🤝｜tech-support

[wet magnet]

the nvidia 5000 series should be out by the end of hte year

[rose ferry]

Ohhh

[wet magnet]

I'm waiting for a 5000

[rose ferry]

5090!

[fair oxide]

I have a 6650xt

[rose ferry]

poor thing.

It has no cuda

therefore, switch to CPU

[wet magnet]

I have a 3080 and was going to get a 4080 but I think it's best to wait

[rose ferry]

I am not sure what I am doing wrong with SD3 locally, only different is file name without sdxl_base on clip

[fair oxide]

I dont have " webui-user.sh"

[rose ferry]

thats ok, it automatically create when you launchg

Apparent you havent run it yet

[fair oxide]

But I cant launch it

Due to the error

[rose ferry]

my bad, i misunderstood from memory thats venv folder. But anyway....

webui-user.sh is for A1111

my bad, let me look

[rose ferry]

no, the sd3 medium is checkpoint

t5xx goes clip

[fair oxide]

So should I switch to the auto one? @rose ferry

[wet magnet]

I feel like I have one too many

[rose ferry]

Look for a script named start.sh, start.bat, or a similarly named file. This script is often responsible for launching ComfyUI.

no, sd3 is works best on comfyUI

I got myself confused because I use both. lol

[rose ferry]

open with notepad?

[glossy lynx]

not sure if i can ask for help here

i just got the swarm loaded with SD3

get this error that the backend dont have the model

[fair oxide]

REM Check if venv directory exists
IF NOT EXIST "venv" (
    echo Creating virtual environment...
    python -m venv venv

    REM Activate the virtual environment
    echo Activating virtual environment...
    call venv\Scripts\activate

    REM Upgrade pip to the latest version
    echo Upgrading pip...
    python -m pip install --upgrade pip

    REM Install dependencies from requirements.txt
    IF EXIST "requirements.txt" (
        echo Installing dependencies from requirements.txt...
        pip install -r requirements.txt
    ) ELSE (
        echo requirements.txt not found. Skipping dependency installation.
    )

    REM Install torch, torchvision, and torchaudio with specific index URL
    echo Installing torch, torchvision, and torchaudio...
    pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
) ELSE (
    REM Activate the virtual environment
    echo Activating virtual environment...
    call venv\Scripts\activate
)

REM Run the main.py script with --auto-launch argument
echo Running main.py with --auto-launch argument...
python main.py --auto-launch

REM Deactivate the virtual environment
echo Deactivating virtual environment...
deactivate```

[glossy lynx]

anyone know how to show the backend where to find my models?

i take it the backend is the Comfy-UI

where do we config the ComfyUI?

[ornate elk]

@fair oxide what's the issue?

[fair oxide]

I cant use my gpu

[ornate elk]

At which step are you from my guide?

[fair oxide]

I followed the whole guide

[wet magnet]

oh shit it worked.... what folder do my images go to?

[rose ferry]

output folder

[fair oxide]

https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides#amd-comfyui-with-zluda

[rose ferry]

happy for you.

[ornate elk]

Okay then show me your environment path entrys

[glossy lynx]

My Models load fine in SWARM but error is the model is not available in the backend?

[ornate elk]

The zluda folder path entry is missing

[fair oxide]

It didnt get added

[honest tendon]

Error occurred when executing FaceDetailer:

[ornate elk]

You have to add it

[fair oxide]

I tried

Ill try again

[brittle lantern]

avg speed for my m2 pro correct or can it be fatser?

[glossy lynx]

damn nearly 5s/it

[brittle lantern]

bad damn or good?

[glossy lynx]

seems slow

[brittle lantern]

for m2 pro or just in gen

[glossy lynx]

in gen, but then again i just got hit with a bombshell an hour ago, i was told SD Forge is depricated and i was getting +1it/s not +1s/it

[fair oxide]

@ornate elk

[glossy lynx]

SD Forge was giving me incredible speeeds

[brittle lantern]

well yh ill assume cuz its a macbook with 16gb of ram shared

[rose ferry]

SD3 is designed for 64VRAM

[glossy lynx]

i thought m2 pro was a cooker on ai inference

[fair oxide]

@ornate elk Thanks that fixxed it, I edited the wrong thing

[ornate elk]

Make sure its in the system environment and not in the user environment

[glossy lynx]

where you read that?

[brittle lantern]

until draw things update for sd3 then thats when ill get quite fast speeds on macbook

[ornate elk]

Nice!, remember first image can take 15-40 minutes.

[honest tendon]

idk what went wrong

[fair oxide]

Thanks for informing me of this

[brittle lantern]

question fren cfg scale 7 is fine for sd3?

[brittle lantern]

alr and 5 its seems to be the highest i get on comfy with my m2 pro is there anyway i can optimise it? @ornate elk

or i just have to use as it

[glossy lynx]

@ornate elk do u have any ideas on my issue?

[honest tendon]

https://www.youtube.com/watch?v=849xBkgpF3E
I follow his guide, the first part worked but suddenly just showed me

[worn yacht]

Is there something as easy to use as a diffusers python demo pipeline to load the model and run pipe() for SD3? That should be the bare minimum.

[ornate elk]

Nope sry, didn't used swarm that much

[glossy lynx]

thx

[brittle lantern]

i wanna see what Juggernaught will cook with sd3 (i spelt that wrong)

[rose ferry]

my bad, i say it wrong.

"6GB VRAM can probably run this easily"

[ornate elk]

Comfyui already uses the best settings by default. But idk what you could change specifically for M chips

[brittle lantern]

i see

[fair oxide]

Hey, i got a error

[brittle lantern]

hope draw things updates fast and hopefully auto1111 makes my life easier

[rose ferry]

keep pushing. new error is a progress.

[brittle lantern]

what samplers people using rn

[glossy lynx]

maybe im just being a retard here.

i found in swarm the comfy interface

just trying to figure out how to select a model in this node based interface?

[ornate elk]

When do you get that?

[glossy lynx]

no matter where i click on load checkpoint i cant get it to let me specify a checkpoint / model

ah i imported from generate tab in swarm and it looks like its loaded the SD3

[opaque cove]

Just a FYI: If you get an 'Error while deserializing header: InvalidHeaderDeserialization'
when loading the T5_fp8 or the SD3 model containing the T5_fp8.
It seems like in some cases updating ComfyUI might not be enough.
-> Downloading the latest version of comfy portable fixed it for me!

[ornate elk]

@fair oxide make sure you replaced the necessary files from zluda inside torch/lib

[glossy lynx]

even just queueing prompt from comfy without swarm trigger i get these errors

[ornate elk]

Cfg 4-6 should be good too with sd3

[wet magnet]

is there a way to add a "system prompt" to my overall workflow?

[fair oxide]

I need to switch to cpu

how would i do that

[ornate elk]

Why? What are you trying to do?

[fair oxide]

I have a 6650xt

[ornate elk]

Yes

[fair oxide]

Someone said I could not run it

[ornate elk]

Run what?

[fair oxide]

cuda

[ornate elk]

It can with zluda

But you need to make sure you did everything correctly

[ornate elk]

Looks good. And you replaced the files in the library one with the download Rocm libs?

[fair oxide]

I think so

[ornate elk]

And you replaced the Cublas and cuparse DLLs?

[ornate elk]

And you copied them to the torch/lib path?

[fair oxide]

?

Ohhh

[ornate elk]

Step 5 of the Zluda part

From the Guide

[solemn atlas]

Hii, i have a question, how can I get access to stable diffusion 3? I signed up to hugging face . Is this also for automatic111

[ornate elk]

Auto1111 first needs an update to support it. So we have to wait a few days or more

[solemn atlas]

I understand, so I didn't needed to sign up I guess

[ornate elk]

You can still download the model and safe it for later or use it on Comfyui

[ornate elk]

Looks good

[fair oxide]

Didnt work

[ornate elk]

Have you restarted the PC after changing the library files?

[fair oxide]

yes

Just did

[ornate elk]

And your trying the Dreamshaper model or sd3?

