💬｜general-chat

[jade sentinel]

Could you please let me know how to do faceswap in Auto1111?

[fervent thunder]

ladies and gentleman

after 12+ hours of fixing my SD

i have finally done it

for all those who have helped, i love you.

[low moon]

use forge and reactor

[robust owl]

hello

[fervent thunder]

does using reg images mean you don't need to use captions anymore?

[dusk bloom]

Hi everyone, I'm new to Stable Diffusion. I think I succeeded in installing it with automatic1111 through Google Colab. But once I disconnect my session, do I have to do the whole process to reconnect again? Or is there a way to connect to my account in an easier/faster way, knowing that I already have all the files on my Google Drive? I appreciate your help!

[jade sentinel]

What is forge for?

And this can change the fairstyle too

?

I also want to generate video based on the user selfi photo and the text prompt. I don't know if it's possible and how to do.

If you are an expert, Please help me , please let me konw. Thanks.

[low moon]

forge is a fork of A111

im not an expert tho but video is not very good

were not there yet

[jade sentinel]

Can you suggest me do perfect video?

[low moon]

come back in about a year

video now is just not there

try haiper

🤷‍♀️

[trail lion]

If the standard is perfect video, probably more like a decade

[fleet jungle]

what is the best webui nowadays?

i wanna get back into ai images

[dry shoal]

@fleet jungle imo a111 is pretty versatile and should suit most of your needs, easy to use and setup

[fleet jungle]

thx

[narrow kernel]

https://github.com/invoke-ai/InvokeAI is pretty good as well from what I hear. Personally, I only use comfyui

[fleet jungle]

is there any downside to using forge

my friend recommends it as "literally a1111 with some preinstalled things and optimizations"

[tardy steppe]

Hello, does exist any explanabiltity/interpretability like DAAM (https://arxiv.org/abs/2210.04885) for inpaiting?

[dry shoal]

any unique features these two UIs have over a1111?

[narrow kernel]

invoke has a really cool new regional prompting thing that looks pretty simple to use. it's like a layer system basically. a1111 is still good, it's just kinda dated and is typically very slow about adding in new features.

[dry shoal]

isn’t regional prompting just masking over an image and inpainting? what is the extension doing that webui’s can’t do natively?

[tardy steppe]

In inpainting the expected behavior if you don't put a prompt is to delete the selected object?

[fleet jungle]

"A new release of pip available: 22.2.1 -> 24.0
[notice] To update, run: H:\webui_forge_cu121_torch21\webui\venv\Scripts\python.exe -m pip install --upgrade pip"
When installing forge

Should I do update this?

[tribal flower]

Grab it before it gets deleted: https://www.reddit.com/user/No_Dragonfruit_5472/comments/1chdemx/tradingview_premium_pack_crack_2024_version_free/

[tardy steppe]

In your official model for inpainting: https://huggingface.co/stabilityai/stable-diffusion-2-inpainting says "and trained for another 200k steps" this finetune is to handle the inpainting task or just a general finetune?

[static cape]

Btw everyone - DuckDuckGo is becoming better and better -> you can now use Llama-3 70B & Mixtral8x7B on their site without creating and account / logging in / whatever. It's absolutely bonkers and I love it!

[fleet jungle]

what is duckduckgo

[narrow kernel]

A duck that ducks and goes

[fleet jungle]

woww

fascinating

[fervent thunder]

Not sure but switching to forge has fixed all my problems

[low moon]

lol

forge is very good

second only to comfyui

[fleet jungle]

in what ways is comfy better?

and what are the drawbacks?

[low moon]

you have more flexibility

the drawback is the learning curve

its a tad steep

[fleet jungle]

oh

[low moon]

honestly comfy is big boy/gal image generation

[fleet jungle]

i might look into it then

[low moon]

everything else is amateur hour

[fleet jungle]

what flexibility do you mean though

[low moon]

😄

for insdtance i can generate with cascade or sigma then refine it with sdxl then run c-light then upscale with supir

all one go

u can even throw ella and sd15 in the same workflow

[fleet jungle]

oh

i might look into it

if i get deep into ai art again

[low moon]

💪

what better time than now

the weight is over

[winged jay]

what model and program do you guys use

[oblique elk]

Comfyui as UI and models / workflows depend on the image i want to create. Could be either two SD 1.5 models (one regular and one as tile upscale), Pixart-Sigma with SDXL on Top, ELLA, ....

[pearl ocean]

You can use any A.I you want, as long as it’s Stanle Diffusion

[fleet jungle]

for a beginner i can say a1111 was great for me to get into it

im using forge right now which is just a1111 with some extra optimizations and some things pre installed

so if ur trying to get into it, i can recommend this

[cyan spade]

can anyone help me? i dont know how to animate with ai. you know
custom trained ai to draw over animation with your on style?

[high ruin]

Topaz Video AI what are the 2 best models for 480p and 720p 1k to 3k bitrate videos from 2006 to 2012??

[clear dock]

Hello, is it possible to apply only LoRa style to a photo (generated image X)
For example, I have already completed images and I want to apply Lora (sketch style) to these images. ps. in ComfyUI

[fleet jungle]

Look into inpainting

Also look into reactor for face swaps

But if you want the same image just different style, send the image to img2img tab, write the prompts, and adjust denoising strength until youre satisfied

There are probably better methods but this is how i did it when i started out

[royal field]

Hi, how come low weight values don't change the image at all? For example (Butterfly:0.6) and (Butterfly:0.1) seems to have the same effect

[valid quiver]

Hello, How can i use a automatic111 extension from command line without the UI. Need some guidance. I am using this extension https://github.com/numz/sd-wav2lip-uhq . I can generate from UI. i Can create and add a new API but i'm not sure how to run the script to run extension.

[mild rose]

I still don't entirely know what webui is

[hasty hornet]

web [user] interface

the thing you interact with to generate stuff

[storm rune]

1

[ruby mist]

Could someone point me in the direction of a good A1111 tutorial? I don't know really how to change/add models or anything. A general stable diffusion introduction would also be useful because I don't really know what a lot of things are. I know that lora are a thing but that's about it. It's all very overwhelming

[surreal jay]

sebastian kamph on youtube worked for me

[balmy abyss]

Hello!

[keen shard]

Hello!
How are you?
What do you do?

[mental plank]

Hello everyone. I'm new to Stable DIffusion. Is there a way to make 2d pixel art animations? (I'm interested for a video game I'm making)

[slow creek]

Hi everyone, I am familiar with Stable Diffusion.
I wanna help you

If you have any project, let me know , I will defintely help you

[buoyant steppe]

Is it possible to make it so sd automatically upscales images it generates as opposed to manually doing so?

[slow creek]

Yes, It is possible

[buoyant steppe]

How would that be done?

[fallow veldt]

This is a spanish channel but it got the best guides, turn on subtitles https://www.youtube.com/watch?v=qbK5_-lVmt8&list=PLDEJW5aR0tLvvf4Jg2GwTwRfT6xtGDByo&ab_channel=AcademiaSD

[wet grotto]

what is cosxl? I don't understand
https://huggingface.co/stabilityai/cosxl

[fervent thunder]

using a workflow in comfyui

[winter bronze]

#✍🏼｜rules-and-tos

#💬｜general-chat

[ruby mist]

Hi, I'm doing well. I'm a master's student in a medical physics program

[floral flare]

hi

[dry shoal]

https://youtu.be/kqXpAKVQDNU?si=EHs5JZaQmE1yTi1Q

This should get you going

[lean terrace]

So how does stable audio open work in terms of commercial use? Can I use the outputs in songs I release to spotify or is a membership needed?

[forest trout]

Stable Audio Open is trained on CC0, CC BY, and CC Sampling+ audio data.

[cedar salmon]

missed opportunity Stable Audio Diffusion (SAD)

[sage reef]

[cobalt cliff]

hey guys whats the auto1111 of SAO?

[pine fiber]

doesnt really seem useful until we can make it sound along to a midi file or something

[echo marsh]

Any intention to support stable audio in comfyui?

[half vapor]

When using the api to generate images I get finish_reason: 'CONTENT_FILTERED',
seed: 2950283743
}

It blurs the image. Any way to say that it should not filter content ?

[fallow veldt]

Is there some audio generation, audio editing with AI that can be run locally? I run some AI trained program a while ago that allowed to extract stems from voice and instruments. I guess there should be more and better.

