🤝｜tech-support

[manic matrix]

Bro what’s it gonna do differently lmao

[ornate elk]

It has more power 😛

[manic matrix]

Wait no I can see online it says to do something diskpart to fix this anyway

I need to create a simple volume using it

I fucking guess

[ornate elk]

Exactly

You need to to list disk

Select disk etc

[manic matrix]

Ok I’m caught up now what

[ornate elk]

Be sure to select the right disk

[manic matrix]

Disk 2

[ornate elk]

Does it shows * ?

[manic matrix]

Yep

[ornate elk]

Okay then type
Active

[manic matrix]

There is no partition selected

How do I select partition

[ornate elk]

Ah well

Then first:
create partition primary

[manic matrix]

Worked

[ornate elk]

Then;
select partition 1

Then:
Active

[manic matrix]

The selected disk is not a fixed mbr disk

The active command can only be used on fixed mbr disks

[ornate elk]

Ah okay

[manic matrix]

I sure hope you know what you’re doing 😭

This was 100 bucks

[ornate elk]

Okay so type list disk again

[manic matrix]

Ok

[ornate elk]

To be sure we are on disk 2

[manic matrix]

Yep

[ornate elk]

Okay then type again:
Select partition 1

[manic matrix]

Yep

[ornate elk]

And then:
format fs=NTFS label=External quick

[manic matrix]

There is no volume selected

LMAO

Let me guess

Select disk

Select volume 1?

[ornate elk]

Select disk 2

Select partition 1

[manic matrix]

But it says no volume detected

[ornate elk]

Then try
Create partition primary

[manic matrix]

Not enough usable space for this operation

Didn’t we already do that

[ornate elk]

Yea

[manic matrix]

So I have partition one selected now what

[ornate elk]

This

[manic matrix]

Please select a volume and try again

I think we need to select a volume

[ornate elk]

But it didnt worked

To select the volume

[manic matrix]

I didn’t try that

You didn’t tell me to do it lol

[ornate elk]

We tried these steps:
https://www.windowscentral.com/how-clean-and-format-storage-drive-using-diskpart-windows-10#section-how-to-fix-drive-gpt-problems-with-diskpart-on-windows-10

[manic matrix]

After I selected a volume, it says format is not allowed on the current boot, system, page file, crash dump or hibernation volume

I have windows 11

Not allowed***

In total these were the steps I did

List drive

Select drive 2

Create partition primary

Select partition 1

Select volume 1

Then the command you typed

Now I have the format is now allowed message

[ornate elk]

Yea dont use select volume 1

Its 2am here

[manic matrix]

When I don’t select volume one it says no selected volume

[ornate elk]

Hmm

How does it looks in disk management right now?

[manic matrix]

It won’t even let me select new spanned or striped volume

Only thing highlighted now is convert to dynamic disk

[ornate elk]

Can you screenshot it ?

[manic matrix]

Wait it looks like I have to type in volume 0

Wait

No volume 6

IT WORKED

IM A FUCKING GENIUS

[ornate elk]

Why 6 ?

[manic matrix]

Below that it says successfully formatted the volume

[ornate elk]

Perfect

[manic matrix]

Because every other volume was being used

So now what

[ornate elk]

Now
assign letter=E

Or any other letter

[manic matrix]

External E notification popped up

Which seems good

[ornate elk]

Sounds good

Check if you can create a folder on it

[manic matrix]

I CAN

HOLY SHIT

[ornate elk]

Perfect

What a ride xD

[manic matrix]

Jesus thank you bro

Bro helped me more than school ever could

Christ

[ornate elk]

Haha no problem at all ^^

Now you shouldn't get any error messages regarding file permissions when installing Stable-diffusion

[manic matrix]

If this works where I do I submit a dono of gratitude bro

[ornate elk]

On my main github page ^^ where the install guides are

[manic matrix]

On it chief

[ornate elk]

The link is in the pinned messages

[manic matrix]

I’m on the page but idk where to dono

[ornate elk]

Click on the top left on "stable diffusion info"

Then you should land on the main page

[manic matrix]

Boom

Done

[ornate elk]

Thanks a lot!

[manic matrix]

All good I really needed someone patient help my dumb ass

Crystal wizard helped me all of last night to get it installed on my other drive but that had like no space so I had to get the ssd today and do it again

[ornate elk]

Ahh I see

Yea crystal is a big help too

[manic matrix]

Now how the fuck do I get more models and bases and checkpoints 😭😭😭😭😭😭😭😭😭😭😭

[ornate elk]

From Civitai.com

Its the largest model database

[manic matrix]

Do I need to do anything special with those or do I just click download and it works

Lemme guess it’s the more stupid, first half

[ornate elk]

You need to read the info box of the files.
Some are "checkpoints" (models) which have to be moved into the stable-diffusion-webui/models/stable-diffusion folder.
Some other files are Loras and go into models/lora

[manic matrix]

Okay 🥲

Let’s see how I mess this up

[ornate elk]

You can't do much wrong. And if its not working, just ask here again.
I'm going to sleep now. Its really late here.
But I can help tomorrow again. Just @ me and I will answer

[manic matrix]

Thanks so much bro, will do

[tender jay]

Hi I'm watching some youtube tutorials on Flux and I see that they have Comfy Manager that looks like this.

But mine look like this?

Is there a reason why mine is so limited?

[onyx solar]

When was the last time you updated the manager itself?

[tender jay]

I just installed it yesterday from Git

I pressed the updated a few times since.

[onyx solar]

Update ComfyUI does not update the add-ons.

But that looks like a really old version installed.

[tender jay]

I get this when I press fetch updates.

but I dont' know what do after this?

[onyx solar]

Open a command prompt, go to your comfyui-manager folder and run git pull from within that folder.

[tender jay]

from this folder?

Run git pull from this folder?

?

[onyx solar]

No, from the comfyui-manager folder. That would be inside ComfyUI, then inside custom_nodes

[tender jay]

oh ok

like this right?

[onyx solar]

yup

[tender jay]

That's what's suppose to happen right?

[lavish void]

looking to buy a pc that can run stable diffusion locally. found this one for 600€ used online, will that run it smoothly?

[onyx solar]

lol...yeah...you were juuuuuuuuuust a bit behind...

So completely shut down ComfyUI and start it again. (Not just the browser.)

[tender jay]

ok

[onyx solar]

Then when that's done and it's up and running, re-open ComfyUI Manager and hit that Update All button that you'll then have. That'll get everything else up to date.

[tender jay]

uhh....

still the old one?

I close out everything and started it back up?

[onyx solar]

So you actually closed the command window that was running ComfyUI and not just the browser?

[tender jay]

yeah

I closed it and the browser and started it back up

going to do it again.

yup still the old one....

should I restart my PC?

[onyx solar]

Wouldn't hurt.

[tender jay]

ok going to restart the pc.

brb

[onyx solar]

If this doesn't work, my advice would probably be to just blow away the comfyui-manager folder, then run git clone https://github.com/ltdrdata/ComfyUI-Manager.git from the custom-nodes folder to re-create the manager add-on at the latest version from scratch.

[tender jay]

I didn't work....

going to have to delete the folder.

question before I do.

[ruby yoke]

i tend to jsut go find the github page for stuff like that and install from their install instructions, not from manager

[tender jay]

are these suppose to me in the custom nodes folder?

[ruby yoke]

yes

[tender jay]

oh ok so just delete the manager folder then.

[ruby yoke]

comfy manager is a custom node and the other is a cache file location

[onyx solar]

Yeah, I'd probably just re-install the manager and go from there.

[ruby yoke]

before you do that, what exactly is the issue

[onyx solar]

He's getting the old small menu...was WAY out of date

Updated manually, but it was too far out of date.

[tender jay]

#🤝｜tech-support message

opps wrong reply.

[ruby yoke]

follow these steps: kill the browser. kill the command line. open a new command line. go into /comfyUI/update. run update.py

[tender jay]

I'm getting the small menu

[ruby yoke]

reboot

start comfy

[tender jay]

did all that.

still small menu kind of late now. I have to redownload since I deleted the folder already....

[ruby yoke]

okay then yeah, get rid of manager and reinstall it

[onyx solar]

That's where I was at with it. 🙂

Literally like thousands of changes out of date for the manager.

