its settings in the metadata
6/ Maybe you're using different versions of some extensions/auto1111/lora/models/etc that yield different results
7/ Are you using the same random number generator ? GPU / CPU / NV (Settings->Stable Diffusion->"Random number generator source")
8/ Maybe you're using a newer version than OP. One that includes a "seed breaking change", cf https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Seed-breaking-changes

storm shard
lunar nebula
#

What does that mean "turbo model"? Please excuse my cluelessnes šŸ˜‰

storm shard
lunar nebula
#

OK. But when I use the viewer steps the picture looks not finished ...

#

It has rougher graphics and colors.

#

And when I can ask, what is the drawback of that turbo-model instead of the normal model?

#

OK, set down the scale to two and now it gets a good image. Reading is an important skill though ...

naive frost
#

Is anyone running deepspeed in WSL but the rest of their stuff like comfyui whatever in Windows? I am sick of trying to get deepspeed to work in windows but I know wsl is like a 15-20% performance drop for this

# 
(base) PS Z:\stable_diffusion\DeepSpeed-0.13.1> mklink /D .\deepspeed\accelerator ..\accelerator
mklink : The term 'mklink' is not recognized as the name of a cmdlet, function, script file, or operable program. Check the spelling of the name, or if a path was included, verify that the path is correct and try
again.
At line:1 char:1
+ mklink /D .\deepspeed\accelerator ..\accelerator
+ ~~~~~~
    + CategoryInfo          : ObjectNotFound: (mklink:String) [], CommandNotFoundException
    + FullyQualifiedErrorId : CommandNotFoundException

(base) PS Z:\stable_diffusion\DeepSpeed-0.13.1> ln /D .\deepspeed\accelerator ..\accelerator
ln : The term 'ln' is not recognized as the name of a cmdlet, function, script file, or operable program. Check the spelling of the name, or if a path was included, verify that the path is correct and try again.
At line:1 char:1
+ ln /D .\deepspeed\accelerator ..\accelerator
+ ~~
    + CategoryInfo          : ObjectNotFound: (ln:String) [], CommandNotFoundException
    + FullyQualifiedErrorId : CommandNotFoundException

(base) PS Z:\stable_diffusion\DeepSpeed-0.13.1> New-Item -ItemType SymbolicLink -Path ".\deepspeed\accelerator" -Value "..\accelerator"
>>
New-Item : Cannot find path 'Z:\stable_diffusion\accelerator' because it does not exist.```

I hate this os so much
#

literally running entire shell as admin it even does a preliminary check to see if i AM admin

young karma
#

where are the bots ?

still glacier
bleak matrix
#

Good afternoon, everyone! How are we all today?

naive frost
#

good other than wantong to throw computer out window

outer crescent
#

did they remove the philosophy channel or am i blind?

hot harness
#

Hey (just asking) why does Bing AI generate better images (i'm not talking about AI artifacts) from input than midjourney and is there a way I can change midjourney to behave more like the Bing AI gen or develop some midware between the initial prompt and the midjourney to do this?

outer crescent
#

you cant look under the hood, but its assumed that each take your prompts and polish them nicely with refiners.

hot harness
#

I was thinking is theres a textual midware that fills in random stuff based on some other model

idle heart
#

Hello! Can someone tell me where I can generate ai pictures? Is this still possible on this discord platform?

outer crescent
#

thats the assumption we all work on, mr numbers

hot harness
#

Right, but I'm asking if anyones knows (or has ideas) why Bing AI behaves the way it does

outer crescent
#

if you want similar to that you can do it locally but your mileage will vary. the a1111 extension dynamic prompts, lets you download small language models that spice up your prompts.

hot harness
#

Thanks will be taking a look

storm shard
outer crescent
#

well in fairness the philosophy channel was just whining but slowly šŸ™‚

#

thanks eloyse šŸ™‚

forest trout
#

They have removed a few channels actually.

#

Looks like the stability.ai side projects channel got removed too.

hot harness
outer crescent
#

different things. controlnet is very, VERY good and you should have it. look up some tutorials on it. its super powerful and will help you with stuff you didnt know you needed help with

hot harness
#

Yeah

#

I was just unsur what you exactly meant by dynamic prompts

#

Is it a model made by a1111?

#

Or what?

outer crescent
#

dynamic prompts will add words to what you say based on it. like if you asked for a "fat cat on a mat" it would add in "closeup intricate photo showing a cat's fur as it sits looking at the camera, on a lovely persian rug, polaroid photograph, film grain" šŸ™‚

hot harness
#

Ok thanks man will take a look ;)

#

Where do you get it though?

#

I couldn't find it on the a1111 repos list

twilit mural
#

i am back

outer crescent
#

i'd suggest playing with controlnet this weekend first. and watching some tutorials.

twilit mural
#

time to try to fix my webui/also try comfy

outer crescent
#

one at a time, in case one causes annoying errors šŸ™‚

marble hawk
#

Would sd-turbo (not xl) work okay on a laptop for prompt practice??

#

Or would it still be super slow?

nova zodiac
marble hawk
#

No, laptop is basic 4500u APU 16GB shared RAM

#

Was just wondering how light the models can run nowadays

#

Guessing AMD support still lacks then?

#

Been sticking to using bing image creators, mage space, and playground while the discord bots are down.

Was just thinking if a local option was viable for prompt practice only.

nova zodiac
#

Depends on the q

nova zodiac
warm junco
nova zodiac
#

Probably #šŸ¤ļ½œtech-support or here then.. first thought is that your best bet is to use photoshop to ā€œput the liquid inā€ by a rough painting it in, then doing img2img on that using controlnet canny with the image of the empty bottle as the reference image

marble hawk
nova zodiac
marble hawk
#

dang

#

Wonder how much more efficient it will be in a few years?

nova zodiac
#

If you are just practicinng prompts and arent too fussed about not having a hires fix workflow with upscalers - try the generator on the Civitai site

warm junco
marble hawk
wraith olive
#

When will the bots be available ?

nova zodiac
#

In short, dunno

hallow sable
#

Chief nz pretty good guy

nova zodiac
#

Thanks šŸ™‚

wraith olive
nova zodiac
#

No worries

marble hawk
#

super slow

#

maybe they have temporary issues

amber shuttle
#

Hello everyone! I have a question..