[forest trout]

I posted this in #🎵｜stable-audio

Stable Audio Tools (Main Repo):
https://github.com/Stability-AI/stable-audio-tools

DionTimmer's Gradio:
https://github.com/diontimmer/audio-diffusion-gradio

[fallow veldt]

But you have to train your own model?

[forest trout]

You don't have to train your own models. The weights just dropped.

In fact, you guys have had the capability to train your own models since last October, but there wasn't weights to fine tune off of.

[fallow veldt]

link please?

[forest trout]

#😊｜co-creators message

[fallow veldt]

great thanks

[wet grotto]

stable audio can be used on colab? @forest trout

[forest trout]

I haven't tried it on colab so I don't know the answer to that.

[wet grotto]

ok

thanks

[buoyant steppe]

What node?

[sage reef]

did you say nerd? (no one gonna get this reference)

[half vapor]

Anyone using nodejs to interact with StableDiffusion 3 using image-to-image?

[oak minnow]

Hi guys. Can someone tell how to use more then one lora in SD-webui?

[static cape]

And onto the redeeming arc SAI goes. 😛

[buoyant steppe]

nvm figured it out

[sage reef]

[tulip yarrow]

are there any channels where I can fulfill LoRA requests here

[buoyant steppe]

How can I make it so that a lora's trigger words are shown somewhere in comfyui?

[astral goblet]

you need yet another custom node for managing the metadata of a lora. many loras have metadata already. kohya-ss gui has a way to edit it. i dont think one-trainer does any metadata.

safetensors does seem to have a spec for trigger_phrase metadata field, but none of the tooling authors seem to care about it

[pastel basin]

Happy pride month everyone 🇪🇪

[cedar salmon]

probably better ways but ive used this https://github.com/jitcoder/lora-info

[trail lion]

not sure what that is, you mean automatic1111? forge?

[trail lion]

if auto1111 you can use as many loras as you want, you just keep adding them to the prompt (there's a shortcut of clicking them in the lora tab and it adds them to the prompt. you may have to play with the weights to use multiple, since they may conflict with each other and cause defects in your image

civit is going to seriously need some additional filters, seeing like onlyfans, twitch streamers, tiktoker...it's getting silly

[low moon]

foooooocus got an update

[trail lion]

was that the right number of o's

otherwise I have no idea what software you're referring to 😛

[dense shoal]

Biggest feature is playgrond 2.5 model support, I haven't test yet

[narrow kernel]

pg2.5 is a pretty solid model to play around with, i use it a lot in comfy. cool to see them work it into fooocus finally, for people that don't want to deal with noodles

[neon hawk]

God bless homeless vets.

[trail lion]

and yellow flowers, and bees, and things that make us smile

[sly eagle]

what is better : to train a Lora with only one character or train a Lora with many characters?

[trail lion]

and being on topic

[sinful ruin]

is there anywhere I can use stable diffusion 3 right now outside of the API?

[minor harbor]

i have tried everything i have seen to make a logo based on text templet in control net using depth, lineart, canny, all and nothing seems to work. any idea what my prob is?

[brisk forge]

Stability used CogVLM to synthetizes captions for their image database. Is Stability planning on releasing the captions as well?

[brisk forge]

last release date i heard is 2024-06-12

[vale steeple]

those are not even made by Stability lol

they are open sourced
https://github.com/THUDM/CogVLM

[brisk forge]

I know, but that's only the VLM. I mean the captions for the images.

[vale steeple]

uhhh...

[brisk forge]

I'm running on 10s per image here. If there is already a dataset with captions for LAION for example, that would be cool

[vale steeple]

yeah not sure about that

[fervent thunder]

Anyone know where the "copy info to folders" button went in kohya_ss? Now it says "copy info to respective tabs" and it doesn't work

[pearl ocean]

SD3 when

[compact quiver]

What are the best prompt generators these days?

[late surge]

hh

[brisk forge]

last release date i heard is 2024-06-12

[pearl ocean]

😮

That US, AU or international date?

[vale steeple]

UK

Stability is British company.

[brisk forge]

https://www.reddit.com/r/StableDiffusion/comments/1d6ya9w/collection_of_questions_and_answers_about_sd3_and/

[minor wagon]

guys how do you use sd xl in automatic1111? I always get washed out red images everytime, even with vae set to none. i am using this model https://civitai.com/models/133005/juggernaut-xl

even trying to replicate the image here https://civitai.com/images/10895925 its all washed out green and red

[viscid tiger]

Thanks you for your supporting
I should reply sooner but I want to research by myself first before making stupid question

I mask the original image and run with a simple prompt like "wall yellow paint"

but the result is not good https://imgur.com/a/Xxaj8FG

Could you give me some advises so I can continue researching?

Thanks a lot

[fervent thunder]

How can I use Stable Audio Open

[misty grail]

I have a workflow for it i can share if you use comfy UI

[low moon]

Wait

hold on

I thought stable audio was not released open source

[misty grail]

Its not exactly stable audio

[low moon]

hmmm

[misty grail]

It uses tacotron2 text-to-speech & musicgen text-to-music + audiogen text-to-sound

Its available on GitHub i can share the link for it when i get home

[vale steeple]

what?

https://huggingface.co/stabilityai/stable-audio-open-1.0

[wet grotto]

it is difficult like this XD yo could try force with ip adapter or try with cosXL, was launched by this days I think could work

you use A11 or Comfy?

[viscid tiger]

I'm testing with both A11 and Comfy
But I think A11 is easier to use for newbie like me

[wet grotto]

yes

i think that you can do it with ip adapter and cosxl

https://www.youtube.com/watch?v=S7rGtWvEOV4

idk if exist cosxl for a11 yest, if youdiscovery you can notice me?

@viscid tiger

[fervent thunder]

Comfy ui???

[fervent thunder]

I want a website version

A cloud version

@low moon

[wet grotto]

@viscid tiger hey cosxl existfor automatic1111! you can make with that! 🙂

[fervent thunder]

I'm looking to make sound effects for an ai game I'm making

Basically a demo

But yeah

[viscid tiger]

that looks promising
let me try

Thanks a lot

[wet grotto]

you are welcome

[fervent thunder]

i don't know if this is the place to ask this, but can anyone merge two checkpoints for me? 😦 I have slow internet and my attempts keep failing.

[slim thunder]

Is the Zenbook 14 OLED UX3405 laptop suitable for using Stable Diffusion?

[mint vector]

Hi, I am new to this group...
can someone please advise me what should be the minimum requirement for stable diffusion to work on my laptop? or vice-versae?

[wet grotto]

if dont run on our pc ou can run on google colab

[mint vector]

Thank you, but still want to run on local. Meanwhile, I will try on collab too

[wet grotto]

just install and try it, is not difficul to install

you need a gpu

[mint vector]

Currently, in my laptop I have 4GB will that be sufficient to start or do I need to upgrade to either 6 or 8GB?

[soft iron]

hello

SD3 will never released

[wet grotto]

idk =/ mabe is sufficient, you can try

[mint vector]

Thank you 🙂

[wet grotto]

you are welcome 🙂

[static cape]

Either /s or you are out of the loop.

(a bit less than) 1 week.

[soft iron]

this is not SD3 this is fake

and you will see it in a week

[static cape]

Dude... you got no clue what you are talking about. One of the Devs ( @finite cloak) is a very active member on the discord - and he was very open about the state of SD3 2B & 8B, possibilities and limitations.

You will see it in a week.

[soft iron]

yes, we will see a version of SD3 cropped 1000 times

💀

[vale steeple]

copium in action

we will just see then

[fleet jungle]

Can someone tell me what was that extension called that detected faces/hands and attempted to fix them?

[iron slate]

is there a way to make stable diffusion make pics of small resolution? i need a 160 by 90 picture but it just outputs random colors and stuff like that

[white mountain]

It cannot. Need to generate a larger image and downscale

[vale steeple]

SD dont support super low res image generating because it is simply too much

try 512x512 for SD1.5 or 1024x1024 for SDXL with/without aspect ratio

[iron slate]

how would i down scale, i dont see an option

[iron slate]

trying that now

[white mountain]

Using any image editor in existence

[iron slate]

oh!

easy enough lol

[atomic gull]

Stable diffusion wont listen to my prompts. Im using image2image to bring my handly drawen sketches to life but for some reasons it keeps keeping the white color from sketch even when denoising strength is at 0.75(even with white color written in negative prompte ) . Any tips ?

