🤝｜tech-support

[ornate elk]

Can you show how it looks like in the config?

[stark yacht]

https://pastebin.com/b3Q6hKbm

[ornate elk]

Thats not the full log. What does it now?

[ornate elk]

Ah the folder name has a space in it. So you have to wrap the path in " "

[celest delta]

"AI ART" like that?

[ornate elk]

The whole path

[stark yacht]

do you need a log from the venv installation? or image generation

[ornate elk]

Image generation

[vocal burrow]

didn't follow the conversation, but having space in a URL/path is not a good idea.

[celest delta]

i did this and doesn't work either (i'm sure the solution is really simple, lol)

[vocal burrow]

might have to put it into quotes

yes something like that

[ornate elk]

Change the \ to /

In the path

[ornate elk]

Your las log showed that it tried to load the model. Did it worked?

[celest delta]

still doesn't work :p

i'll try removing the space in the folder name and remove the quotes

nope, still not working 💀

[ornate elk]

Will check my config when I'm at home

Then I can show how it should look like

[celest delta]

thanks

[stark yacht]

fixed

[celest delta]

mmmh

the models showup in comfy workflow, but not in the normal generate tab

[zenith seal]

hello

i got a issue when starting my ui today :/

Seems that I can't connect to it

[zenith seal]

How did you fix it i probably have the same problem

[stark yacht]

.

and delete env folder

[ornate elk]

What's your GPU?

[zenith seal]

6800xt

[ornate elk]

Okay, can you show the full cmd log?

[zenith seal]

when i strat the iu?

start

[ornate elk]

The extensions Mov2mov, ebsynth and booru2prompt show errors.
I would delete these and then delete the venv folder and relaunch the webui-user.bat

Also your on an older hip SDK version

[zenith seal]

oh okey how do i change the hip version?

[zenith seal]

i delete them from the ext folder right?

can't find the ebsynth :/

[ornate elk]

Ah true the ebsynth is inside mov2mov

[zenith seal]

oh i just delete it

im trying now

[ornate elk]

Danbooru prompt dont need to be deleted

Only the booru2promp

[zenith seal]

it's not selected

[ornate elk]

Ah ok

[zenith seal]

my mouse was just on it ^^

okey reinstaling venv rn

let's see

[ornate elk]

Scroll down to:
How to upgrade HIP SDK and Zluda:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides#troubleshooting-tips

Hip sdk 6.1 is much better for 6000 and 7000 series

[zenith seal]

so i need to do all these steps right

[zenith seal]

crashed the ui again

[woeful falcon]

Hi CS1o, I hope you're available now 🙂

I got this upscaler problem, still persistent. Before experimenting more, I guess I need to rent a GPU T4 instance on GCloud. It takes too much time currently.

[ornate elk]

Hmm looks normal

How much ram do you have?

[zenith seal]

16

[zenith seal]

do i also unistall this?

[ornate elk]

nope

only hip sdk stuff

[zenith seal]

okey i just did

launching the installer

[ornate elk]

ah okay, then you also need to increase the windows pagefile, to be able to run larger models

[zenith seal]

but i was doing it normaly yesterday :/

[ornate elk]

did you load an other model maybe sdxl or pony?

[zenith seal]

nop didn't add anymodel

[ornate elk]

hmm very strange then

[zenith seal]

well i gonna do the hip sdk thing

and see

[ornate elk]

yea maybe its a hip sdk 5.7 load bug

or maybe your C drive got fuller and that srhunk the pagefile

which can also cause the loading error

[zenith seal]

roc 5 or 6?

oh 6 for sdk 6.1?

[ornate elk]

exactly

[zenith seal]

tbh i have this but im on the ui

[ornate elk]

yea it will take 15-40 minutes after you pressed generate

[zenith seal]

oh yeah right ^^ thanks a lot

i will tell if it work

[zenith seal]

Well for some reason my first gen made my computer freeze :/

i mean the one after the compiling one

[ornate elk]

did you used an pony or sdxl model?

then you still have to increase the windows pagefile
https://www.tomshardware.com/news/how-to-manage-virtual-memory-pagefile-windows-10,36929.html
Enable it only for the C drive and disable it for any other drive.
Set it to customized: 16000 min and 24000max.
Then restart the PC.

[woeful falcon]

Okay, the reason why 3.12 is problematic is that some dependencies do not support 3.12 but > 3.8 🙂

Requires-Python <3.12,>=3.8;

I will need to create venv for 3.10 and start again.

[ornate elk]

3.10.11 has the best compability

[woeful falcon]

I've found this project for clothing, it looks promising but I see a lot of complicated workflows on YouTube. Do I really need that kind of a workflow? https://github.com/frankchieng/ComfyUI_MagicClothing

[celest delta]

@ornate elk hey again, sorry to bother you, i didn't think to ask when you would be home to check my stuff (as said earlier, the workflow tab detect the models, but not the normal image generation tab)

[ornate elk]

Thats for SwarmUI

[celest delta]

not sure what i'm supposed to do here

it's already set that way

[ornate elk]

you have to input the paths

to the folder

[balmy monolith]

I'm trying to load the comfyui extension into auto1111, but it gives an error

anyone able to assist with this please. I have programmed python before, but this is beyond my experience with python

I opened auto1111 and installed the comfyui from the extensions page.

google doesn't tell me much about the error

[ornate elk]

hey, can you show the full log?

[balmy monolith]

where can i find the log file, will it be somewhere in the comfyui folder?

[ornate elk]

no i mean just copy the cmd log

also its better to use comfyui as standalone webui instead of using the extension for auto

[balmy monolith]

mabye this

I'll download it from github and try that

[ornate elk]

yea it will work much better than the extension

[balmy monolith]

can that be used alongside auto1111?

[ornate elk]

yep you can also link the models

Here is a Guide for every webui and also how to use the models from autos directory in comfyui
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[balmy monolith]

I'll read through that, thanks

[ornate elk]

no problem

[celest delta]

how can i effectively remove details with inpainting, like hair? Like i inpainted a character with shorthair on the image, but the longer hair is still visible, i trying inpainting it but it stays there, any advide? (prompt, or settings)

[ornate elk]

the thing with the clothing ai stuff is pretty difficult

you probably only can use these big comfyui workflows to get it work or the VTON project

[celest delta]

uuuugh, why am i so dumb, i don't get it, i successfully inpaint a character with lora, then proceed to do the second character and inpainting with another lora, and suddenly it doesn't work anymore

it just redo the same character like i didn't inpaint

[ornate elk]

to get 2 lora chars together you can use an extension like regional prompter

[celest delta]

i tried and it's too complicated to make it work properly, i find that just inpainting an image is easier, but why it works one time and suddenly it doesn't? i just don't get it

[ornate elk]

make sure you always set the Inpaint Area to "Mask only"

then set the resolution to a sqaure

[celest delta]

okay i think the thing just glitched and i need to restart, it doesn't even modify the image at all now, regardless of what i inpaint

[woeful falcon]

Thank you! I guess I need to keep grinding. Trying the models in a simple matter and get to big workflows.

I've seen some CatVTON videos, this seems to be the easiest but it's interesting that VTON is dominated by Chinese researchers 😄

[zenith seal]

hello again

mhh im having cuda error every time now :/

nvm didn't launch this time

[woeful falcon]

It doesn't look like a CUDA error. It cannot download from huggingface due to permissions

[zenith seal]

download what from hugging face?

[woeful falcon]

I don't know, the URL it's trying to download is written in the error report: Repository Not Found for url: https://huggingface.co/None/resolve/main/config.json

None is the culprit here. It seems like a weird programming error. I don't really know how to debug it

[zenith seal]

Yeah thats pretty strange i have no clue why i need hugging face for this

also one error say it's about a model

[woeful falcon]

Maybe it's trying to download a model? I've seen plugins automatically downloading a model. Seems like stable-diffusion-webui-amdgpu is the problem here.

[zenith seal]

i mean yeah probably it's the ui folder

[zenith seal]

i think that caused the problem,i mean i can't load a model without crash now :/

maybe cause my models are on the d drive

[ornate elk]

Oh you can try revert the pagefile setting to Automatic for the C drive and then restart the PC

But normaly 16000min and 24000max is good

Which model did you tried to load?
It doesn't matter where they are located.

[ornate elk]

Thats a "to less memory" error

[ornate elk]

That error can be ignored as it get fixed with the next Auto1111 update

[woeful falcon]

Oh I see. It's an intermittent issue then

[zenith seal]

i did load a illu model

it worked now but i can use high resolution for some reason

[white thorn]

hey guys, im new on comfyUI and got myself a custom workflow but I dont get there is a node called SamplercustomAdvanced where it load pretty fast the first time but if im doing the exact same thing again take forever. any idea ? feels like it has something to do with unloading vram maybe .. something like that ?