[fair oxide]

Dreamshaper

[fair oxide]

Should I use sd3?

[ornate elk]

Nope

[fair oxide]

ok

[ornate elk]

It won't work until I update my guide

[fair oxide]

ok

Is there anything that you would sugest?

[ornate elk]

Please recheck this

Normaly zluda and the rocm path should be both there

But there is only the zluda folder path

Make sure your in the System variables and not in the User variables

[waxen badger]

Hi guys

[waxen badger]

I put the model in correct place in SwarmUI

When I go to SwarmUI in browser, models tab, I can't see a model at all.

What did I miss?

[ornate elk]

Okay that looks good.

Is your AMD driver updated?

[ornate elk]

Thats outdated

[waxen badger]

Nevermind I see it

[fair oxide]

It dosnt say I can update

It also made me download the pro vesrion

[ornate elk]

Then uninstall the pro drjver

[fair oxide]

How?

[ornate elk]

The same way as you uninstall other programms

In the system settings

Under Apps

[fair oxide]

ok

doing

[ornate elk]

Then download the latest AMD driver for your GPU

From AMDs website

And then install it and restart the PC

[ornate latch]

@ornate elk Hello! I was the guy yesterday that asked about my pc freezing when my render hit 70% loaded, you told me to add some lines to my command args. I didn't get to see if that fixed my problem because those lines caused my pc to freeze and slow down just from running normal images, and made it take several minutes to load a lora

[ornate elk]

You mean --medvram-sdxl?

And hey

[fair oxide]

It still didnt work

Should I js try [AMD] ComfyUI with DirectML

[ornate elk]

Nope don't xD

[fair oxide]

ok

[ornate elk]

Better try automatic 1111 with zluda

Its the easiest and the one I can help easier than with comfyui.

[ornate latch]

Yeah you told me to add --xformers --medvram-sdxl --no-half-vae

[ornate elk]

Is your webui updated?

It shouldn't freeze with these arguments

But you can try without --medvram-sdxl

[ornate latch]

It should be I reinstalled it 2 days ago. I'll do that

[ornate elk]

Ah alright.
Then its updated

Then you need to the the tiled vae extension

From the extensions tab

[ornate latch]

What does it do?

[ornate elk]

The VAE step is the last step of the image generation.
Thats the thing that needs the most resources and causes the Freeze.
What tiled vae extension does is to tile the VAE into multiple tiles. That reduces the load and shouldn't freeze

[ornate latch]

Gotcha, so it that the step that happens at 70% on Highres fix? Like I said it never happens when I make images, but if I use hires or image 2 image it has a chance of freezing

[ornate elk]

@fair oxide I will redo my comfyui guide for AMD Zluda, when I'm back home on Sunday. So best for your GPU is to try the Automatic1111 guide with Zluda.

[ornate elk]

Yep, when upscaling the VAE gets larger

[fair oxide]

ok

[ornate latch]

Okay that makes sense. Update removing the --medvram stopped the lora slowing down. I'll download that extension and try it after work. Thank you!

[ornate elk]

NP, and only enable tiled vae and let the settings on default

[chrome rampart]

hello all. anybody idea hw to solve in comy error TripleCLIPLoader:
- Value not in list: clip_name1: 'clip_g.safetensors' not in []
- Value not in list: clip_name2: 'clip_l.safetensors' not in []
- Value not in list: clip_name3: 't5xxl_fp8_e4m3fn.safetensors' not in []

[ornate latch]

@ornate elk Is the extension called Multidiffusion Upscaler? That's what comes up when I search for Tiled vae extension

[ornate elk]

Yep

It has two names xD

[viral marsh]

I need help with v1 image to image

def edit_image(image, prompt):
    try: 
        response = requests.post(
            f"{API_HOST}/v1/generation/{ENGINE_ID}/image-to-image",
            headers={
                "Accept": "application/json",
                "Authorization": f"Bearer {API_KEY}"
            },
            files={
                "init_image": image,
            },
            data={
                "text_prompts[0][text]": prompt,
            }
        )
        response.raise_for_status()
        data = response.json()
        
        # Extract and return the first image artifact
        image_data = data["artifacts"][0]["base64"]
        return base64.b64decode(image_data)
    
    except Exception as e:
        print(f"Error in edit_image: {e}")
        return None 

[fair oxide]

stats = memory_stats_as_nested_dict(device=device)
File "C:\SD-Zluda\stable-diffusion-webui-amdgpu\venv\lib\site-packages\torch\cuda\memory.py", line 270, in memory_stats_as_nested_dict
return torch._C._cuda_memoryStats(device)
RuntimeError: invalid argument to memory_allocated

[ornate elk]

Can you share the whole cmd log

[viral marsh]

@ornate elk the parameters are correct formated? Also what type of image data needed

[ornate elk]

Sry, can't help with Api stuff

[fair oxide]

Im waiting for it launch again

[fair oxide]

ITs just sitting like this

[ornate elk]

What does the cmd shows?

[ornate elk]

Perfect it works.
Remember the first generation can take up 15-40 minutes

[fair oxide]

Ohh ok

Thanks

[ornate elk]

NP, just let it load until you get an image 🙂

[fair oxide]

🥳

[ornate elk]

And if that works you can also try the comfyui zluda guide again then it should be easier to get it running too

[fair oxide]

I didnt change zluda tho

[ornate elk]

Yea but you can delete the Comfyui folder again and follow the guide again.
If Automatic1111 works, you can skip the steps 4 and 4.5 of it.
But first Auto1111 should work.

[honest brook]

https://youtu.be/VOg8XFR47p4?si=MnR7q-BVhl0eJDYZ

[honest tendon]

what does custom build mean?

[ornate elk]

Where?

[ornate elk]

Don't know but seems something called NMS of torch can't be run by cuda

[honest tendon]

how can that be fixed though

do i download the new SD3 Zluda thing?

[ornate elk]

Are You on amd?

Then you need to git pull

[honest tendon]

YES

[honest tendon]

will this fix it?

[night orchid]

How can I use SD3 on my local installation ? should I download the models and use them as a normal model

[ornate elk]

Update your comfyui first

[ornate elk]

Nope, it should work on 11.8, but you can try if you want

[ornate elk]

SD3 currently only works with Comfyui

[night orchid]

ah see

does comfy update it self, or is it something I have to do on my part?

[ornate elk]

You have to run the update.bat if sou have one

If not then you need to run git pull with a cmd inside the comfyui folder

@fair oxide did it worked?

[fair oxide]

Its generating but the outputs are poor

[ornate elk]

Thats normal. You have to download other checkpoints from civitai.com like Dreamshaper or Juggernaut or Anime models

[fair oxide]

What site should I look on?

[ornate elk]

The base 1.5 model is bad

[ornate elk]

Civitai.com

Its the largest model database

[night orchid]

good explainantion here: https://www.youtube.com/watch?v=9zfF7Jt-JnU

[sudden helm]

what is the easiest way to toggle on and off certain sections of a workflow ? (ie upscaling)

[pure forge]

So for SD3 I need the model without text encoders along with g, l, and the fp16? I have a 4090.

[spare spruce]

I'm still have problems with that

[sudden helm]

yeah that worked for me

[mystic tinsel]

Hi guys, please help me how do you install this enhancer?

[cyan patio]

Do any of you know why my credits are below 0 in the stability.ai api?

@limpid whale idk if it is supposed to be as it is but i can use the ultra model as often as i want even after having my credits below 0

[fair oxide]

@ornate elk This is me

[ornate elk]

Cute

[pallid abyss]

hey anyone knows how to fix this error?
error: RPC failed; curl 92 HTTP/2 stream 5 was not closed cleanly: CANCEL (err 8)
error: 5177 bytes of body are still expected
fetch-pack: unexpected disconnect while reading sideband packet
fatal: early EOF
fatal: fetch-pack: invalid index-pack output

[pure forge]

where are you even seeing this

[pallid abyss]

git

when i'm trying to download the stable diffusion

[pure forge]

I just downloaded from hugging face

[ornate elk]

If your from asia, Iran, Russia you may need a VPN for the download

[pallid abyss]

asia

right

oh

[pure forge]

damn an entire continent blocked

[pallid abyss]

even lgbt people have more freedom, right?