[ruby yoke]

personaly, i'm not ever updating comfy again until there's something i need the updated version for. too many bugs in the recent updates

and i hate that new menu

[onyx solar]

Change is change. 🤷

[tender jay]

well this is my first time using comfyui

I'm just trying to do images with better print words.

ok clone done let me start up again.

[ruby yoke]

not if it's not needed

[ruby yoke]

you'll want either SD3 or Flux for that

[onyx solar]

That's a bit subjective.

[tender jay]

hmmmm....

[ruby yoke]

i'm a hardware tech. not subjective at all

[tender jay]

do I update in here?

it's still coming up small menu?

[ruby yoke]

you update if you need to, and don't if you don't need to

[ruby yoke]

go into your settings and turn the thing on

[tender jay]

what on?

what am I looking for?

that's my comfyui

[onyx solar]

That doesn't change how the manager looks. Also, that looks like an old ComfyUI itself to me.

[tender jay]

.........

[onyx solar]

What version are you on for Comfy?

[tender jay]

No clue...

I followed a youtube video to install comfy....

[onyx solar]

Click the gear. Share the menu you see.

[tender jay]

https://www.youtube.com/watch?v=KTPLOqAMR0s&t=521s

https://github.com/comfyanonymous/ComfyUI

[onyx solar]

I get that videos are nice, but following them can be bad advice if they're not recent videos.

[tender jay]

from that git

[onyx solar]

That video is a year old.

[tender jay]

https://tenor.com/view/why-me-facepalm-stressed-star-trek-gif-14598957

what about the git?

[onyx solar]

Might have some old stuff in it. I'm not gonna watch it to verify if the instructions are good or not, but I wouldn't use it.

[tender jay]

is the git a year old??

[onyx solar]

The git is the live one.

[tender jay]

Well I'm using that git...

how is it wrong?

[onyx solar]

Using the git isn't the problem. There may have been some steps that were wrong/out of date in the other video.

[tender jay]

??But there's was like only 2 things I did to get it working?? All I ddi was download the zip exported it to my drive and run the bat file?

how it's that a complicated step?

uhhh by the way there isn't any instruction in that video for installing comfyui?

So how do I get the newest ComfyUi?

[onyx solar]

I'd personally just blow away the whole thing and start again, ensuring that you are doing it as a non-administrator user and putting it into a folder you have full permissions to. I wouldn't watch any video; I'd just follow the instructions on the git itself. (I can't explain why you're having the issues you're having at this point or guarantee that my suggestion will work, I'm just sharing the next steps I'd take if I was in your position.)

[tender jay]

But? The video doesn't have any instruction for installing ComfyUI?

You said I don't have the latest version. So how do I get the latest version??

The video only shows you how to use ComfyUI. But I dont' have the latest version and I can't install the latest manager??

I can't watch a video if I dont' have the latest version to follow along with a tutorial??

From where?

[brittle lantern]

installing focus using zluda and got this during step 6 on the pinned tutorial page

whats the fix?

[tender jay]

Because I downloaded my ComfyUI from that git.

and it's supposedly old....

so where do I get the latest version?

Is that the same place I got it from? #🤝｜tech-support message

this is what I used to install it.

Is this outdated?

[tender jay]

Well that was the comfy I installed.

Now reinstalling...

[lean fulcrum]

Hello frens. I have a technical issue I was hoping someone might be able to shed some light on for me.

I downloaded webui forge as a suggestion from a friend. I'm extremely new to this whole process but was wondering if someone might be able to lend me some pointers.

I updated the program, inserted the files where they needed to go and clicked run.

I got this command prompt message.

Does this mean I need to use something different?

[tender jay]

Ok I did a reinstall of ComfyUi

is that the latest version?

I understand. I'm just trying to get the stuff working.

I'm just trying to get the Manager to be the full size menu.

[ruby yoke]

comfy has recently changed the interface and he was guessing why you were seeing things you saw

[ruby yoke]

i don't think that's availble any more.

if you go to Sebastian's channel - the guy you were watching the install video from, and watch his latest video, he shows you what the new interface looks like

that is one reason why i'm not going to update comfy unless i have to

[tender jay]

I got it to work now.

[ruby yoke]

excellent. now, please watch scott's comfyUI series, it'll give you a nice kickstart

[tender jay]

I will. I added it to my Stable Diffusion playlist.

[ruby yoke]

🙂

[ruby yoke]

one thing to be aware of - it's extremely easy to run out of space because it's very tempting to go try and download every lora and checkpoint in existance. but you can point comfy to a differnt directory for models

[tender jay]

Well I been meaning to ask about something like that. How do I tell Comfy to save the images to a specific folder of my choosing? I don't want it in the Comfy output folder.

or is that in the videos tutorials.

[subtle quarry]

how do i download controlnets from huggingface? Im so confused

https://huggingface.co/xinsir/controlnet-openpose-sdxl-1.0/tree/main

trying to download this

and i have no idea how.,

nevermind im just really stupid

[ruby yoke]

you can't. it saves them in /comfyUI/output. you can't change that. but you can create a symlink to some other directory if you want

[ruby yoke]

at the bottom of the list on that page

the 2 .sasfetensors models are what you download

[subtle quarry]

so i need both?

[tender jay]

Oh. I only ask because Auto1111 I have my images going to a special drive. But oh well.

[ruby yoke]

yeah, and people have asked comfy to put that in, so far he hasn't

[ruby yoke]

they're not that large, i would download them both

[subtle quarry]

do i name them close to the same thing?

im confused lmao.

i hate how unclear these things are sometimes

[ruby yoke]

that's just openpose though

https://huggingface.co/xinsir

[subtle quarry]

ik, i just want openpose

[tender jay]

I guess it's time to make a batch command to move all images to said directory.

[ruby yoke]

you don't change their names at all. and no, i'm sorry, i don't know what directory they go into.

[ruby yoke]

just symlink /comfyUI/output to the directory you want them in

[subtle quarry]

but there's like 50 other things named the same thing out there

[ruby yoke]

okay well, if you change their names, don't change the .safetensors file extenstion and plan on having to tell comfy what they are every time you run someone's workflow that wants them and it can't find them

[subtle quarry]

alr.

[tender jay]

ok

[subtle quarry]

alright so now my workflow is making this weird image

it really shouldn't be doing that, and it's doing it with the open pose i switched to, since xinsir was doing the same thing

[subtle quarry]

changing the model to a different one fixed it... but...

okay changing it back seemed to fix it?

I don't know how that works...

it's not listening at all to my controlnet though.

i may need to learn sdxl or something to use controlnets. ....fuck.

HAA what the FUCK is this?

i got two images to print just fine with sdxl, and then this shit comes out!

what the fuck is going on? Is my model just repeatedly corrupting?

[ruby yoke]

there are a number of things that can cause this

[ruby yoke]

you need to start by making sure you're using the controlnet for the base model you are using. if you're using sdxl, is the controlnet you're using also for sdxl?

[ruby yoke]

upload the image that comfy generated please - not a screen shot - so we can look at your workflow

[ruby yoke]

don't use pony models for this

[subtle quarry]

i didn't let it go through to the image save, i have an image preview before i send it to the upscaler

but i want to use a pony model for controlnet, and it sucks that i can't...ever get shit to work.

[ruby yoke]

don't use pony models - they have their own set up and don't work with much. use a normal checkpoint or use the base model

you'd have to have a controlnet created FOR pony

[subtle quarry]

i use pony because it's the closest quality i'll ever get to a better 1.5, and sdxl has a fucking learning curve with the wording.

i've had a couple people here tell me to learn how to write sdxl.

and writing like 1.5 still works.

