🤝｜tech-support

[pulsar grail]

Thanks a bunch. Just got a new comp and loved SD on my old one. Can't wait to see what difference an I-9 makes over an i5-6600k lol

[fair oxide]

It's actually longer but it wouldn't let me upload it

[ornate elk]

NP, for SD the CPU doesn't make to much of a difference

[ornate elk]

What model did you tried to use?

[pulsar grail]

You're right of course, but at least I can load models in 20 seconds instead of 5 minutes now. 🙂

[fair oxide]

v1-5-pruned-emaonly

stable-diffusion-v1-5

[ornate elk]

Okay delete that model and download any 1.5 based model from Civitai.com
For example Dreamshaper v8

[fair oxide]

I'm downloading now

SD-Zluda\stable-diffusion-webui-amdgpu\models\Stable-diffusion Am I loading the model to the correct location?

[ornate elk]

Yes

[fair oxide]

.safetensors Is it the right extension?

[fair oxide]

RuntimeError: "LayerNormKernelImpl" not implemented for 'Half'

I got a long error message again

[ornate elk]

Okay try add --no-half to the webui-user.bat

Then relaunch

[fair oxide]

The error seems to be fixed, I will share the result.

[fierce hollow]

how do I run Stable Diffusion 4d?

[ornate elk]

Did it worked?

[fair oxide]

yes it worked thanks

[ornate elk]

No problem 🙂

[void sun]

Hey guys! Pretty new to SD and PCs in general. Right now I'm borrowing a friends PC for SD with a 1080 TI GPU. My other friend offered me to use his computer with 4090 instead. Maybe a stupid question, will I notice any other differances other than speed?

[karmic crown]

only speed, but the speed difference is drastic

[void sun]

Ah ok! Thanks! So no difference in image quality?

[karmic crown]

nope

but 4090 is few times faster

like seconds vs minutes

[void sun]

Oh wow.

So the only thing that matters for Stable Diffusion is GPU?

[karmic crown]

so its almost 10 times faster

[void sun]

Oh god haha

[karmic crown]

45 it/s = 1 image per 1 or 2 seconds

[karmic crown]

yes

but, you need to have enough RAM too

[void sun]

The one im using has 64gb.

Which is guess is enough?

And would you say that 4090 is optimal for SD? Or is there anything "better"

[karmic crown]

for home users, no

NVIDIA A100 is faster 🙂

if you have spear $10000, go and buy A100 🙂

32GB of RAM and 4090 will work great

[void sun]

So A100 is better than H100?

[karmic crown]

no

[void sun]

Or maybe it's not for home/office computers?

[karmic crown]

H100 is more faster and expensive

[void sun]

But again, the only thing that will improve is speed?

[karmic crown]

yes

[void sun]

So like h100 would be for massive around the clock generation basically

[karmic crown]

yes

[void sun]

And same guess for running LLM?

goes*

Like 4090 will still be good?

[ornate elk]

Its the best consumer GPU for Ai stuff right now so yes

The more VRAM you have the more advanced tasks you can do with ai

Not only speed is better but you can also do video generation on it as example

[void sun]

Ah I see! Thank you so much!

[karmic crown]

i am sorry, i had to urgently go earlier

[sinful marten]

hey guys so i have tydiffusion inside 3ds max now, is there a resource to explain all the special characters and symbols and what they do for sd. Like [ ] --1, things like that not needing a list of how to prompt but how to usde the special charachters and symbols to add more dept to your image, hope this makes sense?

[karmic crown]

you should check their documentation since that is not a "normal" way of using comfy

[sinful marten]

thnx

[karmic crown]

do you have any examples?

By using the {scheduler} keyword, you can control where (within your main prompt) the scheduled prompts will appear. If the {scheduler}} keyword is not included in your main prompt, scheduled prompts will be appended (rather than inserted) into the main prompt.

i was afraid it's something like this

in normal case you select scheduler from a dropdown

@sinful marten my advice would be to install https://github.com/AUTOMATIC1111/stable-diffusion-webui since it is easer to use than comfy, you will understand most thing very quickly, later you can play with comfy, and you will undestand your tool properly

[sinful marten]

this is great i certainly appreciate it, i will try it but i believe comfy is the only one tydiffusion works with but i will look into it, thanks again

i think for me confy is easy as hell but il check the other, its a built in gui inside max so its not a BROWSER BASED NODE SYSTEM....

its all internal within max

one click download for all models accepted

[worthy sapphire]

I tried running webui.bat today and everything basically reinstalled. Now it just seems to work for some reason. The first render took about 10 minutes but I remember reading that that is normal. I don't think I did anything to change the way the program runs.

[oak isle]

Hi, I started using automatic1111 today, but when I try to generate an image, whatever the prompt, size selected, etc., I always get the out of memory error message. I've tried to read a bit about this error, but I must admit I'm a bit (a lot) lost... I was wondering if someone might have some time to help me solve it... Thank you in advance.

[unique field]

Is it possible to send or pull data in terms of a prompt to ComfyUI or have ComfyUI send out results to another program or process on your LAN?

[unique field]

How much RAM do you have? Also which GPU are you using? Also, what size images are you trying to make? Which model? 1.5? SDXL?

[rich garden]

I generated this first image and it looks good but there’s some errors like 5 fingers on one hand and some other errors in the full original generated image (this one was zoomed in to hide most of the lewd stuff) so I tried doing img2img and it fixes a couple of things but one thing that’s been annoying me is that it won’t regenerate the hand the same as the original. I’m trying to regenerate this image with the same hand that has 2 fingers on her cheek and removing the extra 5th finger. Can someone plz help?

PING ME!!

[hallow latch]

I updated my Python and get this error, how do I solve it easy?

[unique field]

If you want to keep the face and part of the hand the same (the fingers) use masking, only mask the areas you want changed. Img2Img just uses the previous one as a refrence point and make something somewhat similar, and not identical

Make sure to have enabled the mask inpainting

[restive sparrow]

in paint or image to image at at least 50denoise with some hand fixing negative prompts

[oak isle]

Hi, thank you for you reply. Do you think I can dm you those infos?

[unique field]

sure, go ahead

[rich garden]

I’ve never used the masking tool so I’ll probably need some guidance with that later when I’m on my PC. Thanks for letting me know

[rich garden]

I’ve tried changing the prompts for fixing the hand but it doesn’t seem to work out so well cuz it gives me results like you see in the second image with the one index finger. What prompts do you recommend I add/edit?

[unique field]

inpaint/mask is the key bit

[fair oxide]

spare

this is a spear

one might rather spear somone for spare $10 grand

[unique field]

tbh, that's beyond overkill

unless you are doing a LOT of AI art nonstop as a business.

But even then, a 4090 or 3090 will most likely get the job done fine

If you have to question "Do I need that?" You dont need it.

[ornate elk]

Ah okay, but good that it works now

[ornate elk]

Hey, what's your GPU?

[karmic crown]

english is my 3rd language, so i should be pardoned for misspelling some things from time to time 🙃

[fierce hollow]

how do I run Stable Diffusion 4d?

@ornate elk help..

[ornate elk]

I don't know

Didnt looked into it yet

[hollow quartz]

Hi, I tried to convert a video into an AI video using controlnet and animateDiff, but seem to get this error:
"2024-07-30 09:56:55,808 - AnimateDiff - ERROR - [AnimateDiff] Error extracting frames via ffmpeg: [WinError 2] Das System kann die angegebene Datei nicht finden, fall back to OpenCV." (Btw, the sentence "Das System kann die angegebene Datei nicht finden" is German and translates to "The system cannot find the specified file") Does anyone happen to know what I have done wrong?

[hollow quartz]

Btw this is the message that came before the error: "Attempting to extract frames via ffmpeg from C:\Users\Anwender\AppData\Local\Temp\gradio\42a37a1b2d90164defdf9f652dc6420cfb0e3cdb\ShortRandomTestVideo.mp4 to I:\A1111\stable-diffusion-webui/tmp/animatediff-frames\ShortRandomTestVideo-b5af14f8"

[ornate elk]

Do you have ffmpeg installed and in your Windows environment path added?

[hollow quartz]

Tbh i don't even know what ffmpeg is lmao

where can I check if I have this added in my path?

[honest tendon]

https://www.youtube.com/watch?v=5qjR9aFRg1A

is there a way to download this for ZLUDA?

i need it to access my ROCM stuff

[ornate elk]

I can check it later

[fierce hollow]

Where can I read about it? At least what environment to run it in, it doesn't work in auto1111( Direct at least...

[ornate elk]

Its not an extension

Checkout the github page of that project

[calm cairn]

coming here because auto update seems to have ruined something

I installed Forge last night and was surprised how much faster it was than 1111. I messed with some settings (directory defailts for models, dark theme) and then ran it, and it updated at least 3 different things, and then it said "import failed" and "press any key to continue", which ends webui-user-bat program.

When I run Forge now, it does this again, so I cannot use it. What went wrong and how do I fix it? Thanks in advance

[ornate elk]

Hey, can you show the cmd log?

[calm cairn]

how do I find that?

Ah never mind, I can run "run" and get in. Should I be worried that webui-user.bat doesn't work? Or should I assume "run.bat" is OK to use and ignore webui-user unless I want to update? I am eager to use Forge but I don't wanna shoot myself in the foot if webui-user doesn't work anymore

[ornate elk]

Copy the text from cmd when launching the webui-user.bat

[hollow quartz]

Hi, I switched my Animate Diff motion model and now I can create Animate Diff videos, although I still get the ffmpeg error. Now I tried to generate a larger video and got another error: In Stable Diffusion it says: "OutOfMemoryError: CUDA out of memory. Tried to allocate 1.10 GiB (GPU 0; 8.00 GiB total capacity; 12.37 GiB already allocated; 0 bytes free; 14.65 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF"

[calm cairn]

seems that if I hit "ctrl+a" it counts as pressing any key to continue, and then it closes. I also cannot select text with mouse cursor. Should I just type it in manually?

ok I just typed it out manually, except shortened some long paths, can re-include those if you think they might be helpful

file "path"...params.py", line 4, in <module>
from pydantic.fields import fieldinfo, undefined
ImportError: cannot import name 'Undefined' from 'pydantic.fields' (path...webui_forge_cu121_torch21/webui/venv/lib/site-packages/pydantic/fields.py
Press any key to continue . . .