[pure forge]

wdym even

[pallid abyss]

nothing happy pride sis

[pure forge]

sis???

maybe asia should be blocked if they act like this smh

[pallid abyss]

https://tenor.com/view/thisisalecx-block-blocked-queen-of-brooklyn-gif-22784328

like that?

[pure forge]

man this is tech support not shitpostland

[pallid abyss]

oh sorry

[ornate elk]

You need a VPN for the first git clone

Then you should be able to launch it while the VPN is still on

And if everything is installed it should work without it

[quartz lark]

really?

[pallid abyss]

thanks im gonna try

[ornate elk]

Yes, we have to wait for Auto1111 update

[quartz lark]

i guess i'll try comfy UI today, maybe i'll like it

[ornate elk]

Always worth a try ^^

[quartz lark]

can they both pull from the same model folders?

[pure forge]

yes

[quartz lark]

sick

[pure forge]

https://www.reddit.com/r/StableDiffusion/comments/17630s0/help_getting_automatic1111_and_comfyui_to_use_the/

[frail umbra]

Got the latest ComfyUI but all I see is the node for the API. Any ideas why I can't see the local SD3 node?

[fair oxide]

justemery

justemery

justemery

justemery

justemery

justemery

[tropic shale]

@tropic shale

[frail umbra]

Uh...

[ornate elk]

Idk xD

[ornate elk]

Try input a example workflow from sd3

[frail umbra]

Good idea

[fair oxide]

Are there any modles that are good at everthing

[quartz lark]

i also am using 1.9.4 or something, is A111 actually better?

[ornate elk]

1.9.4 is the latest version of A1111

[ornate elk]

Nope. But there are very good models for different styles.
For example Juggernaut Reborn is good for Realistic stuff.
And Dreamshaper v8 is an good Allrounder.
For Anime checkout Divine Anime mix or sudachi

[quartz lark]

ah ok

so A111 is just the base UI?

[ornate elk]

Yea Automatic1111 is just a webui like comfyui which can load and used Stable diffusion models

Auto1111 is also the most used one

[pallid abyss]

it works lol.. the funniest thing that i was thinking that i prob need vpn but i choose the hard way lmao

[quartz lark]

so why did they make this work for comfyUI first?

[ornate elk]

Ah nice ^^ yea sometimes the first guess is the right one

[pallid abyss]

i downloaded it a couple of months ago and it worked without vpn 😦

[ornate elk]

Because the Dev of Comfyui works for Stability ai since a few months

[quartz lark]

oh

where do you get comfy UI?

i think i found it

[ornate elk]

In the pinned messages of this channel are install links and guides

[quartz lark]

thanks!

[ornate elk]

No problem

[quartz lark]

I used one of the pre-made workflows recommended on the sd3 page and im getting this error. Seems related to this clip loader, but when i click it, it doesn't let me input anything.

using comfy_example_workflows_sd3_medium_example_workflow_basic

so i kinda don't know what else to do here. it won't let me select anything in those "undefined" boxes

im using the model that supposedly includes clips

but don't know how to load them

ok wait i think these are the text encoders

[full steeple]

I have created an installation video of SD3 Medium , which channel should I share the link to it here please? Thanks.

[strange sapphire]

Hi everyone... I am wondering if there's anyone who would be willing to help me figure out an issue I am having opening stable diffusion

When I try to go to http://127.0.0.1:7860/ I am getting Error: Connection Refused

Is that a problem with my firewall?

It seems like the firewall has that open - I even manually checked and it looks good. What am I missing?

[terse plaza]

is there an official guide on how to install SD locally in ubuntu?

[buoyant carbon]

Hello guys, I'm new in comfyUI and I installed SD3 to it. I get this error message while running SD3, can someone help me about it? thanks

[topaz socket]

Trying to use ponyrealism in comfy ui but I'm getting this

Here are the nodes

seemed to be an issue with the sampler I was using?

[solemn moon]

Out of interest, what image size is 3?

Also looks like ze bokeh is gone yay!!

[tacit brook]

Is there a way to disable Facechain so it stops trying to load models on startup? (A1111) It's unchecked in extensions. (I've installed Cuda Toolkit, Visual Studio + Build Tools and still getting the "no module named mmcv")

nevermind, seems to be disabled now that I asked for help. 🙄

[limpid whale]

Hi, I don't work for Stability anymore, sorry 🙃

[topaz socket]

What is causing the image to be so noisy?

increasing sample count doesn't really do anything

[kindred meadow]

Hello WOuld this be the right channel to ask about stable diffusion in krita?

On Krita i am usaing the stable diffusion ai generative tool but i keep getting this error, Anyone know swhy?

[atomic mauve]

trying to test the new sd3medium with the code from huggingface and nothing works, it fails on loading: Exception: data did not match any variant of untagged enum PyPreTokenizerTypeWrapper at line 960 column 3 any idea?

[pallid abyss]

hey someone knows how to fix it?
File "C:\Stable Diffusion\stable-diffusion-webui\modules\sd_models.py", line 304, in read_state_dict
pl_sd = safetensors.torch.load_file(checkpoint_file, device=device)
File "C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\safetensors\torch.py", line 308, in load_file
with safe_open(filename, framework="pt", device=device) as f:
safetensors_rust.SafetensorError: Error while deserializing header: MetadataIncompleteBuffer

Stable diffusion model failed to load

[ornate elk]

Hey, would need the full CMD log

Ah okay, download the Dreamshaper v8 model from Civitai.com
And put that into the models/stable-diffusion folder.
There delete the 1.5-ema-pruned model.
Then restart and try again.
Mostly the 1.5 model file was broken

[pallid abyss]

lmao what would i do without you?! SLAY

[pallid abyss]

i don't think i have the "1.5-ema-pruned model." I tried to find in search bar but it's not there def

[ornate elk]

What's in the stable-diffusion folder?

[pallid abyss]

everything i guess, cause there a lot of folders

[ornate elk]

No I mean in your screenshot

There is a folder named stable-diffusion

[pallid abyss]

models

[ornate elk]

Stable-diffusion-webui/models/stable-diffusion

[pallid abyss]

OPS

i got it

yes

sorry

im stupid

deleted

see

[pallid abyss]

Oh, it works, no, you're very smart, I'm going to vote for you for president in the future

[ornate elk]

Haha thanks xD

[grave condor]

If I have an SSD right now and go buy an external hard drive to run stable diffusion on and store everything will it effect its performance?

[fierce hollow]

I often see this. And what is it? and how to connect it to automatic1111?

[grave condor]

comfyui

[fierce hollow]

?
https://imgur.com/a/BBBslkP

https://imgur.com/a/akUPkGl

[quartz lark]

it was because i didn't have text CLIP files installed and chosen

[fierce hollow]

I ask in general, what does this extension do? What is it for?

[delicate urchin]

Running the basic SD3 Workflow from huggingface and getting this error. What is the fix?

[ornate elk]

It will affect the model load time. But not the generation speed as that is done on the GPU

[ornate elk]

You need to follow my AMD Zluda install guide for AMD GPUs.
Its In the pinned messages of this channel

[storm cosmos]

Hi! there!

[honest tendon]

I wanna update to SD3 how do I do that?

6950xt AMD

https://youtu.be/VOg8XFR47p4?si=KIVp0EwTGOEOfBLi

is this tutorial good?

[ornate elk]

Its okay but I would still recommend using my written guide.

[honest tendon]

Did you write a new one?

DAMN

[ornate elk]

I wrote multiple.
In the pinned tab you find the link to all my guides

[honest tendon]

Wait i already did the ZLUDA thing from that

how do i download the new one though

do i have to uninstall everything

how do i even check my current version of SD

[ornate elk]

Do you have Automatic1111 or comfyui installed?