[ornate elk]

You can or you can't?

[zenith seal]

can't sorry

i mean is still have some frezze when i do it

[ornate elk]

how much free space do you have on C ? and which model do you try to load?

[zenith seal]

40 g

[teal creek]

I cannot find the COMMANDLINE_ARGS file, did i miss smt?

[ruby yoke]

it's not a file. in the configuration file there's a variable called commandline_args. add those flags to that line

[teal creek]

where do i find the config file?
by looking around i cant seen to find it

[ornate elk]

Its in the webui-user.bat file

What's your GPU?

[ornate elk]

Ah okay then the CPU mode should work

Or directml if your GPUs has ,4gb vram

For CPU mode add:
--skip-torch-cuda-test --use-cpu all --no-half --precision full --opt-sub-quad-attention

[teal creek]

Like this?
Does it need a space betwen the ARGS= and the --skip?

[ornate elk]

The space between = and --skip doesn't matter

[primal wigeon]

In general, how do you know when you'll need a lora and when you won't

[ruby yoke]

in general, try with just the base and/or fine tuned checkpoint. if you absolutely can't get something that way, then you go looking for a lora to fill in the gaps. or if you want to make images of something the AI can't possibly know about - like your dog or yourself.

[primal wigeon]

but it would know what a dog was?

[quaint basalt]

Hello everyone
how can I use SD3.5 large for sketch to image?

@vocal burrow, @rocky ore

[terse birch]

using forge, trying an illustrious model for the first time. When I inpaint the image gets all desaturated though. any ideas on how to fix it? here's a picture of what it's doing

model is madly mix btw

[ornate elk]

Make sure you set the Inpaint area to "mask only"

[terse birch]

shoulda posted this first, heres my settings

[ornate elk]

Did you Inpaint with the same model?

[terse birch]

yeah

[ruby yoke]

sure. but not your personal pet dog

[olive lintel]

Has anyone managed to get authentication free port forwarding to work on vast.ai?
I want to forward a local port (via ssh) to vast.ai A1111 instance so I can use SD Photoshop plugins.

I have that working on ThinkDiffusion but TD is 4x the cost of vast.ai

[white thorn]

Hello there I have a problem with a comfyui workflow i got online. lets say i queue 10 batch count, first render at 2s/it ( RTX 4080 16Gbvram) but after that other render goes up to 50 or even 100s/it. I tested other workflow with no issue. Is that a way to simply deactivate some part of the workflow to debug this and find out wich node is involved?

[ruby yoke]

1. drop your batch down to 1 and 2. make sure nothing else that will use your GPU is running (like games, discord app, and so on) and then see how things function

if you still have issues, then start tearing the workflow apart. it's far more likely, hoewver, that there's processing going on that takes a lot of resources

[white thorn]

its probably controlnet but I need to verify it.

[ornate ruin]

Hi:)
Im still using Automatic1111 and SD1.5 checkpoints. The other day I learned that Pony is something completely different from SD, and so is Flux right?

[ruby yoke]

you're right, yes. pony is a base model, so is flux. and you can't use them with your sd.15 stuff

[ornate ruin]

I wonder what the best base models are. So far I mostly preferred SD1.5 over SDXL. So to 'upgrade' my generations I could maybe try out SD3, but I barely meet the requirements for generations already

[ruby yoke]

that depends on what you want to do.

[ornate ruin]

Does the improved model also consume more resources on my laptop like newer videogames would for example? Lol

[ruby yoke]

um yes and no. some models are larger than others. the issue is how much vram you have. and laptops are not designed to do heavy graphics work.

[ornate ruin]

8gb VRAM

So not very much sadly, RTX4060 mobile

[ruby yoke]

you can get the fp8 and gguf models for SD3.5 and they would run fine

[ornate ruin]

fp8 and gguf, doesn't tell me anything but there should probably pop up something about it when I google it. And it all works fine with automatic1111?

[ruby yoke]

every version of stable diffusion, however, like all models, speak a different language. my SD1.5 prompts work fine with SD3.5 - but not too well with sd 2.1 - etc

[ruby yoke]

no. auto1111 doesn't have support for a lot of the models out there

if you're going to use something other than sd.15, firs make sure auto1111 supports it

[ornate ruin]

Can you recommend me an alternative to a1111 that. Could try for the newer base models like 3.5?

Probably Forge? That's something I've read a lot about as well.

[ruby yoke]

what i would recommend would be that you install SwarmUI and allow it to deal with all the technical stuff and just generate inside it. i can give you the link to their discord if you want, but we also ahve a #🐝｜swarm-ui channel

[ruby yoke]

if you'd like the swarm discord invite link, i'll need to DM it to you as we can't post discord links here

[ornate ruin]

Checking out swarm right now first, thanks!

Also - is there somewhere a list of base models that are supported by various UIs? Like automatic1111, can it only run XL and 1.5?

Since there are so many I don't know which I even can/should use before I just try out every single one 😅😃

[ruby yoke]

if you run swarm, that's probably not an issue

[ornate ruin]

Currently still using a1111. But year.. I'll just try out swarm, see if I get in installed and everything.

[winged dust]

Using comfyui on an AMD rx6700 xt, what arguments can I use that could help me generate faster? Appreciate it

Right now I just have —directml and —normalvram

[ruby yoke]

start by reading the pinned posts and guides in this channel

[winged dust]

Good idea, my fault OG

[sleek kiln]

if i want install animateddiff it says JSONDecodeError: Expecting value: line 8 column 1 (char 7)

what can i do here

[winged dust]

I'm using the Efficient loader and Ksampler (Efficient) nodes in comfyui, keep getting this issue when tyring to generate images

TypeError: Cannot handle this data type: (1, 1, 3), <f4

[ruby yoke]

you should probably ask this in the #🧣｜comfy-ui channel

[sleek kiln]

i installed this one now

but i dont seee it here

[ornate elk]

You would need to use ZLUDA with comfyui or any other webui

Instead of directml

Its much faster and allows for larger resolution

In the pinned tabs you find the link to my AMD Guides

[ornate elk]

You need the version for forge:
https://github.com/continue-revolution/sd-forge-animatediff

[teal creek]

No real clue what i may have done wrong but i'm getting these errors

i've downloaded some models from civitai and placed them "D:\AI\stable-diffusion-webui\models\Stable-diffusion"

[winged dust]

My images come out like this when I use VAE, but are fine without vae. Is there a way to fix this?

[ornate elk]

What's your GPU?

It says it didnt found an nvidia driver

[ornate elk]

Try an other VAE

They only work with a model if they both are the same base type

[winged dust]

oh like if both are sdxl?

thank you

[teal creek]

this one

maybe my pc is too low end for it :^(

[ornate elk]

Yep thats not that good for stable diffusion.
You can only use it in the slow CPU mode

You would need a PC with a dedicated GPU

To use the CPU mode you need to edit the webui-user.bat and add
--skip-torch-cuda-test --use-cpu all --no-half --precision full --opt-sub-quad-attention
At the line commandline_args=

[teal creek]

got it

another question, i tryied to dowload some models from* civitai
Where do i place them?

[ornate elk]

If they are 2gb or larger, the go into models/stable-diffusion

Smaller files like Loras go into models/lora

[winged dust]

Anyway I can stop this issue?

RuntimeError: Could not allocate tensor with 528482304 bytes. There is not enough GPU video memory available!

I have 12 GB vram

rx 6700 XT GPU

[ruby yoke]

what's your batch size?

[winged dust]

just 1 image

30 steps, 896 x 1152

[ruby yoke]

how much diskspace on your system drive is free?

[winged dust]

like amount of GB I have left? 55 Gb

[ruby yoke]

read through the pinned messages and guides in this channel and make sure you have that AMD gpu set up right

[winged dust]

Should have followed everything right. I didn't follow the ZLUDA guide if that makes a difference?

ZLUDA didn't work for me so I just went with the previous guide

[ruby yoke]

it makes a difference.

[winged dust]

That big of a difference though? I've seen someone with a worse gpu than mine generate images

[ruby yoke]

okay. you'll probably need to wait for @ornate elk to log back on. my understanding is not as good as his for amd gpus

[winged dust]

oh alright

[woeful falcon]

Cursor AI tells me that nvidia-driver-545 is the most compatible driver for CUDA 12.4 and PyTorch supports CUDA 12.4 out-of-the-box. Do I need to lower CUDA version or PyTorch for ComfyUI? What are the best versions to use?

[ornate elk]

Yes it makes a big difference

The vram management is much better with ZLUDA.
Directml will always use the Max amount and then goes out of memory.
Zluda is faster and allows for larger resolution and bigger models.

I wouldn't use comfyui with 12gb vram, but you can.
Better use Auto1111 or Forge with Zluda for the best experience.

