🤝｜tech-support

[lost crescent]

If you are using an sdxl model, only sdxl loras will show.

If you are using a 1.5 model, only 1.5 loras will show ...

#🤝｜tech-support message

[hallow latch]

Quick question, I have the live server option on.. But it says 72 hours, but twice now I have gotten the error like it cannot connect. Both times it has been max 24 hours.

[ornate elk]

Yea the gradio link isn't the best option

If you know portforwarding you can use --listen for a permanent solution

[rancid knot]

Hi, I have a question regarding creating embeddings, specifically preprocessing images. In my automatic1111 Web UI I am missing the preprocess images tab. Does anyone know this issue or how to enable that specific tab? Is it checkpoint / HW dependant?

[hallow latch]

Yeah, I was thinking off opening up a port instead as this happens twice now.

[ornate elk]

Hey, do you have a screenshot of the ui?

[ornate elk]

Interesting it vanished on 1.7.0

[karmic crown]

let's be honest: I could easily make the same script for Windows that will use winget, but I honestly have no interest in bothering with Windows 🙃

but, i might make one for ubuntu next year

looks like it is not mac specific, check out this https://github.com/bmaltais/kohya_ss/issues/208

[rancid knot]

That was a good hint, I checked the release notes, seems like it was integrated to the extras tab, thx!

[ornate elk]

ah, good to know, no problem

[hallow latch]

I put this in the webui-user.bat or in the webui.py?

[karmic crown]

please, do not touch .py files 🙂

[hallow latch]

I added share=True in the .py to enable online thing

[karmic crown]

remove that and revert it to be as it was and just add --share to webui-user.bat

[hallow latch]

Tested at home now, just 6 images with 512x515, 30 steps
** Time taken: 14.6 sec. - A: 3.25 GB, R: 4.27 GB, Sys: 5.4/11 GB (49.4%) **

[hallow latch]

Alright, I will add --listen as mentioned a bit up for a permant solution 🙂

[karmic crown]

listen will give you access from local network, if that is enough, just listen

if you want internet access, you need to use --share or --listen + reverse proxy

i have --listen on my son's pc, so i can access that one from my mac
i tried to use --listen + reverse proxy on my linux server, but i failed to run properly, it is possible that is related to cloudflare that i have, but it can aslo be some mess i made in apache

btw --share has the same effect as share=True, but is a recommended way

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings

[ornate elk]

nice, but test sdxl speed with 1024x1024, batch size and count 1

[karmic crown]

@ornate elk is there anythin specific on linux for amd vs nvidia? except command line params?

xformers i guess

[ornate elk]

amd need rocm drivers to work with SD on linux

[hallow latch]

sdxl speed? Just any model and 1024x1024?

Time taken: 16.0 sec.

A: 5.51 GB, R: 8.06 GB, Sys: 9.9/11 GB (89.7%)

[karmic crown]

i just checked my script, and what i would need to change to make it run on Ubuntu
basically, i just need to make a new function that will install python and git with apt get
and to run that one instead of homebrew installation if linux is detected
the rest would work as it is
so i am asking if there is anything else that i should take care of
i would not like to bother with drivers, but i guess i could add a check if they are installed or not

[ornate elk]

okay thats good

[hallow latch]

I notice with just 4 images the /it is at 1.50-1.60s/it

[ornate elk]

yea its faster with batch size 1, and then increase batch count

[ornate elk]

hmm, nothing special besides the driver

[hallow latch]

Batch count 1 and batch size 4?

or batch bount 4 and batch size 1?

[ornate elk]

batch count 4

size 1

[hallow latch]

Ohh, why? 😮

[karmic crown]

4x1 is faster than 1x4

thats why

[hallow latch]

1:4 = 54,7
4:1 = 1min 1.7 sec

[karmic crown]

as i just wrote 🙂

[hallow latch]

But you said batch count 4 and 1 ..

That was slower.

[karmic crown]

?!

[hallow latch]

Batch count 4 / Batch size 1 = 1 min and 1.7 sec

Batch count 1 / Batch size 4 = 54,7 sec

[ornate elk]

i would recommend using batch count instead of batch size if you gonna use upscaling or refiner as that increases the vram usage

[karmic crown]

strange, it is usually the other way

[hallow latch]

Maybe better to use that way so it's less strain? I dunno

But I will do so in the future 🙂

Even if it's slower for me 🙂

[karmic crown]

cs1o has a point

[hallow latch]

Maybe less freeze and crashes 🙂 for me

[karmic crown]

memory usage is lower

and that is probably the reson why it faster on some systems with lower vram

but it is posible that is faster if you have enough vram (which i do not have) 🙂

[hallow latch]

How much do you have?