[honest tendon]

comfyui

[ornate elk]

Okay with my guide or an other one?

[honest tendon]

ur guide

[ornate elk]

So you have the Start-comfyui.bat file?

[honest tendon]

Yes

[ornate elk]

Okay then its easy

[honest tendon]

idk what to git pull

you mentioned yesterday

but i fell asleep since it was like 4am

[ornate elk]

When your in the Comfyui folder.
Click into the explorer bar (not searchbar)
Then type cmd and hit enter.
In the cmd type
Git pull

Then relaunch the start-comfyui.bat

[fair oxide]

does swarm ui support adetailer?

[ornate elk]

As it uses comfyui in the backend I don't think so

[fair oxide]

i see ty, that's a problem for me and not having sd3 supported on auto atm

[ornate elk]

Seems like it worked

[honest tendon]

will i fuck up the system if i re git pull?

[ornate elk]

Yep, we have to wait a bit for auto1111

[ornate elk]

Nope, it then should say already uptodate

[honest tendon]

it only works with comfy?

thank god i switched to comfy lol

[ornate elk]

For now yes. But I guess it will work in Auto1111 in a few days too

I'm hyped for the next auto1111 version.
Forge got deprecated because the dev build of auto implemented a lot of stuff for better performance

[signal torrent]

A1111 with the same speed as forge would be a dream

Doubt it'll happen in the next update though

[honest tendon]

bit off topic was about to say thanks but then my tummy rumbled so bad i had to go to the bathroom, on my way back my pants got wet so i ran to my room without noticing the blinds were open that my neighbor crush saw me but ass naked

thanks though @ornate elk

[ornate elk]

Np xD

[earnest jackal]

read the model description for the recommended samplers, some do not work

[fair oxide]

@ornate elk i ran some basic samples on sd3 using comfyui, nothing too impressive tbh, its a base model, i was also having issues with face detailer on comfyui, and didnt find much good workflow for it

[ornate elk]

Yea remembers me of the sdxl release

[autumn stag]

hello i am trying to use sd3 with diffusers on kaggle notebook but when I try to load the pipeline I get this exception :

code : ```python
import torch
from diffusers import StableDiffusion3Pipeline
from huggingface_hub import notebook_login
notebook_login()

pipe = StableDiffusion3Pipeline.from_pretrained("stabilityai/stable-diffusion-3-medium-diffusers", torch_dtype=torch.float16)
pipe.enable_model_cpu_offload()

exception

403 Forbidden: Please enable access to public gated repositories in your fine-grained token settings to view this repository..
Cannot access content at: https://huggingface.co/api/models/stabilityai/stable-diffusion-3-medium-diffusers.
If you are trying to create or update content,make sure you have a token with the write role..
Will try to load from local cache.

[fair oxide]

lol yeah

[autumn stag]

I accepted the license terms and created a token beforehand, but it still doesn't work

[honest tendon]

this face detailer has been fucking my ass up for a day now

how do i fix it

i keep trying nothing works

[fair oxide]

sounds terrible, you might wanna purge cache then resinstall comfyui, here is a good guide https://youtu.be/aL4xrbg4NjU

[ornate elk]

Maybe delete the nodes and reinstall them

Then check for the errors

[honest tendon]

I did that

[light geyser]

what do i need to place here?

[honest tendon]

and even refreshed

[ornate elk]

Some nodes just won't work with AMD Zluda or require manual fixes

I can try to get them running but I'm not at home until Saturday

[honest tendon]

heres the txt file

[runic shuttle]

hello. total noob here. anyone got a SD3 colab notebook?

[ornate elk]

The Clip text encoders

Available on the same site as the sd3 model

[fair oxide]

for a base model this is nice .. but have some technical restrictions with human figures

[swift barn]

is there a guide for installing SD in ubuntu?

the ones i saw on the pin seem to be for linux

i tried to do it following the Linux section from this

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

but im getting two errors

during the instalation process i got

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 20.00 MiB. GPU 0 has a total capacty of 3.81 GiB of which 18.69 MiB is free. Including non-PyTorch memory, this process has 3.46 GiB memory in use. Of the allocated memory 3.35 GiB is allocated by PyTorch, and 54.54 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

but the web page http://127.0.0.1:7860/ can run

but when i try to test a simple prompt i get

AttributeError: 'NoneType' object has no attribute 'lowvram'

this is the log with both errors

[ornate elk]

What's your GPU?

[rare wraith]

@ornate elk hey man it's working thanks also I wanted yo know is there any documentation for creating a custom offline model and train it

[swift barn]

glxinfo | grep -E -i 'device| memory'
Dedicated video memory: 4096 MB
Total available memory: 4096 MB
Currently available dedicated video memory: 3340 MB

[ornate elk]

@swift barn You need to edit the webui-user.sh
At the line commandline_args you need to add:
--medvram --opt-sub-quad-attention --no-half-vae

[swift barn]

i edited the file, killed the process and ran webui.sh again

but i have the same issue

do i need to do something extra for the change to take effect?

[ornate elk]

Download the Dreamshaper v8 model from Civitai.com and delete the 1.5 EMA pruned one

Then try again.

Sometimes the base model is broken

[swift barn]

this https://civitai.com/models/4384/dreamshaper

?

[ornate elk]

The terminal should show a line "launching webui with following arguments: --medvram --opt-sub-quad-attention --no-half-vae"

Then you know its working

[swift barn]

should that be in the beginning?

because it doesnt print that

[ornate elk]

Yep so the file isnt correctly edited

You need to remove the # in front of the Commandline_args line

[swift barn]

🤦‍♂️ rookie mistake

ok, that OutOfMemoryError is gone

yes, its working now!

[ornate elk]

Perfect 🙂

[swift barn]

if i want to uninstall so that i can do an installation without errors, the process its just to delete the folders?

[ornate elk]

Yep deleting the stable-diffusion-webui folder is enough.
But you can also run a
Pip cache purge to remove the temp files too

[swift barn]

good

and i saw in the install logs that this installs SDXL, if i want to use SD3 how do i change that?

[ornate elk]

You installed Automatic1111, its a webui that can run any model.
1.5, SDXL and SD3.
But for SD3 there is an update needed so we have to wait a few days.

Then its just putting in the model inside the models/stable-diffusion folder

And selecting it at the dropdown

But with a 4gb GPU its not that fun to use SDXL or SD3

[swift barn]

ah, SD its just a model?

i though it would have more code an login inside

(new in both AI in general and SD in particular)

[swift barn]

is this because of low quality of the generated image or just the loading time?

[ornate elk]

Loading time mostly. For sdxl you would need to change --medvram to --lowvram

[ornate elk]

Stable diffusion is the process to generate images from Noise.
But the source for that are the models.
And so Stable Diffusion 1.5, SDXL and SD3 are models needed for that process

The code for that process is inside the webui

[swift barn]

and all the custom models that you download have that model "inside"

[ornate elk]

Yep they are based on one of the base models

[swift barn]

ok, perfect

and if i want to also have comfy in the same machine, can i have it use the same model folder?

[ornate elk]

Yes, comfyui has a file you have to rename and edit. Its the "extra_model_path.yaml" you need to remove the .example.
Then in the file you only need to esit the base path to the Auto1111 install path

[swift barn]

ok, thanks!

now i just have to search how to create models to play with that

[fair oxide]

any idea how to fix the segm detector that's crossed out in comfyui

[ornate elk]

You don't need to create them.
Go on Civitai.com and download the ones you like
Its the largest database for models

[fair oxide]

getting this error right when face detailer section starts

im guessing i might be missing some cuda related files?

[swift barn]

im doing this as a challenge, so i want to try my hand with everything

is there a way to train models locally?

im guessing 4gb VRAM is not enough

[ornate elk]

Yep 4gb is not enough :/

[swift barn]

but there is a way to do it locally?

[fierce hollow]

Is there an analogue of Hires. fix in the ComfyUI plugin?