[sleek kiln]

i have it now but doesnt show up either

[ornate elk]

You first need to delete the other one and then delete the venv folder and relaunch

But the extension didnt got updated since 8 months so maybe it wont work too

Reading the issues on github, it seems it won't work at all with Forge right now

[sleek kiln]

mhh how guys then create AI Videos if it doesnt work

im getting many of these

*** Error loading script: animatediff_infotext.py
Traceback (most recent call last):
File "D:\A1111\Data\Packages\Stable Diffusion WebUI Forge\modules\scripts.py", line 525, in load_scripts
script_module = script_loading.load_module(scriptfile.path)
File "D:\A1111\Data\Packages\Stable Diffusion WebUI Forge\modules\script_loading.py", line 13, in load_module
module_spec.loader.exec_module(module)
File "<frozen importlib._bootstrap_external>", line 883, in exec_module
File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
File "D:\A1111\Data\Packages\Stable Diffusion WebUI Forge\extensions\sd-forge-animatediff\scripts\animatediff_infotext.py", line 6, in <module>
from scripts.animatediff_ui import AnimateDiffProcess
File "D:\A1111\Data\Packages\Stable Diffusion WebUI Forge\extensions\sd-forge-animatediff\scripts\animatediff_ui.py", line 12, in <module>
from scripts.animatediff_mm import mm_animatediff as motion_module
File "D:\A1111\Data\Packages\Stable Diffusion WebUI Forge\extensions\sd-forge-animatediff\scripts\animatediff_mm.py", line 5, in <module>
from modules_forge.unet_patcher import UnetPatcher
ModuleNotFoundError: No module named 'modules_forge.unet_patcher'

[gritty storm]

hello frineds
what version of stabble you would reccomend me? Optimal for functionalit maybe

[ornate elk]

In auto1111 it works

[ornate elk]

Hey, what's your GPU?

[sleek kiln]

i have the automatic1111 :/

[ornate elk]

You have Forge

Its a fork of Auto1111

[sleek kiln]

i got the feature now

but here another problem i cant choose any models in controlnet even if i download it

which path i need to safe the downloaded models ?

[ornate elk]

Models/controlnet

And they need to be the version with the same base model

For sdxl you need sdxl Controlnets

[chilly knot]

VRAM leak when working with ipadapter automatic 11111, is there any advice on how to fix it?

when i try use lowvram the image starts to generate but at the end the same out of memory

4gb vram

sd 15

before it works

[chilly knot]

max_split_size_mb what i need set in there

what number should I put?

[chrome rampart]

Hi all and hope doing well. Question in forge (locally) today I keep getting in lets say last step> connection is timing out. I see build of image but right at the end errors comes. Any ideas

[chilly knot]

check cmd

[chrome rampart]

yes i did but will do again and maybe restart fully to make sure> thks for now

[chilly knot]

try different browsers

[chrome rampart]

it seems related to shuttle-3-diffusion model the other flux-related version work. Will need to check more.> thks

[ornate elk]

You shouldn't use it

[ornate elk]

Make sure you enable lowvram at the Controlnet

[chilly knot]

its not help me?

[chilly knot]

i set low vram in bat file

its different/

[ornate elk]

Yea lowvram in bat helps too but you have to check "lowvram" at the Controlnet Settings

[chilly knot]

thx i check it

before i use ipadapter without lowvram(

what i change or update

last i install cudnn

[ornate elk]

Cudnn?

How?

[chilly knot]

Downloading cuDNN for Windows
cuDNN is available at https://developer.nvidia.com/cudnn. Click on the green buttons that describe your target platform and choose one of the following installer formats:

[ornate elk]

Why did you need it?
Normaly your fine with the latest normal nvidia drivers

[chilly knot]

i dont know i saw tutorial on youtube

my cuda provider not works in facefusion

and i do step from tutorial

[ornate elk]

What's your GPU?

[chilly knot]

1050 ti

now cuda provider works

but i think i broke something🙂

[ornate elk]

You can try to delete the venv folder and relaunch the webui-user.bat

[chilly knot]

on 99 %

works when i set split size to 32

set PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.9,max_split_size_mb:32
set COMMANDLINE_ARGS=--theme=dark --xformers --lowvram --no-half-vae

dont know what help split size or lowvram in cn but thx😎

oh second generation with same problem .I'll try different options

[ornate elk]

what are the other txt2img settings ?

resolution etc

[chilly knot]

how use think problem that in bat file i not set venv folder

i have facefusion icon

when start automatic

[ornate elk]

thats not a problem and shouldnt bet set

[ornate elk]

thats because you didnt closed your browser or cleared your cookies between using facefusion and auto1111

[true sphinx]

for some reason im getting this error and SD wont open

[white thorn]

Hello, is anyone familiar with controlnet on flux ? im trying to use an openpose with a pose sheet image, but I cant get consistant result. sometimes i get faces without hair, somtimes hair without faces, i tried to play with the strenght but it just make thing worse

[ornate elk]

delete the SD-CN-Animation extension, and then the venv folder
then relaunch the wbeui-user.bat

[true sphinx]

any idea why these extension just break automatic 1111 and thanks for the help

[ornate elk]

np, and because the extension was updated the last over a year ago

[true sphinx]

god damnnn i thought i was good about updating these extensions but apparently im not

[ornate elk]

Yea but if the dev doesn't update them its over at some point xD

[true sphinx]

yea guess i really gotta follow that more closely

do you have any reccomendations for other working animation SD tools cuz i liked using that

[ornate elk]

AnimateDiff would be one

[true sphinx]

perf thanks!

[winged dust]

I tried following the pinned guide for ZLUDA, but it didn’t work for some reason I can’t remember so I ended up deleting it and just following the older guide for directML

[ruby yoke]

directML is the problem, though

[winged dust]

Yeah but zluda didn’t work either

[ornate elk]

You can ask me if you have an issue with the ZLUDA guide

We can easily fix it

[winged dust]

Alright, I’ll following the guide again. I do remember that the very first issue I encountered in the guide was where it said launching the webui-user bat file would result in an error but that’s okay, however I didn’t get the error and it loaded up the website alright. Then I got an error, something like “failed to load model, will retry using slow method”

[calm cipher]

Trying to use SwarmUI control nets. I only have one control net downloaded, but my preprocessor looks like this. But anytime I try to use any of them I get an error like the one shown. Does anyone know how I download these pre processors so they actually work OR remove them from the list so Swarm stops lying to me?

[winged dust]

@ornate elk before I start the zluda guide for automatic 1111, should I delete this folder since I need the 6.1 instead?

[ornate elk]

You need to uninstall hip sdk) rocm 5.7

[winged dust]

Gotcha, that probably caused an issue last time

[ornate elk]

You want 6.1

[winged dust]

Yeah I just had both 6.1 and 5.7 last time

[ornate elk]

And start with the Automatic1111 with ZLUDA guide.
Its easier to get working

[winged dust]

Will do, thank you

[ornate elk]

NP, ask me if any step isnt clear

[winged dust]

Ok so I launched the bat file and the checkpoint finished downloading, got these errors, is this normal?

[ornate elk]

can you show the full cmd log?

but yea they look normal

[winged dust]

Oh i closed the terminal, want me to launch the bat file again?

[ornate elk]

its fine when you first do the guide

and then show me any errors ^^

[winged dust]

gotcha, thank you

[ornate elk]

if the webui still opened normaly its good

[winged dust]

oh ok yeah it did

[winged dust]

Where it says to install AMD HIP SDK 6.1.2, it says "Don't check the pro driver", just to be sure, this is what that means?

[ornate elk]

exactly

[winged dust]

Thank you

[winged dust]

Uh so I finished the guide and this is what showed up when I launched the bat file

that's the whole log, from start to end of opening the bat file

[ornate elk]

looks normal

does the webui opens?

[winged dust]

yeah it opens but i got this issue when trying to generate images

RuntimeError: Input type (float) and bias type (struct c10::Half) should be the same

[ornate elk]

which model did you tried?

[winged dust]

the default one

[ornate elk]

with the default settings?

[winged dust]

yeah nothing changed

[ornate elk]

ah i see

it installed the wrong torch version

[winged dust]

oh

[ornate elk]

to fix this:
opens up a cmd and run
pip cache purge
Then delete the venv folder from the stable-diffusion-webui folder and relaunch the webui-user.bat

[winged dust]

Will try that, thank you

[ornate elk]

np

[plain pond]

how u change the theme into dark mode?

[winged dust]

it's just default for me

[plain pond]

ow

[ornate elk]

setting windows theme to dark

that applies it to the browser + the webui

you can also add --theme-dark to the webui-user.bat then its only for the webui

[plain pond]

wow thank you..

[plain pond]

yea i need to try later.. thanks

its work thanks

[winged dust]

Hey, can you help me with this problem?

[ornate elk]

when do you get this?

can you copy the cmd log?