[karmic crown]

m1 mac 🙂

if i say none i would lie, but i would also lie if i say 16, since that is ram 🤣

who knows with silicon macs 🤣

[hallow latch]

Testing with 10 batch count and Hires fix with 1 upscale and 20 hires steps 🙂

Time taken: 4 min. 25.5 sec.

🙂

[karmic crown]

i found a comparation 🙂

[hallow latch]

I checked the preview files, they are the same name..

I mentioned this a bit earlier regarding the previews being broken.

[fallen nymph]

@ornate elk broski am i able to get better image quality using the ai

[quartz brook]

anyone else having issues with clip encode? I keep erroring out on comfyui whenever I change the text of the prompt, no other changes to the workflow, just the text. Restart Comfyui and it works again until I change the text
[9:02 AM]
"Error occurred when executing CLIPTextEncode:

Cannot copy out of meta tensor; no data!" is the error message

[night locust]

How do you guys deal with juggling different python versions on windows?

[faint night]

https://cdn.discordapp.com/attachments/795355171420504094/1187081356811976704/image.png?ex=659596aa&is=658321aa&hm=110b8f62d4477615685c4cd84d2d2db344ea31199cafb3024cfd26f7b97863c0&

how can we enable the control net on the code on any safetensors like should we modify the model and add the control net to its py files or how? (enabling control net like exactly the webui)

[wind sun]

OutOfMemoryError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 4.00 GiB total capacity; 3.43 GiB already allocated; 0 bytes free; 3.48 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 does anyone know what this error means after i click generate

[karmic crown]

what do you want do to? to have different versions for comfy and a1111?

[ornate elk]

you can specify which python SD should use if you edit the webui-user.bat (if you use auto1111)

i only have one python version installed, but with that i could use more

[ornate elk]

if you use auto1111 that means your running out of vram.
you can fix that by editing the webui-user.bat
you need to add: --medvram --xformers --no-half-vae
to your webui-user.bat
at the line Commanline_ARGS=
Then save and relaunch

[fair oxide]

@ornate elk whats this error?

[ornate elk]

when do you get this error? what model did you tried?

[fair oxide]

none, how do I add a model? I'm downloading one

[ornate elk]

for example download dremshaper v8 and put it into models/stable-diffusion

[fair oxide]

ohh i get it, i dont know that i need to put into this folder

im already downloading a model

so i just need to put it into models/sd? @ornate elk

[opaque spindle]

I saw this Lora tutorial from our resources channel: #1080946152318443610 message
The instruction is pip install git+https://github.com/cloneofsimo/lora.git.

How to combine this to the original a1111 script?

I don't need to follow the exact resource but to create a consistent character, what's the easiest way on a 4GB vram Mac with a1111/automatic1111 on it. Thanks. Appreciate if anyone reply.

[ornate elk]

yes models (checkpoints) that are larger than 1.98gb go into models/stable-diffusion

everything below 1.98gb isnt a model

[fair oxide]

mine is 6.6gb haha

[fair oxide]

oh

[ornate elk]

thats an SDXL model, and if your on AMD you run into problems if you have to less vram for it

try 768x768 with sdxl models

if you only have 8gb vram your out of luck

[opaque spindle]

Also, I saw this tutorial, https://stable-diffusion-art.com/train-lora/, however, the ui looks different from the one I installed with a1111/automatic1111, do I need another extension?

[fair oxide]

do u know do I change faces in photos? @ornate elk

[ornate elk]

you mean make them better quality?

[fair oxide]

no

like put billie eilish's face on jenna ortega's body

[ornate elk]

that can be done with inpainting or with extensions like faceswap

[fair oxide]

oh okay thanks

[fair oxide]

okay but i will get low quality?

[ornate elk]

whats your GPU?

[fair oxide]

can i get more quality?

[fair oxide]

it's a gpu that comes with the processor

[ornate elk]

ohh okay, so integrated amd gpu

i dont know if sdxl models would work for you, but try with a 1.5, 2gb model to test

then you can try an sdxl model on 768x768

[fair oxide]

okay

thx

[karmic crown]

where did you see that? i cant find it

[opaque spindle]

Let me send you the link again.