[crimson blaze]

Hi! I have a trouble with MeshGraphormer Hand Refiner. Tried to reinstall nodes but it didn't help. I didn't find any thread exept one unreplied about this issue. Please help.

[fair oxide]

so this is what chatgpt is suggesting from the error i posted above...

would appreciate some guidance, this was the actual error on comfyui #🤝｜tech-support message

[ornate elk]

Yes it can done with the tools Khoya_ss and OneTrainer

[rose ferry]

Question about Stable Assistant. How do you get it reveal the prompt from the image generated inside?

[fair oxide]

well what do you know ....

[rose ferry]

you dont have anything with torch and torchvision so you cant uninstall it.

[fair oxide]

yeah i see that now after checking, which might have been causing error for the face detailer on comfyui

[rose ferry]

I think it cant be run without them at any point, maybe you put it in wrong enviroment.

[fair oxide]

well i had workflow set up, but was having error when parsing face detailer section

downloading and intalling the torch vision now, fingers crossed, hope this can fix the error

[rose ferry]

yeah thats what i would... i am no tech support.

[rose ferry]

waiting for guides...

[fair oxide]

got this bunch of error in red after running ... pip install torch==1.13.0 torchvision==0.14.0+cu113 -f https://download.pytorch.org/whl/torch_stable.html

[ornate elk]

Does your comfyui has a venv folder?

You can also try run the install_torch.bat

If you want a venv for comfyui you can checkout my install guide on github

[fair oxide]

yes it has a venv folder

@ornate elk should i be deleting venv?

[fair oxide]

i dont see install_torch.bat under comfyui root folder

how do i check what cuda version i have on my system?

oooh this command displays the whole things in tables inside console nvidia-smi

[quiet gust]

can anyone help me with opening stable diffusion?

i keep getting this message

[wet magnet]

it's something with the command

I don't remember having to put any kind of directory before the pip command

[solemn moon]

The error message you're encountering seems related to the installation of a Python package where distutils is causing issues. Here's a detailed breakdown and potential steps to resolve the problem:

[quiet gust]

like what do I do

[solemn moon]

Steps to Resolve
Ensure setuptools and wheel are Updated:
Sometimes, older versions of setuptools or wheel can cause issues. Try upgrading them:

pip install --upgrade setuptools wheel

[quiet gust]

on the command promp?

[solemn moon]

Install distutils:
On some systems, distutils might not be included. You can install it manually:

[solemn moon]

mustbe, im using gpt 😅

[solemn moon]

pip install distutils

Check Python Version Compatibility:
Ensure the Python version you are using is compatible with the package you are trying to install. The package might have specific version requirements.

[quiet gust]

it says cannot find any downloads that satisfy reuirements

[solemn moon]

[quiet gust]

well do you know of any?

[ornate elk]

Are you trying go install stable diffusion for the first time?

[quiet gust]

yes

i'm installing forge

cause I was told it's faster

[ornate elk]

Don't do that, forge is deprecated

Please follow my install guide from the pinned messages of this channel

[quiet gust]

which version cause I want one that generates images fast

[ornate elk]

Automatic1111 will generate fast if you follow the guide

[quiet gust]

where?

[ornate elk]

Here:
#🤝｜tech-support message

Or when you click the Pin needle. There are all guides linked

[quiet gust]

ok i downloaded it on the cmd where should it end up?

[ornate elk]

If you done like the guide says. Then its in the folder you created.
But if you just opened a cmd then it could be anywhere

[quiet gust]

shit

[fair oxide]

holy crap, i finally managed to solve the face detailer issue conflicting with cuda and torchvision on my system, if anyone facing an issue where it throws error like this #🤝｜tech-support message i've attached a txt file with the solution.

ignore the top seciton

python "cuda avail.py" ("" used for space between filename)
otherwise:
python cuda_avail.py```

[quiet gust]

ok i made a folder and it's downloading

@ornate elk so it says my GPU can run up to 1.8/7.8 GB which vram is good for that?

[fierce hollow]

Is there a way to save images automatically in ComfyUI? How does SD do this to the outputs folder

[ornate elk]

What's your gpu?

With 8gb vram you need
--xformers --medvram-sdxl --no-half-vae

[ornate elk]

Comfyui also saves automatically in an folder

[quiet gust]

1.8/7.8GB that's what it says

[fierce hollow]

In which folder in outputs there is no photo

[ornate elk]

Okay then you have a GPU with 8gb vram

[ornate elk]

Ohhh thats not an nvidia or AMD GPU

With that Intel UHD GPU chip you can't run Stable-diffusion localy

Only on slow CPU mode

[quiet gust]

oh do you have anything for that?

will it still generate fast though?

what command args do I put in then?

[ornate elk]

It will be slow

[quiet gust]

can I upgrade it?

[ornate elk]

You would need:
--use-cpu all --opt-sub-quad-attention --no-half --upcast-sampling

[ornate elk]

I guess your using a Laptop, then nope.
Only If the Laptop supports Thunderbolt connection you can plug in an external GPU

[quiet gust]

i'm using an hp laptop

like which version of stable diffusion can run on my gpu?

[ornate elk]

That Intel UHD graphics isnt a dedicated GPU.
Its an so called igpu (a GPU chip from the CPU)

So you have to use CPU

[quiet gust]

which version for cpu can i use?

[ornate elk]

?

The one you installed can work with CPU

You only need to add the launch args from above

[solemn atlas]

somehow my textual inversion does not show up and there is also an error

[quiet gust]

where is that?

[ornate elk]

You See the webui-user.bat?

In the folder you created

There should be a stable-diffusion-webui folder

[quiet gust]

yes and which file in the folder do I go

[solemn atlas]

and another problem I have is, my PC is crashing down. I just crashed down when I wanted to generate something. That happens a lot in last days and generating XL takes also a lot of time too

[quiet gust]

@ornate elk i'm still getting the error message

hello?

[ornate elk]

What error?

[ornate elk]

Remove that embedding. Its probably a lora

[quiet gust]

this

[ornate elk]

Make sure you run the webui-user.bat and not webui.bat
Also make sure to have tiled VAE enabled

[quiet gust]

i can show you my notepad to see if I got anything wrong

this is my notepad setup

[ornate elk]

Remove --xformers --skip-torch-cuda-test --lowvram

Then delete the venv folder

Then open up a cmd and run
Pip cache purge

Then launch the webui-user.bat

[solemn atlas]

thank you, so I looked it up. I use webui.user.bat and also enabled the tiled VAE. Should the hires steps also be at 10 for XL too?

[quiet gust]

capital P

[solemn atlas]

but I don't have any embedding

[ornate elk]

Yep

[solemn atlas]

I want to learn how to use regional prompter. So I used an already posted image from the tutorial and use the png info from it.

[ornate elk]

It says d3t41l3dXLP.safetensor

Check your embeddings folder

[solemn atlas]

yes it's in there

I will delate it and see if it works

[quiet gust]

@ornate elk I put in pip cache purge and it says no command exists

[ornate elk]

Oh but you installed Python 3.10.11 before?

And checked "add python to path" ?

[quiet gust]

I have python 3.10.6

[ornate elk]

Okay then just run the python installer again.
Click on modify, next, and select "add python to system variables"

After that you should be able to run
Pip cache purge

[solemn atlas]

now the error is not there anymore but generation takes very long. I wonder if it has something to do with XL in general

and my textual inversion also doesn't show expect for one

i have much more than one

[ornate elk]

Your using regional prompter with hires fix

That takes longer than normal

[quiet gust]

where's modify?

[ornate elk]

It should be there when using the python installer again

[solemn atlas]

I understand. A few weeks ago it was not that slow like it is now.
And about my textual inversion embeddings. I sadly can't see any of it

Right know I don't use hires.fix but it still needs a long time unsual to what I'm used it

to*

[ornate elk]

What's your GPU again?

[solemn atlas]

4070s

[ornate elk]

You can try adding --medvram-sdxl to the webui-user.bat

And test if it helps

If not just make a normal restart and test a 1024x1024 image without anything enabled

[solemn atlas]

like this?