[quick pivot]

Hey guys quick question, which python version do I need to run stable diffusion? I have 3.10.9

[calm cairn]

i ran it on an older version for a long time and it worked fine, in fact even after my last update I'm still using 3.10.6 right now and it works great

(also there exists code to automatically update when you run it, if you want to not have to worry)

[quick pivot]

I'm gonna try to install it with 3.10.9 to see what happens

[ornate elk]

That will work but best right now is 3.10.11 64bit

[quick pivot]

Ah okay, I will try that then

[ornate elk]

Ahh the pydantic fields error.
Thats common for forge. I have to look the fix up again

@calm cairn
you need to do this to fix it:
put albumentations==1.4.3 and pydantic==1.10.15 in the requirements_versions.txt file
Then relaunch or delete the venv

[quick pivot]

I get an error when I execute the webui-user, it says this:

Traceback (most recent call last):
File "D:\ai\stable-diffusion-webui\launch.py", line 48, in <module>
main()
File "D:\ai\stable-diffusion-webui\launch.py", line 39, in main
prepare_environment()
File "D:\ai\stable-diffusion-webui\modules\launch_utils.py", line 381, in prepare_environment
run(f'"{python}" -m {torch_command}', "Installing torch and torchvision", "Couldn't install torch", live=True)
File "D:\ai\stable-diffusion-webui\modules\launch_utils.py", line 116, in run
raise RuntimeError("\n".join(error_bits))
RuntimeError: Couldn't install torch.
Command: "D:\ai\stable-diffusion-webui\venv\Scripts\python.exe" -m pip install torch==2.1.2 torchvision==0.16.2 --extra-index-url https://download.pytorch.org/whl/cu121
Error code: 2

I get it when the pytorch download is happening

[calm cairn]

I added the text to the requirements_versions file, but I don't know how to relaunch a virtual environment. Should I run the install again? (it was an automated package)

[vital nexus]

Where can I cancel my subscription

[ornate elk]

Then just delete the venv folder abd relaunch the webui-user.bat

[ornate elk]

Hey, what's the full cmd log

[calm cairn]

ok, thank you

[fair oxide]

Helloooo not sure if this is the correct channel but I wanted to know if there's a guide I can follow to install Stable Diffusion please!
Or if someone would walk me through the process I'd appreciate that more 🥲

guys

Amuse 2.0 is out and it has a video depth/line converter

my question is how to use the depth video , whats the workflow ?

[ornate elk]

Hey checkout the first link from the pinned messages of this channel

There are all guides

[fair oxide]

What's the difference between each of these and which should I pick?

[fair oxide]

i think my destiny was with confyui.

i just hope i dont regret it.

and it seem they are arranged according to the best webui hehe

[fair oxide]

Soo Automatic1111 would be the best choice?

i think because of model manager of confyui , confy is the best.

the manager can download missing nodes, install models directly.

TBH i use confyui because Tydifusion is based on it.

[ornate elk]

Yes

Best for starting with SD.

[fair oxide]

yeah i think a1111 have all u need in one place especially if u r not a fan of nodes.

[fair oxide]

I don't know what nodes are 🙂

[fair oxide]

u are late to the ai party just like me ?

its been only 15 days i got into sd ai

i no longer sleep lol

these are comfyui nodes, a1111 have interface with tabs.

[calm cairn]

hmm
"RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!"
What is this?

[hollow quartz]

This is the entire error log I was talking about in #💬｜general-chat

[calm cairn]

Can't reproduce someone else's gen with prompt because of a strange error, please let me know what this means and/or how to solve it

[vocal burrow]

Most likely they re running on some kind of laptop with igpu/video chipset. Force auto1111 to use one specific device with the --device-id xxxx argument where x is the dedicated GPU device id (usually 0 or 1)

@calm cairn

[calm cairn]

hmm. It's a desktop, but yeah I probably have both integrated graphics on motherboard and a video card. Will try solution, thank you

Oh--how do I check which number (0 or 1) my GPU is?

[valid gorge]

Hey guys, anyone remeber a guide / tutorial about model merging, and it explained some kind of different approach to normal merging. It was about block / segment / something merging of different models. Does this ring a bell to anymore?

[valid gorge]

Found it! https://rentry.org/BlockMergeExplained

[plush spire]

I can't get a couple of Loras to show up in SD. Not sure what I'm doing wrong. I've loaded them into /stable-diffusion-webui/models/Lora and reloaded and refreshed the UI. Anyone got any ideas?

[vocal burrow]

Should be indicated in the GPU tab of resource manager

[karmic crown]

1.5 model and xl lora, or vice versa

this will always show all, but... many will not work with incompatible model

[plush spire]

1.5 model with 1.5 Loras

I'll try the setting

Ok, enabling the setting fixed it

thanks

[quick forge]

how do I access stable diffusion remotely? I have it set to --listen, but when I try to go to my ip address instead of localhost, it doesn't work. yes I specified port.

[green marsh]

Guys, what is wrong ? what can i do ?

[shy kettle]

alright this is driving me insane, i followed the pinned guide for installing webui on nvidia, did everything and got the panel open in my web browser but i'm having the exact same problem as @ivory bramble my computer just shuts off the instant i click the generate button on txt2img. i have the same specs as him too, gtx 1080 with 8gb vram and 16gbs of ram, and this is my command args in webui-user.bat

set COMMANDLINE_ARGS=--xformers --no-half-vae --medvram-sdxl --lowram

can anyone help?

[plucky ore]

I've been trying to install SD for years now, but I can't do it.

I got very close this time, but I got hit with a huge line of code.

It says 'timeout error.' What does that mean?

[karmic crown]

you installed python 3.12, you need to install python 3.10

find appropriate guide for your graphics here https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides

[karmic crown]

try to install in on D: or C: in some folder outside of your home folder
do not install in folders synced with one drive, google drive and similar

[green marsh]

yes, but aparentely the SD is not recognizing the Python now... (i'm setting the right path)

[plucky ore]

I installed the file through an internet browser instead, and it worked. Where would I then put the file?

[karmic crown]

as you can see you are using 3.12

[green marsh]

ohhh

wtf... why is taking the wrong python ?

[karmic crown]

uninstall 3.12, install 3.10
remove vevn folder
run webui-user.bat

[green marsh]

ok ok, i'll try

[karmic crown]

it will create new vevn with right version

[green marsh]

i deleted the vevn, i think now will work

[plucky ore]

When I run the webui-user.bat file, it still makes me install it again.

[karmic crown]

do you have C:\Users\swirl\ synced with any cloud drive?

[plucky ore]

Nope.
Not that I know of.

How would I check?

[karmic crown]

if you do not have onedrive, googledrive, dropbox or similar app, than its fine

let it install what it needs to install

[plucky ore]

But I already installed the file itself.

[plucky ore]

So where would I put it?

[karmic crown]

it had 1241241241 other files

torch has 58 dirs inside

[plucky ore]

Is this right?

I see lib, but I don't see python.

[karmic crown]

you are on windows

i am on mac

structure is not identical

[plucky ore]

Oh, gotcha.
Sorry for your loss. LOL

[karmic crown]

but anyway, you should have more than 200 dirs in site-packages

[plucky ore]

This is what I see inside.

[karmic crown]

exactly what i said

let it install everything

it will take a while

btw, i would remove venv completely, and let it create a new one, so you can be sure that everything was installed correctly

[plucky ore]

The folder, venv?

[karmic crown]

yes

when you run webui-user.bat it will create a new one if not exist

installation takes a few minutes on my computer
but depending on you internet and computer speed it can take a while

i have 600Mbps

[plucky ore]

I have, like, 5Mbps.

[karmic crown]

it should download 5-6GB in total

[green marsh]

finally worked, ty @karmic crown

[calm cairn]

hmm, the fixes from here and from elsewhere on the web are not functioning for me.

Right now I am testing this:
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
input_tensor = input_tensor.to(device)
output_tensor = output_tensor.to(device)

I just stuck it in webui-user.bat because I dont know what else to do with it

the hope is that it will do what it has to, putting everything on the graphics card (apparently can't have some on CPU and some on GPU or it wont work)

nope, code isn't doing it, although I don't know if its because the code doesn't work inside webui-user.bat or because the code is the wrong code

still getting error:
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

can't gen anything, not sure what next plan is

[plucky ore]

Right, so I moved the stable diffusion folder to a USB drive, since I don't have room on my main PC. It gave me this.

[misty marten]

I installed openpose, but its tab does not appear in the web-ui

[calm cairn]

update: It seems I can do normal txt to img gens, and I get the error on the img to img gens. I'm not sure why it wasn't a problem earlier today but I'll just do normal txt gens until someone comes by who might know how to resolve it

[ornate elk]

Remove --lowram and then you need to increase your Windows Pagefile

[ornate elk]

Check the cmd. That extension is outdated and requires that you manually install basicsr inside the venv

[ornate elk]

Can you shoe the img2img settings ?

[ornate elk]

Best would be to move it back. The webui itself needs only 8gb.
You can then store the models on the USB drive and link them in the webui-user.bat via --models-dir "D:\PathTo USB"

[hollow quartz]

These are my settings that I used before btw

[ornate elk]

Yea too high resolution and with that much frames and 60fps + controlnet it will take hours

To test try 512x512 24 frames on 8fps

Context Batch size 4

If that works you can go higher in batch size

Or higher in resolution but not above 768x768

[hollow quartz]

Btw what exactly does batch size do?

[ornate elk]

It defines how many images the GPU will generate at the same time.
That uses a lot more vram than a normal image generation and is the cause why it didnt worked for you. A GTX can't do 16 images at once at this resolution.

[hollow quartz]

Wait, so setting batch size from 1 to 2 or 4 would work better for me?

[hollow quartz]

Yeah, I thought it probably would take ages, was planning to let it run overnight, every time I go to bed, though sadly I keep getting these errors. I'll try reducing the fps though 60 is far too much and I'll also try to increase the batch size

[ornate elk]

No

A higher batch size won't work for animatediff

With 8gb vram

[hollow quartz]

Oh, ok, cause my batch size is always 1 usually

Then I'll just try to reduce the quality, fps

[ornate elk]

Like I said before I'm not talking about the normal batch size.

Look at AnimateDiff

There is context batch size

You have it on 16

That won't work

[hollow quartz]

Oh yeah you are right, my bad for mixing up the two, sorry

[ornate elk]

Np

Context Batch size of 16 uses 24gb vram on my AMD GPU.

So you should try it on 4

If that won't work lower it by 1 and test again

But always restart when you get the out of memory error

[hollow quartz]

Ok great, I'll try setting the context batch size to 4 and I'll also reduce the fps a bit and then I'll try again, thanks for the help 🙂 👍

Btw, this will take longer then, correct?