[winged dust]

I get it when I try to generate images

[winged dust]

Sure

Oh wait

[winged dust]

It works now after restarting webui

[ornate elk]

ah okay good!

[winged dust]

I can actually generate 896 x 1152 images now 😮

and faster than when I generated 512x512 images

Thanks so much man

[ornate elk]

yea and you can easily upscale them too now

no problem

a good extension you should install is the Tiled VAE extension.
That helps when upscaling

[true sphinx]

Does anyone have experience with both AMD and Nvida cards for SD im looking to upgrade soon and those Radeon Cards have insane vram in comparison but I am more familiar with Nvidia cards and when i google this issue its hard to find a clear answer
like the 7900xtx has 24 gigs of vram compaired to a 4080 super with around 16 plz let me know if i should change the text channel for this question as well

i dont want to break the rules

current gpu is 3070ti

[ruby yoke]

should ask this in the #🐝｜swarm-ui channel

[ornate elk]

i have the 7900XTX
Its a beast for Gaming. But for AI stuff its not as good as a 4080.
Most AI stuff is made with Cuda Support which is from Nvidia.
With ZLUDA, AMD cards have a tool to emulate the cuda support on Windows.
On Linux you have native ROCm Drives which have the best Performance for AMD.

Even i knew that AMD cards are not the best for AI, i bought it mainly for Gaming and for the 24GB Vram. And i didnt like Nvidias Price Policy.
With Zluda on windows the performance in stable diffusion is still great. I can use mostly everything and Flux and SD3.5 too.
For LLMs the AMD Cards are beasts. With Ollama its insanely fast.

So if you mainly focus on AI, go for a Nvidia (4070 ti super or 4080 super)
If Gaming is your main Focus and AI Stuff just a hobby and you like to tinker, or liking Linux then go For AMD.

[chilly knot]

network error install torch , i download it torch-2.0.1+cu118-cp310-cp310-win_amd64.whl
where i need put it for install from automatic cmd

[ornate elk]

it goes into the main folder where the webui-user.bat is

[chilly knot]

you sure?

i put it in folder where bat

and automatic again try download it

[ornate elk]

whats the error you get?

[ornate elk]

yes you have to manually install it with 2 commands

in cmd

it wont get used automaticaly

[chilly knot]

where i need put it for instal manually

[ornate elk]

where i said to

[chilly knot]

after this i need go to script

.\activate

and go to root folder

[ornate elk]

then click into the file explorer bar and type cmd and hit enter.
then run these 2 commands:
venv\Scripts\activate.bat
pip install torch-2.0.1+cu118-cp310-cp310-win_amd64.whl

[chilly knot]

it take it from root folder ?

ARNING: There was an error checking the latest version of pip.

pip updgade need first

ERROR: Wheel 'torch' located at "folder where whl file" is invalid

[ornate elk]

then you can try delete the venv and let the webui redownbload torch

[chilly knot]

i cant

network problem

[ornate elk]

change the dns server then

your download speed looks good

[chilly knot]

yes it works 30 second)

[winged dust]

Just got it. Do I use it by sending the image I want to upscale to img2img or can I upscale it from the text2img page too?

[delicate citrus]

@ornate elk can you recomend me a processor to accompany the RTX 4060 Ti 16GB VRAM graphics card?

[ruby yoke]

best intel you can afford

[ornate elk]

you can upscale it in txt2img by enabling the Hires Fix.
Set the Hires Steps to 10,
Denois to 0.5
Upscale by 2
and as Upscaler choose Resrgan4x+
Then also Enable Tiled VAE and let it on defautl settings

[winged dust]

Ohh got it. Thank you so much for all the help 🙏

[ornate elk]

For Gaming or workstation stuff (rendering etc)

[delicate citrus]

for AI and gaming

[ornate elk]

okay and whats your budget? you also need a new mainboard for a new cpu

[delicate citrus]

@ornate elk I don't know, you can recomend me one and I will check... maybe 510 USD? some ryzen 9? I don't understand to much about computer parts

yes, the mainboard too

[chilly knot]

cs10 when automatic download torch where he put it ?

tmp folder

[delicate citrus]

intel is better then ryzen?

[ruby yoke]

yes

[delicate citrus]

can you suggeste me one?

[ruby yoke]

the best one you can afford. it's going to be running your system, your gpu is going to be doing doing the work with the AI - so you need a good cpu that works with your mainboard and runs your system well. and you want an nvidia GPU with at least 16 gig vram

[delicate citrus]

ok. With 510 USD can you give me a name of a intel preprocessor? will helop me to compare and know what can I buy

[ornate elk]

Nope

AMD ryzen is the best you can buy as CPU

Intel did really bad stuff with their latest 13 and 14 gen CPUs.
And now their new Ultra core CPU is worse than the 14 gen and worse than AMD in Games and costs even 150€ more xD

Wouldn't recommend anyone buying Intel CPUs currently

Only if you really need 24 cores for Blender or Cinema4D rendering etc

Also by going for an AMD CPU now you can use the same Mainboard for their next CPU release instead of needing something new.

The Ryzen 9800X3D is the best Gaming CPU right now

[ruby yoke]

it has serious issues for people that don't know what they are doing.

[ornate elk]

Not at all lol

You don't need to do anything with your CPU

[ruby yoke]

yeah. becaue of iGPU

[chilly knot]

I bought Intel just out of interest, all my life I bought only AMD

[ruby yoke]

every day, several times a day - "why is my cpu, not my gpu..."

[ornate elk]

Thats not an issue lol, you can deactivate if needed

[ruby yoke]

i've always bought amd too - but i would not have an amd with this work if you paid me

i't snot worth it

[ornate elk]

Yea but this can be fixed in seconds. And not by buying a overpriced CPU with less performance

[ruby yoke]

https://www.newegg.com/abs-aqa14700kf4060ti16g-stratos-aqua/p/N82E16883360436 go to the right place, full systems aren't that expensive

[delicate citrus]

@ornate elk this one?

[ornate elk]

Thats the last gen 16 core version

[delicate citrus]

it's is good, so?

[ornate elk]

Thats the new one:
https://www.amd.com/en/products/processors/desktops/ryzen/9000-series/amd-ryzen-7-9800x3d.html

[ornate elk]

Its a good CPU yes but a 7800X3D would be cheaper and enough too

If you can't find a 9800x3d

[ornate elk]

Yea but the intel 14 gen has the known problems that can only be fixed with a BIOS update.
If the manufacturer didnt did that its worse for people to buy it blindly

[chilly knot]

sometimes you read a technical forum about processors and are scared to buy something after that, but when you buy it it’s not so bad

[delicate citrus]

I will take this one, thanks a lot 😄

[ornate elk]

No problem, its the best you can get. It got released a week ago.
You will need an AM5 Mainboard and DDR5 Ram for it too.
For Mainboards go with a B850 or B650.
For RAM the CPU likes 2x16 or 2x32 GB (up to you), 6000 Mhz CL30-32

[ruby yoke]

haven't had a single problem

[delicate citrus]

nice, thanxs

[ornate elk]

Thats good then. But a lot of people did. Thats why they released the BIOS update.

[ornate elk]

NP, you can cool the CPU with something like a BeQuiet Dark Rock 5 or Pro, Fan or any 360 Aio watercooler.

It doesn't get hot so any mid tier fan should be enough

[delicate citrus]

perfect, help me a lot too ^^

[ornate elk]

NP, and for the CPU as its so new you maybe have to preorder or just wait a few weeks/months until the shops have enough of them in stock

[delicate citrus]

I found that one here in Brazil to buy on internet

[delicate citrus]

what is this? "6000 Mhz CL30-32"

[winged dust]

OutOfMemoryError: CUDA out of memory. Tried to allocate 2.21 GiB. GPU

Get this error when trying to generate images

was working fine earlier

tried a different checkpoint and thats when it started happening i thik

[ruby yoke]

try a reboot, might have something taking up memory and not releasing right

[winged dust]

reboot pc or webui?

[ruby yoke]

if it was me, i'd reboot my system

[winged dust]

Will do that]

[winged dust]

Forgot to say but that worked

Thank you 🙂

[remote timber]

hi - so, im not very good with computers, so feel free to tell me if this is the wrong place to be asking? but im trying to run SV3D, and having some trouble setting it up. would anybody be able/willing to help me out a bit?

i downloaded this repository, and this one too - im not sure how to make them properly work together though? the second repository also seems to be a tutorial for just 4d, whcih i dont want

so im not sure if i need to download that repository too?