[ornate elk]

Worth a try, if it doesn't change anything, remove it again

@quiet gust did you found it?

[solemn atlas]

So I tried it and I'm impressed, it really fast now. I hope it stays like that.

Also is there any chance I get to see my embeddings on sd?

[ornate elk]

They will show up if they are compatible with the model.
Then click the embeddings refresh button a few times

[quiet gust]

yeah it's launching

[solemn atlas]

sadly it's doesn't show at all. Doesn't fastneg. work with xl?

[quiet gust]

pytorch is downloading right now

[ornate elk]

Strange, idk, is fastneg an extension?

[solemn atlas]

It's an embedding, I just looked it up to be sure. It worked all the time but when my SD got slow it also disappeared. Now that it's fast again thanks to you, it's still not showing

[ornate elk]

FastNeg is for 1.5

Checked it on civitai

[solemn atlas]

I tried to use now normal checkpoints with vae (no XL) but the textual inversion still doesn't show somehow

Ahhh now it worked

So I need embeddings for XL too

[quiet gust]

ok now I need a venv

[ornate elk]

? It should have created one

[ornate elk]

Yep

[quiet gust]

maybe i did wrong

i'll try downloading python 3.11 again

[ornate elk]

No

Not 3.11

You need 3.10.11

[quiet gust]

ok

how do you launch python again?

@ornate elk can you run through the process again like i'm sorry for sounding stupid but i'm still new to stable diffusion

[ornate elk]

Download the Python 3.10.11 installer and run it

Then delete the venv folder thats inside the stable-diffusion-webui folder.

Open up a CMD and type
Pip cache purge

[quiet gust]

it says invalid syntax

[ornate elk]

Then run the python install again

And click modify

Then next

Then select "add Python to system variables"

[quiet gust]

it says add python to enviornement variables

[ornate elk]

Yea this

[quiet gust]

do I check py.launcher?

[ornate elk]

Nope

Just leave the rest default

[quiet gust]

even the install location?

ok it says it was successful

which cmd do i go to type pip cache purge?

[ornate elk]

Doesn't matter in that case

[quiet gust]

it says this

[ornate elk]

Close that

Thats a python terminal

Open up a cmd and just type
Pip cache purge

Then close it

Then go into the stable-diffusion-webui folder

There delete the venv folder

And then launch the webui-user.bat

[quiet gust]

it's already deleted the venv folder

do I press enter in the command prompt?

it says no code object available for venv

[ornate elk]

Can you screenshot the cmd?

[ornate elk]

The guide says to not install it on desktop

And why is there a python.exe in nsfw/anaconda folder?

[quiet gust]

oh that's for personal stuff

where do I install it then?

[ornate elk]

So you have anaconda installed too

Then you need to edit the webui-user.bat again

At the line PYTHON=
You need to add the path to the python.exe of 3.10.11

[quiet gust]

the windows batch file?

now it says this

[ornate elk]

Yea mostly because It doesn't have the permission for the desktop

[quiet gust]

where do I go?

to allow it?

[ornate elk]

Best would be to create a new folder for it not on desktop and git clone the webui again

[quiet gust]

ok so the folder on a different place?

[ornate elk]

Just follow the nvidia install guide but when it comes to the launch args you use the ones I gave you instead

[quiet gust]

now it says this

[ornate elk]

Okay then edit the webui-user.bat again and there add
--skip-torch-cuda-test

Then relaunch

[quiet gust]

ok

[ornate elk]

I'm afk now

[sullen valve]

Hi, I've been getting this message... again and again, how can I fix it?

[ornate elk]

What's your GPU?

Also forge got deprecated

[sullen valve]

Radeon RX 590 Series

[sullen valve]

What should I do?
I'd like to train a model on my own source images without uploading them online

[ornate elk]

You need to follow my Automatic1111 AMD Zluda install guide from the pinned messages of this channel

[ornate elk]

When you installed the webui with my guide. You can then install the OneTrainer tool to train

[north cedar]

Hii@vocal burrow :), do you know if Stability plans to include SD 3 in the DreamStudio UI?

[quiet gust]

ok i was able to get stable diffusion up and running again

[fallow heath]

cant seem to use sd3, any idea why? got 8 gb of vram

[ornate elk]

Nice

[ornate elk]

Auto1111 doesn't support SD3 currently.
We need to wait for an update

[fallow heath]

only comfy ui for now?

thx in advance

[ornate elk]

Yep for now only comfyui and stableswarm

[thin nexus]

the best option in local still be the sd a11111 or somthing like that?

[wet cedar]

So how do we use Stable Diffusion 3? Do I delete the current set-up or uninstall it? And then download and set up Stable-Diffusion 3 or does it auto-integrate?

[lament tiger]

Hello, I have a question. Recently, I was using SD Automatic1111 and installed the CivitAI Browser+ extension. When I restarted the Docker, the new extension (Colab Command) started, and my RTX began mining due to a script called "miner." Has anyone else experienced this problem? I have now deleted that extension and removed the files from my Docker, but I’m not sure what happened.

[ornate elk]

SD3 is a model you can download and use inside the webui

But for now it only works in comfyui.
Auto1111 needs a few days or so

[wet cedar]

Oh. So it's not a completely brand new thing?

Gotcha. Thank you so much! I gotta wait for Auto

[ornate elk]

Its a brand new model bases on a different architecture

[elfin wyvern]

Heyo! Is anyone else on apple silicon experiencing results like these, even though you follow Lykon ' s exact workflow json and the recommended settings? Is comfy to blame?

[ancient rapids]

I've just started AI image generation and I've set up Stable diffusion on ComfyUI correctly (I think) however, at the moment I'm getting very very low quality images all the time. Such as this one. I'm not sure how to fix this.

Models and embeddings I use don't seem to change anything.
At the moment I'm downloading them and placing them into the appropriate folder.
Are there steps im missing?

[ornate elk]

If your using an sdxl model you need to change the resolution to 1024x1024

[ancient rapids]

using SD 1.5 atm, Should I swap to sdxl?

[terse plaza]

What are the reccomended system requirements for local install?

[ornate elk]

You don't need to but you can try. What model are you currently using?

[ornate elk]

A dedicated GPU with 4gb vram

16gb RAM

[ancient rapids]

I've used 1.5 SD and SD 3 and all of them gave me very low quality images. Similar to the one before.

I've just gotten deliberate just now hoping that might change something.

[terse plaza]

I have that but im getting errors

Like the pc restarts in the middle of the generation

Or it freezes

[ornate elk]

What's your GPU?

[terse plaza]

4gb

[ornate elk]

And what's inside your webui-user.bat

[ornate elk]

Brand

And model

[terse plaza]

It says effective 3.7

Nvidia

[ornate elk]

Yea or try Dreamshaper v8

[terse plaza]

1650

[ornate elk]

Okay then edit the webui-user.bat
At the line commandline_args=
You need to add: --xformers --medvram --no-half
Then save and relaunch the webui-user.bat

[terse plaza]

And the webi has --lowram --opt-sub-quad-attention --no-half-value --no-half

[ornate elk]

Remove --no-half-vae and --opt-sub-quad-attention and --lowvram

And add --xformers --medvram

Then it shouldn't freeze

[terse plaza]

Couldnt install xformers

And the webi.sh crashes

[grave condor]

@ornate elk is the model load time slow with an external hard drive or its fine

[ornate elk]

Are you on Linux?

[ornate elk]

Depends on the hard drive and your connection to the PC.
For examaple:
HDD with 5400rpm over USB2.0 = Slow
SSD over USB 3.0 = fast

[magic hollow]

i think i installed the wrong version for my card, im not sure which verison of theses might work on an amd 580

[ornate elk]

You need the RocmLibs.zip

Sry, I didn't mentioned your GPU because there was no one with an 580 to test but it should work

[magic hollow]

its ok but its good to know that.

[ornate elk]

I will edit once I'm home and when its working for you

[magic hollow]

well you dont have to im not forcign you.