[ornate elk]

Yep

[calm cairn]

does that mean "show"? if so idk how, you want ss of SD GUI? or cmd output?

[calm cairn]

I screencapped in client settings for img2img, if thats correct then cool, if you want something else (like from cmd) then let me know

while you are away, I'm turning color correction off to see if that matters

[ornate elk]

I mean the img2img tab

Your resolution, steps etc

[calm cairn]

oh ok

I am running forge (not 1111) which is why I have all the extra, unused settings at the bottom

[vernal thorn]

maybe I am an idiot, but I logged in a few times now, and I can't see the first letter of my email to go to my account

[ornate elk]

Your using inpaint?

[blissful zenith]

hi someone can help me run IDM VTON localy on google colab? https://huggingface.co/yisol/IDM-VTON

[split comet]

Are these speeds normal on a RTX 3080?

[ornate elk]

Your using sdxl?

Then its normal

Make sure you have --xformers --medvram-sdxl --no-half-vae
In the webui-user.bat

[split comet]

Yeah SDXL, im using xformers, what do the others do?

[ornate elk]

--medvram-sdxl speeds up the sdxl generation on GPUs with less than 12gb vram

--no-half-vae is needed for VAE compatibility

[split comet]

I have 32gb of ram

[split comet]

Ok thnx

[ornate elk]

Vram = GPU memory

3080 = 10gb vram

[split comet]

Ah ok

[ornate elk]

And if you use hires fix make sure to set the hires steps to 10 or 15

[split comet]

Does it have a big difference? I have mine on 0 which would be 25

Those settings seem to be slower for me, the gen stops at every interval, if i batch 4 it stops at 24%, 49% etc

[ornate elk]

Yea with 10 its much faster and 25 is to much. Not needed

[ornate elk]

Use batch count and not batch size

[ornate elk]

You can also try to remove --no-half-vae
And see if it's still the same

[split comet]

Seems to be faster with just xformers, adding the others makes it slower

10 steps helps a bit though

[ornate elk]

Oh okay

[proven turtle]

Hey, I want to perform image reconstruction. I am trying to condition a diffusion model using the latent embeddings of the image I want to reconstruct, sourced from a pretrained encoder . Does anyone know how I should approach this task? Any suggestion on which Diffusion Model I could try to use? I was thinking of learning a mapping between the vae's encoder and my own encoder, then using the decoder from the diffusion model for the reconstruction

[maiden bramble]

Is there a pretrained controlnet (with canny support) for stable video diffusion.

[calm cairn]

I'm sorry, I fell asleep. I value your help and didn't mean to.
Yes, I'm trying to inpaint and getting the same error over and over.

[ornate elk]

If your using inpaint you need to set the Inpaint Area to "masked only"
And then you need to set the resolution to a sqaure.
If your using an 1.5 model set it to 512x512 and for sdxl to 1024x1024

[calm cairn]

ok i'll try that now

nope, same error

[calm cairn]

I often get the error on txt2img btw, so I'm not sure what is going on. Perhaps it started when I asked for help yesterday and had to remake the venv so I deleted the folder?

I'm thinking about deleting everything and reinstalling Forge to see if it works that way

have not checked if my 1111 install is affected but I assume it is not

I deleted Forge and am attempting to reinstall, I still had the self-install package on my desktop from 2 days ago.

ah, it also could have been when I installed adetailer, I've heard that some installs can give that error

[oblique vapor]

It physically pains me to ask this cause it feels like a very stupid question but:

Currently I am trying to get ComfyUI. I downloaded the files okay, but something strange is happening.

I went under ComfyUI>.ci>windows_base_files to click on run_nvidia_gpu.bat... and an error message saying "They system cannot find the path specified" And shuts down.

Now, I've gathered that main.py is in the ComfyUI folder, but it seems no amount of moving of things fixes the issue... I'm pretty sure I have to change the file path, but copyxpasting the directory with main didn't seem to work... unless I did it wrong... So My questions is: What am I doing stupid?

[shy kettle]

did both, still shutting down

[ornate elk]

What's your PSU wattage?

[ruby yoke]

did you ever put your directories into your sytem path?

[ornate elk]

Maybe it doesn't find your python path

Run the python 3.10.11 64bit installer again. Click Modify, Next, and select "add python to environment variable's"

[calm cairn]

I deleted Forge and reinstalled, and was able to gen about 6 pictures in txt2img, before I got the error again. I don't know what to do if the fresh install won't work right. Got any idea what is wrong?

[ornate elk]

Did you changed anything after the 6 image?

[oblique vapor]

It appears that I do have enviroment variables.

[oblique vapor]

Let me show you its current state

.\python_embeded\python.exe -s C:\Users\MYNAME\Documents\Coding\Python\AI Python Projects\NewProjects\ComfyUI\ComfyUI\main.py --windows-standalone-build
pause

[ruby yoke]

show me your actual system path statement

[oblique vapor]

This is the path to the run_nvidia_gpu.bat file:
C:\Users\MYNAME\Documents\Coding\Python\AI Python Projects\NewProjects\ComfyUI\ComfyUI\.ci\windows_base_files

This path has main.py:
C:\Users\MYNAME\Documents\Coding\Python\AI Python Projects\NewProjects\ComfyUI\ComfyUI

[misty marten]

this appeared

[ruby yoke]

you said you moved things around. what did you move?

[oblique vapor]

The windows_base_files, but I moved everything back to be the way it was

[ruby yoke]

did you reboot after you did that?

[oblique vapor]

I did not, I haven't rebooted (the entire computer) in a minute.

I've run run_nvidia_gpu.bat multiple times though

I could try removing what I have, get it back off git, reboot the computer, and try running again. Do you think that'd work?

[ruby yoke]

please do so, so that all the paths and stuff are reloaded into memory

[oblique vapor]

Gotcha. I will give that a try.

[ruby yoke]

i think what will work is for you to make sure everything is back where it should be, then shut down, leave it off for a minute, turn it on and then try running the comfy startup bat file

[oblique vapor]

I'll give it a whirl

[ruby yoke]

(thankfully, moving stuff around in windows doesn't change permissions)

[calm cairn]

nope, just went to CivitAI page for Animagine XL and copied positive and negative prompts, guidance, steps, sampler, and seeds, and manually pasted it in each time

[ornate elk]

Do you use a lora?

Or embeddings?

[calm cairn]

no, just text in prompt box, no special stuff

[ruby yoke]

why are you runing the nvidia bat file instead of just running the main Comfy startup file?

[ornate elk]

I can send you the two commands to fix it later.
Also there is a better editor now.
Its included In Controlnet where you can use openpose and edit the Skeleton

[oblique vapor]

Because the video told me to.

Also I have an Nvidia graphics card. 😅

Maybe that was silly of me. Let me try just running main.

[ruby yoke]

well don't. just run the main startup bat file

[ruby yoke]

i do too

[calm cairn]

in fact heres the page w image

https://civitai.com/images/8301359

as you can see it isnt even calling a lora i dont have

[oblique vapor]

That seems to be making progress!

I got a whole host of test this time instead of just "Can't find file" so forward progress is forward progress.

[ruby yoke]

so a couple of things: 1. don't move stuff around from an application install if you don't know what you're doing and 2. always run the main startup file

[oblique vapor]

Good deal. I'll keep that in mind.

Nooooww... I have to observe the import fails.

[ruby yoke]

what import?

[oblique vapor]

nodes_audio.py

[ruby yoke]

what's that for?

[oblique vapor]

No idea, I've never seen it before. It came with the code I got from git.

[ruby yoke]

go back and read the git page

[maiden bramble]

Is there a pretrained controlnet (with canny support) for stable video diffusion.

[misty marten]

ok, I'll wait

[oblique vapor]

Good deal. There appear to be a couple steps I may have been forgetting so we;ll see.

Major progress!! It seems that I can gain access to the UI!!!

I guess atm I have one final question:
When I hit main, the python window pops up, some text scrolls, and it says "to see gui, go here"
I go there and Bing! I have a UI!!

My only question is if there's a way to pull that up automatically? 🤔

[ruby yoke]

um, well, on my system it just opens that window automatically. loads my browser if it's not loaded, and opens it. so - make sure you don't have popup blockers in your browser. i'm on chrome, in case you're not

[frozen raft]

hello, does anyone know if it is possible to do 1000x1000 images with a 1050ti 4gb? whenever i try to do it i get this error

[calm cairn]

now I restarted and pasted the same stuff in, and it generated an image. I really have no idea what to do

I'll just pay a lot more attention to what i am changing, if anything, when it fails again

[split comet]

I swapped to forge and holy moly its so much faster

[ornate elk]

Yes but you need to use --xformers --medvram --no-half-vae
And then you need to install the tiled vae extension

[sand dawn]

I've been running around in circles, chasing my tail on this for hours, days...
At some point, ReActor (CUDA) and InstantID ControlNet models both stopped working for me in A1111, Forge, and ReForge.
I get this error in CMD Window:

    RuntimeError: D:\a\_work\1\s\onnxruntime\python\onnxruntime_pybind_state.cc:891 onnxruntime::python::CreateExecutionProviderInstance CUDA_PATH is set but CUDA wasnt able to be loaded. Please install the correct version of CUDA andcuDNN as mentioned in the GPU requirements page  (https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html#requirements),  make sure they're in the PATH, and that your GPU is supported.

My build is < 2 yrs old - 4070ti on Windows 11.

• VS Community 2022 is installed.
• Current version of A1111 / Forge / ReForge
  I followed the instructions for CuDNN:
• Latest nVidia drivers
• Latest CUDA Toolkit
• Latest CuDNN

I check my environment variables, I have one CUDA version listed, and one CuDNN listed, both pointing to correct locations

I've looked around and for every instance where someone reported this error, they received no answer

[ornate elk]

What's the full cmd log?
And are the extension updated?

[sand dawn]

Yep, extensions up to date.
I'll share log in a moment...

https://pastebin.com/QsbcR9X4

If I delete the onnx related packages from site-packages, ReActor reinstalls them on next launch.
I then close and re launch the WebUI... same error.

[ornate elk]

Have you tried to delete the venv folder and relaunch the webui-user.bat?

[sand dawn]

I'll do it just for the sake that you are here and offering assistance, but it's not going to do anything - it's happening in all the webUIs

Shouldn't take long either... super fast speed here...