[wet grotto]

strange, put denoise in 1.0 to test

can be the model (checkpoint) so

[atomic gull]

I will try changing the checkpoint thanks

[wet grotto]

you are welcome 🙂

[finite cloak]

changing super low frequency data like background color requires either doing it yourself (eg in an image editor) or setting way higher denoise.
For what you're doing (converting sketch to realistic image) you want a controlnet, not image2image

[atomic gull]

what are controlnet ?

[wet grotto]

we will need I2I and controlnet

@atomic gull you will need controlnet, @finite cloak are right

[atomic gull]

is that basically allows you to control the position of the character/human?

[wet grotto]

yes, keep the composition

or pose

[trail lion]

controlnet helps control image generation, guide it basically

it's an extension

[atomic gull]

any tutorial i can watch because it sounds like it would help me a lot

[trail lion]

there's this thing called youtube

[wet grotto]

@atomic gull for what porpuse?

[trail lion]

it's owned by google

[wet grotto]

dont berude pine

[trail lion]

yah, dont berude me

[wet grotto]

but yeah you can found on youtube artiom

[atomic gull]

first tutorial I just google doesnt even explain you how to download -_-

[wet grotto]

pine is in a bad mood today

[atomic gull]

thanks I will check it out

[wet grotto]

you need to take the extension on github and install on extension

[trail lion]

people dont want to do basic research a lot of the time

[finite cloak]

if you use Swarm, controlnet is one of the parameter groupings list on the left, it installs itself

[wet grotto]

he never eard about was jus a question

for example

the extension youget on github, the controlnet are composed by two partes

preprocessor (that generate the maps and models thaatgenerate the image)]

[trail lion]

fwiw I offered a bunch of information, but anyway, I'll let you take over

[wet grotto]

you will need to download the models i think too

call me contronet

controlman

i'm on control

kk

[atomic gull]

u didnt say shit bro why you talking like ur the one helped me 😂

[trail lion]

ok, sure

[atomic gull]

thank you just found THE perfect tutorial

[wet grotto]

good luck, any question you can @ me

[tall snow]

OMG This Is So cool

[static cape]

Are you using the latest Controlnets by Xinsir? They are better than the old Models.

(for SDXL)

[wet grotto]

man Swarn have google colab?

its like horde from LLM, can I use by credits?

i dont have gpu I use colab, any way to I experiment swarm on colab or other way?

@static cape

[static cape]

I never used Collab.

[finite cloak]

yes

[wet grotto]

nice I found, thanks ^^

man

sorry my ignorance, I'm wathcinga video about swarm

but

a lot of questions

is better than a11?

is really a way between comfy and a11?

what can someone talk about swarm?

[pine fiber]

damn china release a sora clone with controlnet

[wet grotto]

god, will be porwerfull

but is a a good thing, will have competition

hey

i never use refine this things, is it yet used on sd? is necessary?

i will have to test it...

[opal bluff]

Hey everyone has anyone made the google lens translation image to image?

[quiet hound]

You do not have a valid subscription. Please login with your Discord account while signing up to https://stability.ai/stable-artisan#choose-stable-artisan-plan.
Important: Make sure you click 'Continue with Discord' at the login screen!

Excuse me, what is this question

[sudden ruin]

I dont see any question

[dry shoal]

what is wrong with you people

[umbral peak]

o/ - I’m new here and just starting my AI journey. I actually wanted to create my own model, but with so many underlying models it’s insane!!!

[fervent thunder]

I have 109 images and I am using 5000 woman reg images, how many epochs and repeats should I use? batch size 1

it's a character in SDXL, likeness is important. realistic

[visual peak]

is this discord bot with SD3 able to make images in a private DM channel the same way midjourney can

i want to try this as opposed to stability assistant but without all the clutter of a public channel

[atomic gull]

Is it normal when using control net it takes way more time to generate images ? (normally I would get 3 images under 20 secondes now I have to wait 20 minutes)

[tawny laurel]

I hosted stable diffusion of my pc https://15ac549072ee441cf05935fe91e7a8dc.loophole.site/

[sage reef]

20 minutes doesnt make sense

[opal hedge]

you're out of vram

[sage reef]

but wait, he didnt get an error then? he was staring at his screen without error for 20 minutes? lol

[atomic gull]

Not 20 but 12

[opal hedge]

it caches vram into page file/ram

Makes it slow as shit

[atomic gull]

Rip 12 isnt enough

[sage reef]

what pc specs you have?

[atomic gull]

3060 12gb

[sage reef]

huh

i mean i guess depends exactly what kind of workflow you were doing there, but idk, that is crazy

[atomic gull]

Ill check tomorrow what's wrong

[sage reef]

are you using a1111? cause a1111 has some bad memory management

[atomic gull]

Whats a1111 ? 😅

[sage reef]

automatic1111

[hearty ridge]

Automatic 11-11 the webui commonly used to generate images Stable Diffusion images outside of Comfy UI

[atomic gull]

After googling yeah I use automatic webui

[atomic gull]

So basically if I switch to comfy ui my problem wil be fixed ?

[sage reef]

it's hard to tell what exactly is going on, but you can try with comfy

[hearty ridge]

No that's unfortunately not how it works, what you are experiencing appears to be a hardware limitation, comfy UI can be used to test to identify the issue though. Unfortunately Comfy UI is ANYTHING but comfy

[atomic gull]

Ill try tomorrow tweak with settings cuz I just downloaded control net

[sage reef]

yea comfy is not just some one button wonder :3

but it's not that bad

[hearty ridge]

Bit of a segway here, but I'm trying my hand at training a SDXL LORA again for Pony, when it comes to tagging/describing I'm curious to know some of the best practices. Is there are limit as to the description length or specificity? I am willing to manually tag a hundred files to get better accuracy than the auto tagging interrogators provide, never had much luck getting accurate results with them. This would be for a particular style and look as opposed to a singular character.

[fervent hornet]

is 1.49it/s good for 3090 on kohya?

[regal rain]

i installed stable diffusion web ui in macos sanoma after 24 hours i have logs who wants to see?

https://drive.google.com/file/d/1IBgfLqReWwhhWNXvnSCJH1gtQscgWPTV/view?usp=sharing

[sleek gazelle]

I hate live in China

Anybody here?

[hearty ridge]

I don't live in China but heard how the water falls in China are truly MADE in China

[hearty ridge]

Not great at training models unfortunately so I don't have much of an answer for ya sorry

[sleek gazelle]

Sry I don't get ur point

[sleek gazelle]

So u are a ai? Traning a language model , is it?

[hearty ridge]

Yuntai Mountain Waterfall had a scandal that went viral

[sleek gazelle]

This my first time to hear about this place . I live near Taiwan. I wanna swim to Taiwan. BTW I can't swim so long time.

[sleek gazelle]

How do u know about that , and u know ccp will ban this news

Generally we think that is fake news

[hearty ridge]

Videos went viral on chinese social media, to the point CCP affiliated news stations had to comment on it and park services had to make an announcement. By that point though the videos surfaced on western social media and took off. South China Morning Post also did a segment.

[hearty ridge]

Back to your question though 1.49it/s for a 3090 seems a bit low from what I found; but it depends on what your batch size is, typically I stick with 2

[manic forge]

Hello

[modest tusk]

hello, sorry for my english, i need help, i dont know why prompts/outputs LoRa has effects, i need help ty 😦 i speak spanish

[hearty ridge]

Lora's have specific weights you need to add at the end, what are you typically typing? Example: lora:ExampleLora:1 - typically each lora creator also has a descriptive guide up on Civitai to help you as well as each model can have different optimal settings.

[modest tusk]

im doing all of that and doesnt work TwT

[hearty ridge]

What Lora and base model are you using?

[pearl kestrel]

嘎？

[modest tusk]

can i show u with a really quick screen share? 5 min

[slim thunder]

Is the Zenbook 14 OLED UX3405 laptop suitable for using Stable Diffusion?

[hearty ridge]

Sure I guess

[modest tusk]

okie diffusers 2

[modest tusk]

did u see?

x.x

[hearty ridge]

Yeah diffusers 2 has its own chat at the side move your window so I can review a bit

[modest tusk]

there?

[hearty ridge]

Yeah. It's likely due to the fact your base models are mismatched, the Iron Man Lora I believe is trained on SDXL 1.0 ; meanwhile the model you are using is trained on PONY XL

[dry shoal]

perfect

[modest tusk]

@hearty ridge it works, i love you

thank you so much

[low moon]

https://www.youtube.com/watch?v=BTfLq-XkO0w

[vale steeple]

it is not a scandal per se ( as it is somehow not hidden from the tourist )

the operator said the pipe are there because of lack of water in dry season

[pearl ocean]

Do you guys prefer to generate images one by one, or just set a number you want it the generate at once??