[ruby yoke]

i'm sorry, but you would have to get someone to train you a controlnet for pony

[subtle quarry]

yeah i know

[ruby yoke]

1.5 does work, and works well, if you understand how to prompt it

[subtle quarry]

i know how to do 1.5. i started on 1.5

i keep prompting sdxl like 1.5

[ruby yoke]

i do too

sdxl usually understands

[subtle quarry]

It does, and it pisses me off when people tell me to change something that works perfectly fine

[ruby yoke]

just thank them for their input and ignore them

[subtle quarry]

alright, using sdxl. it's ignoring my thing still

[ruby yoke]

nice image though

[subtle quarry]

it's very nice, yeah but i'm a perfectionist. i want my poses

[ruby yoke]

you should probably ask in the L2 discord. or maybe @fair oxide has some suggestions

[fair oxide]

this is what my download scripts look like:

# Set the working directory to where the checkpoints will be saved
cd /workspace/ComfyUI/models/checkpoints && \

# Download the CinenautXLv2 model
wget -O CinenautXLv2.safetensors "https://civitai.com/api/download/models/607422?type=Model&format=SafeTensor&size=full&fp=fp16&token=YOURTOKEN" && \

# Download the Epic Photonism model
wget -O Epic_Photonism.safetensors "https://civitai.com/api/download/models/355189?type=Model&format=SafeTensor&size=pruned&fp=fp16&token=YOURTOKEN" && \

# Navigate to the LoRA folder
cd /workspace/ComfyUI/models/loras && \

# Download the desired LoRA model
wget -O Detail_Tweaker_XL.safetensors "https://civitai.com/api/download/models/135867?type=Model&format=SafeTensor&token=YOURTOKEN" && \

# If all downloads are successful, restart the ComfyUI service
if [ $? -eq 0 ]; then
    sudo supervisorctl restart comfyui
else
    echo "Download failed, restart aborted."
fi

so what you need is to copy the clean download link

get your login token

pick a name that is generally just similar to the name of the thing and then .safetensors

when you go into huggingface, civit or github pages

the vast majority of the time what you want is just the largest .safetensors file

this is SDXL though, there is kinda one king control net for SDXL

the Pro Max model on this page: https://huggingface.co/xinsir/controlnet-union-sdxl-1.0/tree/main

its by a long way the best control net I have seen so you should take advantage of the fact that SDXL has it 🙂

you can also use git instead of wget if you prefer that

the actual layout of your comfy folders funnily enough doesn't matter

because of this https://github.com/comfyanonymous/ComfyUI/blob/master/extra_model_paths.yaml.example

sadly its not really possible to just write one really good install script that will work on all computers and servers

they are just set up too differently

[subtle quarry]

struggling to figure out how to get embeddings to show up in forge

because im stupid. i figured it out

okay, okay so it's kind of working. it's listening but it's turning them backwards

[fair oxide]

I haven't gotten into embeddings yet

they kinda lost popularity after more loras came out

but I think embeddings are probably under-rated at this point

they have a big advantage that they can mess things up less than loras can

[subtle quarry]

i got the openpose working with pony. had to do it on forge, but it works

[fair oxide]

nice

if it worked on forge that implies the file was a correct one

was likely either an issue of nodes or an issue of extra_model_paths.yaml

but if you like Forge that's fine, its getting good these days

[subtle quarry]

probably a nodes issue.

[kindred dragon]

I keep getting this error whenever I try to generate anything, I just installed SD like an hour ago so there is a chance I messed something up.

I'm using this arguments:

[ornate elk]

What's your GPU?

[kindred dragon]

RX6600

[ornate elk]

Then you need to follow my AMD Automatic1111 with ZLUDA install guide.
Its on the first link of the pinned messages.

[kindred dragon]

Gotcha, will do so later. Thanks!

[ornate elk]

NP, feel free to ask If any step isnt clear

[kindred dragon]

Might want to update the post, HIP 6.1 is suggested for ZLUDA

[ornate elk]

I said the guide from the link ^^ its the updated one

But yes I will remove the old one

When I'm back home I will also edit the main guide again to only suggest HIP 6.1

[kindred dragon]

oops, my bad

thankfully the only thing that changed from what I can see is the HIP version, which I think I installed correctly anyways

[ornate elk]

It could be that you need some other rocm library files.
Because they need to be supported for hip SDK 6

[kindred dragon]

Yeah I realized that too and installed the right ones (I think)

[ornate elk]

Okay, let me know if it works or if you need a link to the other files

[kindred dragon]

Uhh
Exception Code: 0xC0000005

Probably fucked something up

[ornate elk]

Can you show the full cmd log?

You need these rocm library files:
https://github.com/likelovewant/ROCmLibs-for-gfx1103-AMD780M-APU/releases/download/v0.6.1.2/rocm.gfx1032.for.hip.sdk.6.1.2.7z

[kindred dragon]

Yeah those are the ones I downloaded pretty sure

[ornate elk]

Thats not the full log

But yea you also need to make a PC restart after changing the files

[kindred dragon]

yup, did that

[ornate elk]

You also need to put the rocm library DLL into the Library folder

[kindred dragon]

Yup, did that too

[kindred dragon]

here's full

[ornate elk]

And your on the latest amd adrenalin version ?

[kindred dragon]

Yeah

[ornate elk]

How do your environment path entry look like?

[ornate elk]

Delete the venv folder and the .zluda folder and relaunch the webui-user.bat

[kindred dragon]

Huh? I could've sworn I already put the paths what the fuck

the paths aren't there

uhh let me try putting the paths and try again

well that didn't work

the error message is formatted differently though for some reason

I think it's the same info as last time just formatted differently for whatever reason

[kindred dragon]

trying this now

[ornate elk]

Update your python to 3.10.11 64bit then delete the venv and the .zluda folder

[kindred dragon]

Here goes nothing

[kindred dragon]

No dice, same error

[ornate elk]

Hmm

And your environment paths?

How do they look?

[kindred dragon]

Are they right?

[ornate elk]

If your in the Path setting then yes

A bigger screenshot would help

[ornate elk]

Looks correct

[kindred dragon]

So it's not that then

[ornate elk]

Probably something with the replaced files

[lean fulcrum]

Yes

[worthy spindle]

Can I bother someone who knows kohya Lora training? I'm having a few issues. Tutorials lose me on a few spots and I cannot understand some of lingo nor trouble shoot. What I'm trying to do is make lora's off a Pony models. ||(IF you want, I could drop a bit of money for your time. Just Dm me.)||

[quasi pine]

Nice to meet you!
I have been using this product since the other day. I have a question about a color problem after output.
I have a 512x768 image output with txt2img → upscaled by 2x with img2img. 1024x1536 (using tiled diffusion,Contorol net) → upscaled by 2x again. 2048x3072 and then the colors are wrong as in the image. After running 2048x3072, the colors are not right as shown in the image.
The first upscale works fine.

[ornate elk]

What VAE do you use?

[vast pulsar]

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument index in method wrapper_CUDA__index_select)

anybody know what's happening? started happening after i tried to use flux in automatic111111 webui (i quickly learned it wasn't a normal checkpoint lol)

also idk why it says i got 2 gpus, i don't

[ruby yoke]

@ornate elk would you mind posting your guides on the stable diffusion subreddit?

[brittle lantern]

Anyone got a tutorial or link to one for ComfyUI installation with ROCm on Linux

[vast pulsar]

managed to fix it

i literally did the meme and it worked

"tried turning it off and on?"

well fuck, it was enough lol

[subtle quarry]

trying to get a lora to work and it's just straight up not working how it should

trying to get image 1 to have the same look as image 2

1st one has the workflow inluded

maybe i'm just being overly picky... i just want the eyes to match.

[fair oxide]

Installed krita stable diffusion and got everything downloaded. When i try circling what I want to change
and clicking fill i get the following error

"Server execution error: Numpy is not available"

Going into the serer.log file it says

2024-08-29 22:55:31,338 INFO A module that was compiled using NumPy 1.x cannot be run in
2024-08-29 22:55:31,338 INFO NumPy 2.1.0 as it may crash. To support both 1.x and 2.x
2024-08-29 22:55:31,338 INFO versions of NumPy, modules must be compiled with NumPy 2.0.
2024-08-29 22:55:31,339 INFO Some module may need to rebuild instead e.g. with 'pybind11>=2.12'.
2024-08-29 22:55:31,339 INFO 
2024-08-29 22:55:31,339 INFO If you are a user of the module, the easiest solution will be to
2024-08-29 22:55:31,339 INFO downgrade to 'numpy<2' or try to upgrade the affected module.
2024-08-29 22:55:31,339 INFO We expect that some modules will need time to support NumPy 2.
2024-08-29 22:55:31,339 INFO ```

I looked in the site-packages folder and found the numpy module and looked to see what version it was running. Its running 2.10. How do i downgrade it? I tried downloaded it through pip and cutting and pasting the pip downloaded module and putting it in place of 2.1.0 but then the server wouldnt even run

[white stirrup]

how does one get flux to work on a1111 or comfyUI

[ruby yoke]

with comfy, you just need to download the flux models, comfy has flux support built into it.