And it's not restricted to ReActor - same error seems to occur for anything using insightface (InstantID model via ControlNet)

venv rebuilt...

indeed, same error

[ornate elk]

@sand dawn On AMD I had the same error. Face ID didn't worked because reactor was installed. Reactor installs onnxruntime-gpu which is for nvidia but Face ID needs onnxruntime

[sand dawn]

I'll try disabling ReActor, deleting the onnx libs, and see what happens when I run InstantID

HMMM

[oblique vapor]

Hmm that could be a thing. I will try having chrome open

[sand dawn]

So it said Module not found: onxx (because I deleted the lib, and apparently it won't install itself.
So I activated the venv and pip install onxx

InstantID is working

[oblique vapor]

Okay... something got broke. Main.py is now flashing a Py terminal and then immediatly closing... All I did was change my default browser to chrome... even changing it back didn't fix it... what'd I do?

[frozen raft]

thankyou bro

[ornate elk]

go into the stable-diffusion-webui folder.
There click into the explorer bar (not searchbar)
type cmd and hit enter.
then you have to run these two commands:
venv\Scripts\activate.bat
pip install basicsr
then relaunch the webui-user.bat

[oblique vapor]

Deleting files and then re-cloning git does not fix the issue 😓
Update: Running from a comandprompt appears to yeild that there is no module named "torch". So I guess I will try and fix that

[ornate elk]

yea and if you install reactor it will break it because it installs onnxruntime-gpu

[sand dawn]

Indeed. This is some BS XD

[misty marten]

It worked, thank you very much

[ruby yoke]

you made a mess. if it was me, i'd uninstall everything, including python, and reinstall. remember that comfy uses an embedded version of python. that's why people usualy set up virtual environments - so that what they have installed for comfy isnt' going to affect the rest of their sytem

[calm cairn]

alright I am beginning to think my problem has something to do with memory allocation
given I had no problem with gens in automatic1111, and now have the problem after swapping to Forge, which allocates mem differently
given that nothing I am doing seems to be causing it
It just started failing again, when my vram was stretched to the limit

I'll test to see if clearing out the memory fixes the problem, if you know any strings of code/settings that can help me with that, then plz let me know, could save me time

[ornate elk]

what forge launch args do you use?

there are some which can help

[calm cairn]

i dont use anything substantial, used more on A1111
--theme=dark --ckpt-dir "E:/Life/Projects/ART/Stable_Diffusion/Models/" --lora-dir "E:/Life/Projects/ART/Stable_Diffusion/LoRA/"

so just theme and directories

[ornate elk]

forge needs these for max performance:
--cuda-malloc --cuda-stream --pin-shared-memory

[sand dawn]

Awhile ago, I forked --sd-webui-ar-plus and improved it very much.
https://github.com/altoiddealer/--sd-webui-ar-plusplus
After checking out the new version of Forge, which is using Gradio 4, my extension is only one out of 2 from my extensions folder that is failing to load.

Seeing if anyone is savvy to what changes in Gradio 4 that could cause this error:
https://pastebin.com/iACS5udu

[oblique vapor]

I took your que and that looks like it fixed it. I can now get to the UI. Still wants me to copy and paste a link to it... which is weird but maybe I should just deal with it for now.

I can see the UI so that's a win

[ruby yoke]

you are running it on what? what's the link?

[karmic crown]

find an extension that works, and see how they created ui, do the same as they did

this is how I fixed mov2mov to work on 1.8.0
https://github.com/Scholar01/sd-webui-mov2mov/compare/master...viking1304:sd-webui-mov2mov:master

[oblique vapor]

If I throw a link like this:
http://127.0.0.1:81##
(Two more numbers instead of hashtags)
It brings up the GUI. I can throw that into Microsoft edge

[ruby yoke]

wha'ts the number where the hashtags are please

the entire URL it gives you. i want to explain what you're looking at

[oblique vapor]

Sure sure, let me que up another one

http://127.0.0.1:8188

[sand dawn]

Of course, none of the extensions that are working have any ToolButton components 😛

[karmic crown]

this is not the same as
http://127.0.0.1:81/xx

this is the same as
http://127.0.0.1:81xx/

[ruby yoke]

okay, to explain - 127.0.0.1 is the IP that's set asside for the internal local host of every network adapter. and the :8188 is the port that it's running on

To see the GUI go to: http://127.0.0.1:8188
FETCH DATA from: C:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Manager\extension-node-map.json [DONE]

[karmic crown]

and that is a drastic difference

[ruby yoke]

that's the same that i get. comfy runs on local host on port 8188

not a clue what the /## you mentioned might be (other than typo?). but you can just bookmark http://127.0.0.1:8188 and just go to that bookmark if you need to

[oblique vapor]

Ahh okay. Cool, glad I'm not the only one who's gotta copy paste that then.

I put in the ## signs just because I was weirded out about throwing it in normally, that was my bad.
So, dumb question, will me posting that link allow someone else to modify what's in my GUI?

[karmic crown]

i was a mixup with http://127.0.0.1:81xx/

[oblique vapor]

I put the ## there, I was modifying it because it didn't feel wise to just post something that could be a vulnerability

[ruby yoke]

no one else copies and pastes it. for most of us, the browser just opens a new tab and loads. again, make sure you don't have popup blockers and things that prevent your browser from opening new tabs

[oblique vapor]

Let me check. It worked for Auto1111 so I am surprised, but I shall

[ruby yoke]

127.0.0.1 is the local host IP of every single network adapter in the world and only reachable IF you are on the machine itself

[oblique vapor]

Ahh well that's good news 🙂

[ruby yoke]

when you see something like that with any ip, take for instance http://270.45.2.654:8877 you are looking at an actual website address then the : tells you what port the application you are using is running on

[signal wasp]

Hello! Not sure if this is the right place to ask, but can someone help me with LivePortrait Video2Video?

It works fine, but my output videos in the Video Combine window are only 5 seconds long, how do I get the full video?

Thank you 🙂

[ruby yoke]

purz has a really good tutorial for this on his youtube channel

[signal wasp]

Cool I will have a look thank you!

[oblique vapor]

Interesting, so then what in this case would the port go to? You said it was local

[ruby yoke]

when you start up any application or process, you have to tell it what port to run on. you can think of ports like little rooms in a huge mansion. the local host ip assigned to your adapter is the house number that no one but you can use. but in that mansion, at address 127.0.0.1 there are 10,000 rooms. so in order for your machine to know which room the application you want to use is in, you have to tell it the room number (port number) - in the case of comfyui, when it's running, it's located at the home address of 127.0.0.1 (your private mansion) and in room number(port) 8188

normally, you just run an application and all the startup code takes care of assiging the room number(port) for you, but you could, technically if you wanted to, assign it to a different port. if you do that, make sure nothing else is running on that port, or you'll ahve a huge problem

[oblique vapor]

I get it now. That makes sense!! Thank you very very much for the explanation!

[ruby yoke]

🙂 you're welcome

[quick forge]

How do I access SD from another machine? I've tried using ipaddress:port but it doesn't work, and I do have --listen enabled.

[ruby yoke]

you have it running on a public IP - so you'd have to have a web server set up on your host machine and and your client would need to access it like you would any website OR you'd need to write some sort of server for it that could make it available to your client

[quick forge]

im using A1111, so that's how I wanted to access it.....

[ruby yoke]

you would be using that on your client. i don't think auto1111 has a built in server, however

[quick forge]

well that kind of sucks.... i pretty much just wanted to access it from my windows machine

so i don't have to keep switching between my HDMI inputs lol

I have SD running on a linux machine in my bedroom

Launching Web UI with arguments: --listen --enable-insecure-extension-access

[ruby yoke]

i know, standard client/server tech here. you have to have a server of some sort running on the host machine that makes the application available to the client. @ornate elk might know

[quick forge]

well poop lol

not really that important right now though I guess

I'm a little curious why sometimes my loras all load under the lora tab, and then sometimes they don't and they don't show.

[ruby yoke]

do you have them in the wrong spot?

[pearl musk]

Yo guys, can I use an RTX 3090 on a 650W 80plus Gold PSU? I would reduce the TDP by 10% manually, but before buying it I want to be 100% sure: I have a ryzen 5 5600G, 32GB of RAM and 4 1TB ssds

[quick forge]

Shouldn't be, they're in the lora folder and they'

are loras

[ruby yoke]

are they loras built for the model you have loaded?

[quick forge]

that could be part of the problem, but it seems inconsistent in that they won't show sometimes when it is the right model loaded.....but this might also be user error. I'm actually trying to sort through my models, and put them in folders with their graphics and such, and make sure they're all in the right folders themselves, like checkpoints or loras.

ok so I think I need "LiberteRedmond SD 1.5" but I can't seem to find it anywhere? or im misunderstanding what that is.

[ruby yoke]

okay, well, all i can do is guess, but that's the most common cause. you've got, say, an sdxl model loaded, and you're looking for an SD 1.5 lora

[vast pulsar]

hey guys, kinda new to this and was hoping someone could tell me what an "activation token" is

[ruby yoke]

where'd you see the phrase?

[vast pulsar]

one of the models i wanted to try out suggested to use "4rchon" activation token

[ruby yoke]

which model, please

[vast pulsar]

idk if i can post links so i imagine this is what ya want?

[vast pulsar]

wherever i look i only find the terms "token merging" not sure that's what i'm looking for

also if it helps: using webui

[ruby yoke]

okay, that's the keyword you put in your prompt. so your prompt might be: 4rchon a man eating a hot dog

without that keyword, the lora will be ignored

[vast pulsar]

oooh, alright! tought i had to download smth. thnx

[ruby yoke]

make sure you put lORAs into the LORA folder, not the models folder

[misty marten]

Zluda error

[vast pulsar]

yea, ik that part, was only missing this detail

[ruby yoke]

you might search this discord for that error, it looks familiar

[gaunt crystal]

hey gang, i used Nightcafe's "creative upscaler" to get this result, anyone have any idea how i can do the same thing in stable? (auto1111)

(image on left is upscaled, lol)

[mossy pulsar]

Hello. Does anyone here know how to generate a specific objects and apply in the prompts? I'm struggling with this.

I'm new in generating AI images, so, I tried to put some ideas in the generator but it didn't work.

I was trying to make an image of a prussian soldier, and it didn't recognize the helmet - it has a lot of details - and I want the most similar possible.

This is the helmet.

[misty marten]

zluda gives an error saying that a file is missing, but it is in the folder

[ruby yoke]

that's going to depend on the model you're using, whether it was trained on data with that specific label or not.