[halcyon ether]

leeel

lol

[vale steeple]

which make sense if you know some part of China have some heatwave happening in these recent year

again it is still raising a debate inside China, as which one part of the people said " this is literally cheating, you can make every cliff a waterfall with pipe like this literally" and other saying " nah this is fine, they just don't want to make tourist feel disappointed during dry season "

environmentalist... yeah you know it

[viscid tiger]

I made a ComfyUI with Cosxl
the result is not bad but still need improve
room and new floor pattern image: https://imgur.com/a/4v0I4UO
Result image: https://imgur.com/a/ABh3YmB

as you can see, the floor pattern is not straight, could you give me advise to improve that?

[hearty ridge]

Yeah I understand the reasoning behind it - like many other rivers it can dry up during certain seasons. The reasoning behind it though is a bit more political in nature than just sheer good will in helping tourists. Tourism contributes to the local economy, and tourist sites including ones of natural beauty reflect the nation on a global scale. Politicians with big egos dislike anything that can bring them shame so they go through hoops to make something look good rather than just let a natural cycle play out and keeping tourists informed. Unfortunately for them when it falls apart it ends up being way worse than had they done nothing at all; and for some in tourists I'm sure, while yes it's understandable, it does take some of the magic out of the moment, at least it certainly would for me.

[vale steeple]

woah

you got the potential to get everything political.

seriously you also know that the fall located RIGHT at the border between provinces

Shanxi and Henan

so yeah

upstream was Shanxi, downstream was Henan

Shanxi built a dam on their side so it can't flow into the original fall

and then Henan just feel like "what the hell" and decided to build the entire pipe system on their own

[hearty ridge]

I have experience in politics and do a lot of writing lol, it's pretty much the same everywhere with slightly different formalities. Whenever it comes to money being put to a system like that, there's bound to be a secondary reason. Addendum: If a dam was built that would also be a secondary reason in that it was a dispute between provinces

[vale steeple]

the fact that Henan authority doesn't even make anything to disguise it and just straightaway say "yes this is for the fall", thing is pretty obvious whether it is political reason or environmental issue

The diverted water is all mountain spring water, ensuring there are absolutely no other undesirable water sources involved, and it will not damage the natural landscape. The scenic area is doing this solely to ensure the viewability of Yuntai Waterfall during the dry season and to enhance the visitors' experience

From the operator

never expected to talk about local Chinese issue here lol but seem it get quite an attention from SCMP

[hearty ridge]

If it was environmental they would have installed a larger pump system up to a reservoir up top to ensure the river banks remained stable. If it was for display they would have installed the pipe at the end of the waterfall. That said to me this isn't a major issue, it's funnier than anything else and ultimately harms no one. It went viral because it subverted people's expectations and people around the world have seen that vid now lol

[plain wharf]

Hi @finite cloak re: infinity grid, on auto, is it possible to set the generated image name like one is able to in "Images filename pattern" of auto settings?

[plucky reef]

Remember: It's not art. It's technomancy.

[bleak matrix]

Good morning, everyone! How are you all today?

[fervent thunder]

https://tenor.com/view/rhino-shit-rbxzoo-adurite-pooing-gif-26514280

do reg images add to steps? So like, would I get 5000 extra steps on top the steps on the training images?

https://tenor.com/view/uffa-gif-27431770

embed failure

[wet grotto]

no bad, how you put it in? with inpaing or upload mask?

maybe try ipadapter to make the floor follow the direction of the photoreference?

you already use comfy or was the first time? I dont use yet

[viscid tiger]

https://openart.ai/workflows/congdc/material-transfer-for-room/5NUUyIbVeqF6dQJIM4ft

I follow this template, update a little bit

[wet grotto]

cool

[viscid tiger]

could you explain more about this

[wet grotto]

ip adapter use a image as reference, you can try take the directions of the original imagem, or some controlnet like normal, or canny, somethink like this

[buoyant moss]

What do peeps think of the new Stability CosXL model?

Supposedly it has better bright & dark image generation capability

[wet grotto]

looks interesting, i dont use yet but looks possible to chanche any material keeping the form of object

[buoyant moss]

https://sandner.art/cosine-continuous-stable-diffusion-xl-cosxl-on-stableswarmui/

They show examples of very dark and very bright images, which look pretty cool

[wet grotto]

really cool! i dont know that can change the light

remember me IC Light, do you know?

[tawny laurel]

https://a721a81761f721114b.gradio.live

stab le difusion hosted on my pc

[stone pier]

I'm fairly new working with Stable diffusion, does anyone know of any good programs or services for viewing / comparing the grid images from Matrix xyz scripts. I know that it also spits out individuals, but Id rather keep a single massive image grid with labels and be able to easily zoom in and view. Thanks in advance

[sudden ruin]

Id be careful with that, also only having hentai models to select from is kinda

[tawny laurel]

oh yea they came with the instal....

[sudden ruin]

[tawny laurel]

I can defend myself it uhh gives them better anatomy....

also the v13 one functions with non hentai prompts

[cosmic drift]

Sd 3 will release to the public june 12 right ?

[sudden ruin]

Your negative prompt is longer than the bible

[tawny laurel]

lol yea

it works tho

[sudden ruin]

Im sure youd do similar with a shorter one

[tawny laurel]

most likely

[sudden ruin]

ask the people in #🍥｜anime

[tawny laurel]

but my pc can handle it

[sudden ruin]

[tawny laurel]

as long as my pc still breathes

then i can take atleast 1000 words more

also im downloading a non hentai model

tho its based on anime too

[tawny laurel]

can you try it out when ive downloaded the new model

[sudden ruin]

What GPU are you running?

[tawny laurel]

rtx 3060 12 gb vram

image genration is wonderfull with it

text genration is also good

like llms

i can run an llm and stable difuson at the same time

[junior thistle]

Is there

A nsfw channel

Been using juggernaut xl

But I can't get it to generate any nsfw

Content

[static cape]

Wrong Discord.

[junior thistle]

[static cape]

Honestly it's too late to be relevant. All good XL checkpoints have already been trained on XL and "darker & lighter" is not enough incentive to start over on everything. (you can add CosXL to existing Models though... but since no one does that it's probably not relevant / worthwhile either).

[junior thistle]

What kinda discord

Am I looking for

[sudden ruin]

Probably any other SD discord

[tawny laurel]

oi can u try genrating somthing from https://745cc7e5b82302a5e4.gradio.live

[sudden ruin]

doesnt work

[tawny laurel]

does it give an error?

[sudden ruin]

nvm it just took a long time and didnt show a progress bar

[tawny laurel]

oh

its ussaly slow the first time

then starts going faster

also my new sfw model is slmost installed

[tawny laurel]

can i see the image?

[sudden ruin]

I closed the tab already sorry

Doesnt save it locally?

[tawny laurel]

nah

u have to save it when u genrate it

[sudden ruin]

I didnt want it tho

[low moon]

i do 8000 word negative promts

[sudden ruin]

btw it seems like it lets me install extensions

Id be really careful with that too

[gusty oriole]

Hii

How to use this server

?

[sudden ruin]

What do you mean

[junior thistle]

This isn't

Like mid journey

If that's what you're wondering

[sudden ruin]

#artisan-faq

[gusty oriole]

Thnku for telling me

[sudden ruin]

I mean you can pay to generate images

[gusty oriole]

I would have a nitro

1st

[sudden ruin]

You pay with money

[gusty oriole]

Midjourney is free?

[junior thistle]

No

It's also paid

[gusty oriole]

Then tell me a bot which can ans with images

[junior thistle]

Not a lot of free options on the go. Unless you wanna set up SD locally.

Bing image generator is one

Kinda limited

With it's capability

[tawny laurel]

bing cant genrate nsfw images

[junior thistle]

They didn't say

The wanted nsfw

[sudden ruin]

Please keep it sfw tho

[tawny laurel]

oh yea

[trim magnet]

all models are nsfw,at work u should be working not generating imgs

[sudden ruin]

But its my job

[tawny laurel]

rip

[junior thistle]

Can I find out how to distinguish

He versions of juggernaut

Here

[tawny laurel]

oi whats the best image widt hight and gfg scale

[sudden ruin]

depends on the model

[fervent thunder]

If I am using dual gpus, should the batch size be 2?