[white stirrup]

for some reason it threw out massive errors when i tried to use it

[ruby yoke]

read through this https://comfyui-wiki.com/tutorial/advanced/flux1-comfyui-guide-workflow-and-examples

[white stirrup]

thanks

[white stirrup]

i get this error now, i have debugged a few errors i got before but idk what to do here

im using the schnell workflow from the site

[ruby yoke]

did you fill in the information in the two nodes it mentions in that error?

[white stirrup]

CivitAI

From what i assume was the official account

Actually no

This specific instance is a copy from my a1111 model folder that has also been renamed

Could that be an issue?

[white stirrup]

Ok ill try thanks

[balmy matrix]

So, I'm trying to use the ComfyUI extension in auto1111. However, I cannot use ComfyUI and auto1111 at the same time, because one requires Rocm6.1 (auto1111) and the other uses 5.7 (comfyUI). In the HIP_Path system variable, I can only set it to one of those two at a time. Any way to get around this?

[ornate elk]

That makes no sense. Comfyui can work with hip 6.1 too

But you should set comfyui up manually instead of using the extension in auto1111

[balmy matrix]

I did that, it works just fine, I just like using auto1111 on my phone when I'm not at my pc. Comfyui basically doesn't work on mobile, I can't seem to move the manager around and when I switch to the newer mode the top bar disappears and I can't get it back, I've tried on multiple mobile browsers.

[balmy matrix]

Why does it not when I set my HIP PATH to 6.1 then? Is there a way to remedy that?

Also I noticed when I use ComfyUI after some two dozen or so generations my computer crashes. I don't know why. However I can generate as much as I like with auto1111 and my pc doesn't crash.

[ornate elk]

Auto1111 has better optimisations with Zluda than comfyui

Comfyui can easily max out your vram

[elder mauve]

getting this error

[ornate elk]

What was your GPU again ?

[elder mauve]

7800xt

[ornate elk]

Ah okay

[elder mauve]

last time u said i need some special files

[ornate elk]

Yep

[elder mauve]

which files

[ornate elk]

The gfx1101 files from here:
https://github.com/likelovewant/ROCmLibs-for-gfx1103-AMD780M-APU/releases/download/v0.6.1.2/rocBLAS-HIP6.1.2.7z

[elder mauve]

i downloaded these ones

🤔 and replaced the old files

[ornate elk]

And the library DLL too ?

[elder mauve]

which library DLL?

[ornate elk]

There should be a DLL inside too

But I can't check rn

[elder mauve]

none with .dll

[ornate elk]

Hmm okay

I have a 7800xt in a friends PC, maybe I can get it to work at the weekend, and then tell you how to

[elder mauve]

thanks

[ornate elk]

Can you give me the full cmd log

[heady pine]

I wanted to check out this new Flux checkpoint, so I updated ComfyUI since I had all these errors pop up when trying to use it, but now when I go to open ComfyUI, I am met with a white page. How can I fix this?

[ornate elk]

What does the cmd shows?

[balmy matrix]

What is Swarm? Does it have as good of optimization as auto1111?

Okay I was about to ask, I saw there were two different versions of it.

[heady pine]

I don't know what the cmd said before, but ComfyUI was working fine before I updated it

[ornate elk]

Swarm is a frontend ui for Comfyui so no it has the same performance

[balmy matrix]

ah, damn.

and right now I dunno how to fix ComfyUI running at 5.7

Maybe that's why my PC crashes, 5.7 just ain't as efficient as 6.1

[ornate elk]

It makes no sense that you even have hip SDK 5 and 6 installed

Only 6 and replacing the comfyui files should be enough

[balmy matrix]

Well, the reason I have 5.7 installed is because when I went to install ComfyUI-Zluda, it says right there on the Github "USE 5.7, NOT 6.1"

So I did as was told lol

[ornate elk]

Hmm looks normal. Only core nodes has an error, maybe delete that from custom nodes

[ornate elk]

Ah so you followed the other comfyui zluda guide

Not mine

[balmy matrix]

I didn't know about yours until afterward

[ornate elk]

I'm running every webui with rocm 6.1 without problems

[balmy matrix]

So: uninstall 5.7, then delete the Comfy UI venv folder

Reinstall

Then see what happens?

[ornate elk]

Yea and then you need to replace some zluda files in the new venv

Manually

Like stated in my guide

[balmy matrix]

Your pinned guide says to install 5.7 too btw

It's correct on the github link but incorrect in the discord pin

[ornate elk]

Yea the discord pin will be deleted when I'm back home

[balmy matrix]

Reinstalling everything as exactly per your guide now 🙏

Foolish me, thinking the github from where I got ComfyUI-Zluda would have the correct install information

😂

[ornate elk]

I mean it works, they just use the old sdk

And they automated everything by scripts.

[balmy matrix]

It worked perfectly.

No issues.

[balmy matrix]

However, with SwarmUI, I get this error:

This is what's in the cmd log

[ornate elk]

I have no guide for swarm yet as it requires additional steps to make it work

[chrome rampart]

hi all hope doing well. Question can i de-install Python and do re-install or risky? Seems to have issue that comfy/SD look for python on drive i want to remove

[formal trout]

Hi, I couldn't run flux on Forge, I was able to generate SD XL

I'm getting

return torch.nn.functional.linear(x, weight, bias) RuntimeError: mat1 and mat2 shapes cannot be multiplied (4096x64 and 1x98304) mat1 and mat2 shapes cannot be multiplied (4096x64 and 1x98304)

in Forge while trying to use the flux1-dev-bnb-nf4.safetensors model

no idea really

[fierce hollow]

So, there is SD 1.5 which is trained on 512x512 images
There is SDXL which is trained on 1024x1024.
Does anyone have any information if SD models will ever be trained on 2048x2048 images?

[plush spire]

I appear to have lost the Gallery tab in Automatic1111, anyone know how to get it back?

[trail wadi]

which model do i need to get for automatic 1111

its not letting me generate images because it cant find a checkpoint

[formal trout]

any SD 1.5 or SD XL model, download one from civit ai

[formal trout]

usually it downloads a model when you first install and you don't have a checkpoint

[trail wadi]

so how do i get the checkpoint?

this is the error i got

[trail wadi]

so i should download any of these?

[plush spire]

If there is a specific image or style you want to recreate from civitai, look at the information for that image.

it'll tell you what checkpoint was used and Loras

[trail wadi]

sorry im very very new to this. all i've done so far is follow the guide on the github, but now i've run into this error

i dont know what checkpoint i need

[plush spire]

depends what you want to create

as I said, look on Civitai, see what image takes your fancy and download the checkpoint used for that image

[trail wadi]

okay

all of them work for stable diffusion? like once i download it i just have to move it into the stable diffusion model folder right?

[plush spire]

just make sure the version is right for your installation of SD

[trail wadi]

wdym?

[plush spire]

but yes, they all work if you have the right version but they all give different results. You'll struggle to create anything yourself to start with so just copy what other people have done on civitai

[trail wadi]

idk what version i have

[formal trout]

go juggernaut SD XL

[plush spire]

I mean there are checkpoints for SD versions SD XL is the latest I believe, but there is also SD 1.5 etc

[trail wadi]

how do i check which sd version i have

[plush spire]

it'll say on github

probably you have XL

[trail wadi]

https://github.com/AUTOMATIC1111/stable-diffusion-webui?tab=readme-ov-file

this is the one i got

[trail wadi]

could u help me check which version this is

i think this is gonna have to be trial and error

i'll just get juggernaut for now and see how that works. thanks for ur help guys

[plush spire]

you have SD3

[formal trout]

nah Automatic can handle several SD versions

you have to put thee model in the checkpoints folder

and then selected, it should load, then generate

[trail wadi]

so i got this error

do i need more ram?

[ornate elk]

What's your GPU?

[trail wadi]

rtx 4070

[ornate elk]

Thats a tool called stable-diffusion-webui it can use different versions of stable diffusion models like sdxl or 1.5

[ornate elk]

Then you need to edit the webui-user.bat, at the line commandline_args=
You need to add: --xformers
Then save and relaunch

[trail wadi]

was just about to say that whenever i run it it says no module xformers

[ornate elk]

Yea, thats why you need to enable it.
Its a performance boost and needs less vram

So you can generate at higher resolution or do heavier tasks

[trail wadi]

so like this?