[sharp junco]

Who knows why it goes Red and Cursed at the last 5% of rendering?

maybe because of vae it could be thou

[vale veldt]

Anyone know why my speeds seem to be flucating so wildly? These are same prompt, models, and settings. Just different speeds:

[unborn bramble]

Colab Lora trainer gives me this...

I swear it was working yesterday. What happened to it?

[fluid grove]

Anyone have issues when installing SD with pinnochio? I can't get realistic faces anymore... thinking of installing manually again

[ornate elk]

Better install it manually via the guides from the pinned messages

[fluid grove]

Made a ticket, we sorted it out, I was making a rookie mistake with my cehckpoints resolution

ty tho

[plush spire]

I'm using the wget command to get files from Civit. Sometimes the get works and then other times it asks me for a username and password. I tried adding my username and password as paramaters to the wget command but it still failed with the issue. Anyone know how to get around this?

[karmic crown]

give me the link that gives you the error

[plush spire]

--2024-08-01 14:41:56-- https://civitai.com/api/download/models/424720
Resolving civitai.com (civitai.com)... 104.22.18.237, 104.22.19.237, 172.67.12.143, ...
Connecting to civitai.com (civitai.com)|104.22.18.237|:443... connected.
HTTP request sent, awaiting response... 401 Unauthorized

Username/Password Authentication Failed.

[fair oxide]

https://civitai.com/models/288584?modelVersionId=324619 hi i want to use this checkpoint and when i downloaded it, it said i had to install xformers. I installed it then it said i need CUDA for it to work and i installed it. But then i ended up with this message

i need help pls

i also get this message

[shut geyser]

amd 7800 xt, trying to run this https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs

everything works fine but when i try to generate an image, this error happens

[gentle lava]

if i upload a image can someone tell me what is the model or style?

[ornate elk]

Please follow my AMD Zluda guide.
You can find it in the first link of the pinned messages

[ornate elk]

Can you show the full cmd log?

[fair oxide]

@ornate elkbtw i installed the newest version of python but it says i have the other one which is weird save with cuda i got the latest version and it shows the wrong version
do you know how i can fix it?

[ornate elk]

You need to uninstall python 3.12
Then install python 3.10.11 64bit
Then delete the venv folder
Finally launch the webui-user.bat

[fair oxide]

why can't it work with the latest version of python?

[ornate elk]

because its not coded to work with it

3.12 is not supported same goes for older versions like python 3.9

[fair oxide]

oh okay

@ornate elk btw i'm downloading the correct cuda version i also wanted to ask
why do i have to delete the venv folder?

[ornate elk]

because if you wouldnt. it would always still use python 3.12
the venv folder needs to be rebuild with the compatible python then torch and xformers will work too

[ornate elk]

maybe in #📝｜prompting-help

[fair oxide]

i see thanks
i'm installing pytorch now and it should be good to go i think

[ornate elk]

whats your gpu btw?

[fair oxide]

rtx 3070

[ornate elk]

ah then make sure you have --xformers --medvram-sdxl --no-half-vae in the webui-user.bat

for the best performance and compatiblity

[fair oxide]

i'm gonna downlaod xformers after my downlaod is done

[ornate elk]

you dont have to download and install aynthing manually

dont do that

the venv will do this all for you

[fair oxide]

wait really?

[ornate elk]

yep

pleas dont install pytorch or xformers manually
that will break things

[fair oxide]

i saw videos of people explaing it was mandatory

[ornate elk]

its not

but feel free to share me that video

just deleting the venv and having the right commandline args in the webui-user.bat is enough to setup everything.
and of course the right python version

[fair oxide]

https://youtu.be/ZVqalCax6MA

[ornate elk]

this video is more than a year old and even there it was outdated after 2 months xD
dont follow old stable diffusion videos, everything changed

[fair oxide]

@ornate elk https://civitai.com/models/257749/pony-diffusion-v6-xl btw this is the checkpoint i want to use but it says i had to use xformers

[ornate elk]

yea that will work with your gpu but only if you use the commandline args i mentioned

also you need 16gb RAM or more

[fair oxide]

i have 32gb ram

[ornate elk]

then edit the webui-user.bat, delete the venv and your ready

[fair oxide]

okay so i just need to insintall pytorch now?

[ornate elk]

Python 3.10.11 64bit

not pytorch

[fair oxide]

it said this when i installed pytorch

the web ui launched

[ornate elk]

you didnt editet the webui-user.bat

or didnt editet it correctly

[fair oxide]

Oh right

i forgot yes

[ornate elk]

then do that and then delete the venv folder again

then your fine

set Commandline_ARGS= --xformers --medvram-sdxl --no-half-vae

[fair oxide]

done

so now i delete the venv folder and then launch the web ui?

[ornate elk]

yep

[fair oxide]

it tells me this now...

[ornate elk]

full cmd log pls

[ornate elk]

it seems you have added a space in the webui-user.bat after the line
set VENV_DIR=

delete the space after the =

then save and relaunch the webui-user.bat

[fair oxide]

now i get these codes

[ornate elk]

thats fine

its rebuilding the venv now

and will launch then

[fair oxide]

ah okay it launched

now i need to see if the checkpoint works

@ornate elk thank you a lot for the help btw

[ornate elk]

no problem ^^

[misty marten]

This error appears when I open the web-ui with zluda

I looked in the folder, and this file "nvrtc64_112_0" is there

[ornate elk]

have you replaced the rocm libs files?

like stated in my guide

[misty marten]

yes, I copied the library folder and renamed it to old library, downloaded the 6600 rocm files and placed them in the copied library, without the old_

[ornate elk]

You need to install rocm 6.1 instead of 5.7

And then add the hip path again and replace the Library files again

[misty marten]

Which of these Path's should I put?

[ornate elk]

In the bottom one

Its system variables

The top one is called user variables

[fast loom]

Can I ask about stable-fast-3d here? Or is this not the correct discord for it?

Does it take time for the authentication of the model (Gated model You have been granted access to this model )to propagate?

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
...
...

    raise GatedRepoError(message, response) from e
huggingface_hub.utils._errors.GatedRepoError: 401 Client Error. (Request ID: Root=1-66ac054e-2c8906bb336f3fb337447d7f;905cdb37-e025-49cd-9092-3f68190982f7)

Cannot access gated repo for url https://huggingface.co/stabilityai/stable-fast-3d/resolve/main/config.yaml.
Access to model stabilityai/stable-fast-3d is restricted. You must be authenticated to access it.

[ornate elk]

You have to download the model manually from huggingface because its behind a login

[fast loom]

Oh, okay. Even though I have logged in with my token/account?

Do you know where to put it?

[ornate elk]

What service/Programm do you use?

[fast loom]

Windows 11, just CMD client thing ^^

I've set up a venv, installed all requirements (and some more that wasn't listed), and logged in with my HF account from this env. And it still displays that error message.

I have also downloaded the model now, model.safetensors, but not sure where it is supposed to go.

Here's the complete console log

[ornate elk]

Hmm I don't know where the model goes into

Sry

[misty marten]

like this?

[ornate elk]

Yes

Its around midnight here I'm off.
I can help tomorrow if you have any issue.
Make sure to restart the PC after Installing HIP SDK 6.1 and replacing the Library files.

[formal trout]

Hi, I'm trying to set directory for CLIP models for Automatic1111 in another folder, for SD3, it always try to download the clip models anyway

I'm using --clip-models-path /media/mario/Ultra-Zen/SD3/models/CLIP/

but I don't know if I'm doing something wrong there as it doesn't seem to detect the models or use the folder

It detects the checkpoint model in the other disk just fine with --ckpt-dir /media/mario/Ultra-Zen/SD3/models/

mmm it seems it is trying to use other models as I put the clip sd3 models I have and it started downloading this one instead

So maybe I let it download the clip models and then test with the alternative dir

[hallow latch]

Hey, I need some help.. I am not able to use certain ControlNet function anymore and I have tried to update ControlNet but still getting this weird Error..

RuntimeError: The size of tensor a (80) must match the size of tensor b (64) at non-singleton dimension 3

I have been trying to google but cannot find any working solution..

[fast loom]

Can you check to make sure you are using an SDXL ControlNet when using an SDXL model, and SD1.5 ControlNet when using an SD1.5 model?

[hallow latch]

Uhm, how can I check that.. I have been able to use this controlnet before with the model.

Or are you refering to this one?

[fast loom]

Those are both for SD1.5

What checkpoint/base model are you trying to use to generate the image with?

[hallow latch]

It's called TFM_Cutesy

It's an SD 1.5

[fast loom]

Okay, then it should work.
The error of the size of the tensors not matching is sometimes because you're trying to use incompatible models.

You're not using a <lora> for SDXL or something in your prompt, right?

Just making sure you aren't loading any incompatible models anywhere 🙂

[hallow latch]

Might have found the issue

I had a --non ..something .. Because I had issue with AnimateDiff

I removed it and now it worked.

Nvm.. It's the .. Sampling Method

I can't run Euler SMEA Dy with OpenPose. ...

[fast loom]

Ah, good to know!

[formal trout]

I'm trying to run SD3 on Automatic1111, it seems I'm 28 mb short

[ruby yoke]

how much Vram do you have?

[formal trout]

12 GB VRAM

I tried to run it with "lowvram" and got a lot of errors, it seems it doesn't work for AMD or soemthing

[ruby yoke]

ouch. @ornate elk might have a solution for that

[formal trout]

I don't know if it needs different flags or if it isn't supported for AMD yet

I have a RX 6700, running on Ubuntu

[lament fog]

auto supports it already? Sweet

[misty marten]

same error again :\

[calm cairn]

Hi there, I fell asleep last night receiving tech support from CS1o. I was advised to add

--cuda-malloc --cuda-strea --pin-shared-memory

to my arguments, but upon running Forge, I am now told that it's launch.py error, unrecognized arguments

What file am I supposed to put these arguments in? Did I type this wrong? I thought I copy/pasted

[ruby yoke]

he hasn't been on today, can you wait till he's back online?

[calm cairn]

yeah, I'm assuming he'll probably be back tonight, seems like thats the pattern

[ruby yoke]

he also works, so he's here when he's not at work

[lost tinsel]

Anyone know where the PBRemTools Segment Anything & Clip Model needs to be placed? Or how to install the relevant models with stability matrix?

I've tried adding "sam_vit_h_4b8939.pth" and "sam_vit_l_0b3195.pth" to a 'sam' folder and the 'CLIP' folder in D:\Stable Diffusion\Stability Matrix\StabilityMatrix-win-x64\Data\Models - but I'm more just guessing at this point

[ruby yoke]

nothing on the page you downloaded it from?