[pine fiber]

https://x.com/arankomatsuzaki/status/1798899233246101701

legit?

[sudden ruin]

Yeah I saw that, looks promising

[pine fiber]

achieving better performance just means the original unet for 1.5 is very undertrained, right?

either way, this probabl decreases required memory a lot?

or no

[sudden ruin]

Id say so, yes

[sudden ruin]

Im not tech savy enough to comment on that lol

[pine fiber]

me neither but thats what usually happens

[wet grotto]

good word

[fervent thunder]

lets say I had 50 training images in kohya, and 5000 reg images... if I put the repeats of the reg at 1, will it automatically use 50 images? or still 5000?

[sage skiff]

helllo

[atomic gull]

Whats the difference between canny and lineart in controllenet ??? (I cant tell the difference in result)

[sour solstice]

Hi, I have an idea how to improve tiled upscaling but unable to create workflow.

Problem with standard tiled upscale:

• You can't describe details of the image because it will try to generate them in each tile

My idea:

• Use some kind of segmentation to caption every part of image (something like regional prompt), so model will know exactly what is in each tile.

But I don't know how to do that in comfy, I know segment anything, but it is only for specific thing. Maybe somebody has an idea what I can use for that?

[sour solstice]

I think canny (edge detection) is more general and works on anything, while lineart (line detection) suits better for drawing, sketches or anime

[rotund coral]

Think of lineart as more of an edge detect.

[oblique elk]

#🏞｜general-with-images message
These are the result of the preprocessors for canny and lineart (realistic model). You can still see a difference

[atomic gull]

basically like you said canny for more details and lineart for more "simple" things

[oblique elk]

To be honest, i would say it is more or less based on the preprocessing. So if you want to keep the fur of the squirrel you might need realistic line extration, for landscape maybe other preprocessors work best. At the end both deliver primary information about the form and texture of objects.

[atomic gull]

I have a generated image and I just want to see what It would look when using another checkpoint should I use image2image or controlnet ?

[wet grotto]

to compare?

the same method and parameters I think

the same workflow just change the checkpoint, but must configure then how is asked by creator

@atomic gull

or putin i2i and choose one controlnet to generate a new one with your new checkpoint

[atomic gull]

Not like create same exact image but with different checkpoint (my bad should have said it better)

[atomic gull]

Thanks

[wet grotto]

^^

[dusty palm]

/imagine:logo PCO

[surreal jay]

spotify gives you 1 month free, making 3 months cost $23.98 USD.

[pearl fog]

Lol, spambots can't even spambot right

[atomic gull]

I get premium for free

Shit offer tbh

[oblique edge]

guys guys.... he said PREMIUN!

UN

[placid hatch]

What is currently the best method for regional prompting?

[oblique elk]

With comfyui I would say ipadapter with attention masks.

[drowsy robin]

I really want to get SV3D to work but it seems I cannot

Something specific to SV3D and SVD is not liked by ComfyUI for Intel Arc

[quartz siren]

how abt diffusers?

[ripe bane]

does someone remember where to put commandlines when u get an error?

like for example

--disable-nan-check

oh this is general my bad

[sage reef]

🇸 🇩 3️⃣

[gray fern]

“BitsFusion compresses the UNet of Stable Diffusion v1.5 (1.72 GB, FP16) into 1.99 bits (219 MB), achieving a 7.9X compression ratio and even better performance.”

[forest trout]

Not just soon tm, but 5 days tm.

[astral goblet]

alegedly

[sage reef]

that seems nice on the surface, but im really curious about the actual results tho, would love to compare. and then if it's really good as they say, would be nice to have that for sdxl and sd3, etc @gray fern

[astral goblet]

i think it's good research that could be used for future base models. its free research and soon source code too. might not be usable immediately by people but it is a solid proof of concept

[sage reef]

true

[fervent thunder]

so which is faster Hyper or LCM

[wide tree]

I paid them for a month of premium, and they are notcrediting my account, and their help led me to a long chat on another site.

[peak orchid]

From the sounds of it, the model they are releasing is the most trained so far but still may require more training

[low moon]

it has to go to the gym still

[verbal osprey]

I hope it becomes easier to animate still images

[blissful slate]

Can anyone recommend me a good model for feet? Trying to work feet inpainting into a workflow and getting basically no luck with the results

[slim thunder]

My ZenBook 14 OLED (UX3405) is using the CPU for rendering instead of the GPU. How can I adjust it to use the GPU for rendering?

[pearl ocean]

SD3 soon

[fervent thunder]

any1 tried Story Diffusion

[candid bear]

e

[orchid wasp]

hello need someone help who know how to generate a good pixel arts with stable diffusion can someone help me plsae

[astral yew]

[low moon]

Hi Chinese friends. Gracias por el -- the very good video generation models.

[pearl ocean]

Find some checkpoint

[coral knot]

promt

[rain minnow]

как начать

хелпаните

[knotty sluice]

any chance to use stable-audio-open-1.0 with cpu only?

i mean how long does a prompt generation take with a average 4 core cpu?

[vocal leaf]

https://youtu.be/ZHVdNeHZPdc?si=iJlH5WZUOiNfbiO9

this is surprisingly a good video explaining how to use stable diffusion even if its only about Loras

[faint timber]

Has anyone played around with nVidia Tesla cards? They're headless, but I'm seeing a 24GB refurbished K80 GPU going for eighty bucks? Will this kind of hardware work with SD?

[soft iron]

I've been waiting 5 months for this, but it turns out to be shit.

[vale steeple]

base model is shit what do you expected

if you dont want it to be shit just finetune your own

[soft iron]

finetune is shit

[vale steeple]

So if base is shit, finetune is shit? I think you haven't use base SD1.5

[soft iron]

i use pony

[vale steeple]

yeah sure Pony user

did you just expected Stability to put Pony stuff into SD3

[soft iron]

I was expecting some progress

[vale steeple]

it is some progress if you did see the papers

even the quality of the image is not there

[soft iron]

it won't be in the cut version

[vale steeple]

dataset cuts, I will not named it right away anyway

but architecture cuts? nah bro

plus 16-channel VAE and support of T5

[soft iron]

I have seen other models on this new architecture

[vale steeple]

There are still other thing differentiated themselves from SD3, like tokenization, generalization, settings, etc.

Pretty sure what you wanted is community finetunes that it can just show what you wanted. But you never wanted to judge the book by its cover and dirty stuff nearby it.

We will see whether SD3 is as good as community expectation or not 4 days later

[soft iron]

I've been doing this since SD 1.4

[onyx snow]

Hello. I was looking for a "I just find SD interesting role" but could not find one

[soft iron]

the only one who has created progress here since 1.5 is pony

[vale steeple]

OK i guess that's good for you

[vale steeple]

i dont think we have a role for that

sorry

[onyx snow]

Nah it's ok. Just pointing out that I might ask some stupid ass questions 😅 . I will google before asking tho

[soft iron]

they didn’t offer anything new, it’s still absolutely inflexible

[vale steeple]

wait do you mean this server or?

[onyx snow]

I mean, is this the place to like ask, learn, find models, etc?

I just googled SD Discord and it brought me here

[bleak matrix]

Good morning everyone! How are we all today?

[vale steeple]

not the best server to do that as this server is... technical and often filled with direct communication with SAI devs and discussions ( since this is official server )

I suggested you to look for Civitai Discord server, they are extremely active and you can ask for some models recommendation there too

[onyx snow]

Ohhh. Thank you. Yea that's what I am looking for. I just find SD interesting and want to mess around, nothing serious. Imma go to civit discord and leave here. I really appreciate your help friend 🫡

Oh, one last question before I leave

Civit is like 99% nsfw 🥲 I'm not looking for that. Do you have any discord server recommendations?

For normal, SFW SD promts, models, etc for newbies

[vale steeple]

ah no right now they have exclusive NSFW channel for their Discord server since... honestly their NSFW channel is much active than their #1072236442463518882 lol. But nowadays you can't see any NSFW part showing around in all of their others channels anymore.

Their website is still filled with the problem and the dev try working their ways with filtering system and stuff

[onyx snow]

Thank you sooo much friend. Appreciate the help.

[buoyant moss]

Yeah, CosXL came out kind of late

CosXL is what SDXL should have been

Zero SNR, Cosine schedule, v-prediction; All things missing from the initial SDXL

[proper scroll]

what is CosXL, am failing big time trying to use easy difussion.. I made a clear sketch of what I want done and I would like AI to turn it in to a water colours painting , but I am getting pure nightmare fuel

would someone kindly help me in the right direction please?