I only use online tools for generating images and training Lora's..
Now I want to use something like BLIP-2 for tagging my images.. Is there good website (or Mac app) to do this?

#

Tagging everything by hand can be a pain stometimes. (especially in the Civit lora training tool)

#

I know tagging everything yourself can make it better, but I wanna let the AI do the heavy lifting you know šŸ™‚

honest dagger
#

Pessoa, por onde fazemos as imagens?

nova zodiac
nova zodiac
twilit mural
#

How well does SD run on Apple silicon? like base m2

#

im assuming not terrible but not great either

steep star
twilit mural
#

makes sense

#

i have an m2 but i also have a 1080 so i think im gonna keep using the 1080 šŸ’€

twilit mural
#

holy crap comfyui is FAST

hallow sable
#

can dreambooth work with comfy ui?

#

when I open auto11 it uses 6gb alone

nova zodiac
hallow sable
#

yeah I keep meaning to look into kohya. does training images 768res work with 1.5 model?

#

ive heard some people say that there is a few 768 images in 1.5

nova zodiac
hallow sable
#

oh awesome, Ive always thought that 2.x was for 768 and XL was for 1024 or bigger

wind island
#

anyone wanna help me make some banners for a discord server. im new to this stuff and stuggle a bit to get what i want haha

shut frost
#

What are some good male character trained anime style models?

true locust
shut frost
true locust
twilit mural
#

is there an instantid alternative for comfyui since the official implementation isnt outyet

#

im not gonna use the unofficial implementation seeing as apparently it takes 40 mins to generate

crisp verge
#

I was not at all active

sharp hemlock
#

Any configuration that makes it go faster? I have a 4090 and Ryzen 9, but I don't know if something can be configured to make it go faster.

true locust
hallow sable
#

not even worth it to overclock anything today really

#

everything is done automatically

true locust
#

Okay I didn't know that, good to know

naive frost
#

6 hours of writing go today to close work laptop and open fun machine to broken deepspeed python

#

What parts of stable diffusion are faster on windows? I know deepspeed gets a performance hit in wsl, but I think I read that a lot of sd stuff nets you 10-20% faster on windows. i have no idea if i care about these numbers or not. I think it was related to nvidia/cuda drivers being better on windows? could be way off

#

i think at this point i'll take 10-20% perf hit to run all of this in containers

fervent thunder
#

Wtf

#

Yapathon happenin

naive frost
#

woah someone talked in a chat room alert batman

nova zodiac
naive frost
#

are any of these developer chans generic they look like projects

twilit mural
#

should i be using 11/16gb of system memory when iā€™m generating

#

is that cuda fallback?

nova zodiac
twilit mural
#

i have it set to preferred off

left gulch
#

Hey fellow AI enthusiasts, hope you can support me by subscribing to my youtube channel. I create documentary content on AI news, current events, business, travel, etc. Hereā€™s one of the videos I made šŸ™‚ https://youtu.be/zfpKeujiaVE?si=Z-dfAArKeXcLGz7a

nova zodiac
#

Bye spammer!!

twilit mural
#

I think itā€™s because i need to unload models after using one and switching

#

yeah that fixed it

sullen temple
#

hey anyone here played much with CLIP Interrogator?

#

wondering if there is some tuning i can do to get better prompt ideas, i've got this pretty sweet img2img music visualizer going that takes album art, turns it into a prompt with clip interrogator

nova zodiac
#

Its good for getting tags you wouldnt normally think of, but its not as good as it could be

sullen temple
#

yeah it really loves to say arafed

#

plus omg if you haven't played with it yet, streamdiffusion is black magic. sdturbo, 768x448 , and 70+ fps doing img2img with a 4090 and a 14900k

#

it does use a slightly gimped VAE though

pallid sable
#

i ned dis bro

#

plz turn it back on

sullen temple
vestal dew
#

Anyone know of something like SadTalker, but better, to make a video appear to speak - talking head. SadTalker's results are disappointing.

nova zodiac
vestal dew
#

D-ID?

nova zodiac
#

Yep

#

Its an online tool (not a local install) but its best for that task

vestal dew
#

that looks like pre created avatars. I want to take images I make - like SadTalker does

trim magnet
#

u can upload your own pics with chars u make

vestal dew
#

I see

#

it is much better than ST. However, it doesn't let me crop the image as I want in the app. That's a bother.

#

Would be nice to find something like that to run locally. Their pricing is limited minutes.

trim magnet
#

for free sadtalker is the best u can get

vestal dew
#

such a shame. It doesn't do a very good job

#

thanks for telling me about it.

quaint saffron
#

My instand ID broke or something :/

frank rampart
#

hey diffusers. i got a weird error when launching my webUI today "ModuleNotFoundError: No module named 'optimum'".

Ive been told "Simplest fix would be to just go into the webUI directory, activate the venv and just pip install optimum" but i have no idea what most of that means

fallow veldt
woeful quail
#

šŸ˜Š

nova zodiac
deft depot
#

I also just got Errors today, when i opend up my UI, directly Error messages and when i try to use controllnet also error, Any suggestions? I didnt nothing just started this morning nothing works šŸ˜„

nova zodiac
#

Simple answer is to open the command window, go to the webui folder then the venv subfolder then scripts (or bin for linux) then type activate.