[ruby yoke]

this is the right place, but I dunno if we have anyone here that runs SV3D

[remote timber]

oh. really?? isnt this server really big?

would you be able to help at all? or no clue really

i really just dont even know how to start install

[ruby yoke]

yes, but most people are just making images or videos. i create 3D but usually use blender

give me a minute to look at the repos please

[remote timber]

im a painter, and trying to generate the back side of an image.. is there something that would be more painless?

thank you - sorry for the bother

[ruby yoke]

the back side of an image? explain, please

[remote timber]

i have an image, and i want a 3d model for me to look at the back of what it generates

or even just a gif or something

[ruby yoke]

you're not going to be able to do that with SV3D - you'd put in a prompt and it would make you a mesh object. so you're wanting help visualzing the back side of an exsiting image?

[remote timber]

oh... id misunderstood :(

yeah i have an existing image

[ruby yoke]

would you be willing to discuss this with me in DMs?

[remote timber]

absolutely!

[ruby yoke]

dm sent

[tough patio]

Is this the right place to get tech support?

[ruby yoke]

depending on what you need tech support for, yes

[tough patio]

I am using Stability AI's API.

[ruby yoke]

what seems to be the problem?

[tough patio]

I bought some credits.
After some testing I ran out of credits.
And it started throwing errors that I don't have enough credits.

So insted of getting the results, I was getting the errors (of not having the credits).

But when I looked after about 12-15 tests without credits, there were -192 credits.

So I was charged, even tho I didn't get the results (and instead got the error).

[ruby yoke]

okay, that is, unfortunately, a customer service issue and not something we can solve on this discord. you'll need to contact customer support via the contact form here https://stability.ai/contact

[tough patio]

Alright, thank you.

[ruby yoke]

you're welcome.

[tough patio]

I thought you were the tech suport.
Sorry.

[ruby yoke]

not as far as fixing the issue you've got goes, no. i'm sorry, you'll ahve to talk to customer service

[tough patio]

Yeah

[ornate elk]

Its the RAMs frequency and speed

[fathom oasis]

Hey So Can i Run stable diffusion smoothly on Mac Mini m4 16GB variant?

[ornate elk]

Yep shouldn't be a problem

For Mac there is an all in one installation script:
[MAC] Installing Automatic1111 or Forge webui on Apple MAC OS:
https://github.com/viking1304/a1111-setup

[fathom oasis]

Image and text ai generation simultaneously?

[ornate elk]

I don't think that will work as both use the vram or in this case the System RAM

Maybe if you have 64gb

[fathom oasis]

If i had 24GB, it works?

[fathom oasis]

Ahh nvm it needs that much

[ornate elk]

Yea only run one ai tool at a time

Maybe it can work if you use a smaller llm model

I can try that later maybe.

[ornate elk]

24gb would be better as its shared memory so you can use larger models

[fathom oasis]

Hmm i have to take a cancel my shipment, take a loan of $200 to get the 24GB

as i'm a student. its kinda hard. No parents support

[ornate elk]

Oh thats hard

Apple prices are also not helping in that situation xD

[fathom oasis]

lol True

[delicate citrus]

@ornate elk can I ask you about the mainboard too?

[ornate elk]

Sure

[delicate citrus]

what do you recommend me? ^^

[true sphinx]

facts current gen ryzen is pretty dope I got a ryzen 7 7700x for like 100$ used and its a beast

and thanks for the advice i knew that Nvida was faster supposedly but i didnt know if the vram diff would make up for it

/better

[true sphinx]

I highly reccomend looking on the used market for parts as well if your willing to take a little risk, my entire rig is used except for my GPU, and PSU, and you can end up getting some very good parts for very cheap. Ebay and Facebook market have been good to me

[lusty moss]

if i drag my SD files from one pc to another, will SD work on my new pc?

[true sphinx]

it should i have done that before and have had no immediate issues

[ornate elk]

Yea but it will break the git update function probably.
Its fixable.

[lusty moss]

awesome, thank you two

[iron lantern]

is this importent to add into the webui or its not needed ?

[vocal burrow]

It depends what you re trying to do.
Trying to recreate the picture ? Yes (and you ll probably be missing some settings here too)
Jus trying to use the model ? Not really, those are the settings used for that picture, you re free to use them or not. Just like you re free to pick such or such brush in photoshop.

different settings => different results

[tropic torrent]

Hello, I've generated an image in 512x512 and I'd like to generate the same image but uncropped at the top.
Can someone explain me how to do that ?

[ruby yoke]

thsoe are the settings they are suggesting.

[ruby yoke]

you'd have to know the seed, and the exact settings, but if you change the aspect ratio, it'll change the image. so what you want to do is pull that image into something you can outpaint with and outpaint the top

[tropic torrent]

ok I will try it thank you

[ornate ruin]

Hi! I'm so sorry, i was here for this just yesterday basically, you recommended me Swarm.ui right? I had to fully factory reset my laptop because of a bug. Now I try to see it as a good thing and treat it like a fresh restart lol..

Can you again let me know why you are recommending me swarm.ui? Does it even support IP Adapter? I saw control net in it yesterday when I played with it a bit but no IP adapter, I was probably just missing something right?

Alternatively I'll just reinstall Automatic1111 which was working fine for me so far. Again, 8GB VRAM if that matters for the choice of .ui

[ruby yoke]

yes, i recommended swarmUI - let it do all the technical stuff and generate inside it. you can run ComfyUI inside swarm, and comfy supports IPadapter if you want to have to do something more technical than what swarm does nativily

[ornate ruin]

But both comfy and swarm are in no relationship with stable diffusion so for any further assistance with those I'd have to head to their respective communities like the discord you sent me an invite link yesterday right?

[ruby yoke]

you have to run stable in something unless you want to be right at the commandline. and we have both a #🧣｜comfy-ui and a #🐝｜swarm-ui channel here

[calm river]

Hi everyone. Does anyone have recommendations for installing A1111 using Docker?
I've found some images so far, but many seem behind

Maybe it's not a conventional way of installing stable diffusion then?

[ruby yoke]

all i know is that @floral cradle battled with docker for a couple days and i think, gave up

[floral cradle]

Correct, was much easier to run in a separate LXC container, but that's only good to you if you're running Proxmox.

[iron lantern]

why is this going on using forge couple

[delicate citrus]

wich mainboard do you reccomend me? the config will be something like this:

The Ryzen 9800X3D
RTX 4060 Ti 16GB VRAM
AM5 Mainboard and DDR5 Ram for it too
Mainboards go with a B850 or B650
RAM the CPU likes 2x48 GB, 6000 Mhz CL30-32
cool the CPU with something like a BeQuiet Dark Rock 5 or Pro, Fan or any 360 Aio watercooler

well, I think you already told it to me, right? xD

[ornate elk]

What's your budget? I would go for an MSI Board like a B650 because the B850 boards will be released in the coming months.

[true sphinx]

i agree I have been very happy with my MSI b650-p

granted I would wait to upgrade as BF sales and all the holiday sales are coming

Also CS1o is right that with the 850s coming soon the 650s may drop

in price

[noble iron]

I notice that prompt weights dont really work in SDXL and SD3, like they did in SD1.5. By prompt weights I mean the brackets ( ) and [ ] around certain words in the prompt to emphasis or de-emphasis their priority in image generation. Is this by design or am I missing something here?

[ruby yoke]

prompt weights don't work in sd1.5 either, unless you are using a front end that deals with them and passes that on to the AI. Stable Diffusion uses attention - the weighting is from first to last, with the first token in the prompt having the most weight.

[noble belfry]

if youre versed in it, what about the BREAK keyword? is that even a thing?

also which frontends dont use them?

[ruby yoke]

there is no such thing. there are hard token limits and people have been trying to use stuff like that to get around them since sd.15 came out. IF you are using an interface, such as what nightcafe created, where you can put in multiple prompts and set the weights on each of those, what happens is that on the backend those are dealt with and then handed to the AI. but if you are just using stable diffusion in something like comfyUI or at the command line, or on a website that doesn't make that available, you can't use that. all that the word BREAK does in those cases is just act like any other token.

[noble belfry]

well i did know that the [] and () werent raw wasnt sure about BREAK. Theres so much thats woo, isn't there? Should make a differentiation though as you say by abstraction. if you're not using a website directly but through a local UI you have an extra layer that DOES add in the weighting. but the weighting doesnt mean much if youre using say lightning models etc.

i think a lot of it is cargo culting and im as guilty as anyone else

[ruby yoke]

stuff like [ ] and ( ) do have an effect, just not the way a lot of people think they do. i'll use punctuation as a means of specific noise injection to tweak the final result

[noble belfry]

oh yeah, i call it salting, dfkjlhfjkhjhjgk has the same effect.

[ruby yoke]

lighting models are optimized for speed, that has nothing to do with weights. the 'weight' a token has is the IMPORTANCE it has. the more weight, the more it's important, and the more the AI will pay attention to it

and we're now well past technical support and into a prompting discussion so we should move this to the #💬｜general-chat channel

[noble belfry]

its maybe not support, but fair enough, i think we covered it all.

probably people reading here would glean more than in there 😉 but thanks, wasnt sure about BREAK, confirmed about the other woo 🙂

[last oak]

hello is anyone avl to support?