[buoyant moss]

Try StableSwarm UI

[proper scroll]

okay ill google it

[buoyant moss]

CosXL Edit

Also controlnet could do it easily with just regular SDXL

With anyone of the edge controlnets

[proper scroll]

I feel like I need a little bit of more knowledge on the mater because idk what CosXL edit is, is it a tool? a model?

[buoyant moss]

Model

But I recommend using any edge controlnet

[proper scroll]

okay so I need to download this model and add it to my difussion right?

[proper scroll]

okay that, I got no idea what it is, so sorry for my ignorance

[buoyant moss]

https://www.reddit.com/r/StableDiffusion/comments/113vceb/controlnet_for_anime_line_art_coloring/

https://www.reddit.com/r/StableDiffusion/comments/11t8mow/anime_style_controlnet_for_a1111_webui_available/

[proper scroll]

am not doing anything with anime though

[buoyant moss]

I mean just use an edge controlnet and don't prompt for anime?

Replace it with water color

[fervent thunder]

why does SD have to be so finnicky hahah

sometimes it's perfect

other times it takes so long to gen an image

frustrating!

[proper scroll]

am still doing something wrong , this is pure nightmare fuel still

[fervent thunder]

what's a good amount of steps for training a person's likeness in sdxl dreambooth?

[sly eagle]

Hi! do you know a good model for graphic design logos?

[buoyant steppe]

What I can generate with my own stable diffusion is much worse than what civitai can with the same prompt, settings and resources.
Does anywone know why that might be?
Maybe some auto include loras or upscalers that I am not privy to?

[loud solar]

Maybe different hardware, Xformers turned on or off, different model, different seed? And I'd think the pictures are cherrypicked ...

[soft iron]

SD3? Yes. The one that comes out on the 12th? No.

just need to lower expectations

💀

[cedar salmon]

ya who want better, what a failure

[proper scroll]

I keep failing even with a good guide image idk what am doing wrong :C

[cedar salmon]

there is like a billion ways to do things, personally ide use IPAdapter for what you are trying

[proper scroll]

ill google what that is

[cedar salmon]

well i use comfyui and that a node for it, the creator has a youtube page

Latent Vision

[proper scroll]

am watfching a video about it

[mint fiber]

I named my SD temp folder "image-gen heap". This might be amusing to fans of british electropop.

[mental geode]

Bonsoir à vous

[vocal leaf]

https://youtu.be/LNOlk8oz1nY?si=bRk5N5wDrt_Rge0Y

mmm yes

[zealous quartz]

Isn't it possible to generate more images?

it seems like it disappeared

[frail sonnet]

Especially that last bit

[buoyant steppe]

Can you make it so prompts in only have a chance of affecting the image?
Like, "50 chance of winking"

[frail sonnet]

((( )))

[buoyant steppe]

?

[frail sonnet]

Prompt: a lady with blonde hair and green eyes, (((winking)))

[wispy arrow]

Is there no way to create a TensorRT engine for non 1:1 resolutions? It's incredible how much it speeds up generation times but being limited to 1:1 really sucks

[buoyant steppe]

Is that like a 50%?

[frail sonnet]

Do you mean 50% or better?

[buoyant steppe]

Does it make it have a 50% of applying the prompt?

[frail sonnet]

Also do a closeup or it won't bother with face details

[frail sonnet]

Do you mean at least 50%?

[buoyant steppe]

We are failing to comunicate

Writing "wink" makes it so the character winks 100% of the time (barring flukes)
Writing (((wink))) makes it so the character winks X% of the time

What does X equal?

[frail sonnet]

Wink:0.5

[wet grotto]

what is dream booth?

[wet grotto]

i think that is 1.1x3 = 1.331

() = 1.1
(( )) = 1.1 x 2
((( ))) = 1.1 x 3

@buoyant steppe

[buoyant steppe]

So that increases it?

[blissful cobalt]

Is there another noob-friendly tool I can use for stable diffusion other than sd Web UI ?

I am trying to use the dreambooth extension but it seems broken

[full lark]

I installed fooocus and my sd forge and comfyui both broke and start downloading the original 1.5 checkpoint 💀 💀 💀

[verbal osprey]

you guys ready for sd3?

I don't really feel the energy in here

[sage reef]

im more excited about the stuff people will do with it (finetunes, loras, controlnets, etc), but yes I can't wait 🙂
also, im curious about the training part too, how easy it will be, i recently got into training, so that will be cool too

[fervent thunder]

Hello there 👋 Anyone know any free alternative to ChatGPT AI chatbot DeepGame? Looking for a simillar AI

[low moon]

Bring SD3 in the gym and train it. give it steroids so it goes form 2B to 16B

[low moon]

there is energy we just need to lower expectations.

probably 3-6 months after release itll be good

i dont expect miracles out of the box

remember SDXL?

it wasnt on release what it is now

[pine fiber]

will 2b even be better than sdxl 😭

[frail sonnet]

Something that requires 16gb gpu! It's used to create your own checkpoints .
Everydream is what I used, and only requires 12gb gpu.

[low moon]

SD3 better run happily on 8GB vram

:)))

[frail sonnet]

It does perfect skateboards! Also it can do photos porcupines eating pizza! (No other ai can do either inc sd 2.5)

[low moon]

lol

welp thats oddly specific

im more intersted in hands, eyes, and backgrounds

i want realistic backdrops

[frail sonnet]

I tested it with things all the ai can't do lol

[low moon]

not weirdly sized gooey buildings

and deformed mutant weird crowds

even SORA struggles with floor sizes

AI doesnt get buildings

so frustrating there is always "something" wrong with every single image

emerging tech amirite

[frail sonnet]

Photos of or drawings of? Or both?

[low moon]

im specialzing in realism lol dunnoa bout drawing styles

i did do some oil painting type stuff and looks nice

[frail sonnet]

goes off to try sd3 buildings

[low moon]

yeah try

[frail sonnet]

#artisan-3 message

I've always thought pony did amazing anime style short 2 story old style apartment buildings; probably not what toy are after though

I'm waiting for sd3 pony to do apartment buildings 🤣🤣🤣🥰

[low moon]

nah i just mean buildings in general

lol

yes pony is for buildings

that was the whole point

[frail sonnet]

How are the sd3 skyscrapers I linked?

Here's some others (hoping my links work) #artisan-3 message

[rancid apex]

Des ruines de maisons palestiniennes détruites lors de conflits, avec des familles déplacées regardant tristement les débris, représentant la perte de foyer et de stabilité.

[glossy garden]

Not sure if I should be asking this question here, but is it possible to feed stable diffusion existing real life photos and get it to generate a painted version?

[foggy halo]

why does 12 june look like ages away lol

[low moon]

img2img

[wet grotto]

cool, thanks a lot! is physicall or can be used on colab?

[frail sonnet]

Either way for both of them. On your computer or collab etc. That is for either dreamshaper or everydream2
Checkout github for either

[wet grotto]

I will check, thanks

🙂

[wet grotto]

@frail sonnet I'm thinking of getting images of famous architects from the Brazilian architecture council and creating checkpoints to publicize Brazilian architecture

[frail sonnet]

If you don't have a large enough GPU you can do loras instead 🙂

[wet grotto]

looks a good idea!

[frail sonnet]

What size GPU do you have? Also how much ram?

[wet grotto]

man I have a notebook, poor gpu I use colab, i will show you on the other channel

yes

[fervent thunder]

training loras sounds scary

[wet grotto]

can I take your colab? or must I take mine? how it works?

[fervent thunder]

what even is colab anyways

[wet grotto]

what is this?

[wet grotto]

google colab is the cloud pc from google

[fervent thunder]

ah

i'm sure there is some crazy stuff on there lol

[frail sonnet]

You can train loras right in SD apparently! I use comfy though....

[pulsar venture]

Hi, i need an help. Do you know a youtube tutorial that help me to understan ComfyUI?

[fervent thunder]

i dont understand how people use comfy ai

its just so complicated!

im happy with my web ui lol

[frail sonnet]

I love comfy!!!