[undone urchin]

Hello, I'm getting a "model not found" error with the animation SDK using the SDXL- 1024 model and was wondering if anyone has info about it. Haven't used the SDK in a while and got this error out of nowhere when running my exact same configs on the same project as before.
Only happens after 1 frame has rendered. Log: """
8%|█▏ | 1/12 [00:07<01:22, 7.50s/it]Received RpcError: <_MultiThreadedRendezvous of RPC that terminated with:
status = StatusCode.NOT_FOUND
details = "The specified engine (ID stable-diffusion-xl-1024-v1-0-comfy) was not found."
"""

[karmic crown]

i see now, you saw that on their github page

that would be fine if you want to use command line, but i am not sure how that will work with a1111 without plugin

[opaque spindle]

https://stable-diffusion-art.com/train-lora/
LoRA: fine-tuning with a smaller size
LORA is a type of embedding training method, which functions similarly to Dreambooth. Although it might have slightly lower quality results, it compensates with a much faster training process and smaller file sizes than Dreambooth. Currently, it's regarded as one of the best options available in terms of balancing quality and requirements.

Difficulty: easy-hard (less likely to affect the base model)
Style training: good quality
Subject training: good quality
GPU cost: medium (8GB or more)

LoRA Support in Diffusers: https://huggingface.co/docs/diffusers/training/lora
Github: https://github.com/cloneofsimo/lora

[karmic crown]

so i would stay away from it

[opaque spindle]

ok ok

[opaque spindle]

I am trying to create a comic with consistent characters, faces, what's my best bait?

[karmic crown]

i would ask in https://discord.com/channels/1002292111942635562/1004159122335354970

[opaque spindle]

Thanks a lot. I am new. Just got a1111 install.

ok

[karmic crown]

some of those guys does not follow this channel, but they know much more about that than me

[opaque spindle]

I know this channel is about installation. I am following this tutorial but the UI looks different. Wondering if I should add extensions?

[karmic crown]

as i can see there is a google colab that you run on google cloud

https://andrewongai.gumroad.com/l/lora_training

but as i can see you need to pay for it

as i said, other channels are much better choice to ask about lora training

[ornate elk]

depends on the settings

[undone urchin]

I am getting "engine not found" for every model I try to use for animation SDK in the gradio ui. anyone else?

[mighty rivet]

for automatic1111 api, is it possible to insert the image and mask from the 8080/docs

[hallow latch]

If the console get stuck, can I somehow like "unstuck" it or should I just close it and open a new one?

[wind sun]

thank you but i was doing the auto1111 for a friend and this poped up, when i checked his nvidia drivers, they were outdated by 2 years xD... I downloaded the latest and everything now is honey

[verbal burrow]

Uh, what do I do when I get this error
AssertionError: We do not support vanilla attention in 1.13.1+cu117 anymore, as it is too expensive. Please install xformers via e.g. 'pip install xformers==0.0.16'

Cause I swear I already installed xformers awhile back

[ornate elk]

okay, he still should do steps i mentioned too, so he can generate much faster and higher res

[ornate elk]

your webui is outdated

[verbal burrow]

Oh

If I add git pull will it update?

[ornate elk]

yes and then delete the venv folder before launching

[verbal burrow]

What is the venv folder by the way?

You've mentioned it before to me a month ago, but forgot to ask

[ornate elk]

its the folder inside the stable-diffusion-webui folder

it cointains the outdated stuff and will be regenrated

[verbal burrow]

Ahh, okay

[karmic crown]

remove that folder

[verbal burrow]

Yep, just waiting for it to finish up

[wind sun]

ohh rightt

[ornate elk]

if you have an nvidia gpu you should use xformers too

[verbal burrow]

@ornate elk Prompt says Stable Diffusion Model failed to load

[ornate elk]

screenshot of the full cmd after launching it pls

[ornate elk]

its not updated

how did you added the git pull ?

[verbal burrow]

I'm on Notepad++

[ornate elk]

close everything and relaunch the webui-user.bat
then make a full screenshot of the cmd

[verbal burrow]

Oh

OHH

I've been launching webui.py

[ornate elk]

... xD

[verbal burrow]

Okay, it's installing

It's doing stuff

[ornate elk]

ok we wait how it turns out

then we can do the next step,

[verbal burrow]

https://tenor.com/view/survive-live-win-tropicthunder-rdj-gif-5129499

Alright, hopefully it won't take too long

What are the specs that are most recommended for StableDiffusion?