[iron lantern]

what prompts can i use for a few dogs in an image like in the background

[vast crest]

hi today stable diffusion (i use it on chrome) stopped by itself 3 times, here is the error: Out of Memory

[ornate elk]

Hey, what's your GPU?

[vast crest]

RTX 4070 OC 12 GB

[ornate elk]

And which webui do you use?

[vast crest]

automatic 1111

[true sphinx]

12 should be plenty for auto1111 is anything running in the background or something that could be memory leaking

[vast crest]

since I've been using it (for 2 years) it has never given me this strange error

😦

[ornate elk]

Okay, what's in your webui-user.bat?

[vast crest]

this: git pull

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --xformers --no-half-vae --autolaunch

call webui.bat

[ornate elk]

Looks normal

What did you tried as you got the error?

Txt2img? Which settings, resolution etc

[vast crest]

if it were that I have been using for years honestly. the only thing that has changed, and that I downloaded some checkpoints and now illustrious, that's all. the page goes black with that error, and I have to restart the page, losing all the work I was doing

[ornate elk]

make sure the webui is whitelisted in any browser adblocker

also is your webui updated?

[vast crest]

How can I check on both sides if it's updated? I just can't remember how to do it XD

[ornate elk]

you can screenshot the bottom of your webui for me to see the version

[vast crest]

ok,1 moment

[ornate elk]

seems updated

did you tried to upscale an image maybe?

and then got out of vram

[vast crest]

all the photos I create are: 640 x 960 dpm2 sde - karras if I remember correctly

What exactly is the point of upscaling? to see if I get the error?

[ornate elk]

Better resolution and better quality

[delicate citrus]

I have a budget to this parts, is difficult to me convert to you correctly BRL on USD, but I have. I will check it, thanks again ^^

[true sphinx]

1 BRL is like 17 cents

unless things have changed since last month

[ruby knot]

Hi can you ask hardware questions here too? Because I got a gpu and a egpu enclosure so I can do better ai stuff too and I have a question about one of the power cables and if they are compatible with eachother?

The gpu has some kind of powercable splitter or whatever it’s called and the egpu enclosure has 1 power cable that has the correct amount of bits on the connecter and 1 cable that has half as many on on it so I’m not sure what to do?

[ruby yoke]

this is a question I'd ask @floral cradle since he's a hardware guy

[ruby knot]

Thanks I’ll try

[ruby knot]

Does anyone happen to have any idea if I’d be able to do Ai audio on my 4070ti 16gb super Ai gpu and if so would it be worth trying it or would it take forever to do stuff?

[ornate elk]

Should work good I guess

[vast crest]

hi, i forgot to ask you something yesterday. are there any better programs than webui 1111?

[cedar oxide]

I think webui forge is better than webui 1111 because it can support the new model - FLUX

[ornate elk]

a converter helps in this case ^^

[true sphinx]

there are different ones many people love something called comfy UI

but I would not go as far as to say "better" as much as personal preferance

https://github.com/comfyanonymous/ComfyUI

[vast crest]

without me even doing anything, it crashed

[tight depot]

Been looking online to try and fix this launch error, I used to be able to launch no issues. Getting the press any key to continue error. Already updated VC-redist 64x

[ornate elk]

hey did you installed the webui today or over a year ago?

because your Webui Version is older than a year

[ornate elk]

does it still crash?

[vast crest]

so far it has only done it 1 time. at the moment it is working as usual

[tight depot]

last year correct, I have it on multiple machines, tested 2 so far, and the other one launched and I was able to update and relaunch

should I just delete the folder and re-install?

[ornate elk]

Yea would be the best option.
A clean install also with Python 3.10.11 64bit

You can checkout my install guides from the pinned messages of this channel

[tight depot]

ty

[floral cradle]

WebUI is good for LLM integration, but I ended up hosting ComfyAI as a separate container and ignored doing anything image related in WebUI

[tight depot]

I make large batches of images using hundreds of prompts, would that also be better under comfyAI?

[ruby yoke]

that would be better at the command line with scripting

[tight depot]

interesting, not familiar on that method but makes sense

[ruby yoke]

you'd probaly have to write custom scripts to do everything, but for batches like you described that's the only wan I can think of things would run smoothly

[deft blaze]

anyone know how to fix this error the stable diffusion open in web but i can't generate anythink it say :
AssertionError: Torch not compiled with CUDA enabled

[ornate elk]

hey, can you show the full cmd log?

and whats your gpu?

[ruby knot]

Does anyone know how much slowdown or performance percentage loss or whatever I’ll get using my 4070ti super 16gb Ai gpu in a legion egpu enclosure (thunderbolt 3) with 500w psu and a 3.5” hdd in it too, compared to in a pc build which I can’t afford atm?

[deft blaze]

i have rx 5600xt and here is the cmd logs

@ornate elk

[ornate elk]

you downloaded the wrong webui version (the one for nvidia or cpu)

you can use my Automatic1111 with ZLUDA Guide for the AMD Version:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[deft blaze]

should i delete mine and download this ? and do you have video guide on how to install this

[ruby knot]

Well damn please can someone advise me on my egpu ? I plugged it all in and it’s mostly seeming dead, the hdd in it doesn’t show in windows and the fans very slowly on and I have no idea if it’s picking up the gpu

[ruby yoke]

any errors in your system logs?

[ruby knot]

How do I see?

[ruby yoke]

are you using windows? mac? linux? something else?

[ruby knot]

Oh yeah my bad windows 11

[ruby yoke]

https://www.solarwinds.com/resources/it-glossary/windows-event-log

[ruby knot]

How do I access it?

[ruby yoke]

you read the information on that post, it'll explain how to access it and use it

[ruby knot]

Also before when I only put the hdd in it the built in fan sounded like a tornado now it’s amost off can not hear it and barley feel it

[ruby knot]

Oh yeah sorry I missed that tab on the list

Wel that’s an issue I don’t have a keyboard on my pc atm I mostly use the software one

[ruby yoke]

not sure what to tell you there. other than... i would expect the software version to be able to do what the hardware version does?

[ruby knot]

I tried scanning for new hardware with device manager ?

[ruby knot]

Not pressing multiple keys at the same time as far as I know

[ruby yoke]

if you didn't just put new hardware in, manager shouldn't find anything. did it?

[ruby knot]

I did tho egpu thunderbolt 3

[ruby yoke]

probably needs drivers as well then

[ruby knot]

Like I said before when it was just the hdd in it it worked fine

[ruby knot]

Sure for the gpu but as I said it’s not even showing the hdd and it did before

[ruby knot]

I’m going to try taking the hdd out and try scan it in device manager again?

[ornate elk]

Yes delete the one and start with the Automatic1111 with ZLUDA from my Guide

[ruby knot]

Surely it would at least pick up the enclosure tho right? I mean one of my nvme usbc things did even tho it wouldn’t show the storage for some bs reason

[ruby yoke]

if it was me, i'd stop right there and make sure all the cables were tight

[ruby knot]

I did

[ruby knot]

I also had a guy at a computer shop fit the gpu for me and he did it in a few seconds

[ruby yoke]

i think i'd be calling him up about now and asking him to help me get it working then

[ruby knot]

He wouldn’t be any help and charge a lot, he advices against it even tho it should work.

[ruby knot]

I’m going to restart my laptop I don’t know if that will help but still

[ornate elk]

also open up a cmd and run sfc /scannow that fixes windows OS bugs.

and use the Restart button instead of shutdown so it reloads the drivers correctly

[ruby knot]

I did restart instead of shutdown

[ruby knot]

I don’t see how the gpu enclosure could go from working to 99% seeming dead by having the gpu put in it

[deft blaze]

i will try it thank you

[ruby knot]

It’s like the laptop doesn’t know it exists

[ornate elk]

Is it powered ?

By a PSU?

Does it have enough wattage?

[ruby knot]

Yeah of course no way would it work otherwise

[ruby knot]

2 people told me it should do since it’s not in a full pc

[ruby knot]

The gpu should be around 280-300w or something I forget and the psu is 500

[ornate elk]

Hmm ok

[ruby knot]

I’m getting really fed up etc tho nothing but issues

[ruby yoke]

welcome to the world of IT ;)

[ruby knot]

I’ve had computers since I was young XD

[ruby yoke]

:) welcome to the world of IT support

[ruby knot]

I’ve not tried to do this before tho

[ruby yoke]

if it was me, i'd ditch the laptop and just get a real desktop system

(which i did actually do)

[ruby knot]

An article I read said maybe the bios needs a setting changed but it was talking about internal gpu s so would it apply to an egpu too?

[ruby knot]

Money?