[lost tinsel]

not really, I just installed what I found in forge extensions. This is the only doc and it doesn't really explain where the models go/where to get them https://github.com/mattyamonaca/PBRemTools

[lost tinsel]

doh, lemme give that a try

noice, we got it - thanks @ruby yoke (I might need to invest in some glasses xD)

[ruby yoke]

🙂 you're welcome

[lost tinsel]

I think I scrolled down to the installation part and just skimmed the rest haha

[ruby yoke]

we all do that I think. just glad you got it working

[gaunt crystal]

i cant seem to get stable to recieve my controlnet sketch prompt, it says it does but the little sketch image wont appear next to my finished generations, also none of the sliders on controlnet seem to affect the output image

[ornate elk]

Thats a different error. Try delete the venv folder and the .zluda folder.
Both are in the stable-diffusion-webui-amdgpu folder

[ornate elk]

It goes in the line with the Commandline_args= at the top

[swift garden]

With what i can train Lora using AMD RX 6600 XT?

Any suggestions? I was trying OneTrainer, Kohya...but no luck since i have AMD and not Nvidia.

[ornate elk]

Oh hey, your from Github.
With OneTrainer it should work because it can use Zluda.
I haven't tried it myself but I can this weekend and then make a guide for it

[swift garden]

I've tried to use OneTrainer.
First of all it throws error

Second it can't generate captions (text files that generates pormpt for each image)

And i've tried also Kohya and that works on Nvidia

[gusty salmon]

Im trying to swap the head out on a photo, but without manual inpainting cause i want to automate the process. I found this post on reddit: https://www.reddit.com/r/StableDiffusion/comments/18z70b1/roop_but_for_the_whole_head/

Now how exactly do i keep the body but place the new hair+face from my reference images on top of it? Here are my current options and results (face swapped but ne hair and body):

[ornate elk]

You have to not use Face ID plus.
you need to use Face Full to get the hairstyle etc or additional IP adapter plus

Face ID is like Reactor and Swaps only the face

[ionic widget]

helo guys can anyone help me how to fix this problem?

[gusty salmon]

Not sure if the result can be considered any better. Cab you give me some hints with the settings? The hair looks different, some random mutation and the body/clothes shouldnt change

I assume i have to choose a different preprocessor too? But i feel like i have a bunch of settings set wrongly

[ornate elk]

I can try later and send you an example. But I'm wondering what's your webui version

And why it takes so long for you to generate

[gusty salmon]

Im running on a 3060

Lets see if i can find webui version somewhere

[gusty salmon]

that would be awesome

[ornate elk]

Oh your on the latest version vut your not using xformers

Edit the webui-user.bat and at the line commandline_args= you add: --xformers
Then save and relaunch. That will increase the speed alot

[gusty salmon]

alright, lets see

[solemn moon]

Open the .bat file with notepad

[gusty salmon]

already replaced, about to make a new run

[ornate elk]

Also interesting that you got Roop to work

Its outdated by 10 months now

Normaly people use Reactor for faceswap

[gusty salmon]

i actually had a error with roop:

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

--no-half fixed it

should i use reactor then?

[ornate elk]

Oh dont use --no-half

It slows down the generation speed a lot and uses more vram

Yea better use Reactor

[gusty salmon]

alright

[ornate elk]

Its does the same but gets updated frequently

And has more features

So you only should have --xformers and maybe --no-half-vae
In your webui-user.bat

[fair oxide]

is there way to make stable diffusion go idle when no promp ahs ben send for 10 min and to dump from vram

[gusty salmon]

I removed --no-half and now im getting this again:

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

Will --no-half-vae fix it?

[gusty salmon]

it did fix it

also --xformers made it significantly faster lol

[calm cairn]

Hi there, I fell asleep last night receiving tech support from CS1o. I was advised to add

--cuda-malloc --cuda-strea --pin-shared-memory

to my arguments, but upon running Forge, I am now told that it's launch.py error, unrecognized arguments

What file am I supposed to put these arguments in? Did I type this wrong? I thought I copy/pasted

Also I believe the errors I am having (cpu cuda:0 thing) is probably something to do with memory/VRAM, because when I swap to Edge (yuck), the errors go away (normally using Chrome when I got the errors)

[ornate elk]

can you show the content of webui-user.bat?

[calm cairn]

At what point? Do you mean get it to crash in Chrome, then post that?

oh nvm

you wanted the txt not a log

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--theme=dark --ckpt-dir "E:/Life/Projects/ART/Stable_Diffusion/Models/" --lora-dir "E:/Life/Projects/ART/Stable_Diffusion/LoRA/" --cuda-malloc --cuda-strea --pin-shared-memory

@REM Uncomment following code to reference an existing A1111 checkout.
@REM set A1111_HOME=Your A1111 checkout dir
@REM
@REM set VENV_DIR=%A1111_HOME%/venv
@REM set COMMANDLINE_ARGS=%COMMANDLINE_ARGS%^
@REM --ckpt-dir %A1111_HOME%/models/Stable-diffusion^
@REM --hypernetwork-dir %A1111_HOME%/models/hypernetworks^
@REM --embeddings-dir %A1111_HOME%/models/embeddings^
@REM --lora-dir %A1111_HOME%/models/Lora

call webui.bat

[ornate elk]

there is a typo

you have: --cuda-malloc --cuda-strea --pin-shared-memory
but you need: --cuda-malloc --cuda-stream --pin-shared-memory

[calm cairn]

ok will try that, one minute

launch.py: error: unrecognized arguments: --cuda-malloc --cuda-stream --pin-shared-memory
Press any key to continue . . .

in case it is relevant, it is Forge, not normal A1111

[shut geyser]

im using your guide for comfyui with zluda, i just did the first steps, and this pops up

[ornate elk]

hmm maybe he removed these launch args in the latest version

[ornate elk]

hey, at which step are you?

[shut geyser]

launching comfy ui , step 6 🙂

[ornate elk]

okay then read step 7 ^^

zluda needs to be setup

then it will work

let me know if any step isnt clear

[shut geyser]

yeah but i didnt try generating an image, the webui doesnt even open

[ornate elk]

yea i will edit that part. in the past it launched. but now it does not it seems

[shut geyser]

also im installing amd hip sdk right now and i dont see an option amd pro driver

[ornate elk]

its at the bottom. its disabled by default so your fine

[shut geyser]

i dont even see it but ill trust

[calm cairn]

whats the latest version? I only recently installed Forge, but I don't know how old the installer was, only when I got it

I can say my Python is 3.10.6, torch 2.1.2+cu121

[ornate elk]

your running the latest version

1.10

[ornate elk]

thx for letting me know about step 6/7 , i editet the part that launching fails.
btw whats your gpu?

[calm cairn]

ah ok, well then I'll assume my problem is, tentatively, solved
I have not had any trouble running it in Edge, although I don't like Edge, it is functional
Thanks for your help, I don't understand the problem really (cpu/cuda:0 error message), but using Edge is a successful workaround, so I'll just keep using that, unless you really want to solve with more troubleshooting

[ornate elk]

okay, maybe check if you have extensions in chrome that could cause errors. sometimes adblockers do

also if you want you can try if it works on firefox

[calm cairn]

yeah I'll try firefox later, i have no addons for chrome because I prefer Firefox but I have a lot of tabs in Chrome right now so I'm closing all of those first, before I go back

Again, thanks for your help, I'm just gonna gen all day now 😛

[pure glade]

guys, how can i change the settings used by this function ?

[ornate elk]

Enabling hires fix

Changing the vaules

And then disable

[pure glade]

thanks!!

[shut geyser]

its rx 7900 xt ^^ thank you for the help btw, i launched it and it showed up in my browser now ima get the manager

is this how its supossed to look like with the manager? can i somehow make the ui more friendly like it is in automatic 1111?

[onyx solar]

The UI in Comfy is friendly. But it's node-based. If you're not used to it, take time to understand it and you will be able to wield a more powerful tool than A1111. If you don't want to do that, go back to A1111 (or use tools that mask the ComfyUI interface; though by doing that you're defeating the purpose).

[shut geyser]

is there a guide to the interface somewhere? im up to learn it

[onyx solar]

The official documentation is here:
https://docs.comfy.org/get_started/introduction

There are a ton of YouTube videos out there to walk you through it. I suggest starting on this page of the docs:
https://docs.comfy.org/get_started/gettingstarted

Also, many people embed their workflows into their images when they post here. Download and save the original, full size image, then drag and drop it onto a ComfyUI instance and if the workflow is embedded, it will open up that workflow for you.

[shut geyser]

do you have any workflow recommendation?

[onyx solar]

Here. I get asked this every once in a while. This should be enough to start with, giving you an SDXL image. The missing pieces/stuff you'll need to change will be the drop-down for the Checkpoint and VAE. After that, just put text into the prompt and queue up the image.

That flow used only nodes included by default, so you don't have to download any custom ones to get it to run.

I step you through it in numerical order for how a human thinks, but the other thing to understand is that most node-based systems actually logic-check backwards, in a sense. They're going to look to make sure you have everything from the end moving back to the beginning to fill each step's prerequisites.

[shut geyser]

can i run stable diffusion xl too?

i got 20 gb vram

[onyx solar]

That's what the workflow will do for you; create an SDXL image. You've got more than enough for SDXL.

I only have 10GB of VRAM and generate SDXL images all the time.

[shut geyser]

so i download what you sent, and ill drop it into the website right? and then i do the rest you told me

[onyx solar]

Yup. I recommend putting all your workflow files into a directory that you know where it's at so you can easily access it when needed.

[shut geyser]

this normal?

[ornate elk]

Ah nice I have the same GPU.
Works great with Zluda.
Yes thats Comfyui with the manager.
If you want a Ui similar to Auto1111 you would need to install Stable-Swarm UI.
But I don't have a guide for it yet

[ornate elk]

Yes you can even run SD3

[shut geyser]

your guide was THE ONLY one that worked, literally everything else gives an error

[onyx solar]

Ah...I had some legacy group data in there for some reason. Gimmie a sec.

[shut geyser]

stable diffusion 3 medium?

[ornate elk]

Yea there are so many guides out there. Most have missing steps or are outdated. Thats why I made these guides and tried to get them all working before.

[onyx solar]

Try this one.

[shut geyser]

which one should i download? sorry im really new to this

[ornate elk]

But only use the 5.9gb sd3 model. It has the clip text encoders included.
I can give you my workflow

[shut geyser]

sure, i will try both workflows of soul and yours

[onyx solar]

I do appreciate that you're willing to take the time to learn a bit and not give up. The curve for Comfy is a bit steeper than A1111, but once you're past that, it's tremendously easy and powerful.