I recently bought a 3060 12GB just to enhance my AI results

Been using a 1060 6GB since 2016

[karmic crown]

@ornate elk see this

[ornate elk]

--xformers --no-half-vae are for 3060

[verbal burrow]

Yeah, I remember you told me to write that in the .bat file,

Also here's the results

[ornate elk]

looks good

[ornate elk]

see what ? 😮

[verbal burrow]

I was running StableDiffusion for months without updating it, until last month and now

[karmic crown]

@echo off
winget install -e --id Python.Python.3.10
winget install -e --id Git.Git
mkdir c:\ai\
cd c:\ai\
git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui
cd stable-diffusion-webui
call webui-user.bat

it will work for nvidia without any change 🙂

[ornate elk]

ouhh nice and short

[verbal burrow]

Here we go

[hallow latch]

Currently trying to do 10 images and it froze on "74% ETA: 01:14" can I do something?

[karmic crown]

but it is very limited compared to my mac script

btw winget install python from pyton.org

winget is great

[ornate elk]

winget is awesome

using it to keep track of updates

[ornate elk]

check the cmd, or press interrupt, if nothin helps, close the cmd

[verbal burrow]

@ornate elk Works now, thanks a lot!

[ornate elk]

you need to update your extensions after its done

[verbal burrow]

Oh, how does that work?

[ornate elk]

go to the extensions tab and click search for updates, then click on apply and restart

[verbal burrow]

Alright

Done

[viscid badge]

can someone explain me what is this ;

[verbal burrow]

That's a rabbit hole to go down. In a nutshell: How your image is created by AI

[viscid badge]

is it... a best option ?

[verbal burrow]

https://stable-diffusion-art.com/samplers/#What_is_Sampling

[viscid badge]

Something very use

[verbal burrow]

It's a find out what works best

[viscid badge]

ty !

ty ! and another question

when I try to render something it always giving me 2 characters

[verbal burrow]

Use the negative prompt and specific to not give multiple characters

Maybe use the term Multiple people as an example

[viscid badge]

i'll try and telling u

[karmic crown]

@ornate elk version with sed, that can add command line params

@echo off
winget install -e --id Python.Python.3.10
winget install -e --id Git.Git
winget install -e --id mbuilov.sed
mkdir c:\ai\
cd c:\ai\
git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui
cd stable-diffusion-webui
sed -i "s/set COMMANDLINE_ARGS=/set COMMANDLINE_ARGS=--xformers/" webui-user.bat
call webui-user.bat

[ornate elk]

if your using an 1.5 model you need to generate near the trained resolution of 512x512

[karmic crown]

i will leave further modifications to you 🙂

[viscid badge]

ok ok

[ornate elk]

for portrait use for example 512x768

you can then later upscale the image or use hires fix

[viscid badge]

ok ok !

[hallow latch]

I added --listen but I still get

[violet shadow]

Hi

I m new

I'm a new user and I just installed stable diffiusion for the first time. I wanted to install it extensively, the problem seems to be that I did everything right, the problem is that even though it was installed correctly, it does not display it in the selected place.
Right after that, I installed the control net panel and everything works fine. However, roop still does not display it without any changes, does anyone know what could be the reason?

[ornate elk]

You won't get a link with --listen
It only makes the webui accessible via your local PC IP Adress.
Then you have to make a port forwarding entry in your Routers config.

With that IP + Port

[hallow latch]

Okay, well since I added --listen I cannot do any stuff ..

So strange

[ornate elk]

What do you mean ?

If your at home it should launch normaly

[karmic crown]

what git status gives?

[hallow latch]

I can't get a Interrogate CLip to work ..

It is only loading ..

[ornate elk]

The first time it will download a model in the background

Check the cmd

[hallow latch]

Nothing happends in the CMD and I have already downloaded it ??

[ornate elk]

The cmd does nothing?

Then maybe restart and try again.
Also make sure the webui is whitelisted in any browser adblocker

[hallow latch]

It says this

Okay, now it put through.

I added the ports and restarted router..

Maybe it was blocked?

[ornate elk]

Maybe

Also it needs to be allowed in windows firewall

[hushed orchid]

(1.5 issue)

I keep getting a weird issue when I render. The faces keep getting corrupted during the last few steps.

The issue seems to only be happening on this model. The issue does not show when I use the model pages demo generation data.

The issue happens if I try anything else with the prompt. I read that the sampler might be the issue but it's really hard to say due to the fact that the demos work fine.