[ruby yoke]

i honestly have no idea how well an EGPU is going to work for AI generative stuff. you need cuda, which requires nvidia

[ruby knot]

I’m doing windows updates too see if that helps at all?

[ruby yoke]

is yours nvidia?

[ruby knot]

It is and it has loads

[night swift]

does anyone know what this error is? it was working just fine like some hours ago. now all i get is blurry weird images

[ruby knot]

4070ti super 16gb Ai

[ruby yoke]

you said your machine just found it. i would first go install the latest drivers

[ruby knot]

Found what?

[ruby yoke]

wrong VAE it looks like. also possibly wrong sampler and scheduler

[ruby yoke]

earlier, you said you ran manager and it just found your hardware. or did i misunderstand you?

[ruby knot]

No I tried that and it didn’t work is what I meant

[ruby yoke]

i misunderstood then. i would still try installing the drivers first.

[night swift]

how can that be? since it was working just fine like some hours ago. with same setting and everything. it made like 30 images just fine

[ruby yoke]

did you reboot after that?

[night swift]

nope

[ruby knot]

It DID find it when I only put a HDD in it the other day, now it’s acting dead apart from light on it and slow fan movement after the gpu is installed

[ruby yoke]

try that then. you might have memory corruption going on. and i mean shutdown and leave off for a full 10 seconds so everythign discharges and all cache is cleared. and then boot up and load everything fresh

[night swift]

i see, thanks a ton. ill try that

[ruby yoke]

so it's not finding the gpu. so go to nvidia's website and find the drivers and install them. IF hte computer guy installed them, maybe they either didn't install right or there was some sort of corruption in the drivers he ahd

[ruby knot]

It’s literally not finding anything at all now like the thunderbolt is dead or something is dead, that’s how it’s acting

[ruby yoke]

it's windows. it looses it's mind easily. shutdown and turn off so everything clears. wait a full minute. then boot up and see if the hardware is detected before you do anything else after it boots

[ruby knot]

It’s not the laptop tho I checked

[ruby yoke]

if it's not finding it, you might have a faulty cable

[ruby knot]

Well not the thunderbolt port at least

Thunderbolt cable ?

[ruby yoke]

it could be faulty. the connecters on the ends of it could be faulty. the ports they plug into, in your laptop and/or the thunderbolt could be

ports usually arne't the issue though

it's usually going to be drivers or the cable itself

[ruby knot]

I tried 2 I used thunderbolt 4 cable then I was like oh maybe for some reason that won’t work cos it’s thunderbolt 3? So I tried one of the 3 thunderbolt 3 cables I have

Still nope

[ruby yoke]

i, personally, start by reinstalling drivers. then replacing cables, and then look at the harder to fix hardware

[ruby knot]

The laptop port is fine I tried my hub in it

[ruby yoke]

you might want to contact thunderbolt support then

[ruby knot]

The cables where fine last time I tried them

Like I said it seemed fine until I the gpu was installed into the enclosure, I also tried no hdd just gpu incase it was too much watts still dead

[ruby knot]

Who?

[ruby knot]

It’s like an emotional roller coster atm trying to get this sorted, keep making progress then get 2 steps back.

[ruby yoke]

who manufacturers your hardware?

[ruby knot]

Nvidia it’s Gigabyte one tho, is this normal ? It looks like the circuit board is bulging a bit? That looks worrying to me ? Like the circuit board is bent or something

[ruby yoke]

bulging is never good. looks like it got too hot

[ruby knot]

I’ve never used it tho

Even the back plate looks bent to me?

Fml seems like I got sent a dud?

[ruby yoke]

if it was me, i'd be talking to nvidia first to make sure that is normal. and if not, talking to the hardware tech you paid money too

[ruby knot]

Hope it didn’t kill my egpu enclosure too…

[ruby knot]

He didn’t plug it into any power I was with him the whole time

[ruby yoke]

then definately contact nvidia

[ruby knot]

Couldn’t I re contact Amazon ?

[ruby yoke]

amazon's just a store. they're not going to know if that's normal or not

[ruby knot]

I got 216 off it already because it had some marks on it and didnt seem new, now on closer look it looks like landfill?

[ruby knot]

Yeah but if I tell them I had a closer look at it since and managed to try it out it looks and acts dead?

Annoying as hell tho £219 off would have been great discount if it worked

[ruby yoke]

amazon's just going to tell you to contact the vendor or tell you to send it back for a refund

[ruby knot]

I’d rather have a refund than probberly landfill?

I am upset though

I carnt think of any way it could have gotten too hot in the enclosure since it didn’t even seem to power it up

So I think it must have been like that already and I didn’t notice

[ruby yoke]

did you buy it from nvidia on amazon or a 3rd party reseller

[ruby knot]

No clue

[ruby yoke]

go pull up your order and see

[ruby knot]

The invoice just says sold by Amazon unless I missed something and I got it on pay monthly due to its price

[ruby yoke]

if you go to your amazon account, and click on your orders, you'll see your individual orders. you can then click on the order for the gpu and then that'll pull up the page with it. and at the top you'll see if it's amazon or someone else. and if it's amazon, it likely just came out of their warehouse

[ruby knot]

I’m going to be out £210 if it killed my egpu enclosure

[ruby yoke]

if it is actually amazon, then yeah, you'll need to talk them into a refund

[ruby knot]

I did that but I can not seem to see that bit of info

[ruby yoke]

i guess you have no choice but to contact amazon then

[ruby knot]

How annoying, I was excited to finally be able to try sdxl and flux

[ruby yoke]

@worthy glacier spammer on the discord

[formal trout]

suddenly I open Automatic1111 and I don't see ControlNet, and lot of errors on the console like this

[median geode]

which one should i download

the latest one or standalone one

[ornate elk]

Standalone 64bit

[ornate elk]

@distant plover what's your GPU?

And what's in your webui-user.bat from auto111

[distant plover]

4070

[ornate elk]

With 12gb ?

[distant plover]

yes

I have been running a1111 for years

up until a few days ago, didn't have this 30 minute gen issue

it happens if I change something like even change something in the text prompt

I am using two controlnets...one with multiple images or just one image of IP Adapter Face ID, one for openpose

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=

call webui.bat

That's the bat

[ornate elk]

Oh you should add --xformers to the commandline_args=
That should lower the memory and vram usage and speeds up everything

[distant plover]

Cannot

it errors if I add xformers

[ornate elk]

Then your webui is outdated

Or old torch version

[distant plover]

I just reinstalled everything today

[ornate elk]

Can you show me the error you get?

The whole cmd log

[distant plover]

with xformers?

[ornate elk]

Yes with only that

[distant plover]

WARNING:xformers:WARNING[XFORMERS]: xFormers can't load C++/CUDA extensions. xFormers was built for:
PyTorch 2.1.2+cu121 with CUDA 1201 (you have 2.0.1+cu118)
Python 3.10.11 (you have 3.10.10)
Please reinstall xformers (see https://github.com/facebookresearch/xformers#installing-xformers)
Memory-efficient attention, SwiGLU, sparse and more won't be available.
Set XFORMERS_MORE_DETAILS=1 for more details

[ornate elk]

Okay yes you run old torch version

[distant plover]

why would that be from a new install?

[ornate elk]

So open up a cmd and type
Pip cache purge
Then delete the venv folder

[distant plover]

doesn't it install up to date?

[ornate elk]

It will use cached stuff that is old if available on the system

So a pip cache purge will clean it

And then delete the venv and it will download the New one

[distant plover]

This is a constant battle

I will change this, and then other AI apps will break, I bet

[ornate elk]

Nope

[distant plover]

yes, after I installed it today, my RVC for voice cloning didn't work

[ornate elk]

Hmmm

Which version of rvc?

[distant plover]

venv is gone

RVC I don't know the version, whatever is on the github

Now I run the bat?

[ornate elk]

Have you cleared the pip cache with the command?

[distant plover]

I did that first

[ornate elk]

Ok then it should work

[distant plover]

It's installling requirements now

onyxruntime...that's what's missing in swarm for the ipadapter

it is running now

same issue...if I change something like add another image for ipadapter, it takes a long time to generate the image

3 min 30 sec to generate an image...way too long

[median geode]

what's this

[ornate elk]

Have you updated the extensions?

[ornate elk]

Close the cmd.
Open up a cmd and type pip cache purge.
Then delete the venv folder and relaunch the webui-user.bat

[distant plover]

Yes, I did that before generating again

the next generation took 33 minutes, and counting...it sat at 100% but never generated an image

ah, you were telling the other person

[ornate elk]

What does the cmd shows in that time?