[ornate elk]

here is the SD3 Workflow

and here is an sdxl workflow

[shut geyser]

thank you dude this is much easier with people like you and cs1o to do

without his guide i wouldnt be able to run it, and you guys respond really fast too which is nice, when ill have trouble ill message here

[shut geyser]

thank you a lot

[ornate elk]

can i ask where you found my guide? was it here in the pinned messages?

[shut geyser]

yes it was when you told me to check it out

[ornate elk]

ahhh right

[shut geyser]

how do you import characters? i mean the safetensor file, where do you put it?

[ornate elk]

into ComfyUI\models\checkpoints

if its a lora it goes into models\lora

[onyx solar]

Caution, though; are you downloading a full checkpoint or a LoRA?

LoRAs are not able to be used properly as full checkpoints (models); they are applied on top of checkpoints.

[shut geyser]

i got no idea how to see it

[onyx solar]

Where did you get what you are downloading?

[shut geyser]

nevermind im blind

it says type: lora

😭

civitai

[onyx solar]

Ok, so start by just trying to create what you want via text and the base model, then graduate to adding the LoRA into the workflow.

Just make sure that the LoRA is for a version of model that you're using in the workflow. (1.5 and XL are the most common versions. If you see Pony, that's a special thing for XL based on a popular model that has some quirks to it; I'd recommend staying away from that at the very beginning until you get familiar, then go read up on Pony.)

[shut geyser]

where do images save?

i wanna show you what generated

its very accurate

[onyx solar]

Many workflows will save in the "output" folder. If you can't find them there, look for a temp folder.

[shut geyser]

what is 1-5 pruned emaonly?
whats the difference of

emaonly and just pruned

im trying stable diffusion 1.5 emaonly and it gave a bad result

[onyx solar]

Long story short is that the pruned emaonly one is smaller and uses less VRAM.

[shut geyser]

this is not ankha from animal crossing 😭

[onyx solar]

FYI in case you haven't seen yet, SD 1.5 images should be ¼ megapixel (variations of 512x512) versus SDXL, which is 1 megapixel (variations of 1024x1024). Between size differences and the different VAE, you can get garbage output if incorrectly configured in your workflow.

[shut geyser]

can i use a sd 1.5 model in sdxl?

[onyx solar]

SDXL and SD 1.5 are models.

They're 2 different model versions. Careful not to mistake the model version with an application version. 🙂

[shut geyser]

yeah i meant like
1.5 loras

can i use a 1.5 lora for sdxl?

[onyx solar]

LoRAs only apply to the model version they're made for.

[shut geyser]

okay okay ill download normal 1.5 stable diffusion

not emaonly

[sinful marten]

hey guys good morning, so i was hoping someone could point me to the file i need to edit to be able to use all my cpu cores, its telling me when i start the server its set itself to max of 16 cores to be safe and this is the txt i copied from the cmd prompt

NUMEXPR_MAX_THREADS

[shut geyser]

where do i put v1-5 pruned.safetensors? in the vae folder?

[ornate elk]

No don't download the default model

[shut geyser]

Which one should i get

[ornate elk]

For example download Dreamshaper v8
Its an 1.5 based model.
And for SDXL try albedo base xl

Or juggernaut xl

[shut geyser]

Is there a site for those? Like to see the other models

[ornate elk]

Yep
Civitai.com

Its the biggest model database

There you also see if a model or lora is 1.5 or sdxl based

[young bridge]

Hi all

What tool is that

I would like to try it out, looks really cool

[shut geyser]

do you know how to select loras?

my last 5 images look like this

[ornate elk]

You have to make a lora node

[shut geyser]

is there a vae for dreamshaper? i cant find it

and it needs a vae

do you always need to use a vae?

[ornate elk]

Some models have them included. But better use one

[willow river]

why after inpaint, on image in mask area is a lot of blurry mess?

tried a lot of settings, nothing helps

[ornate elk]

Your AMD user?

[sinful marten]

has sd created a model that will actually take text input and make a legitimate 3d model for use inside apps like blender. i read an article from dec 23 that they were working on one for making polygon models?

would love to talk to anyone using tydiffusion in 3ds max , pretty amazing

[ruby yoke]

no, but there are plenty of others out there working on that

and no one working on 3ds max

[sinful marten]

thnx crystal

well the hottest thing is tyflow for effects and its the only legit thing for 3d and only for max so id look again because im using everything in a real gui insiode 3ds max with no browser at all

now it is not for model creation just effects

[ruby yoke]

i know, but no one i've seen is doing anything for 3ds max. most of them are doing text to 3d in their own interfaces.

[sinful marten]

oh gotcha

crystal heh i remeber we already talked about this the other day my bad

[ruby yoke]

🙂 it's fine. i know you really would like this, i wish i had better news or somewhere to point you to

[sinful marten]

ehh all good itl happen

its so new that tyflow doesnt even have but install instructions thats it, they dont have a server or forum or anything so i created a server here and ive been the only one in the room,

[dim fractal]

Is there a way i can .. make my stable diffusion XL model into an API? this way people cant download it, but they can use it, as long as i let them?

or is there a way i can provide the model without downloads?

[fair oxide]

https://youtu.be/tXO6SJ-6Eb8

huge memory hog atm but quality is what they aimed for with SD3, so it seems

not usable for average users, it requires 24gb vram, and 64gb ram

[crude jasper]

can anyone help me out and list the top models for photorealism

[tepid crypt]

im having some problem using xformers, when running appears this:

    PyTorch 2.1.2+cu121 with CUDA 1201 (you have 2.0.0.post304)
    Python  3.10.13 (you have 3.10.14)```
and when trying to generate something appears this:
```NotImplementedError: No operator found for `memory_efficient_attention_forward` with inputs: query : shape=(2, 3840, 10, 64) (torch.float16) key : shape=(2, 3840, 10, 64) (torch.float16) value : shape=(2, 3840, 10, 64) (torch.float16) attn_bias : <class 'NoneType'> p : 0.0 `decoderF` is not supported because: xFormers wasn't build with CUDA support attn_bias type is <class 'NoneType'> operator wasn't built - see `python -m xformers.info` for more info `flshattF@0.0.0` is not supported because: xFormers wasn't build with CUDA support requires device with capability > (8, 0) but your GPU has capability (7, 5) (too old) operator wasn't built - see `python -m xformers.info` for more info `tritonflashattF` is not supported because: xFormers wasn't build with CUDA support requires device with capability > (8, 0) but your GPU has capability (7, 5) (too old) operator wasn't built - see `python -m xformers.info` for more info triton is not available requires GPU with sm80 minimum compute capacity, e.g., A100/H100/L4 `cutlassF` is not supported because: xFormers wasn't build with CUDA support operator wasn't built - see `python -m xformers.info` for more info `smallkF` is not supported because: max(query.shape[-1] != value.shape[-1]) > 32 xFormers wasn't build with CUDA support dtype=torch.float16 (supported: {torch.float32}) operator wasn't built - see `python -m xformers.info` for more info unsupported embed per head: 64```

i have no idea what to do, please help

[ruby yoke]

no

[ruby yoke]

you're running the wrong version of pytorch

[tepid crypt]

how do i install the correct version in a colab? im using sagemaker

[ruby yoke]

normally, at the start of any colab notebook, you'll find a section that installs stuff, like the version of python and pytorch. your error is "xFormers was built for: PyTorch 2.1.2+cu121 with CUDA 1201 (you have 2.0.0.post304)" <-- that tells you that you are going to need to use PyTorch 2.1.2 - is this your notebook?

[tepid crypt]

can share the github of the colab

i have no idea what version of pytorch is using neither know how to install the one i need

[ruby yoke]

if it's not your notebook, you should probably go to the author of it and report the error

[dim fractal]

Hmm. Well then. Is it possible to rent a GPU. Say vast.ai where it costs me like 15 cents per hour, setup a login for them but only to the webui?

[ruby yoke]

you can't very well run a google notebook somewhere else.

[dim fractal]

huh????

oh i see, yeah you have no idea what youre even saying, you can easily run PCs elsewhere, ill figure it out.

[ruby yoke]

no, i replied to the wrong person. but you have a good life with that attitude

[misty marten]

I did, but it gave the same error

[fair oxide]

https://comfyanonymous.github.io/ComfyUI_examples/flux/

[lunar perch]

so is juggernaughtxl ass?

running it now, insanely slow, at just 20 steps

like 10x slower than pony

[glacial sundial]

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)

help

[tawdry depot]

does anyone know any prompt databases

my stuff keeps comin out rlly weird

[calm cairn]

I save cool prompts from CivitAI when I see them, into folders so I can find relevant stuff later. There are a lot of archived posts online, but IMO all the ones who promote themselves are pretty bad, and image gens with prompts available are better. Hope that helps.

[ruby yoke]

i've got a spreadsheet full of prompts and stuff you can have a copy of if you'd like

[ruby yoke]

you've got to set your system up to use just your gpu or just your cpu. let me guess, you're using an AMD card?

[calm cairn]

I would also like it, is there a link, or is it a DM thing?

[ruby yoke]

i need to DM you, if that's okay

[calm cairn]

yeah man, of course

[tawdry depot]

Thx bro u can post here or dm me

[chrome rampart]

Do you know if not functions or just (real) slow?

[ornate elk]

Do you run the latest AMD driver?

[fair oxide]

it works fairly ok for a large model, i'm using it on rtx 3060 with 12gb vram along with 32gb ram, its not fast but not dead slow either

[chrome rampart]

thks thats the card i have also so can try. Have amazing weekend

[fair oxide]

np, this is the quality for this base model

1. Flux Schnell is a distilled 4 step model.
   You can find the Flux Schnell diffusion model weights in the link below and the file should go in your: ComfyUI/models/unet/ folder.
   https://huggingface.co/black-forest-labs/FLUX.1-schnell/blob/main/flux1-schnell.sft

2. download t5xxl_fp8_e4m3fn.safetensors and clip_l.safetensors in your ComfyUI/models/clip/ directory you can find them on:
   https://huggingface.co/comfyanonymous/flux_text_encoders/tree/main

3. The VAE found in the link below should go in your ComfyUI/models/vae/ folder.
   https://huggingface.co/black-forest-labs/FLUX.1-schnell/blob/main/ae.sft

source: https://comfyanonymous.github.io/ComfyUI_examples/flux/

[chrome rampart]

thks

[manic matrix]

Sorry if this issue happens all the time. I just downloaded the automatic stable diffusion and during its start up in the command prompt, it failed 86% of the way through saying that there was no checkpoint. So I downloaded a checkpoint, but cannot put it in the stable diffusion checkpoints Notepad++ folder due to it being over 2GB. How on earth do I set this up if everything I need is over 2GB? Do I have a fundamental misunderstanding of what to do? (I’m not a tech nerd)

[shut geyser]

Is there a guide how to run stable diffusion 1.5? It gives me VERY bad results, just like dreamshaper

[ornate elk]

You have to put the model into the stable-diffusion-webui/models/stable-diffusion folder.
Dont drag and drop it onto the information text file there

[manic matrix]

Oh ok I’m doing that this time, waiting for results

[ornate elk]

use a good negative prompt, use quality tags for postive prompt. and the rest comes with upscaling

[manic matrix]

Now in command prompt it says Error: could not find a version that satisfies the requirement pytorch_lightning==1.7.6

No matching distribution found for PyTorch_lightning==1.7.6

[ornate elk]

can you show the full cmd text?

[manic matrix]

Yeah one sec

[shut geyser]

how to upscale?

[ornate elk]

in comfyui its a bit difficult for AMD

[shut geyser]

is there a special workflow for that?

if you have a 1.5 workflow i would apprieciate

[manic matrix]

Like idk what I did wrong I followed a guide step by step then did what you just told me so

I feel dumb

[ornate elk]

You followed a wrong guide it seems. (Downloading as zip isnt the right installation method)
Please follow my install guide for nvidia from the pinned messages

[manic matrix]

Ill try this, thanks

Lol already on step two, typing cmd in the Ai folder search bar does not do anything

[shut geyser]

press right click and open with cmd

or copy the directory and type cd and then paste the directory

it works the same

[manic matrix]

do I right click open with terminal?

[shut geyser]

yeah anywhere inside the folder

and if it says the directory when the cmd starts up its good

[manic matrix]

copy pasting the github link gave an error though

PS C:\Ai> git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
git : The term 'git' is not recognized as the name of a cmdlet, function, script file, or operable program. Check the
spelling of the name, or if a path was included, verify that the path is correct and try again.
At line:1 char:1

• git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git

•   + CategoryInfo          : ObjectNotFound: (git:String) [], CommandNotFoundException
    + FullyQualifiedErrorId : CommandNotFoundException
  
  

PS C:\Ai>

[ornate elk]

dont use search bar lol

[manic matrix]

i didn't after that, but once I got the cmd working it's giving me that error when I copy paste

[ornate elk]

you have to install git

and python 3.10.11

the links are provided

[manic matrix]

I did python, I didn't see the git link ig my b

Nope it's still not working even after I set up git

and set up python to path

getting the same error

im losing my mind

bro if this is a hint of what comes after step 2 i give up

[ornate elk]

can you show the error again?

did you installed Git with the default settings?

[ornate elk]

after reading it again, you cant.
You can only generate a new image wit hthe hair+face but you cant swap an existing head without inpainting it

[gusty salmon]

Like there is noooo way at all? No other techniques?

There are some extensions that detect the face+hair and automatically create a mask for I painting. Only problem, if the hair length will always be the same since the mask dictates the hair length ... Maybe you have a solution for that?

[dapper briar]

Hello, after installing pytorch 2.3 I have this, what could be a problem?

and if I use the command --skip-torch-cuda-test i get this

[ornate elk]

Hey, what's your GPU?

[dapper briar]

nvidia rtx 4070ti

[ornate elk]

Okay then open up a cmd and type
Pip cache purge
Then delete the venv folder.
And then relaunch the webui-user.bat

[dapper briar]

seems like it's worked, thanks

[ornate elk]

And edit the webui-user.bat and add only --xformers
As commandline_args

Don't use skip torch cuda test

[dapper briar]

I'll try, thanks again

[gusty salmon]

@ornate elk just in case you didnt see it since i didnt quote you

[ornate elk]

Your meaning Segment Anything ?
Have you tried the extension Inpaint anything?
Maybe that can do it

[gusty salmon]

Cant remember, tried one of them many months ago

The problem is, you cant replace a short hair head with a long hair head. Since the mask will be just as big as the hair.

[ornate elk]

Yep

Because it segments the hair.
Its only possible with a bigger mask

[gusty salmon]

Yea

So you think what im trying to achieve is pretty much impossible, right

[ornate elk]

Not impossible but would require someone with a lot coding experience

To manually make such a script that can do it

[gusty salmon]

Got it

Well thanks for all the advice lol

[ornate elk]

No problem

[glacial sundial]

No, Nvidia

[ruby yoke]

That's what SAM is for

[ruby yoke]

Ok, that's weird, usually you get that with amd

[glacial sundial]

I'm using linux btw

[brittle pagoda]

Hi.... I am a dumb architect who have no knowledge in coding:
I already habe installed swarmUI and I need to install A1111. How can I use the same models that I have of SwarmUI models folder in freshly extracted "stable-diffusion-webui-1.10.0"?

[ruby yoke]

Oof. And I use windows. I'm also not at my desk, I'm going to have to research this and get back to you. However we've had the error before here, and the solution might be in this channel. Use the search function for the specific error and give me an hour or so. Please

[ornate elk]

Yes you can edit the webui-user.bat of auto1111 and there add --ckpt-dir "D:\Path\toModels"
--lora-dir "D:...."

[brittle pagoda]

Many thanks....I will give a try now

[brittle pagoda]

Like this?

[ornate elk]

It needs do be after the =

From commandline_args

And you have to use your path to the models

My path is just an example

[brittle pagoda]

How's it now please?

Wait....

[ornate elk]

Looks good

[glacial sundial]

ok, thx, I appreciate it

[brittle pagoda]

And the loras?....Pardon the annoyance

[brittle pagoda]

Thank you.....Its working !!😃

[ornate elk]

Perfect no problem:)

[ruby yoke]

are you using comfy, auto1111, or webui?

[glacial sundial]

https://github.com/AUTOMATIC1111/stable-diffusion-webui

[ruby yoke]

okay, read through all the replies in this thread https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/2373

[calm cairn]

oh no thats the error I have been having
update: Using a different browser does not, in fact, fix it

Some of the people in the thread are saying add command line args
"--precision full --no-half --lowvram --always-batch-cond-uncond --opt-split-attention"
which I can do, but the other people are saying disable sd refiner, and I don't know what that is or how to disable it. I'll try putting in the command line args now, but I would appreciate help disabling refiner

[ruby yoke]

no, the browser won't. there are command line settings and arguements they discuss

@ornate elk is our Auto1111 expert and might have a solution

[calm cairn]

im mentioning the browser because CS1o previously (yesterday) helped me with this and swapping from Chrome to Edge worked, for around 12 hours

[ruby yoke]

that sounds sort of like a caching issue then

[calm cairn]

I also assume it is related to memory in some way, and actually generating the error happens inconsistently, but often after multiple gens, or after swapping models, which is consistent with a memory caching problem

[ornate elk]

Dont add these args. They won't fix anything

[calm cairn]

ah ok

What is the uninstalling refiner thing? I don't know what it is or how to uninstall it. Would it possibly help?

[plain bronze]

Hi, why is that?

I think it happened after I launched Kaspersky antivirus but I'm not sure. I can still run A1111 but it annoys me.

Kaspersky app didn't warn me about anything.

And there is another problem.

Should I delete and reinstall it?

[ruby yoke]

just ignore it - if it helped the one user on that thread that said they did it, something else is wrong on their machine

[calm cairn]

ok, then I still have the error he has, but neither of the solutions in that thread (command line args, uninstall addon) can help me? Definitely looking for a fix so I can gen again

[ruby yoke]

can you poste your entire commandline interface error log please

[calm cairn]

sure, let me start it up and recreate the error, will post it shortly

[ornate elk]

it just means you should not have this enabled:

[ornate elk]

yea you should install it the right way via git clone

a guide for that is in the pinned messages

[plain bronze]

#🤝｜tech-support message

This?

[ornate elk]

yea on that link you find all guides

[plain bronze]

I think it was because OneDrive.

[ruby yoke]

what about OneDrive?

[plain bronze]

Just stopped syncing and all my synced files gave error but I only noticed webui first.

Just deleted OneDrive and problem solved. Megamind. Tired of OneDrive.

[ornate elk]

yes OneDrive cause issues with SD

[manic matrix]

cs10 this is the message

Windows PowerShell
Copyright (C) Microsoft Corporation. All rights reserved.

Install the latest PowerShell for new features and improvements! https://aka.ms/PSWindows

PS C:> git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
git : The term 'git' is not recognized as the name of a cmdlet, function, script file, or operable program. Check the
spelling of the name, or if a path was included, verify that the path is correct and try again.
At line:1 char:1

• git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git

•   + CategoryInfo          : ObjectNotFound: (git:String) [], CommandNotFoundException
    + FullyQualifiedErrorId : CommandNotFoundException
  
  

PS C:>

[ornate elk]

you have to run git clone in a cmd not in a powershell

thats the issue

[honest tendon]

@ornate elk can i run Flux on ROCM?

[manic matrix]

I right click on the AI folder and the only option I have is “open with terminal”. Is this not what I’m supposed to click

If it’s not idk how to open a cmd

[ornate elk]

Dont right click the ai folder

Go into the ai folder

Then on the explorer bar where the path says C:\ai
You click in that bar. Then just type cmd and hit enter

[ornate elk]

Should work yes

But I didnt tested it

[honest tendon]

ill try it out imma reinstall SD since i installed too many stuff on it

i mean comfyui

[ocean locust]

I've got a DevOps question and it's stumped me.

The ComfyUI AI-Dock image has sample init scripts for HF token passing (https://github.com/ai-dock/comfyui/blob/main/config/provisioning/sd3.sh). I want to extend that functionality to the A1111 script (https://github.com/ai-dock/stable-diffusion-webui/), and wouldn't even know how to start doing that.

Unfortunately, when I contacted Vast.ai support, the person I spoke to wasn't familiar with SD-web-ui's init script structure.

[ornate elk]

Are you on Linux or windows with WSL/Zluda?

[honest tendon]

Windiws ZLUDA

[honest tendon]

im just waiting for an image to pop out

takrs a while for tje first image

[rich garden]

Does stable diffusion save the prompts I’ve used before in the past? I’ve noticed it saves my image generations so I was wondering if it can save my past prompts too.

PLZ PING