Here is the model page and also the demo data that I'm using as well.

https://civitai.com/models/234918/worldoffusch

edgy pinterest woman, 20 years old, wearing a oversized sweater, happy, dimples, natural look, sunny, soft lighting, emotional, grainy
Negative prompt: cartoon, painting, illustration, (worst quality, low quality, normal quality:2)
Steps: 30, RNG: CPU, VAE: vae-ft-mse-840000-ema-pruned.safetensors, Size: 512x768, Seed: 3096873722, Model: WorldOfFusch, Version: v1.7.0, Sampler: DPM++ 2M SDE, VAE hash: 735e4c3a44, CFG scale: 5, Model hash: 89c0b3da52, Hires steps: 15, Hires upscale: 2, Hires upscaler: 4x_NMKD-Superscale-SP_178000_G, Denoising strength: 0.25

[ornate elk]

Best would be to delete roop and instead use Reactor

[violet shadow]

How to delete a roop?

I trying to delete extension and models folders but It doesn't works

What is reactor I spend 2-3hourse in this program and I m totaly noob 😅

[ornate elk]

Delete the folder from the extensions folder

Webui cmd needs to be closed

[ornate elk]

Reactor is the same as roop but better, and updated

[violet shadow]

Ok I will try

Now

[hushed orchid]

@ornate elk Do you think you can help me with my issue?

[violet shadow]

Ok I delted so now I try to instal Reactor

[ornate elk]

Test without using hires fix.
There are some models that have this specific issue. Switching sampler can help.

[hushed orchid]

Already off. Works better with it on actually. (should not be that way)

The sampler was working before with no issues and still does when using the seeds on the demos.

The sampler is used throughout the model page as it works best with their model.

[ornate elk]

Did you tried an other sampler?

Also did you downloaded the 5gb or the 2gb model?

[hushed orchid]

I did. It works fine with other ones. The issue is that it was working fine before and now isn't.

Why would I want to use a different sampler if I was just using one that worked and out of nowhere stopped? I'd rather try to find a way to fix it.

[violet shadow]

Thanks guys ❤️

I don't know what a reactor is but let's say I have 2 photos and I really want to put another person's face as a cartoon character.

Is anyone willing to help me with this?

[hushed orchid]

Not a thing.

I downloaded the 4Gb model.

I have 8Gbs of Vram on my 3070Ti and overall 32Gbs of Ram.

[ornate elk]

Check this:
https://github.com/Gourieff/sd-webui-reactor#usage

[ornate elk]

Have you restarted and tried again?

[hushed orchid]

Several times

[ornate elk]

Did you tried an other upscaler too?

[hushed orchid]

Why would I if I'm not using it?

[ornate elk]

I have a model with the same problem and there is not much you can do about these artefacts, other then changing sampler or upscaler.

Its sometimes there and sometimes it works without artefacts

I think thats because of the block weight merge he mentioned. But I'm not sure.

Some models shouldn't be merged together

[round spade]

i am trying a new model

made for generating pixel art

but all i get are grey squares

no errors

so basically i have no lead on what the issue is

maybe its in the wrong folder?

i have put it in the stable diffusion folder

this is the model im talking about

https://civitai.com/models/120096/pixel-art-xl

[kindred tinsel]

But that ain’t a model it’s a Lora, you gotta put in the prompts and it should pick up, I had run into this issue with orher Lora’s and usually putting the ((black and white)) helped but im@not sure

[kindred tinsel]

It doesn’t go there, it goes in Lora’s folder

[round spade]

@kindred tinsel i have tried putting it there

it doesnt show up when i run a1111

[kindred tinsel]

did you hit the refresh button in the Lora list so that it looks back into the folder and parses it?

Aside from that if the model is a lora it should work, keep in mind i think that is for SDXL so sd1.5 wont support it

Also I have an issue with comfy Ui I was wondering someone might help me out with

I am using this setup

1. Generate image with Efficient Ksampler
2. Generate Upscale Latent
3. Use Advance Efficient Ksampler to Denoise again

Basically creating and then upcaling the same image, the thing is that it works fine for some models but I literally tried one model out that says comes with a built in VAE and it produced horrid images, even when setting in the efficient loaders that the VAE was built in

Im new to this kind of stuff and idk if the issue lies within the way the model is loaded in, since I tried a prompt from the civit ai pagbe and it still couldn't do that....
If interested or useful for this question I can provide the workflow.json thingy !

(also sorry dark mode users for bright comfy UI :c )

[fair oxide]

tried everything there and unfortunately it did not work.

@karmic crown just want to say your script made rendering on SD like at least 4-5x faster. Everyone should have this script

[heady stone]

Is anyone using mac on OS 12, can it work well in image generation? I was having some issues before and figure it because I have to update my mac. but i can't really atm

I got a userWarning about the MPS shader Not working right because it requires OS 13

[fair oxide]

I am on 14 and works here

[ornate elk]

The lora is made for sdxl. It will o my show up when you select an sdxl model

[verbal olive]

When i use "read generation parameters from last generation" it always says "override settings: clip skip 1", even if i've checked and made sure it says clip skip 2, is this just a bug from the read gen prompt option or am i not actually using clip skip 2 for some reason?