[ornate elk]

Yep

[trail wadi]

alr. preciate it

[ornate elk]

Np

[ornate elk]

Go for 1.5 or sdxl/pony based models.
Also always check the little info box on a models site.
Not everything is a Checkpoint (model).
Loras for example go into models/lora folder

[trail wadi]

so i should look at this thing?

[chrome rampart]

hi @ornate elk hi hope doing well. Question can i de-install Python and do re-install or risky? Seems to have issue that comfy/SD look for python on drive i want to remove

[ornate elk]

Hey, you can uninstall and reinstall without problems.
If you check "add python to path" when installing python, comfyui should find it in the new place again

[trail wadi]

any idea why its doing this?

and then when it generates its just a grey square

[chrome rampart]

super thks hopes SD also will then

[ornate elk]

That happens because you moved a lora into the models/stable-diffusio folder instead of models/lora

Loras can't generate images on their own

[trail wadi]

im using flux dev

[ornate elk]

They are like addons to help a model

[trail wadi]

is flux a lora? it didnt say it was

[ornate elk]

Flux dev is not supported with Automatic1111 webui

[trail wadi]

...

so i downloaded that 22gb file for nothing

[ornate elk]

Yep :/

[trail wadi]

wait what is it supported by then? i thought automatic1111 was the only/most popular stable diffusion image generator

[ornate elk]

It still is the most popular tool. Flux is just to new, they need to add support to it.

Currently flux only works in comfyui or Webui forge

[ornate elk]

If not you need to delete the venv folder. Then it will

[trail wadi]

which model do u use? u seem experienced so i'll take the one u recommend

[ornate elk]

I use a lot different models.
For realistic stuff you can't go wrong with Juggernaut XL
For Anime stuff try out AAM Anime Mix

Both are sdxl based

[trail wadi]

thank goodness i have insane internet speeds. all these model files are massive

[trail wadi]

i tried juggernaut xl. i have it rn. it was pretty unrealistic with what i thought was a basic prompt

[ornate elk]

Sdxl models are larger (6gb) than 1.5 based models (2-4gb)

[trail wadi]

how intricate do the prompts have to be

[ornate elk]

Sdxl can generate higher quality by default as its trained on 1024x1024

[ornate elk]

Some models work with only words like,
sunset, beach, waves, etc.
But you can also write like:
A scene of a sunset at a beach with big waves and summer atmosphere

[trail wadi]

could u give me a sample prompt for juggernaut xl? or do u know where i can find sample prompts. i want to generate it and work by editing the prompt so i can get the hang of it

[ornate elk]

Sure, but would need to startup my PC real quick

[calm raptor]

I'm getting this error as of late, I don't even know where to start troubleshooting this. any suggestions?

# ComfyUI Error Report
## Error Details
- **Node Type:** KSamplerAdvanced
- **Exception Type:** UnboundLocalError
- **Exception Message:** local variable 'model_options' referenced before assignment
## Stack Trace

[trail wadi]

is ur pc up and running yet?

[ornate elk]

here:
postapocalyptic, dystopic image, a panther inside a living room, destroyed walls, overgrowing plants, masterpiece, intricate, best quality, highly detailed, low light, dark,

Negative prompt: disfigured, kitsch, ugly, oversaturated, grain, low-res, Deformed, blurry, bad anatomy, disfigured, poorly drawn face, mutation, mutated, extra limb, ugly, poorly drawn hands, missing limb, blurry, floating limbs, disconnected limbs, malformed hands, blur, out of focus, long neck, long body, ugly, disgusting, poorly drawn, childish, mutilated, mangled, old, surreal, text, blurry, b&w, monochrome, conjoined twins, multiple heads, extra legs, extra arms, meme, deformed, elongated, twisted fingers, strabismus, blurred, watermark

Steps: 30, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 7, Seed: 3684358796, Size: 1024x768, Model hash: d91d35736d, Model: juggernautXL_juggernautX

[trail wadi]

what does schedule type and steps do

and what does the seed do

and what does cfg scale do

ugly as a negative prompt is killing me. is it normal to have triple the positive prompts as negative prompts?

[ornate elk]

steps is how often the ai denoises the start noise image. so the more steps the more stuf will change. (more steps are not better), mostly 25-35 is a good start.
cfg is the creativity. the lower, the more it will go for only the prompt.
seed is just a random number.

schedule type is like what math behind it processs the image (the denoising)

euler a and dpm++ 2m are good, leave the scheduler to automatic

[trail wadi]

ETA 20 mins 😭

[ornate elk]

for sdxl models, its not needed to have so much negative prompts, you can also try using the half of it

[ornate elk]

for real?

you have a 4070 with 12gb vram right?

make sure youre not running any Games or Wallpaper Engine in the Background

[trail wadi]

...what's wrong with wallpaper engine?

[ornate elk]

its buggy when running wallpaper engine + stable diffusion at the same time. it can freeze and even crash the whole PC xD

also it slows down the generation

[ornate elk]

ohh you have a laptop 4070 😮

[trail wadi]

si

[ornate elk]

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

[trail wadi]

okay. to what?

[ornate elk]

add --medvram-sdxl to the line where you put xformers to

so its like
--xformers --medvram-sdxl

[trail wadi]

what does that do?

[ornate elk]

that lowers the vram amount of sdxl models by loading them splitwise to speed the gen up

[trail wadi]

real quick where do pony models go

[ornate elk]

models/stable-didfusion

[trail wadi]

yeesh. im scrolling through the gallery of some of these models and there seems to be a certain... theme with the generated images

[ornate elk]

on civitai you also find examples and prompts of images

[trail wadi]

what does this mean in a prompt "score_9, score_8_up, score_7_up"

[ornate elk]

thats pony specific

pony models got trained on "scores"

the higher the better, 9 is "best quality"

[trail wadi]

wait do u use a lora alongside a checkpoint?

[ornate elk]

yep, it helps the model to generate a specific character or style the model wouldnt know

[trail wadi]

ohhhhhhhhhhh

there needs to be like a school or something for stable diffusion i swear

[balmy matrix]

lmao no kiddin'

how do I edit command line arguments with ComfyUI? Here? Or do I do it elsewhere?

[ornate elk]

yep there

[trail wadi]

wow these prompts have to be insanely lengthy and intricate

and the negative prompts are even wilder

[balmy matrix]

They don't have to be

It takes a lot of figuring things out, but not everything has to be an essay

I like the flexibility of Pony, but boy do I wish it understood just full on sentences/prose and the like really well, I'd prefer typing that way.

[lost crescent]

Is it still possible to make textual inversions/embeddings in Automatic1111?

[calm raptor]

I tried all that to fix it, turned it was a node that wasn't updated yet or something that caused the problem. the weird part is that i uninstalled that custom node and my workflow ran without issues so maybe that node wasn't even involved? it's all very strange

I don't remember, didn't pay much attention to that. it was some kind of tile diffuser i think

[chrome rampart]

@ornate elk One more if I may. Best to put/install python on c (bootdrive) or drive were SD and comfy are installed? Or does not matter?

[ornate elk]

i would install it on C at its default path

[chrome rampart]

when i rename my old python (maybe also git i git :'git' is not recognized as an internal or external command,
operable program or batch file.
Couldn't launch python
exit code: 9009
stderr:
Python was not found; run without arguments to install from the Microsoft Store, or disable this shortcut from Settings > Manage App Execution Aliases.
Launch unsuccessful. Exiting.
Press any key to continue . . .

smart to reinstall GIT also?

ok CS i will use C drive

damn maybe issue is git and not python

[formal trout]

my basic structure for a prompt is like some form of sentence

Medium/style, angle/distance, subject (with important description), action, location, lighting, details, style/medium details

example:

80's analog polaroid photo, from side long angle, a cute dog celebrating its birthday at backyard, golden hour, realistic, detailed

[ornate elk]

you also should install git on C

so everything on the system can find git and python

also run the python 3.10.11 installer again,
click modify,
click next,
check "add python to environment variables"

[chrome rampart]

ok will do , first deinstall both en redo. thks CS

[formal trout]

actually there is this channel "Academia SD", SD Academmy. Is in Spanish, but you can follow it with english subtitles. Really doesn't waste your time, gets straight to the point and explains all the steps for installation and operation.

I wouldn't recommended if it wasn't that good. Other channels just makes you waste time with long obnoxious videos.

[glad shale]

Hello, hope its okay to share a post here.