That starts the venv, then put your pip command there

nova zodiac
deft depot
#

Controllnet says, no input image is given, but its there, tried it now with multiple images

#

Solved

nova zodiac
deft depot
#

it enabled a second controllnet unit, but was not showing it as active, toke some time until i found, as it was grey, but the checkbox was on enable

pseudo bough
#

anyone know how you can give priority to programs that arent in the foreground? my stable diffusion applications i notice get a slower iterations per second when the command prompt with the program running in it isn't in the foreground

#

windows 10

graceful peak
#

how come i get terrible looking images even when i use the same exact prompts and settings and seed as other generations

rich kestrel
graceful peak
rich kestrel
#

are you copying from civitai

nova zodiac
rich kestrel
#

cuz those are not always reliable

graceful peak
graceful peak
nova zodiac
#

Could be a number of things

#

The best bet is to open the image up fully in civitai, right click and save it, then drag it into png info and send it to txt2img from there

rich kestrel
#

thats not why

nova zodiac
#

That way if there is metadata there that exists but isnt shown by civitai, youll see it

rich kestrel
#

as someone who posts on civitai, I can tell you is because most of the good pics there arent one-click-and-done

#

theres img2img, inpainting, inpainting sketch, even photoshop involved

graceful peak
nova zodiac
rich kestrel
#

nope

#

civitai just asks you for the prompt

#

a few settings

#

but not much else

#

you can put whatever you want there even if its not the original prompt

graceful peak
rich kestrel
#

it should come with a warning tbh

#

yah so when I post an img that I did a lot of inpainting of, sometimes I forget to put the original prompt

graceful peak
#

i never get inpainting right it always looks out of place

rich kestrel
#

and use the inpainting prompt which is different than original (usually)

#

ur better off following tuts from scratch like from olivio sarikas

#

unlike what most haters will say, AI gen is a skill not unlike photoshop or blender

nova zodiac
rich kestrel
#

gotta practice it, set up the right workplace, it's just a newer tool but one that takes patience nonetheless

nova zodiac
#

Has it there at the top that that image is txt2img

#

But yes, the best images arent straight txt2img

graceful peak
nova zodiac
graceful peak
#

oh also sometimes it provides the seed

#

but i think most images use hires fix and i dont rlly know much about that

nova zodiac
#

Good model makers should be making samples that are using basic workflows that are easily replicatable (ie no inpaint or img2img)

graceful peak
#

cause i just came back after a while and it wasnt there in the older webui version

graceful peak
nova zodiac
pseudo bough
#

well i have made a model and posted it on civitai and i didnt have my prompts saved so i just put up the closest guess of what i used for them in the sample images i provided

nova zodiac
graceful peak
nova zodiac
graceful peak
nova zodiac
#

More denoise, or try a different model

#

Or more steps

rich kestrel
#

they were probably using openpose/controlnet on the civitai prompt

#

if you want good hands you HAVE to use cnet

nova zodiac
graceful peak
#

i upscale it to 1024x1024 or 1120x1824 and use latent cause idk whats good or not

rich kestrel
nova zodiac
#

Latent is fine, just needs lots of steps at a higher denoise

rich kestrel
#

look that one up

#

its good for photorealism

graceful peak
nova zodiac
graceful peak
nova zodiac
#

Its good

graceful peak
#

also would my specs be an issue at all?

nova zodiac
#

Not the best these days, but still worthy

nova zodiac
graceful peak
#

so more samples means better image always or no

#

cause 150 samples takes like forever

nova zodiac
#

There is also things like token merging and hypertiling that can affect reproduction of an image as well

graceful peak
#

so ive heard about controlnet for specific poses and all

#

are the results worth it?

nova zodiac
#

Controlnet is OP

#

Def worth learning

graceful peak
#

where can I find it

#

for the webui

fallow veldt
# graceful peak so more samples means better image always or no

Not always, there is a sweet spot that really depend on many things. Too high CFG with lots of steps usually create a more saturated image. Sometimes the image is better at least steps as SD keeps adding detail and blowing away nicer simple detail, that might worked good for the generation. Dependends also on sampler, model, etc. So it is better to search images on Civitai and copy those parameters, or try several number of steps.

#

Of course a very low number of steps would leave a lot to define, but a high one may saturate everything. A medium CFG scale like beetween 4 and 6 and a medium amount of steps (haven't used 1.5 lately) will be a good start.

#

Euler sample takes more steps, DPM take less

#

If you want to try thing out fast with less steps try the Karras samplers, they perform with less steps

#

CFG is how much SD will do your prompt in every step. So high CFG scale with lot of sampling steps it will try to your prompt a lot and over and over, so that's why it will be saturated.

graceful peak
graceful peak
#

or does it vary by model

nova zodiac
nova zodiac
graceful peak
nova zodiac
#

Defintely worth diggin into some youtube vids on the topic to so you know what you are expecting

terse ice
#

Is anyone active that can help me?

fallow veldt
terse ice
#

ive tried to upscale my video but ive been given this error: OSError: broken data stream when reading image file

sage spade
#

Generate dragon year card

mint axle
#

Bro where do I prompt like in which channel

terse ice
fallow veldt
terse ice
graceful peak
#

how come inpainting takes so long compared to text2image? no matter how little samples or upscale it takes like 3x the time to generate a same detailed image

frank rampart
#

i installed optimum successfully but now got this error
"import of xformers halted; None in sys.modules"

#

like dafuq? this thing was working fine last night...

molten monolith
#

Windows 11 is actually so bad lol
It'll just randomly tell you 2GB of disk space left on boot
then when you restart, it goes back to 44GB free space
then it waits for the next sneaky time it can randomly sabotage you by consuming space and pretending you're out of it

frank rampart
#

okay i think i fixed it...new Python Pip v24.0 is out so had to upgrade

graceful peak
frank rampart
#

wait, nvm, still getting xformers issue šŸ˜¦

#

plz halp

#

"Failed to import diffusers.loaders.ip_adapter because of the following error (look up to see its traceback):
import of xformers halted; None in sys.modules"

#

going to try "pip install --upgrade xformers" command line now...

lunar nebula
#

Hello, I have played yesterday a little bit and I don't understand how to "workflow" properly. I used the dreamshape turbo which is pretty fast and made some nice pictures. But I wonder how the usual workflow is? I generated a picture I like. Is it now the way to remake the image with the normal dreamshaper model or what is the drawback of using the turbo model? I want to understand the basics and choosing and understanding the model I think is an important first step šŸ˜‰ When I use different models I get different styles and pictures since they trained differently but here I have the same model in different variants?