[karmic crown]

can you please show me the warning, pytorch metal acceleration requires 12.3 as far as i know, and as it is written here https://developer.apple.com/metal/pytorch/

[hallow latch]

Is this stable diffusions or another one ? How is it compared to the auto1111 one?

[karmic crown]

users of M2 and newer silicon Macs with 16+ GB RAM should have the most benefit from it
it will make things faster even for Intel Macs with Radeon (like iMac 2019), but not as drastically as in your case
M1 with 1.5 models works nicely, but sdxl is still slow, even if it is faster than with the original setup

[fair oxide]

i saw a lot of users complaining about slow speeds on M1 and M2, they would really benefit yea

Even on intel

[karmic crown]

if this doesn't frighten you, go ahead and try it

you can link models from a1111, so you can save disk space

[fair oxide]

i just have to ask you one question viking, is it possible somehow to make any AI "freeze" like one clothing with a lot of special patterns on it and just render rest of it?

I am training loras but the clothing has a lot of small patterns which are freely regererated by SD even if it over all look is same

I want my clothing/dress to be generated exactly as it is on the cool people SD is generating

[karmic crown]

i know this will sound strange, but i am a really new A1111 user, just a little more than 1 month, so I do not have an answer to your question 🤷‍♂️

but i am a developer with a lot of experience, and i know how to systematically test things and find what works the best
and how to put that in guide and code 🤣

[fair oxide]

ah i understand, i have only been on this ride for 2 days and then SD broke, but my understanding is that it is not possible

[hallow latch]

I am just kinda looking for something that might not just freeze/break to much 😄

[karmic crown]

comfy works great, but it can be overwhelming if you do not know what you are doing
a1111 is recommended for beginners
with right setup a1111 works nicely

[ornate elk]

When do you get a freeze or bug?

[hallow latch]

Well, it can be during a generating images, sometimes I can do like 10 images x 5 so I get 50 images and sometimes 100 and it works.

Yesterday I like 4 images and worked, then I did 8 and it froze at 55%

[ornate elk]

Okay, yea that can happen when using batch size.
As pytorch randomly allocate vram at each SD start it will work or not

[hallow latch]

What to do to avoid it? If it's possible+ 😄

[ornate elk]

You can't avoid it really. Using batch count or using lower batch size would help

[hallow latch]

Alriight, I will try to only use batch count and see how often it happens 🙂

[fair oxide]

I am using Kohya LoRa training with gpu V100 with 16 GB vram, i have network training at 128 and image res 1024,1024 but its running out of vram, i either have to lower network rank or the image res, is that normal?

[verbal olive]

How much time does controlnet pose changing add to your generations?

Wanna find out if this is just a me issue or if it's normal to take 35 minutes to Gen :p

Dropping from 3.5it/s to 130~180s/it

[elder turtle]

Doesnt sound normal to me

[ornate elk]

Thats not normal, in auto1111 ?

[verbal olive]

Yeah, even a pretty simple action using controlnet and it goes from a 8 second generation to well over 30 minutes lol

Probably just another 5700xt issue :p

[ornate elk]

You need to check the lowvram option in controlnet

Also you need to use the 700mb controlnet models and not the 1.5gb ones

[kindred tinsel]

Also I have an issue with comfy Ui I was wondering someone might help me out with

I am using this setup

1. Generate image with Efficient Ksampler
2. Generate Upscale Latent
3. Use Advance Efficient Ksampler to Denoise again

Basically creating and then upcaling the same image, the thing is that it works fine for some models but I literally tried one model out that says comes with a built in VAE and it produced horrid images, even when setting in the efficient loaders that the VAE was built in

Im new to this kind of stuff and idk if the issue lies within the way the model is loaded in, since I tried a prompt from the civit ai pagbe and it still couldn't do that....
If interested or useful for this question I can provide the workflow.json thingy !

[faint shuttle]

Does anyone please know how to change Command-line Arguments for comfy ui DirectML ??
idk where to find the file

[karmic crown]

as far as i know confy automatically decides what is best for your system

but i might be wrong

[faint shuttle]

@karmic crown i need to add --lowvram to it to generate images

[karmic crown]

https://www.reddit.com/r/comfyui/comments/15jxydu/comfyui_command_line_arguments_informational/

add that to the end of a command line

as described here

[faint shuttle]

i saw vids for ppl using nvidia cards who has the option to change that, but since im using amd its not that obvious

[faint shuttle]

ye, the problem is, Idk where to find the file to edit it

[karmic crown]

do not edit nothing

run it like that from cmd

how do you run comfy now?

by some bat?