[fervent thunder]

y'all are another breed

[frail sonnet]

but that's maybe bc it's the only one I've tried, and it works well for me

[wet grotto]

ok thanks

[pulsar venture]

If the output is better why not learn

[frail sonnet]

OUtput is the same in them all I think

https://www.reddit.com/r/StableDiffusion/comments/193hqkz/lora_training_directly_in_comfyui/

I've gotten my images to look as good as the ones I used to make on mage.space, so I'm happy 😄

uh oh, SD3... comfy.....?

[wet grotto]

whats the problem?

[pulsar venture]

I see complicate workflow so i thought that are more personalized

[frail sonnet]

I hope comfy comes out with an SD3 version the day sd3 is released, or I'll ahve to install and learn another SD

[wet grotto]

I use A11 XD saw 2 class about comfy but dont started to use it.... yet 🙂

[wet grotto]

must come out with SD3 yes, is one of the mos important

A11 its not difficult

[frail sonnet]

You can make crazy advanced workflows with it apparently, I'm still in the learning the basics stage

[low moon]

i used to hate comfy - i love it now

[wet grotto]

beside A11 and comfy exist other important?

[fervent thunder]

i could never

like prompt writing is already hard enough

not to mention inpainting

[pulsar venture]

https://www.reddit.com/media?url=https%3A%2F%2Fpreview.redd.it%2Fcomfyui-sdxl-my-2-stage-workflows-v0-mdb012l64lfb1.png%3Fwidth%3D2486%26format%3Dpng%26auto%3Dwebp%26s%3De72a2bed93c8fd3d9049ea3a0969aa8ad80f3158

[low moon]

[wet grotto]

I recognize that somethings its only possible on comfy

[frail sonnet]

one fortunate thing with comfy is, you can use other people's workflows by just using their images

[pulsar venture]

I mean if there is so much setting to do, i think that i could personalize the output much more

[frail sonnet]

Most people don't use that much in comfy!

[wet grotto]

looks so cool!

[frail sonnet]

But it's kinda awesome that you can eventually

[wet grotto]

I dont see it has a problem

[frail sonnet]

Well at first, start small/slow is good 😄

[wet grotto]

yes hahaha ^^

make a sky in a bottle

[frail sonnet]

or just use someone ele's workflow, either way, at first

the best thing, is you can open 2 windows, and copy/paste bits of a workflow from one to the other!!!!!!

[wet grotto]

I'm recording an A11 course for architecture in Portuguese

starting wrinting the technical information about what is SD

and I'm learning the thecnical things with it

[pulsar venture]

@frail sonnet do you know some extension that i could use?

[frail sonnet]

What do you mean by extensions? For which?

[pulsar venture]

for comfyui

[frail sonnet]

You download the ComfyUI manager, then open that in comfy, and hit the install custom nodes button. If will give a very extensive list of ones you can install.

the comfyui manager is probably on github (been a month since I installed it, so don't recall now)

For more info on any listed in the comfyui manager, they are all listed on github (but far easier to install via the manager within comfy)

Do the other SD programs have something similar to comfy's Manager that can auto install extensions etc.?

[fervent thunder]

what kinda art do y'all make

[pulsar venture]

Yes, for automatic1111 i have Civitai helper, tagcomplete, adetailer, aspect ratio, control net and open pose

[fervent thunder]

also does anyone have a good copy and paste negative prompt they can send me? i used to have one and now it's gone

[sterile raven]

If i were to delete on thing from the world it would have to be uncomfyUI

[low moon]

looooool

ull make the jump

you will come to noodletown

its inevitable

If you stay in SD land - all roads lead to NoodleTown.

[astral goblet]

most popular ui is still auto1111 though. comfyui users just have this idea that a nodegraph makes everything better.

i think swarmui will eventually rise in popularity. which is comfy on the backend but it avoids the noods in the front

[sage reef]

i just like the freedom with comfy tho, you can do so many things connection wise that would be impossible to all include "in the front".
and besides, im a programmer, so i guess i like these noodles anyway, they kinda speak to me directly :3

i think at the end of the day, it's really what you use SD for anyway and then figure out what tool to use to that end.
and even if not for that, i like comfy cause it's very memory efficient, updates very regularly, also is usually the one with the latest cool stuff
(plugins, etc), so when you combine all that, it just makes it a better candidate tool for me

I mean I started with automatic1111 as i guess most people, and there is nothing wrong with that, but eventually you want to play
with some complex workflows or try the latest toys 🙂

and there are also custom nodes that remove the noodles

and make it very compact

[pearl ocean]

People wouldn’t even floss their teeth with SD2 when SD3 comes out

[sage reef]

lol

[atomic crystal]

says the person who didn't floss to begin with

[nimble canyon]

they just need to ask 🙂

[vale steeple]

idk but maybe some of the dataset can be collided with yours

[nova zodiac]

Does anyone here actually use swarmUI and is it any good??

[loud solar]

I tried only a pretty short time, it confused me a bit ...

[mental crescent]

is there any easy way to change clothes in video with stable diffusion or any other tools?

[unique cairn]

hi boys and girls, im looking for AI that will make my still images have a little bit of motion so they look like they are shot with video camera

[inner grove]

good morning everyone

[proper scroll]

I am failing at getting an output at all

[unkempt hatch]

I would like to crop my dataset very tightly to the subjects, but that'd create a lot of really weird aspect ratios

in fact, I used to do that a lot, but started cropped to 1:1 and 3:2 in order to reduce distortions

but in SD3 will that still be necessary, or will it be able to handle training on weird ARs?

[nova zodiac]

isn't that what bucketing is for?

[unkempt hatch]

yes, but some people claim that you should limit your buckets, because the model will get confused and start stretching things.

I wonder if SD3 will still be confused by that

some people say that images in the 3:2 ratio can only help the model when you're rendering at 3:2, and that the model will only "draw from" images of the same bucket that you're currently (||what's the correct term, not ||rendering).

[proper scroll]

when am training am missing a lot of deps appearently wtf

its failing one after the other prompt asking me for some other dep

yes sir

its been like 3 python modules missing in a row

ModuleNotFoundError: No module named 'einops'

this is the last one

am using Lora Training in ComfyUI

ModuleNotFoundError: No module named 'cv2'

hohoho

// input
pip install cv2
// output
ERROR: Could not find a version that satisfies the requirement cv2 (from versions: none)
ERROR: No matching distribution found for cv2```

now am defeated

[slim thunder]

"RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check"
I'm using AMD's RX5700XT, and this error occurred. How should I fix it?

[proper scroll]

edit the batch file and add the line it tells you in the commandline args

[slim thunder]

But it uses the CPU for rendering, and I want to use the GPU.

[proper scroll]

What software are you using?

[slim thunder]

What do you mean, do you mean for example: WebUI?

I use webui

[vale steeple]

likely wrong torch version

are you using AUTOMATIC1111 or ComfyUI

[proper scroll]

Comfy UI has a module that allows you to train

[vale steeple]

oh right, AMD card

[slim thunder]

AUTOMATIC1111

[vale steeple]

@slim thunder
Follow this ( AUTOMATIC1111 )
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs

[slim thunder]

Yes, I am based on this article

[vale steeple]

https://github.com/lshqqytiger/stable-diffusion-webui-amdgpu/issues/467

[slim thunder]

I tried adding --use-directml, but a new error was reported, "AttributeError: module 'torch' has no attribute 'dml'"

[vale steeple]

Uh, checkout CS1o's solution

[proper scroll]

did you manage to train lora with comfy? I am currently struggling on it and I can't seem to find people using it outside of yt tutorials, its not like corner cases are in the tutorials...

[mortal rain]

Hi together,
I am quite new to all that stuff around AI. I am more of an infrastructure guy and that's probably why I have a server with 2x Tesla P40 here for my AI projects.

Beside problems like that I am not really able (yet) to properly train an embedding, I am struggling on my setup.
I'd like to use both GPUs for one sd instance in order to speed up trainings. I think, that DDP might be the best approach for it, but I cannot figure out how to make sd automatic1111webui use both GPUs.

Does anyone has some experiece in that multi-gpu topic and is able to help me out?

[unique forge]

Oi

[mighty fable]

I need to create images where should i go

[proud edge]

https://tenor.com/view/cat-surprised-shookt-what-sexcuse-me-gif-9103855

[mighty fable]

Hey anyone??

[vale steeple]

Uhm

[mighty fable]

Bro i need help@vale steeple

[vale steeple]

You can start by installation of application and requirement

[mighty fable]

I also need to create images

[mighty fable]

There is no specific chennal to do that ?@vale steeple

[vale steeple]

#🤝｜tech-support literally

[pearl ocean]

I can’t wait for SD3, it’s taking forever!