frank rampart
# lunar nebula Hello, I have played yesterday a little bit and I don't understand how to "workf...

jea that sounds about right. different models give you different art styles. you can upscale the image to get bettere quality, or take it to photoshop and change it up.
usual work flow is once you get an image rendered- you can render a new one, it usually gets better with multiple renders. or you can use that same image in the img2img tab to try improve it. using prompts like "masterpeice...no artefacts...no static...high quality...."

lunar nebula
frank rampart
#

jea generally, once you got an image you are happy with, then you are finished.

#

if you are happy with it, then to upscale it,take it to EXTRAS not img2img. .

if you take it to img2img, it might change the image slightly, as you seek to improve areas like funny hands, weird eyes, etc.

#

keep messing around and experimenting, i think the workflow is simple but you cn find what works for you

lunar nebula
#

Thank you

frank rampart
#

watching other people go through workflows and talk on youtube is also enlightening

lunar nebula
#

Yeah, I searched for that but didn't find good to follow (have to search more, I know they are out there šŸ˜‰ ).

paper anchor
#

Unable to use the bot. Any help?

chrome barn
#

Idk what channel to ask this in but is there any SDXL model that can rival midjourney v6 in quality (excluding hands cuz models can't do nails)

chrome barn
lunar nebula
#

Another question, please be patient šŸ˜‰ I have a image I like (a knight in powerarmor). There is a crest on the chest which is very unsharp. I want to detail that up and maybe switch it to a crest which already exist. Which tool and words can I google to get some directions?

nova zodiac
nova zodiac
lunar nebula
nova zodiac
#

yeah inpainting a specific logo pretty much isn't going to work without ending up with Homer Simpson's haircut...

#

and for photoshopping, should be plenty of youtube tutorials (or search for krita or gimp if you don't have photoshop as they are free bits of software that do the same thing)

warm junco
#

Xformers only works for Nvidia cards

graceful peak
#

whats the difference between xl and normal models

nova zodiac
graceful peak
vast berry
#

the one time i wanna use the bot its down waaGONE my luck be so bad

nova zodiac
vast berry
#

yeah

#

rip bot

nova zodiac
#

well, more 'til next time'

tranquil mural
#

hey

#

new here! we are a startup running SDXL on aitoolbox org, great to be here!

lone rampart
#

.

fervent thunder
#

hi

fervent thunder
#

hello guys, urgent, where can I upload a photo of me and it will generate a photo of me in formal atire?(for resume purposes) thanksss

gilded lantern
#

šŸ‘ˆ hey everyone i got working multi AI bot check it out in my Disc

gilded lantern
obtuse nexus
#

Hi, is the API down ?

#

my bot can't generate images anymore

marsh plover
#

Hardware question, I'm just using CPU mode with my i5-8500T and 16GB RAM. How much faster, if at all, would if I get a used desktop with a 750Ti SC? ._. Or do APUs work? Just wanna keep playing around with this but not make a big investment.

naive frost
#

does anyone use lobehub SD as their sd webui theme? seems to have made my ui impossibly slow to use, idk if thats expected with tons of exts or what - it could be one of my other exts but I did a ton at once sadge

#

i can't even get to the damn place to disable it

rich kestrel
#

why do ppl complicate something that is already perfect

#

just do vanilla a1111

naive frost
#

don't reply with complete irrelevance

rich kestrel
#

lol then dont complain when ur homebrew works like ass

naive frost
#

i didn't complain, i asked if anyone used it, thanks kid, move along

marble steppe
#

so, stable diffusion running here as a bot, and letting however many people generate images..... what kind of hardware is it running on?

gilded lantern
#

i'm making a bot that has all the AI in one, image, text, voice, music, code, and if i had more hardware avatar, video, simultaneous, also working on task runner, better coder

marble hawk
#

Apparently the data used to train bing image creator doesn't like unicorns, says the results are unsafe...

my prompt was "herd of unicorns enjoying a meadow, unicorns playing, unicorns grazing, unicorns casting magic spells"

The unicorn data must have been tainted by horni-internet. šŸ˜æ

naive frost
#

does it even do just unicorn then?

marble hawk
#

Tried just "unicorns"

2 of 4 images blocked

#

thats pretty sad

naive frost
#

i've only used the bing ai maybe 10-15 times, i use openai all day long at work, maybe im bad at prompting but bard and bing help me so rarely with anything beyond like.. what year was 235 years ago sort of stuff

#

i pair program with chatgpt all day long pepe_wine

#

i wonder if it'll show me a unicorn maybe big unicorn is after you

#

wow ive never been to this thing

marble hawk
#

Tried again with "herd of unicorns enjoying a meadow, unicorns playing, unicorns galloping, unicorns grazing, unicorns casting spells"

3 of 4 images blocked...

#

data tainted

naive frost
#

im trying to test mine but this bing image site is weird

#

mine made them

#

they are SUPER my little pony style

marble hawk
#

guess you got good seed

#

or my seeds were bad luck

#

however it is

#

but unicorn prompts do seem to get blocked more often than other stuff I tried so far

lusty beacon
#

Would it be possible to connect a LLM to Stable Diffusion to generate or improve prompts?

honest kettle
lusty beacon
remote glade
lusty beacon
#

okay thx

nova zodiac
lusty beacon
#

any

nova zodiac
#

There is now a comfy node and a1111 extension

cedar salmon
gilded lantern
#

oh the extension

cedar salmon
#

ya

bold python
#

I'm Taylor David, a skilled AI developer with a focus on developing cutting-edge conversational AI systems. Creating AI chatbots, like ChatGPT, with a focus on voice AI, SMS, cold calling, and Twilio connections is one area of expertise. competent in creating Broadcast Chatbots, Auto-responder Bots, Telemarketing Bots, and Appointment Bots. competent in deploying frontline AI, VoIP sales, IVR, live chat, and conversational AI for Plivo SMS, guaranteeing smooth and efficient interactions.

lusty beacon
nova zodiac
# lusty beacon what can i do with it

The idea behind it is to enhance the prompt before it does the diffusion - kinda like the prompt magic that happens behind the scenes in Midjourney/Fooocus

drifting monolith
#

yy

sterile torrent
#

ok so i have question, what is better, comfy UI or SD A1111?