[faint shuttle]

i paste that in the cmd ?

[faint shuttle]

no, by writing commands in cmd

[karmic crown]

good

instead of
python main.py
just do this
python main.py --lowvram

easy as that

you can check
python main.py -h
to see all available commands

[faint shuttle]

ohhh i see

thanks alot ill check that

sry im pretty new to all of this

[reef smelt]

when to use --opt-sdp-attention or --precision {full,autocast}?

[karmic crown]

many programs have command line params, and you run them exactly as i wrote
we used to run games like that in the last century, but kids nowadays usually want to click 🙃

in the case of a1111, they have a variable in bat file where you add them, so you do not need to add them manually every time

[karmic crown]

what gpu do you have?

[reef smelt]

rtx 2060 (not super)

6gb vram

[karmic crown]

i think that opt-sdp-attention doesn't work anymore (i might be wrong)
and with only 6gb forget precision

CS1o will probably know what are best params for your card

[reef smelt]

hmm

I'm currently using --xformers --upcast-sampling --opt-split-attention --lowvram --autolaunch

from https://www.youtube.com/watch?v=AKBelBkPHYk

[karmic crown]

remove --autolaunch, it will run anyway

[reef smelt]

it's just so that it opens it in my browser

[karmic crown]

it should do it without that

[reef smelt]

it didn't from the start

[karmic crown]

i am talking about 1.6.1 and 1.7.0

[ornate elk]

Here do this:

[reef smelt]

hmm well then I'll try

[faint shuttle]

thanks

[reef smelt]

yes I don't need autolaunch

[karmic crown]

i told you 🙂

[reef smelt]

I wanna generate higher images but my gpu can't do medvram on 1024x1024

[karmic crown]

@ornate elk he has rtx 2060, any advice? (i am pinging you since you watched this thread a minute ago 🙃)

[faint shuttle]

im getting a new error, when i try to generate images, Torch not compiled with CUDA enabled

but i have an amd card, why

[reef smelt]

Is there a way for me to get stable diffusion to prioritize using ram over vram?

[karmic crown]

i guess you can use --skip-torch-cuda-test --no-half --use-cpu all --opt-sub-quad-attention but that will be painfully slow (i am not sure if you need the first one, i had to use it since i do not have any gpu on my linux server)

[reef smelt]

can you run gpu less??

lol

[karmic crown]

10 cores and 60GB RAM with SSD

[reef smelt]

is a gpu just a smaller computer?

[karmic crown]

GPU = graphics card 🙂

that is a linux server

and this is the timing:
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Size: 512x512
Time taken: 2 min. 53.9 sec.

on your pc, you can expect like 5 min for this 🙂

but you can try

[ornate elk]

Only use --medvram --xformers --no-half-vae

[reef smelt]

hmm

[kindred tinsel]

@karmic crown if you are around, do you know why the efficiency nodes were giving me problems on comfy ui when I had a baked in vae?

Should I use normal loaders when it comes to baked in vae models or should I extend the model cable towards the section where it says vae in the other nodes

[ornate elk]

You slowed yourself down with --lowvram

Use upscaling instead of native highres

[reef smelt]

upscaling doesn't use vram?

[ornate elk]

It does

[kindred tinsel]

But lowvram helps stop out of memory errors if you have 8gb or less

[reef smelt]

Well I do have 8gb

6gB*

[ornate elk]

--medvram works perfect for 6gb

[kindred tinsel]

Is it for webui?

[reef smelt]

I mean is there a difference between putting the size to 1024x1024 and 512x512 and using highres?

or what it was called

[ornate elk]

Depends on the model you use.
1.5 models are trained on 512x512
So 1024x1024 would look deformed anyway
Upscaling helps to get higher res ans better quality

[kindred tinsel]

For that try maybe 600*900 upscale of 1.5 then extra upscale

[reef smelt]

Hmm I suppose, but like the hi-res option, is it just upscaling similar to extra?

hires.fix

yes

[ornate elk]

No hires fix is more advanced than extra and can even add details if using latent upscalers

[reef smelt]

Hmm okay

[kindred tinsel]

I would recommend comfy for upscaling. It uses more resources but it’s quite more reliable, you can do two up scales in a row and finish up with extras tab too

Was able to generate a 4096*8096 image yesterday

[radiant stratus]

I was actually just about to ask a question about upscaling

[kindred tinsel]

That is comfy UI, faster, less ram intensive, more GPU intensive, I got temps of up to 91 degrees on constant workload but it’s faster, cleaner. And more stable than the web ui.