[ornate elk]

I want to, because then other rx580 user can use zluda too

[magic hollow]

alrighty

[ornate elk]

So you need the files from RocmLibs and move them into the library folder.
Then restart the PC

[magic hollow]

alright also i do have your guide pulled up aswell.

[ornate elk]

Let me know if it worked 🙂

[magic hollow]

im going to restart my pc.

im back im now going to attempt stable diffusion.

[terse plaza]

Yes

But that issue was because wifi got disconected somehow

[ancient rapids]

I'm currently getting this error

Anyone know the fix?

[ornate elk]

What's your GPU and what's inside your webui-user.bat

[ancient rapids]

1650

I’ll check my file rn

[ancient rapids]

This along with my python path and "echo off"

[ornate elk]

At the line commandline_args=
Add: --xformers --medvram --no-half-vae

[terse plaza]

Now it works but pyton is using 3gb of vram

And generation has out of memory error

[ornate elk]

So your on Linux?
Then you need to edit the webui-user.sh and there add --medvram --xformers --no-half

[terse plaza]

I add it there

[ornate elk]

Ah okay

[terse plaza]

I was able tu run the web

[ancient rapids]

is it fine if i add "--theme dark" after that too?

[ornate elk]

Sure

[terse plaza]

But now the issue is out of memory when running a generation

Says that it wants to use 16mb but 3gb are in use

So only 14mb are free

[ornate elk]

What model did you tried?

Also is xformers active?

Check the bottom of the webui in browser

[terse plaza]

How do you activate?

[ornate elk]

By having --xformers in the webui-user.sh

[terse plaza]

Yes, i have it

Im usin dream shaper 8

[ornate elk]

What settings did you used? Like resolution steps etc, hiresfix ?

[terse plaza]

My home internet keeps going on and off, ill come back when i can have a conversation

[ornate elk]

Alright

[ancient rapids]

I'm still getting the attribute lovram error at the moment. When I try to generate prompt

[ornate elk]

Which model do you use?

[ancient rapids]

It's on SD 3 and It's not letting me swap to anything else at the moment. If I try to click something else it just immediately goes back

[ornate elk]

SD3 isnt supported right now in Auto1111

You can remove it from the models folder until its supported

[ancient rapids]

Ah Okay

[ornate elk]

Then try an 1.5 based model like Dreamshaper v8

[ancient rapids]

Just to make sure I've done this correctly, Dreamshaper v8 should pop up on the "Checkpoint" menu on the top left.

So How do I activate embeddings?

[ornate elk]

Yep

Embeddings go into the embeddings folder.
And then in txt2img you have to click on the embeddings tab, and there select it

Then it will add the embedding trigger word to the prompt

[ancient rapids]

Alr, thanks

[magic hollow]

i feel like i did something wrong, but did i really....

[ornate elk]

Would need to see the whole cmd log

[magic hollow]

alright

im going to copy and paste it so ther might be a wall of text soon.

i think it turned it itno a text file so hoepult it can still be read?

[ornate elk]

Yea thats not an problem

[ornate elk]

Okay please delete the .zluda folder inside the stable-diffusion-webui folder

[magic hollow]

ok i did that what next?

[ornate elk]

Relaunch the webui-user.bat

[ornate elk]

Wait wait wait

[magic hollow]

what?

[ornate elk]

Why you have --skip-torch-cuda-test --opt-split-attention --disable-nan-check --lowvram in your webui-user.bat??

Thats not from my guide and breaks the install part

[magic hollow]

i thought it might work well with my decive but im most liekly wrong

ill correct it if you want me to do so.

[ornate elk]

Yea remove them please

Then delete the venv folder

Then relaunch the webui-user.bat

[magic hollow]

i feel dumb now.

[ornate elk]

Also add --medvram-sdxl

And --no-half-vae

[magic hollow]

also to make sure theses are the right commands? "--use-zluda --medvram-sdxl --update-check --skip-ort --no-half-vae"

[ornate elk]

Yep

Only these

[magic hollow]

alright, im not looking forward to downloading venv requiermetn and stuff but if i must then i will.

[ornate elk]

Most stuff is cached

But the torch version needs to be correct

So delete the venv to be safe

[magic hollow]

will do, also how are you so smart/know all this stuff if i might ask?

[ornate elk]

I'm doing stuff with Stable diffusion since October 2022, and I'm also working in IT. (And i'm PC, tech enthusiast)
So I saw the whole local webui history and encountered bugs and fixes with the community here.
Then I switched from Nvidia to AMD and wanted to get SD running again, so I learned the AMD stuff too over time

And yea then I started to wrote some guides to get AMD people easier into SD

I'm also playing around with different webuis, extensions and models and help here other users to fix bugs and problems

[magic hollow]

thats a cool sort of backstory you got going on.

[ornate elk]

Thanks, indeed ^^ if you want to know something about SD or some features, extensions, Installation or General PC problems, i mostly can help with that 🙂

[magic hollow]

cool, im manly just looking for diffrent uses to use sd for like ive thought of making something like a game or comic but ways to get the same chacrater seems to always change or not work.

[ornate elk]

Ah okay nice ideas

Getting the same character can be difficult

[magic hollow]

i know... sorta

[ornate elk]

But there are some extensions that can help. Mostly Controlnet + IP-Adapter

Btw does your SD launched now?

[magic hollow]

its still installing stuff and using chache, i guees its peak hour where i am.

its like alsmost 4pm where i am.

[ornate elk]

Its 1pm here xD

[magic hollow]

timezones are weird.

[ornate elk]

You don't need to get it running now. The first image generation will also take 15-40 minutes

Only one time

[magic hollow]

i know hopefully it goes well for me adn manges to work.

[ornate elk]

Yep I think it will work now

But I have to go to sleep now

[magic hollow]

ok goodnight to you.

[ornate elk]

Thanks, let me know how it went. If something doenst work we can fix it tomorrow. If the webui loads endlessly at the first image generation that's good, just let it load. Like I said 15-40mins can be normal

Gn and cyu

[grave condor]

I meant in general like I dont mind if its like Anything 1 minute under I just want to know what you mean specifically by "slow" cause if it doesnt effec timage generation speed and thats only my GPU I dont mind really if it takes a bit to start

[balmy shore]

is there a way to check xformers version and update if necessary

https://www.reddit.com/r/StableDiffusion/comments/1bh8vw7/what_is_latest_xformers_update/

ah this sort of answered it

[rancid nymph]

WHAT NOW I HAVE TO INSTALL SD3

[balmy shore]

wats that

[rich garden]

i'm having a glitch where all of my lora's won't load on the page anymore except for one of them (PING ME)

[rich garden]

ok nvm, i didn't know that some loras only appear when using different checkpoints

although is there a way to have all loras show for all checkpoints?

[terse plaza]

hello, sorry for the issues before. My wi fi is working now.

im gettin this error even when the only prompt i use is "couple"

OutOfMemoryError: CUDA out of memory. Tried to allocate 10.00 MiB. GPU 0 has a total capacty of 3.81 GiB of which 17.38 MiB is free. Including non-PyTorch memory, this process has 3.46 GiB memory in use. Of the allocated memory 3.32 GiB is allocated by PyTorch, and 42.95 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

and i have this in the webui-user

export COMMANDLINE_ARGS=" --xformers --medvram --no-half"

i checked, and at the bottom of the page it says xformers:N/A

[terse plaza]

this is with dream shaper 8

[quartz vigil]

I have an issue installing/enabling an extension, Tiled Multidiffusion. I install it successfully, checkmark it, and click "Apply and Restart UI", but when the UI reloads, it is unchecked again and does not appear on my main page. I've tried several browsers, restarting PC, and even using a config state JSON file that I manually edit to set it as enable=true, and it won't work. Says it is latest version when I check extension versions.

[old bane]

looking for free ways to improve my ram. i saw this video when i was looking it up. is there any techheads smarter then me say if this is a good idea? or if it will even work for Stable diffusion use. https://www.youtube.com/watch?v=ZsBDaTXYM2A

[teal sparrow]

Can i use this discord to generate images privatly from another image, or chat with a bot?