nova zodiac
drifting monolith
#

generate a picture of sunset

nova zodiac
#

Comfy was better in terms of lower vram usage, but sd-webui-forge recently got released which is an updated version of a1111 but much more efficient

nova zodiac
lusty beacon
lusty beacon
crystal edge
#

how can i queue a lot of prompts to generate images in a1111

sterile torrent
stuck sun
#

I ordered a MSI - NVIDIA GeForce RTX 4090 GAMING SLIM -- do i need some special power cable with this or am i good to go

valid garnet
#

Anyone play with Photomaker in forge? Installed the photomaker model, but can't figure out how to use the @#%$!@# thing

warm junco
#

sdforge is early acces imo

sterile torrent
#

i read about forge and this is 100% a1111 nothing change less differenc for positive aspects is that "Forge backend removes all WebUI's codes related to resource management and reworked everything. All previous CMD flags like medvram, lowvram, medvram-sdxl, precision full, no half, no half vae, attention_xxx, upcast unet, ... are all REMOVED. Adding these flags will not cause error but they will not do anything now. We highly encourage Forge users to remove all cmd flags and let Forge to decide how to load models."

gilded lantern
#

< hey check out my bot

gilded lantern
slate lintel
#

I've downloaded the repo but having an issue with the open_clip module. Is there a way to generate videos via discord?

#

is that something I have to pay for? maybe some kind of API situation?

valid garnet
#

Holy shit, Forge is a BEAST.

cedar salmon
#

forge is like a combination of A1111 and something that works

#

yes i stole that joke from someone

round wedge
#

Is there a specific place to ask for help with making a dataset to use in stable diffusion? Are we allowed to talk about nsfw content without posting it?

#

also how do I increase the samples?

stuck sun
pseudo cairn
#

I am new using SD, I have a big problem understanding the different versions of sd 1
Xl, 1.5, 2, 2.1, SDXL, witch version is better ? Can I train my Lora using automatic1111? 1xl is the same as sdxl?

stuck sun
pseudo cairn
stuck sun
#

np

nova zodiac
nova zodiac
left gulch
hollow gazelle
#

Hey all, just wondering about upscaling. Is the resolution you can upscale to restricted by GPU VRAM? Is there a way to get around it?

crystal coral
#

Oi, What up mates, Iā€™m completely new to this. Can anyone do a couple images for me?

broken cave
#

but not a wacky high amount

#

there's no application for very high levels of upscaling

broken cave
nova zodiac
#

Most upscalers are optimised for 4x, and limited by vram but can get around this with tiled upscaling

wicked root
#

how much images do I need to make a good lora?

gilded lantern
broken cave
wicked root
#

oh ok

#

thanks for the info

#

i'll try with multiple time with different amount of image then

hollow gazelle
outer crescent
#

1500w? are you powering a time travelling car? in the uk the cost of electric is insane right now and this makes me weep. not joking here, but if the cops ever turn up its because of your utterly insane power useage, stuff like that is normally for secret labs.

nova zodiac
outer crescent
#

lol. it is worth saying though that to up your pc for new GPUs the PSU almost certainly needs updated. I* remember at sdxl launch someone had bought a pretty new card and couldnt figure out why the computer wouldnt run. In this age, they should pair them in descriptions and not leave it in fineprint. still 1500w is ridic. But kinda futureproof, a good psu can outlast a build easily.

hallow sable
#

look at the warranty of a psu and derate them by .8. a 750w should be more than enough for any single gpu build

oblique willow
#

Does models have less data of boys?

#

I do not know if its an error but I have switched models, is on random seed, It keeps generating that one same model of a anime boy.

#

I have generated over 20 images, with revanimated,dreamshaper and reaslitic vision.

They are giving me that same face everytime

#

1boy, yellow shirt, blue shorts, round glasses, black hair

#

.
Ok now I am pretty suree something is wrong.

#

I am using a model which only generates a black outline pencil sketch

#

and it is giving me same model. same full aesthetic colorful thing. with same face shape/

broken cave
oblique willow
#

what should i provide

broken cave
#

it's hard to say because even if you write a reproduction here with all the mundane details it will sound fine. something is telling you an error somewhere, in all likelihood, that you are not noticing

#

there is something in the UI or the console that you are not noticing. such as your seed being the same between generations

sullen fossil
#

can u check the image i sent in tech support

#

real quick

#

cause i feel like im just missing something extremely obvious

oblique willow
#

but any idea how can i fix terrible anime faces i am getting

#

I am using negative embeddings, apart from them ,anything?

valid garnet
fervent pumice
#

i am brand new to this server so idk where id ask about this but

#

is there ANY LoRa / safetensors colab that doesnt auto disconnect that i can gen images in

#

ive made a lora (its a safetensors file) and the colab i was using no longer works

#

i also cant find anywhere on huggingface i can do this

nova zodiac
agile tiger
#

heya, can anyone share the link of the page where I can download the clip vision models from

sweet steppe
#

Did we have a Pantheon channel before? I donā€™t see it now

oblique willow
#

Is there any log file?

#

where i can see all commands I have used so far with sd

agile tiger
#

are there safetensor models for clip vision?

nova zodiac
#

Dont think so, just pth

oblique willow
nova zodiac
oblique willow
#

a111

nova zodiac
#

I dont believe so, it does save the last used command, theres probably an extension for that tho

oblique willow
#

so once i generate something

#

and do not remember the command

#

and do not have image either

#

There is no way to get that prompt back unless i remember it?

#

It doesnt keep any records?

fervent pumice
fading osprey
fervent pumice
#

just curious if theres any alternative at all

fading osprey
#

i think i dont remember his name candemru or smthg go check him out he might have something related to what you want

fervent pumice
#

ill find the colab i used to use rq

#

found it

#

you could put loras straight from your drive

drifting monolith
#

(Masterpiece), (Best quality), (Ultra HD), (Super detail), (Whole body :1.2), 1 girl, Chibi, cute, smile, flowers, outdoors, holding the camera, sitting on the roof looking out into the distance, with mountains in the background, amber, warm yellow, sunset, artistic sense, Quadratic style, white clothes,

stuck sun
#

and a dual rtx 8000

#

my third machine i have a 3090 and just ordered a 4090 for it too, this has a 1500 watt PSU

outer crescent
#

cool. is it actually cost effective versus renting cycles?