Help optimizing existing workflow (Paid Contractor)

I'm looking for someone who can help texture a 3D model using ComfyUI. Currently, I'm employing a workflow with custom nodes that generates a 3D model texture, projects it onto the model, and produces a UV texture map. However, I suspect there's room for optimization at the node code level. Achieving consistent style between the front and back views is another challenge. This is a paid opportunity.

If you have experience or interest in:
- 3D model texture inpainting workflows
- Creating a consistent style across different perspectives of a 3D model
- custom node programming
- UV texture map generation
- Depth map preprocessing
- UV map ControlNets

Please DM if interested. Thank you for reading!

[chrome rampart]

@ornate elk Thanks so much for your advise/help it worked again already for SD and Forge YAYYYYYYYYYYYY🥳 . Now going to try Comfy. Have amazing weekend

[heady pine]

I redownloaded ComfyUI, and I am still facing the same problem. I am pretty lost as to what is causing this; the cmd looks fine as well.

This happens anywhere the folder is located as well, so it's not as if something is interfering with it depending on where it's located; it happens even after being extracted from the 7-zip folder.

I was hoping to check out this new Flux checkpoint since it doesn't work with Automatic1111, but it looks like I'm facing a few problems before I can even start to use it.

I generated a picture I made in Automatic1111 with ComfyUI to see what the difference would be before this, so it was working fine, but then I updated ComfyUI with the manager since errors popped up when trying to use Flux, so I thought ComfyUI must have been outdated since I last used it, and since then it hasn't worked ever since, even when doing a fresh download of it.

Custom_nodes shows again, but this time it doesn't say anything about not being able to import them.

I have done what the guides online have told me to do when it comes to downloading ComfyUI, so I don't think it's something I'm missing.

[lost crescent]

If its not affecting the output, then ignore it?

[heady pine]

I'm only showing the cmd so someone might be able to pick something out if there is anything to pick out, the problem is, this is what I am facing when opening ComfyUI

No matter if it's the folder I have had downloaded from months ago or a fresh download of it today, this is what I get every time.

[lost crescent]

Open CMD navigate to ComfyUI folder - then use git reset --hard
then git pull - and see if that works?

[quasi pine]

I have a question about upscaling.
Even with the same resolution (I am using 4K resolution), does the quality of an image that has been upscaled once and one that has been upscaled twice decrease, even if the final resolution is the same?

If you were to produce a 4K resolution image, would you perform the upscale more than once?

[unique glacier]

it dont have to be lengthy or intricate. really depend on what you are looking for. some people pride themself for making good render with just simple prompt. cuz too many prompt just confuses the model and it started making it garbled.

[lost crescent]

https://www.gimp-forum.net/Thread-How-much-size-upscale-or-size-downscale-before-pixelation-or-degradation-of-image

[ornate elk]

Make sure the webui is whitelisted in any browser adblocker

[heady pine]

It was AdBlock, omfg, all this time, and it was AdBlock, sick. I guess updating ComfyUI must have added something AdBlock didn't like; this has literally never been a problem before now.

Like always, thank you for your help. @ornate elk

And I appreciate your effort to help me as well @lost crescent

[ornate elk]

np, also you shouldnt use multiple adblockers, and also ublock origin is the best adblocker as its opensource and doesnt whitelist ads for moneyl ike the others do. (Adguard doesnt do this too)

[heady pine]

I will have to delete the others then and download that one, thanks.

[ornate elk]

Np, as Google chrome is going against adblockers you may need to switch sometime anyway to an other browser.
But Ublock origin also made a light version which works with chrome in the future.

[heady pine]

Awesome

[proper reef]

I've only ever been able to get a single lora to work with Stabe Diffusion. Whenever I try to run a different one I get a runtime error saying there's not enough memory. I'm already using --lowvram. I have no idea how I got it to work to begin with because it was giving em this same issue when I first started using Stable Diffusion, until one day it magically started working.
Also, I heard you need to use the lora tag, or whatever it's called, for the lora to work. It's where you add the lora file name in the prompt enclosed in <>, but I've never had to do that and my lora works fine, but it's only that one lora and no other.
My system doesn't have a lot of vram (around half a gig) but Stable Diffusion was able to generate multiple 1024x1024 images, or 512x512 images upscaled by 2 over a few hours without problem, so there shouldn't be an issue with memory unless stable diffusion can only keep one lora in memory at a time.

[ornate elk]

hey whats your gpu?

[proper reef]

it's not nvidia if you're wondering. I have an AMD Radeon because Bluestacks wouldn't work with an Intel GPU

[ornate elk]

okay, thats not an issue. but whats the gpu model?

and whats in your webui-user.bat?

[proper reef]

GPU model is...

AMD Radeon, like I said.
webui-user.bat has the following:
--lowvram --precision full --no-half --skip-torch-cuda-test --no-half-vae

[ornate elk]

AMD has a lot of radeon graphic cards

can you open task manager and check the name under the performanmce tab

and there click on the GPU

[proper reef]

It just says AMD Radeon(TM) Graphics...
I think it's a Radeon 7, but the sticker on the machine is so worn down I can't read it to confirm

[ornate elk]

ah so your on a laptop?

[proper reef]

yeah

[ornate elk]

ah alright

what you then can try is to use my AMD DirectML install guide.
Its for AMD GPUs and is faster then using the CPU mode your currently using

you can find it here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides#amd-automatic1111-with-directml

you dont need to delete the other webui, so you can compare the speeds and image sizes

[proper reef]

thanks, I'll try it out in a bit!

[ornate elk]

np, let me know if you have questions

[flat lagoon]

has anyone attempted to train data sets by k-mean cluster groups?

[fair oxide]

how do i combine these into a single model file?

[ornate elk]

where did you get these? xD

[fair oxide]

Hugging face

[brazen bramble]

@ornate elk if you have the time could we set up pony now please ?

[ornate elk]

sure

[ornate elk]

what model is it?

sdxl, pony or something different?

[brazen bramble]

or something like that

i enabled the f8 thing like you said yesterday

[ornate elk]

ok, but the questions wasnt for you

[fair oxide]

Oh I just realized this was stable diffusion lol. It's an LLM

[ornate elk]

okay and whats in your webui-user.bat ?

[brazen bramble]

oh sorry my bad

[ornate elk]

your card was a gtx1060 right?

[brazen bramble]

gtx 1650

[ornate elk]

ah okay

[brazen bramble]

and im currently on SD 1.5 with A1111

[ornate elk]

okay, then you can download a pony model and put it into the models/stable-diffusion folder

[brazen bramble]

ok bet i'll do that now

is there anytong else i need to do ?

[ornate elk]

rn not

if you have only 16gb of normal RAM you may need to increase the windows pagefile

[brazen bramble]

i just added it to the folder

[brazen bramble]

can I still use the loras and textual inversion that I used before or do I need to get new ones ?

[ornate elk]

you need new ones that are made for pony

[brazen bramble]

Good to know thank you for the help!

[ornate elk]

np 🙂 and did it worked?

[brazen bramble]

It worked it’s a little laggy but nothing too unbearable

[ornate elk]

alright nice, if its to lagy you would need to increase the windows pagefile

[brazen bramble]

I checked and it says I only have 8gb of ram so idk if I would be able to

[ornate elk]

oh then you should increase it

[brazen bramble]

Is that hard to do ?

[ornate elk]

not really.
follow this guide until your on the pagefile screen:
https://www.tomshardware.com/news/how-to-manage-virtual-memory-pagefile-windows-10,36929.html
Then make sure its only enabled for the C drive and disabled for any other drive.
Then set it to custom sized and set 16000 Min and 24000 Max
Then save and restart your PC

[brazen bramble]

Gotcha thank you so much for all the help your a legend

[ornate elk]

no problem 🙂

[proper reef]

I'm getting this error after a generation finishes:
RuntimeError: The GPU will not respond to more commands, most likely because some other application submitted invalid commands. The calling application should re-create the device and continue.

[ornate elk]

Can you show the full cmd log?

[proper reef]

I'll need to have stable diffusion try and generate a whole new image

That could take between 15 minutes to an hour

[ornate elk]

Have you tried my install guide?

[ornate elk]

you can just copy the whole cmd text if you didn't closed it already

[proper reef]

I did close it. Multiple times, because I wanted to restart it with different settings to see if it fixed it.

twelve minutes cmd says, 25% done

[ornate elk]

What's in your webui-user.bat right now?