[latent sphinx]

I have 16GB VRAM, will I be able to run the sd3 8b model? Any estimates on its size?

[sick swallow]

I have an image not generate from stable diffusion, but I want to use it to make some changes. how can I generate seed number from the image?

[old ether]

Hi does anyone know how to fix this error when trying to use IP Adapter?

[fierce hollow]

I have 2 questions.

1. Is there an analogue of Hires. fix in the ComfyUI plugin
2. In the screenshot there is a function from ComfyUI called "scheduler" how can I find it in the usual SD automatic111 without this plugin?

[lost crescent]

Errmmm ...

[solemn moon]

okay
my body is ready for 3 on A1

[lost crescent]

How do you update ComfyUI without going via The ComfyUI Manager?

[solemn moon]

the update folder

[lost crescent]

I am using System ComfyUI - and there the Update.bat file only looks for windows_portable ComfyUI?!?!?!?

[viral knot]

can you update with git pull?

[lost crescent]

I will try ... seems that LoRA support has been added for SD3

[viral knot]

yes

[lost crescent]

Git Pull! Who'd have thought?! Thank you.

[viral knot]

🙂

[lost crescent]

How do I solve this?

The B:/custom_nodes file does not exist ...

[vocal burrow]

That s using your hard drive as ram. This is not a proper way to """"""increase your ram""""""", the title is very missleading. It s a good mechanism used but it s only used to prevent your pc from crashing when requesting more ram than it actually has.

you cannot either "download more ram"

the only way to improve ram is to buy more ram and reduce your ram footprint by closing whatever program you re not using.

Also SD runs on your Vram, not your ram.

your ram is only used as a middle man when transfering stuff to your gpu.

[vocal burrow]

1/ it s in a hidden folder so you probably didnt enable "show hidden folder" option in your explorer
2/ unless you really know what you re doing (no offense but you probably dont if you re asking stuff like that) then you should never delete stuff in .git folder
3/ try rebooting your computer and git pull again

[ornate elk]

Its just the model loading speed.
It can take a minutes (1-5) on a slow drive. And seconds on a fast drive

[ornate elk]

So xformers doesn't work.
Can you show your webui-user.sh ?

[lost crescent]

I am now updating - seeing as its taking a month-of-Sundays to d/load, it's prolly re-downloading all the very large model files ...

[ornate elk]

Seems like your using Forge webui.
It has multidiffusion already integrated.
But Forge is deprecated for normal users. Better switch back to Auto1111 and install multidiffusion there

[ornate elk]

There is a setting to always show all loras. But they still wont work with an incompatible model

[lost crescent]

I come to this as an artist with some tech-savvy; and I am always grateful for the help! 😉

[vocal burrow]

unlikely that git pull is redownloading any large model

git pull only download what s on the git repository

and there isnt usually big files in there.

[lost crescent]

My first computer (a BBC MicroComputer) I bought the week after Charles and Diana got married ... and that company which made the computer (Acorn Computers) is now the mega-gigantic ARM

[vocal burrow]

always happy to help as long as you re willing to learn a bit and not just looking for a robot giving you stuff to copy paste blindly :p

[lost crescent]

AS I say, it's already been 20 minutes after I used git pull - the Task Manager shows that the window is still active?!

[vocal burrow]

that s a very long time for a git pull... unless you have the same internet connection as back then, it should not take more than a minute at worse.

does the cmd have the focus ? try clicking the cmd and pressing space just to be sure

sometimes cmd processes pause themselves when they re out of focus (tldr)

@lost crescent ?

[lost crescent]

Yes, I think something's amiss

Ah, it d/loaded t5xxlfp16 - is why it took so long! All resolved 🙂

I've never known how to use SMUDGE to stop large file d/loads 🙂

[vocal burrow]

That s weird. it should nt be in comfyui, a1111 or whatever git repository.

the only explanation I have is that you git cloned the actual huggingface stable diffusion 3 medium repository and it somehow recursively called git pull on it.

git cloning model's repository is not something I d recommend. Just download the files you need from them and call it a day

[flint night]

Hello, I don't know if this is the right place, but I need help.

No matter which xl model I use, when the picture appears, my image looks like the attached one. I searched but couldn't find a clear answer. (Google Translate) (Nvidia RTX 4060 mobile)

[old bane]

basically i was trying to have my webui-forge to setup ponydiffusion to work and i have a crapy system but memory in my performance was maxing out and being unable to finish the setup.

[vocal burrow]

you re probably using a SD1.5 vae with an SDXL model

[flint night]

Yes
I'm ashamed that I didn't pay attention to this

thank you

[rich garden]

Ok thanks

[lost crescent]

I don't know, but anyway it has d/loaded and all is OK!

[lost crescent]

VAE or wrong Scheduler? Or too high/low CFG?

[ornate elk]

Forge will be deprecated. Better switch to auto1111

Then tell us what's your GPU, then I can give you the right launch args for the best performance

[honest tendon]

did you guys download all of this?

[ornate elk]

Go for the 4 or 5gb model.
And you also need to Download the text encoders

[honest tendon]

tyty

[honest tendon]

if i download the 15gb one can i just remove the clips and put it straight here

[ornate elk]

Yes

That also works with the 10gb one

[honest tendon]

wtf

its using my gpu so much

much more than cyberpunk

jeez

[azure magnet]

Hi I'm analyzing an comfyui workflow but it ask me to add some node, but i don't know what extension i need.
The node are this: Control Net Stacker
FileNamePrefix
CR Upscale Image
JWImageContrast
PlaySound|pysssss
Unpack SDXL Tuple
Upscale Model Loader
Pack SDXL Tuple
KSampler SDXL (Eff.)
DF_Text_Box
LoRA Stacker
Preview Chooser
Eff. Loader SDXL
ResolutionSelector
GlobalSeed //Inspire
Someone could help me?

[honest tendon]

can you just install missing nodes?

[uncut heron]

i am interested to know how did you manage to start it while not having nvidia gpu

[honest tendon]

here

@uncut heron

[uncut heron]

yeap this is new to me

[azure magnet]

Sorry i'm a noob so i need help. Where i can find that nodes? They are already in comfyui or should i install some extension?

[uncut heron]

i am using out dated guides

[honest tendon]

download comfyui manager

its in github

[honest tendon]

yeah, apparently ZLUDA is much more efficient

[uncut heron]

can you link it for me?

[honest tendon]

just google it

[ancient rapids]

I've encountered this ERROR on Dreamshaper 8

NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.

[honest tendon]

storage filling up quite quickly lol

[azure magnet]

I already has comfyui manager, so should i choose the option install missing custom nodes?

[ornate elk]

Hey checkout my guides from the pinned messages of this channel

I wrote install guides for everything:
(AMD, Nvidia, Zluda, Directml, Auto1111, Comfyui, Fooocus).
With more guides to follow.
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides

[ornate elk]

Better use the smaller model variant

[ornate elk]

Hey, you need to edit the webui-user.bat and change --no-half-vae to --no-half

[ancient rapids]

Ill try that now

[ornate elk]

Also never add --disable-nan-check, that won't fix anything

[ornate elk]

Update comfyui, then it should support the new node

[azure magnet]

I make a step forward but there is something i don't know how to solve. I need to add an extension but it give me confict. Do you know how to solve it?

[delicate citrus]

hi guys

my kaggle dont open the link to a1111, give me a error msgm
anyone know how can I corret it?
I put the port on firewall but yet keeps blocking =/

[ornate elk]

What extension?

[azure magnet]

I don't know how to manage this.

[ornate elk]

Okay, can't help much with that sry

[azure magnet]

https://tenor.com/view/nani-jjba-millks-jotaro-kujo-star-platinum-gif-24558285

thanks anyway, you're a life saver

[vocal burrow]

seems pretty much self explanatory to me. you re trying to install unmaintained extension and it s redirecting you to the new version