stuck sun
#

i mean i got all the quadros for free so that was pretty cost effective

outer crescent
#

im sure the two lines cross on a graph but i dont know where

#

oh well, the price is right šŸ˜„ just electric then and no big down money

stuck sun
#

yeah my power bills are pretty bad

outer crescent
#

mine is like Ā£70/week and i am not doing serious gpu intensive things. its just expensive in the uk

stuck sun
#

i think mine is about 1000 a month USD

stuck sun
#

so you pay a lot of money for bad renders

#

after a certain scale of course it tips over into not making sense to do it at home

lusty beacon
warm junco
#

What's your GPU?

thick raven
#

how to create image

stark hornet
#

can someone tell me whats new since the last few weeks? i've been off for a while

still glacier
barren mulch
#

do you know a tool similar to Nightshade that works on linux ?

outer crescent
#

im not sure what "works on linux" means in the context of that

#

i assume you mean the proof of concept image poisoning thing

hollow gazelle
#

I'm just wondering if there's a tool that auto chunks larger resolutions so any gpu can handle the processing. I guess it's not easy to do as mentioned earlier because each part has to be denoised according to a dataset?

bold python
#

I'm Taylor David, a skilled AI developer with a focus on developing cutting-edge conversational AI systems. Creating AI chatbots, like ChatGPT, with a focus on voice AI, SMS, cold calling, and Twilio connections is one area of expertise. competent in creating Broadcast Chatbots, Auto-responder Bots, Telemarketing Bots, and Appointment Bots. competent in deploying frontline AI, VoIP sales, IVR, live chat, and conversational AI for Plivo SMS, guaranteeing smooth and efficient interactions.

still copper
#

Iā€™ll teach 15individuals to earn $50k or more within 72 hours but you will pay me 10% of your profit when you receive it. Note only interested people should apply, drop a message let's get started by asking (HOW)

http://t.me/diceem

clever phoenix
#

I hate that the bot is down, I was gonna make eazy e out of cheese and call him cheesey e

naive frost
#

i'm just playin with ya but > focus on developing cutting-edge conversational AI systems > cold calling homer

#

you are the enemy! (so am i, i write advertising code)

#

god when AI can scam grammy over the phone its going to be rough "i sent little timmy 5 grand!" that wasnt timmy!

hallow sable
#

they dont need AI they just use kumar with the allies of chris asking for the itunes gift card

pseudo cairn
#

i am new using sd i have trained my own model using a google collab but now i want to use my own pc, so witch tool is the best to train my own models?

naive frost
#

lol this code gets around every single google link i get is just hosting it

marble steppe
hallow sable
#

server grade gpu that have 64gb or more of vram

#

some times things like the 3090 24gb

fervent thunder
#

Nvidia data center gpu H200 has 141 GB Vram

#

H100 80GB

#

others 48GB and 24GB

naive frost
#

i wonder if theres enough skyrim vr mods to fill 80gb vram

#

i can hit 24gb with like 500

surreal frigate
#

so how can i get my dirty hands on it? : )

surreal frigate
marble steppe
hallow sable
#

multi gpu set ups are so impractical, its purely for very niche cases like renderfarms

pale latch
hallow sable
#

or your a wirdo that has more than 3 displays

marble steppe
#

so SD can use multiple cards??? I was told it cant

#

I have a mining rig I want to convert to an SD machine, but was told it couldn't handle multiple GPUs

pale latch
#

stable swarm ui from stability specializes in doing this. managing a cluster of gpus and servers and poitning them all at the same UI

marble steppe
#

whats with the sick face react??

pale latch
hallow sable
#

mining crypto

pale latch
#

"the environment!" meanwhile, youtube or discord both use more energy

marble steppe
#

too bad for them I guess.

pale latch
#

they're jsut bitter because they didn't catch the wave while it was basically free money

marble steppe
#

yup ming a bit of crypto, wish I did it years ago but got to the party too late

surreal frigate
#

yeah,

#

lol

marble steppe
#

started out at about $25/day

#

quickly dropped to $12

#

then ETH switched to POs

#

now it's like $7/week mining RVN

#

but I have two 5700 XT's and an RX 6800.....

surreal frigate
#

they have build a mining systemin Kazakhstan in the mountains, so it's freaking cold so they dont have to pay for cooling : ) still failed lol

marble steppe
#

I'm going to stick the 6800 into an SD machine, I just picked up a motherboard and stuff the other day for it.

pale latch
#

my brother made 2 machines on bath racks. zap strapped parts to them. enclosed it all and piped the exaust into his central heating. saved money on heating and made money with mining. win win

pale latch
marble steppe
#

what is the rocm library though??

pale latch
naive frost
#

is there any point in a massive cryptomining workflow having a good connection, they're just grabbing a data set and running on it locally so thats convenient for siberia

pale latch
surreal frigate
#

SD doesn't support the use of graphics cards in parallel. Ive learned that lesson as well.

hallow sable
#

no because its the most stupidest thing in the world. your gonna do intensive mathematical operations to validate a transactions? crypto also holds no value, much less so than fiat currency that has value IRL. it was painful when GPU pricing sky rocketed in large part because of crypto. so much better uses of a gpu than mining even to generate income from

pale latch
marble steppe
pale latch
marble steppe
#

If crypto has no value how come I keep getting money from it?

#

How come I was able to make a few thousand?

surreal frigate
marble steppe
#

Where did that money come from then??

hallow sable
#

because pedophiles and drug users use crypto

marble steppe
#

why is BTC on the rise to another ATH due to the halving?

surreal frigate
#

people. lol who think they can cmake moeny. but you are who are making it : )

marble steppe
#

LMFAO HAHAHAHAHA

pale latch
#

gpus' didn't skyrocket because of crypto. the ceo's saw an opportunity to puff value and did. notice how since proof of work has become obsolete and gpus aren't needed for mining anymore, the price hasn't come down

marble steppe
#

peodofiles an drug users also money.....