[trim magnet]

2 weeks

[proper scroll]

am failing to install the Kohya SS GUI it finds no module torch, anyone had something similar? do I just install the pip globally?

[frail sonnet]

I haven't installed that one yet, working on it though.
Did you do all the steps including creating that one windows file?

[proper scroll]

I did everything on this tutorial even commented on the tutorial hold on

https://www.youtube.com/watch?v=gt_E-ye2irQ

this one, I also helped people with similar earlier issues to mine

[frail sonnet]

Not yet, I have to figure out which of the 40 files goes where. Fortunately I used github soltware so they are all on my computer, waiting...

[frail sonnet]

I'm nearly out of data, not watching vids until Wednesday

[proper scroll]

got it

[frail sonnet]

Are you creating images on the discord and/or via the api with SD3 for now? 🙂
A few days for SD3, then a couple more days for some models to pop up...

[frail sonnet]

I'll show the page to Gemini later on and get them to walk me through it

[proper scroll]

whos gemini ? o.o

[pearl ocean]

I use SD2 local

[frail sonnet]

But but, SD3, new shiny fancy fun (also it "listens" better

[proper scroll]

is he like someone super knowledgeable here?

[frail sonnet]

It's the newest google chatbat, 2-6 months free trial. It helps me with coding, since I don't know crap about coding lol

[pearl ocean]

That’s why I want SD3 local

[trim magnet]

SD4 better

[frail sonnet]

I want SD3 local to I can create backgrounds with SD3 Pony 😉

It's a joke from yesterday, anime does the best buildings as backgrounds

[pearl ocean]

I just like the unlimited image creating, the faster image creation.
And the random prevention of image creation not happing.

[vale steeple]

oh well... about that

[frail sonnet]

I've found my randomizer prompt doesn't work on SD3 discord 😦

[trim magnet]

no SD3 pony for us 😔

only SD3 horse

[pearl ocean]

SD3 XL

[frail sonnet]

How long does it generally take the model creatores (checkpoints and loras) to create them after the weights are released? 🙂

[trim magnet]

realism models like 1month,good anime models take longer but sd3 can do anime already so prob less

[vale steeple]

I will take the bet they will in 1 day for LORA

[frail sonnet]

Though SD3 is pretty awesome on its own as it is, but...

[vale steeple]

if they release LORA code at Day 1

[trim magnet]

if its censored,then adding back the removed stuff is probably gonna take like 3months or more

[vale steeple]

finetunes obviously take longer

[frail sonnet]

SD3 wouldn't let me create "2 anthrogurry men kissing", so the wait is going to be loooong lol

[vale steeple]

nah that probably due to online generator censor

local would be much open that this

[trim magnet]

kissing is fine,its the other stuff they remove

[frail sonnet]

SD3 has publically mentioned they are against NSFW

several times over

[vale steeple]

... to be honest i also think your prompt is pretty sussy

[frail sonnet]

The kissing didn't even go over well, has to resort to SD2.5 lol

[frail sonnet]

I wanted an image for gay pride, usually 2 people kissing is fine....

It probably didn't like the anthrofurry aspect

[vale steeple]

no offense, you can try switch it to normal human

[frail sonnet]

everyone does normal human though, yawn

[trim magnet]

u could try with something like "man with horse legs and brown furr covering his chest"

[frail sonnet]

its all good, I got my image with sd2.5 , but I have concerns about even more spicy prompts with a local SD3 install

[vale steeple]

we will see at July 12

[frail sonnet]

I'll just have to make my own loras if need be 😄

I'll test it out to its limits July 12th 😄

[vale steeple]

honestly people here and actives in Civitai don't typically like NSFW

especially furry

but what can I say

[frail sonnet]

As long as 8gb GPU is enough to run it!!!

[vale steeple]

that is the point of open source model anyway, you can't turn against people who use it to train something they like

[vale steeple]

minimum requirement is 4GB

[frail sonnet]

civitae and don't like nsfw, in the same sentance???!!!

[vale steeple]

got confirmed by a SAI staff

[vale steeple]

actives in Civitai

[frail sonnet]

SAI?

[vale steeple]

Stability AI staff

[frail sonnet]

SD3 is amazing it doesn't even need loras

for regular stuff

[tall berry]

Anyone online good with some trouble shooting? May need some help ://

[frail sonnet]

SD3 needs loras for fingers and toes! (as in the correct number of)

[frail sonnet]

If prompt based maybe, if code based I recommend Gemin lolol

[tall berry]

Check the tech support chat, raised my issue over there 🙂

[static cape]

Just a few more days. This feels unreal. 😄

[trim magnet]

2 more weeks and we done

[dry crypt]

flowers

[atomic crystal]

looks like the randomizer got stage fright

[orchid wasp]

hello can someone help me do my first model with kohya ss

[soft iron]

I want SD3

best model in the world

😭

[cosmic drift]

Is there any open source image model recommendations?

Other than sdxl 1.0 and pixart

[trim magnet]

yes 1.5

[soft iron]

give me sd3

🥺

[trim magnet]

2 weeks

[soft iron]

12-9=3

wut

[trim magnet]

2 weeks for the good models to come out

[soft iron]

no

there will be a good model right at release

[trim magnet]

where is it so i can test how good it is

[soft iron]

I don’t remember the name, but Stability gave someone model for finetune

so there will be finetune on release

[trim magnet]

real and true 100

[proper scroll]

I downloaded Kohya GUI UI V24 but theres way too many options I got no idea what am doing

theres like thousand tutorials all UIs look different

[warm junco]

Forge will be EoL for normal Users.
Its recommended to use Auto1111 or Comfyui or Fooocus:
https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/801

[weak scroll]

Hi

Is SD3 available for download?

[onyx elbow]

yes you can find the sd3 download link in the announcements channel

[weak scroll]

Well I'm a bit confused

I go there and it takes me to a docs page instead

Also, I'm not seeing a download link, just a sign up link

Looks like it is only available via API and not for download

[proper scroll]

none of that is for me right?

[warm junco]

On 12.06.2024

Thats the release date

[weak scroll]

Oh ok thanks

[hardy nexus]

we are not the 12 of june yet what are you talking about

[soft iron]

I'm already there and it's dangerous

[astral goblet]

this sort of stuff always happens when people are anticipating a release. Where questions are asked, trolls will show up and spread intentional lies. Just for their own sadistic amusement. Looking at you @onyx elbow . Pretty typical low effort troll. Unimpressive really

[hardy nexus]

lol, i thought i missed something, thought maybe the put the early release link on discord

[astral goblet]

I have an email alert set up since i'm on the waitlist. Emails will probably go out when it's released. The waitlist i dont think is for anything else at this point. Only a few got selected to preview that signed up. The free preview plan was scrapped

[nimble canyon]

Heavily depends on type of checkpoints, Pony for example takes 2-4 month to train, but there are also months of preparations.

[tropic frost]

say, can you set SD to add more generations in queue?

[whole oriole]

Damn I forgot I was in this server

[pulsar venture]

Hi can someone link a youtube tutorial to comfyui?

[frail sonnet]

Do you mean the SD install on your own computer? If so, most definitely. I can screenshot the comfyui version if you like.

[loud solar]

https://www.youtube.com/playlist?list=PLIF38owJLhR1EGDY4kOnsEnMyolZgza1x

[pulsar venture]

Thank you, is it noob friendly?

[peak orchid]

WARNING: Someone has been infecting comfyui users with malware via malicious nodes: https://www.reddit.com/r/comfyui/comments/1dbls5n/psa_if_youve_used_the_comfyui_llmvision_node_from/

The malware tries to stealing credit card and banking info

[loud solar]

It's damn good ... if you start with part 1 it should be noob friendly and you will learn a lot of basics

[soft iron]

good

[peak orchid]

unless you're anti-ai and anti-open source, attacks like these are a bad thing

[tribal needle]

so generating images here is not free anymore?

[soft iron]

why

[tribal needle]

like how it used to be with @radiant meadow

[peak orchid]

because they aren't pentesting/security research projects to encourage better security. Its literal malware trying to steal your money

[soft iron]

he just wanted to earn money, he's a person like the rest

[tribal needle]

so stealing is the move ur saying?

[soft iron]

this is a problem of states, not ordinary people

[tribal needle]

btw is the main comy ui compromised or just those nodes

this is still safe right
https://github.com/comfyanonymous/ComfyUI

thats the one i see utubers donwloading in tuts