[proper reef]

--use-directml --lowvram --opt-sub-quad-attention --opt-split-attention --no-half-vae --upcast-sampling --precision full
I think I added --precision full because it fixed another problem I was having

Here's the error:

[ornate elk]

Dont add --precision full

It slows down

Also make sure you use model files with 2gb size

Like Dreamshaper v8 for example

[proper reef]

the one I was using before was 6GB. I was able to get that to run using my previous ui

[proper reef]

Well I tried using Dreamshaper v8 and I got this error:
RuntimeError: Could not allocate tensor with 134217728 bytes. There is not enough GPU video memory available!

it also can't generate the images I want

[brittle lantern]

need some help using zluda on windows my gen gets to 96% then takes quite a while then when it completes it gives full black gen any fix?

[proper reef]

Went back to my previous setup and found the problem. The model I'm using doesn't work with most loras because it's so different from every other model. I'll try using the setup recommended here for when I want to use a lora.

[vocal burrow]

remove --disable-nan-check

it's doing nothing except hide useful error messages

[timber oak]

hello : D

whats the difference between canny and union?

[ornate elk]

Mostly your trying to use a lora as model.
Remove --disable-nan-check from the webui-user.bat so we can see the real error

[mystic lava]

Hey can anyone help me wit stable diffusion

venv "C:\Users\Operator\Desktop\AI GEN\stable-diffusion-webui-directml\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.10.1-amd-4-gb0d9eb6d
Commit hash: b0d9eb6df1f6631a49988a9f705ff568f908aa2b
Traceback (most recent call last):
File "C:\Users\Operator\Desktop\AI GEN\stable-diffusion-webui-directml\launch.py", line 48, in <module>
main()
File "C:\Users\Operator\Desktop\AI GEN\stable-diffusion-webui-directml\launch.py", line 39, in main
prepare_environment()
File "C:\Users\Operator\Desktop\AI GEN\stable-diffusion-webui-directml\modules\launch_utils.py", line 590, in prepare_environment
raise RuntimeError(
RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check
Press any key to continue . . .

sorry for text wall

😭

[ornate elk]

Hey, what's your GPU?

[mystic lava]

Merc Rx 7900 xt 20gb vram

[ornate elk]

Ah okay, then you need to follow my AMD Automatic1111 with ZLUDA guide.
You can find it on the first link of the pinned messages

[mystic lava]

https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides
this>

?

[ornate elk]

Yep

[mystic lava]

ill give it a shot

[ornate elk]

Your GPU is supported by rocm and zluda so you don't need directml

I'm using a spahire 7900xtx

[ornate elk]

Sure, let me know if any step isnt clear

[mystic lava]

Ive done

[AMD] Automatic1111 with DirectML
Install Automatic1111 Directml Fork on Windows:
GPUs Supported: Every AMD GPU and even AMD Integrated GPUs
DirectML is slower than Zluda and requires a lot more VRAM.

Important: If you tried to install SD before and failed, do this: open up a cmd and type pip cache purge to remove the nvidia stuff.

Check the Preparation

Make a new Folder on your drive (not on Desktop, downloads, documents, Programms, Onedrive) and name it Ai for example: C:\Ai
You go into the folder you created in this case Ai, then click in the File Explorer bar (not searchbar) and type cmd then press enter.
Then you copy and paste this command:
git clone https://github.com/lshqqytiger/stable-diffusion-webui-amdgpu && cd stable-diffusion-webui-amdgpu && git submodule init && git submodule update
Press enter and after its done you can close the cmd.
Edit the webui-user.bat (right click), At the line COMMANDLINE_ARGS= You add: --use-directml --medvram --opt-sub-quad-attention --opt-split-attention --no-half-vae --upcast-sampling
If your GPU has only 8GB VRAM or less and you run into out of memory, you may need to change --medvram to --lowvram.
RX 5xxx cards additional need: --no-half
After that save and launch the webui-user.bat inside the stable-diffusion-webui-ammdgpu folder.
The Installation can take a while and when its done, the Webui will autolaunch in your Default Browser.

done alat

still nothing

[ornate elk]

Yea don't use the directml guide

Its really slow and only for older gpus

[mystic lava]

fire

😭

Which do i use

[ornate elk]

Automatic1111 with ZLUDA

[mystic lava]

Comfyui ?

aight]

[ornate elk]

No

Automatic1111

The first guide from my AMD guides

[mystic lava]

aiight

doi i need to reisntall it all

😭

[ornate elk]

Yes

[mystic lava]

peak!

Lets do it

@ornate elk

Im on the part

where i skip to step 3

what do i do now

wit the rx 7900 xzt

xt

[ornate elk]

You do step 1, 2, 4 and 5

[mystic lava]

bet

This is in getting zluda to run

[mystic lava]

which zula do i download

ZLUDA-linux-amd64.tar.gz
7.48 MB
last week
ZLUDA-windows-rocm5-amd64.zip
13.9 MB
last week
ZLUDA-windows-rocm6-amd64.zip
13.9 MB
last week
Source code
(zip)
last week
Source code
(tar.gz)

rocm 6 or 5

[ornate elk]

the rocm 6

[mystic lava]

done

[ornate elk]

as you installed hip sdk 6

[mystic lava]

is this right

for patsh

@ornate elk

mb for pinbgs

@ornate elk

crash

wait what ur pc shows when ur rebooting thats sick asf

[ornate elk]

No

[mystic lava]

peak

what i do wrong

%HIP_PATH%bin im guessing is wrong

im not sure which part trhgo

tho

[ornate elk]

Then add the 2 new entrys

[mystic lava]

yea

i did

😭

[ornate elk]

Show me

[mystic lava]

I did C:;

[ornate elk]

OK looks right

[mystic lava]

do you know what my issue is then:?

[ornate elk]

Your folder is wrong

You have git cloned into the wrong folder

You have stable-diffusion-directml that folder should be deletet

You only need to have a
Stable-diffusion-webui-amdgpu folder

[mystic lava]

wait wdym

what do i do

should Stable Diffusion webui gen be in base folder

[ornate elk]

You need to delete the stable-diffusion-webui-directml folder

You then need to git clone the webui again

[mystic lava]

where do i clone it to

😭

[ornate elk]

Step 2 from my guide

[mystic lava]

can i just copy the files

to the drive folder

[ornate elk]

No

[mystic lava]

Peak

so i start again

wait no

i just neeed t o reclone

right

[ornate elk]

Yes

[mystic lava]

Local disk c

[ornate elk]

Only git clone into the ai folder

[mystic lava]

Folder

[ornate elk]

Then your done

[mystic lava]

Clone into it

yess?

[ornate elk]

Yes

[mystic lava]

does the folder need to be named anything special

[ornate elk]

Not on desktop

Make a folder on C

For it

Call it ai

[mystic lava]

yea

aight

cloning

C:\Windows\System32\cmd.exe

this is path

C:\ImageGen

I mean

[ornate elk]

No

[mystic lava]

😭

[ornate elk]

Yes

[mystic lava]

Hoorray!

venv "C:\ImageGen\stable-diffusion-webui-amdgpu\venv\Scripts\Python.exe"
ROCm Toolkit 6.1 was found.
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.10.1-amd-4-gb0d9eb6d
Commit hash: b0d9eb6df1f6631a49988a9f705ff568f908aa2b
Using ZLUDA in C:\ImageGen\stable-diffusion-webui-amdgpu.zluda
no module 'xformers'. Processing without...
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.
C:\ImageGen\stable-diffusion-webui-amdgpu\venv\lib\site-packages\pytorch_lightning\utilities\distributed.py:258: LightningDeprecationWarning: pytorch_lightning.utilities.distributed.rank_zero_only has been deprecated in v1.8.1 and will be removed in v2.0.0. You can import it from pytorch_lightning.utilities instead.
rank_zero_deprecation(
Launching Web UI with arguments:
ONNX failed to initialize: Failed to import diffusers.pipelines.auto_pipeline because of the following error (look up to see its traceback):
Failed to import diffusers.pipelines.aura_flow.pipeline_aura_flow because of the following error (look up to see its traceback):
cannot import name 'UMT5EncoderModel' from 'transformers' (C:\ImageGen\stable-diffusion-webui-amdgpu\venv\lib\site-packages\transformers_init_.py)

rocBLAS error: Cannot read C:\Program Files\AMD\ROCm\6.1\bin/rocblas/library/TensileLibrary.dat: No such file or directory for GPU arch : gfx1036

rocBLAS error: Could not initialize Tensile host:
regex_error(error_backref): The expression contained an invalid back reference.
Press any key to continue . . .