#

use money

#

and even candy

surreal frigate
#

have a friend a he lost a fkn 50K on crypto and NFT. and he is mart, but not as amart as you are ; )

pale latch
#

prices of gpus are still high post crypto proof of work, and people are STILL blaming crypto networks for gpu prices surging

hallow sable
#

talk to me when you can buy real assets or even bread with that monopoly money

marble steppe
pale latch
surreal frigate
#

in canada it's easy, just go bankrupt for 7 years that's the downfall. but if you had a lot ot invest inveest millions, and it will be forgiven after 7 years period ; )

marble steppe
#

anyway, obviously some people are very uneducated and jealous when it comes to crypto.....but I'm talking about building a rig for SD.

#

tough shit if you don't like my mining or crypto.

surreal frigate
pale latch
surreal frigate
#

plus the noise and heat lol.

pale latch
#

the electric bill for it is actually significant. i wouldn't brag about theft in public places

#

more on topic, i've seen on kohya training scripts now, there's an option for "multi-gpu" but i haven't looked into it at all

surreal frigate
#

i was doing steraling in a return for my services after hours, so it kind of balanced it all out.

pale latch
#

i know they train the base model of SD on a giant cluster of gpu

pale latch
surreal frigate
#

question for you guys, TI it's kind of the old way of training right? I freaking hate LoRa.

pale latch
surreal frigate
#

no disrespect. off course

hallow sable
#

embeddings are only useful for replacing one face with another. at least thats what I mostly see them used for

surreal frigate
#

but SDXL does not support it.

#

LoRA is good in a way but soo freaking messy. TI is simple.

pale latch
#

SDXL supports embeddings. i've trained them with kohya-ss

surreal frigate
#

kohya-ss i've tried it, ehh i think i might have failed, i need to redo the process.

pale latch
#

not sure about automatic1111 ui, if their training menu, supports xl

#

there's also onetrainer

hallow sable
#

I use dreambooth with auto for training but I havent gotten lora to work with it

surreal frigate
#

you see Lora is a finicky thing : )

pale latch
hallow sable
#

Its easy to create models

surreal frigate
#

never had a dreambooth to work either lol

pale latch
#

i learned how to train models and loras using the extension and when i went to kohya it was like "Well why the fuck were they doing it that way!?" everytime i found a better way kohya did things

#

the dreambooth extension is really inneffecient too since it requires the memory hog auto1111 to be loaded along side it

surreal frigate
#

@pale latch damn man, so for example i want to train a face of a person, who do you use for that? I also want to train the posture as well.

pale latch
#

i'd use lora. kohya scripts. dadaptation optimizer.

surreal frigate
#

Ive trained my TI from 600 pix, and it turned out very good.

#

but new model does not support it.

#

wanted to play around with it, but nope.

pale latch
#

600 pix for one person ? lil over kill. 20 high quality and varied photos will do. i've gone as high as 100

hallow sable
#

dreambooth doesnt work with 2.1 only 1.5 or XL

pale latch
#

i've even trained on as low as 7 images

surreal frigate
#

I know right? that's what everyone is saying but it worked for me : )

pale latch
#

overkill is really effective

surreal frigate
#

and it did work beautifully when it came to the nails. hardest part to get.

pale latch
surreal frigate
#

URPM is freaking amazing, not for the port but for realistically looking person.

pale latch
#

oooh he's needles, not nails. my bad.

#

embed fail too

hallow sable
#

does composite dateset work? like say this is a wheel of car and this car? would it use that wheel with that car???

pale latch
#

i'm really just , i'm so over the pornography for "realism" in stable diffusion. It's not teaching the model anatomy. it's teaching it pornography.

#

big plastic skin effects too since every trained porn photo has hard studio lighting to create that pornstar magic.

surreal frigate
#

dude, i get you but it's trained it to be how i want it to be TI helps me a lot.

#

i specifically negate the nakedness

#

it works pretty good.

pale latch
#

fun fact. pornstars look like a used bag when they are in the real world and not under studio lighting. it's why strip clubs stay dark.

surreal frigate
#

a lot of models are too plastic.

pale latch
#

a lot of model merge pornography in an attempt to learn anatomy

surreal frigate
#

exactly. but if you have a IT with 600 pix, it give you what it is. try it.

hallow sable
#

make up makes all woman look like plastic lmao

#

also the fake tans

surreal frigate
#

made my ex like a superstar šŸ™‚ she loves it.

pale latch
#

ever watch tv? Every single guy is wearing makeup

surreal frigate
#

hahaha yep.
first date needs to be in the pool : )

hallow sable
#

yeah and after years of wearing that junk on their face they look like ghouls naturally

surreal frigate
#

hahaha depends

#

make up is different, you need to get a good brands, and natural oils in it. dont ask me how I know lol

pale latch
surreal frigate
#

sunscreen is the worst thing ever to put on your skin bro.

pale latch
hallow sable
#

why?

pale latch
#

someones watched robocop and believed the disclaimers for the commercials

#

awww embed fail aagain!

hallow sable
#

everything is bad for you but so is the sun

surreal frigate
#

it contains a lot of chemicals in it. best thing is to just get a coconut oil, it will make you tan and it will hydrate your skinn at the same time.

pale latch
hallow sable
#

what about cooking oil??

pale latch
#

i just dont want skin cancer is all

surreal frigate
#

hahah, belive me, i know mu shit. if you want to tan faster, baby oil is your solution.

#

but i cant stand the smell of it.

pale latch
#

its that paltrow girl and barrymore going around saying sunscreen is bad for skin isn't it? kardashian bullshit. beauty product sales shenanigans.

#

"Shut it Barrymore!" - Jean Luc Picard

surreal frigate
#

every beauty product is shit. tbh

pale latch
#

i dont want a tan. i dont tan. runs in the fam. cousin lives in thailand and when he comes back he's paler than me.