Also make sure to whitelist the webui in any browser adblocker

[distant plover]

it shoiws the progress up to 100%, while the ui window shows 95% then stops

its not a problem with whitelist

up until Monday, it worked without these long generate times, It works without long generate times except when I change something or every once in awhile. Many generations will take place fast. I have tried in Chrome, Opera, Brave, Firefox, Edge

Eh, tried to launch it again, but now it won't run. There's an error when I run webui-user.bat

[median geode]

exm fr, i'm trying to download webui on my new PC, but i don't want to install the extensions from gihub one by one. If i use a ssd to move the extensions from old to new pc, i can not update them in webui conveniently. is there any better way?

[ornate elk]

Can you show the error?

[ornate elk]

Why you can't update them?

It should still work

[median geode]

if I move the extensions by an ssd from one pc to another pc, will the url be moved ? If not, how can I update them 🥸

[ornate elk]

If you installed git it will work

[untold mauve]

hi, i'm installing zluda for the 4567 time hoping it works ahahha, i have a rx 6600 should i download gfx1032 normal or optimized version?

[valid mulch]

Is it technically possible to run stable diffusion on more than one video card? Like, if I have a second Nvidia card in my PC would I be able to set things up in some way to use the VRAM from both at once? Or does it all have to be a single card?

[untold mauve]

guys why the ticket support is asking for wallet address?

ok it's a scam, i understand

[olive lintel]

No. If you have multiple cards, you could run multiple batches at once, but sharing the same batch between different gpus is not possible.

[valid mulch]

Thanks for confirming. Getting another 8gb card is much more in my price range than getting a really high vram card. Figured I'd ask anyway! 😄

[ornate elk]

Optimized

[olive lintel]

I solved that problem by renting time from vast.ai. Not quite as convenient as running locally but saves me from having to buy an entire new desktop pc and high end gpu.

[ornate elk]

You can only use one instance for each GPU so you could generate simultaneously

[valid mulch]

Yeah, that's an option. I haven't reached the point where I can justify putting money into this stuff.

[untold mauve]

ok thanks

[olive lintel]

Turns out some services can be dirt cheap. I'm spending maybe $10 / month.

[valid mulch]

Not bad... but still... if I'm making $0 on it, that's -$10 hehe

[olive lintel]

That's a lot of months for the price of even a low end gpu. Just something to keep in mind as an alternative to buying a gpu.

[ornate elk]

NP, in case of zluda and AMD, feel free to ask me

[median geode]

hello @ornate elk , does it have any problem

sth has been overwritten

maybe bcz i closed it once in the middle of downloading, since the internet was too slow.

but i don't know how to handle it rn

[iron lantern]

can you run more then one ui at the same time ... on the same pc ... because images take about 10 mins to make for me since iam going 1080x1080 and have only 4gs of vram on my laptop or how can i produce faster image gens

[median geode]

try using upscaler tools

[iron lantern]

where is that ?

[median geode]

running two generation at the same time doesn't work

[median geode]

generate small images first, then drag it into img2img,

[iron lantern]

ok so the defult size then resize it to 1080x1080 ?

[median geode]

yes, it can save like 40% of time

but the best way is to get a new GPU, that's what i'm doing😂

[iron lantern]

i have a laptop cant 🤣

[median geode]

my old gpu laptop was too slow

[iron lantern]

its impossible to get a new gpu and my laptop is only 1 year old

vitus 15.6 laptop with a 1650gtx but its faster then my pc

[median geode]

why not change the GPU on ur pc

it will like way much faster

[iron lantern]

my laptop is better then my pc ... lol i have a 9600k i5 on my pc with a 970 gtx lol

i have a 12th gen cpu in my laptop lol

is 720x720 much faster then 1080x1080 for image gen ?

[median geode]

that' s fine fr, GPU impacts more on image generation speed. doesn't matter what CPU is

[median geode]

512*512 is better, then upscale it

[iron lantern]

cheeeeeeeeeeeers

[iron lantern]

is it easy to upscale it ?

[median geode]

honestly still slow😂 but at least u can choose ur favourite image in 512512 then upscale them, which is way more faster than derectly generate 10801080 for several times and choose ur fav

[median geode]

or u can search Ultimate sd upscale on youtube, it will be faster

[meager prawn]

i have a problem with my Zluda. I had a1111 with directml before because Zluda did not work now i created a new folder in my Nvme drive G and then followed the guide that is pinned for the AMD cards (i have a rx6600) but now i get the error that my ROCm version is none and before it showed 6.1 and then that it could not performe basic things. Can anybody maybe help me because i do not know ho to fix it

sry for my englisch but i am not an native speaker

[ornate elk]

Your webui can't update because you made changes to the webui-user.bat

What's in there?

[meager prawn]

oh and i get some launch.py errors
Should i upload the cmd image?

[ornate elk]

Yes pls

[iron lantern]

now the images are coming at very low quality XD

[ornate elk]

I can help with the AMD stuff

[meager prawn]

i get this errors after doing all the stuff the guide told told me to do

[ornate elk]

Delete the webui-user.bat.
Then click into the file explorer bar and type cmd and hit enter.
Then type
Git pull
Then it should have recreated the webui-user.bat

[ornate elk]

What's your GPU?
Show me your environment variables where you added the folder path and %hip_path%bin

[meager prawn]

i have an rx 6600 and the newest drivers

and should i screenshot you the enviroment editor or what do you mean

[ornate elk]

Yes

[meager prawn]

ok wait

I hope you meant this

[ornate elk]

Yep and that's not correct

Give me a sec I share how it should look like

[meager prawn]

ok

thank you for helping me

[ornate elk]

Here:
https://github.com/CS1o/Stable-Diffusion-Info/issues/7#issuecomment-2408541054

[median geode]

it turns out like this

[ornate elk]

Seems like you have to git clone the repo freshly again

Something is wrong

Or wait

[median geode]

how🥲 i've delete the user bat

[ornate elk]

Open a cmd/powershell/gitbash console, go to StableDiffusionWebui's directory and run :

git init
git remote add origin https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
git fetch --all
git reset --hard origin/master
git branch --set-upstream-to=origin/master master
git pull

This should fix git stuff and update SD to the last version

[meager prawn]

so i did what the site told me to put it in the path but the cmd still shows this

wait i now added the \before the ZLUDA i try starting it

do i need to restart my pc every time?

[ornate elk]

Have you rebooted the PC?

Yep for the Paths its needed

[meager prawn]

so i did restart but i have the same error

[median geode]

"Open a cmd/powershell/gitbash console, go to StableDiffusionWebui's directory" like this?

[ornate elk]

Can help later, driving rn

[meager prawn]

yea all good

[ornate elk]

No

It has to be in the folder path

[meager prawn]

then i just learn for my exam tomorrow
could you ping me if you are free?

[median geode]

or just click into the file explorer bar and type cmd and hit enter?

[ornate elk]

Yep

[ornate elk]

Have you installed hip SDK 6.1 ?
And can you screenshot the path settings again?

[meager prawn]

yes i did the install without the pro drivers but all the other things checked

[median geode]

🥲

is it finished properly?

[ornate elk]

Doesn't seem so

[ornate elk]

Looks correct

Try delete the venv folder

[median geode]

what should i do next🥲

[ornate elk]

Try git clone the webui in a new folder

[meager prawn]

i did delete the venv folder then i get this

[median geode]

i have to delete the old one?

[ornate elk]

Okay maybe you need to reinstall hip sdk

[meager prawn]

i did the installation then i downloaded the gfx 1032 optimized package and copyed the library named one old_library and in the other lib i did copy the lib files from gfx 1032 optimized

[meager prawn]

ok i do it

[ornate elk]

After you moved the stuff over you need

[ornate elk]

Thats correct

[meager prawn]

its doing the install and it did tick all the things and marked the pro driver as not to install so is it all right if the software did it automatically?

[ornate elk]

I'm back in maybe 1-2 hours

[ornate elk]

Yep thats default

[meager prawn]

ok it is now reconized but why can`t it install pytorch?

[ornate elk]

Open up a cmd and run
Pip cache purge
Then delete the venv folder again.

Oh wait

Nope

You installed python with 32bit

You need 64bit

[median geode]

what else can i move except models and extensions🥲 can any other stuff be moved? has taken me a whole night to clone the webui

[meager prawn]

but i downloaded the 64 package

[ornate elk]

But 32bit is installed it seems

So maybe uninstall every version

[meager prawn]

should i do the 3.10.6 or 3.10.11?

[ornate elk]

And reinstall 3.10.11 64bit

[meager prawn]

ok

can i remove all versions with an command?

[ornate elk]

Outputs if you have one

Or embeddings

Also if your from Asia you may need to use a VPN or not and change the DNS

[ornate elk]

Nope

Uninstall them via system control

[median geode]

what about the xformers or clips or several stuff

[ornate elk]

You can copy over the venv folder if you like

AFK now

[meager prawn]

do i need to uninstall 3.12.0?

[median geode]

can i move torch? it takes me too much time to clone

[meager prawn]

i now get this error

[ornate elk]

Yes