@ornate elk

[stone zodiac]

**Hello. I have a RTX 4080 (16GB VRAM) with 32GB of DDR5 Ram on Windows 11. I'm a bit lost between all the Flux models (Dev, Schnell, FP8, FP16, GGUF, and so on). What should I use for fastest outputs ? I'm on ComfyUI. Thank you. **

[ornate elk]

Ahh that error means it detected your cpus igpu, instead of the 7900xt

You have to disable the Radeon TM Graphics in the device manager

Then relaunch

[ornate elk]

Hey, the nf4 flux dev model ia the faster currently with a good output

[mystic lava]

I shall do it when I go home

what do I need to do then step by step

[ornate elk]

Open the device manager,
Click on display adapters.
Right click on Radeon TM Graphics.
And click disable.

[mystic lava]

aight

where are you from btw

[ornate elk]

Germany

[mystic lava]

and what do U drink there

beer?

[ornate elk]

xD me not but most people do

[mystic lava]

was gonna donate enough for a beer lol

[ornate elk]

xD you can on my github page if you want ^^ there is a donate button on the main page

[mystic lava]

I got you

once I get it working I shall send one thing

something

[ornate elk]

thanks! appreciate it, and let me know when everything is running

also remember the first image will take 15-40 minutes. after that it should take seconds for an image

[mystic lava]

yea isaw why so long for first image?

is it just benchmarking stuff and first time encoding

[ornate elk]

Its processing cuda/zluda files for the first run

[mystic lava]

fair

praying this works when I get home

what can wat I'm installing actually do

text to image only or video and stuff to

[ornate elk]

Txt2img and small video generation

Or video2video conversion

[mystic lava]

Les yo

are there extensions ?

[steady thistle]

I now have an Nvidia card, 4070 super, and I'm ready to do a fresh SD install. I've used A1111, but should I go with Comfy? Other options? I'd like to do video eventually as well.

[mystic lava]

venv "C:\ImageGen\stable-diffusion-webui-amdgpu\venv\Scripts\Python.exe"
ROCm Toolkit 6.1 was found.
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.10.1-amd-4-gb0d9eb6d
Commit hash: b0d9eb6df1f6631a49988a9f705ff568f908aa2b
Using ZLUDA in C:\ImageGen\stable-diffusion-webui-amdgpu.zluda
no module 'xformers'. Processing without...
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.
C:\ImageGen\stable-diffusion-webui-amdgpu\venv\lib\site-packages\pytorch_lightning\utilities\distributed.py:258: LightningDeprecationWarning: pytorch_lightning.utilities.distributed.rank_zero_only has been deprecated in v1.8.1 and will be removed in v2.0.0. You can import it from pytorch_lightning.utilities instead.
rank_zero_deprecation(
Launching Web UI with arguments:
ONNX failed to initialize: Failed to import diffusers.pipelines.auto_pipeline because of the following error (look up to see its traceback):
Failed to import diffusers.pipelines.aura_flow.pipeline_aura_flow because of the following error (look up to see its traceback):
cannot import name 'UMT5EncoderModel' from 'transformers' (C:\ImageGen\stable-diffusion-webui-amdgpu\venv\lib\site-packages\transformers_init_.py)

@ornate elk

Its sstuck

after i launched it

on this

idk if its still loading

doiesnt close when i hit error

nmvm it olaunched

holy moly

its useign no pc resources

ioin think its working correctly

peak

[ornate elk]

You need to download a model

From Civitai.com

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

Put it into the models/stable-diffusion folder

[mystic lava]

which is ur favourite

[ornate elk]

Then select it in the dropdown

[ornate elk]

I don't have a favourite one. For realism or anime I use different models

[mystic lava]

fair

SO just mess abt weit it

[ornate elk]

So try Dreamshaper v8 and then you can download whatever you like and want to test

[mystic lava]

which dropdown

wait do i relaunch it ye

reloading now

[ornate elk]

Dropdown is on the top left

[mystic lava]

aoight go tit

i think

it sstill going

how2 long does that take to process?

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument index in method wrapper_CUDA__index_select)

wtfliop

goated prompt i didnt know what to do

[ornate elk]

Make sure your on the latest AMD driver

Then restart the PC and try again

[mystic lava]

oh i havent restarted pc

[steady thistle]

Hope you're well! I haven't been here for awhile but just got a new rig. Now that I'm on Nvidia, I will be doing a fresh install, but I want to get video as well as images. I've been using A1111, but I'm open to changing. What UI would you recommend?

[mystic lava]

aight restarted

here we go

i think its working

it hasnt stopped

so ima assume itll be working

im not seeing massive gpu usage tho

which i assumed i WOuld see

[ornate elk]

Hey, you can try out Comfyui if you want. Its not easy but supports a lot of stuff

[ornate elk]

Yea the first 15-40mins it won't use the gpu

Check the cmd for any errors

It shouldn't show anything

[mystic lava]

venv "C:\ImageGen\stable-diffusion-webui-amdgpu\venv\Scripts\Python.exe"
ROCm Toolkit 6.1 was found.
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.10.1-amd-4-gb0d9eb6d
Commit hash: b0d9eb6df1f6631a49988a9f705ff568f908aa2b
Using ZLUDA in C:\ImageGen\stable-diffusion-webui-amdgpu.zluda
no module 'xformers'. Processing without...
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.
C:\ImageGen\stable-diffusion-webui-amdgpu\venv\lib\site-packages\pytorch_lightning\utilities\distributed.py:258: LightningDeprecationWarning: pytorch_lightning.utilities.distributed.rank_zero_only has been deprecated in v1.8.1 and will be removed in v2.0.0. You can import it from pytorch_lightning.utilities instead.
rank_zero_deprecation(
Launching Web UI with arguments:
ONNX failed to initialize: Failed to import diffusers.pipelines.auto_pipeline because of the following error (look up to see its traceback):
Failed to import diffusers.pipelines.aura_flow.pipeline_aura_flow because of the following error (look up to see its traceback):
cannot import name 'UMT5EncoderModel' from 'transformers' (C:\ImageGen\stable-diffusion-webui-amdgpu\venv\lib\site-packages\transformers_init_.py)
Loading weights [879db523c3] from C:\ImageGen\stable-diffusion-webui-amdgpu\models\Stable-diffusion\dreamshaper_8.safetensors
Creating model from config: C:\ImageGen\stable-diffusion-webui-amdgpu\configs\v1-inference.yaml
Running on local URL: http://127.0.0.1:7860

To create a public link, set share=True in launch().
C:\ImageGen\stable-diffusion-webui-amdgpu\venv\lib\site-packages\huggingface_hub\file_download.py:1150: FutureWarning: resume_download is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use force_download=True.
warnings.warn(
Startup time: 13.0s (prepare environment: 17.3s, initialize shared: 2.6s, load scripts: 0.4s, create ui: 0.3s, gradio launch: 0.6s).
Applying attention optimization: Doggettx... done.

Nothying

ionm think

i think we're good

if i leave it and no image ima be so sad

😭

[ornate elk]

Looks good!

[ornate elk]

Now just wait until the first image is done

[steady thistle]

Alright, I'll give that a go. Sounds like it's the fastest which will matter for video.

[mystic lava]

peak

also im not seeing a tab for video generationm

[ornate elk]

Its not the fastest but has the most video capable nodes and workflows

[steady thistle]

Ah ok

[ornate elk]

For that you need Extensions

You can install them later

But first start with txt2img

[mystic lava]

yea

lets not jumop the gun

https://www.youtube.com/watch?v=A4WgT0QOXhs

wanna make stuff

similar to this

the ai use i9n this

is incredible

its using midjourney

but i can prolly get similar results with thjis

[steady thistle]

I saw your pinned message on installing ComfyUI, but don't I need to install Python first? And which version should I use?

Also with Git

[mystic lava]

it cooked fr

[ornate elk]

On my github guides its all linked

[ornate elk]

Okay so now your ready

[mystic lava]

it starts HAVILY struggling

once i up the resolution

[ornate elk]

Also don't generate 2048x2048