surreal frigate
#

but girls need to look good. so it's a freaking marketing fiasco.

pale latch
#

i just need to not have cancer

surreal frigate
#

guess what, you already have it . my friend.

#

like all of us.

hallow sable
#

cancer is random for the most part

pale latch
#

everything is marketing. you believe gpus are priced so high because of crypto. even years after crypto stopped needing GPUs for their networks.

hallow sable
#

you can spend all your life smoking out in the sun and not get cancer...

surreal frigate
pale latch
surreal frigate
#

Genetics, Environment. etc.

#

I would rather live fast and die young, than suffer at the end.

hallow sable
#

until you get old when you have a family of your own

surreal frigate
#

I'm working in a medical field and IT so i cant imagine my self be like that dried out poptato person.

hallow sable
#

unless your 80 than your like fuck it

surreal frigate
#

73 for me : )

#

afret 74, sorry I'm out : )

pale latch
#

I'm in Canada so i can trust that if i'm suffering, i have MAID as an option now. I also figure with advancements , i'll only be a quarter way done in life at 40

surreal frigate
#

I'm in Canada too.

pale latch
#

You got MAID. people don't have to suffer in old age now

surreal frigate
#

Cant wait to leave this freaking place.

hallow sable
#

I knew this one co worker who was old around 60, and she kept crying because her mom died... I kept thinking wtf is wrong with her, most peoples parents die when they around 20 to 40, its rare to be 60 and still have one of them

pale latch
#

Barbaric christian policy imo. Keeping sufferiing seniors alive when they feel they're past due. Would be dead if it wasn't for machines. Now they can consent to having a dignified death. Beautiful time we live in

#

i wanna stick around as long as possible. The future is x rated i heard

surreal frigate
#

Yeah Euthanasia is legal now. .......Ive had to pull the plug for my mom.... just to stop her suffering, hardest thing i've ever done in my life.

#

I have a brother and Ive told him to do the same thing for me if cancer gets me.

pale latch
#

youth in asia have a very prodigeous future

hallow sable
#

yeah asian culture is superior that's why Asian kids go blind from staring at books all day and commit suicide when they cant get into the best university after studying 365 days for over a decade

surreal frigate
#

that's Japanese couture in simple description.

pale latch
#

pulling the plug is one thing. People who are only alive and unconcious because of a ventilator. MAID in canada allows you to volunteer willingly.

hallow sable
#

no its china and korean as well

#

south korea is like cyberpunk world in todays era

surreal frigate
#

In Russia, that's where I'm from, they make you like literally make you be better against you will. Tough country for sure.

pale latch
surreal frigate
#

the world is fucked, but we need to find happiness in it regardless. : )

pale latch
#

sometimes i go to the gym and pretend i'm dolph in the montage

hallow sable
#

russians cant speak tho.. they say ah pet instead opet.

surreal frigate
#

hahah, thereis NO Icant do it in Russian vocabulary.

#

yeah we are fucked in a way i agree ; )

hallow sable
#

i kakva ne mogu?

surreal frigate
#

But Dolph is Swedish : )

#

If he dies - he dies LOL

#

simple Russian philosophy

pale latch
#

Dolph is sweedish?! WTF?!

#

The hell have i been doing all these years? Training like a swede in the gym!? WTF!?

surreal frigate
#

aaahahahahahahha

#

boom goes the dinamite. šŸ˜„

#

He is also a chemical engineer by profession : ) booom

#

Ive actually met him several times. he is freaking TAAAL

#

I've met them Wesley Snipes and Dolph at one party, it's like meeting the Rock and the Kevin lol

hallow sable
#

I want to find russian wives

surreal frigate
#

nooooooo

#

they are the wors.

#

t

hallow sable
#

why?

surreal frigate
#

They will suck you dry. like in sex and moneys šŸ™‚ a prenup is needed 100%

hallow sable
#

is that not a universal constant?

surreal frigate
#

not yet. well for most people who have old money and nice taste. šŸ™‚

#

I'm looking for a nice Persian lady at this moment.

#

Slavic girls, are too smart, like tooo smart. But if they fall in love with you, you can run but you cant hide : )

pale latch
jolly berry
#

are the bot channels not coming back?

#

for ai image generation

#

what else can i use for it that is as good as stable diffusion image generation ai for free?

lilac knoll
#

ĀæWhere are the bot channels?

nova zodiac
nova zodiac
cobalt comet
#

Hi everyone, does anybody know what's the best option/model for image faceswapping?

crimson belfry
#

Can we make images here anymore?

hot wing
#

Hi guys! Any tips to have nice upscale using img2img i do want to use resize and fill option and finish of the composition with some experimenting, do you have tips on the settings like denoise etcā€¦? My machine isnā€™t very strong so not even sure I can use CNet too much ( 8gb ram laptop)

pale latch
#

its not really a swapper i guesss, but wow

#

outside of instant-id, everything else is based on insightface 128x128 resolution model, and polished with codeformer

cobalt comet
# pale latch instant-id hands down

I've tried insightface on discord actually, even i paid them but the model doesnt do great job on long bearded people. It works with Midjourney so so i'm searching alternative way to do this on SD side. Thanks for the tip, i'm looking into it!

potent spire
#

time for some Fooocus trial

pale latch
#

not sure how to use it on comfyui. fooocus has nice face swap too with the ip-adapter approach

potent spire
#

oh i didnt speak related to others now xD

#

i just came in

pale latch
#

instant-id is fresh off the press. newest model on the block

pale latch
potent spire
#

im faceswapping in Photoshop xD

cobalt comet
cobalt comet
pale latch
potent spire
#

with automate i dont even mean script

#

there is a "play" function in Photoshop that makes you basically save a process and you click play later and it does that process

#

and generative fill in Photoshop helps me with for example edges between the face swapped stuff and the rest of the body or head

#

but sometimes you gotta put the gloves and edit manually

potent spire
#

anyway, i photoshopped my friend into Hitler a few days ago KEK

sullen fossil
#

why do my models never savef