Although it’s hard to learn and master, even the image I uploaded is simple in terms of what you can do with it

[radiant stratus]

I upscaled this https://media.moddb.com/cache/images/groups/1/3/2715/thumb_620x2000/del_1024_768.jpg (there's actually a better version but give me a sec to find it) with playground on someone's recomendation and the result was pretty good but there's some spots where the upscaling was seeing noise and it took it as distortions.

https://i.imgur.com/Ac9EebP.jpgthis was the result

[reef smelt]

finding the source is better than upscaling?

or is it just a test

[radiant stratus]

so when I used playground I had actually found the source image that didn't go through

[reef smelt]

hmm oay well I suppose

[radiant stratus]

https://tenor.com/view/king-of-the-hill-jpeg-do-i-look-like-i-know-what-a-jpeg-is-distort-koth-gif-27539179

but the true source image is the same resolution

of 1024 by 768

[kindred tinsel]

Well you could use comfy if the image was generated to upscale the latent image directly after generation, kind of like hires in web uou

[reef smelt]

well I don't think going over 1.5k is gonna make a difference

[kindred tinsel]

In webui I would suggest trying upscale of 1.5 depending on your vram if the image is 512*512 you could probably handle 2x but I wouldn’t bank on it

[ornate elk]

--opt-sdp-attention does still work but xformers uses less vram

[kindred tinsel]

A double scaling actually help, in the webui your hires fix is basically just recreating the same ring with noise and changing it ever so slightly, in the comfy ui I can send negative and positive prompts as the image is about to enter upscaling, fix hands, fix hairs, change artstyle

[karmic crown]

@radiant stratus when original picture is resized, many information that original picture has are lost
when ai upscale a picture, it tries with best guess how it should look

clear enough?

imagine this situation
if i tell you a story, and you forget some details
your version of the story will not be the same as mine, even if you guess most forgotten things almost perfectly

it is similar with upscaling the picture that is a smaller version of original picture

[radiant stratus]

I understand that the AI is trying to reintepret the image to fill in the data that it thinks is "lost" (was never there in the first place), I am just inexperienced in how to best go about getting the AI to do this

[kindred tinsel]

Wait but if the image is already done it would be better to pass it through something like 4x ultrasharp which just sharpens the drawing style and doesn’t change the details in it

[radiant stratus]

and how to fix spots where it improperly interpreted the details there as being meaningful and not noise

[karmic crown]

i guess with inpaint

at least i would do that in photoshop 🙂

[kindred tinsel]

Inpaint would be better, far slower but better overall,

[radiant stratus]

such as this area

[karmic crown]

i would use inpaint

[kindred tinsel]

Yeah impaint trying to describe that in a prompt would be too hard, rather focus on what you want to see and only mask the tower or the block

[kindred tinsel]

Bumping this so Viking can see it and doesn’t forget about me

[radiant stratus]

so to be clear the only AI generation tool I've actually used thus far have been playground and a couple of very simple web demonstrators.

[karmic crown]

i used comfy twice, do not expect too much from me 🤷‍♂️

a1111 on macs is my thing 🙃

[kindred tinsel]

I

Damn

I just don’t wanna have to revert back to god awful slow webui

That thing has too slow, memory ineficient and it doesn’t even use all of my resources like comfy does

[radiant stratus]

so what is the best way to do this, should I upscale it then inpaint with masking to prevent it from editing the buildings with a prompt of blue sky with scattered grey clouds?
and what actual tools should i use for this.

[karmic crown]

on all my macs it use all of the resources

see?

95+ gpu

[kindred tinsel]

In my desktop it used from 50 to 80 % of GPU and literally in intervals so like it uses it only when it wants to do an iteration, hence why it took so long to generate a single image

The time the webui took for a 600900 image on this GPU
Was the same time it took for comfy to generate 9001344 first upscale

[kindred tinsel]

You should be very specific with your prompt if you want the sky a certain way before you produce the image, for instance in the generation put that in. For impaint you have to be specific about the region you are going to impaint since it will generate (depending on your settings) latent noise to fill the spot up

I would recommend looking up some tutorials on YouTube to see how it is done as I myself don’t do it too much and mostly overprompt haha

[ornate elk]

What's your GPU and do you use xformers?
Also disable image preview is helpful

[kindred tinsel]

It’s a Radeon GPU, Rx 5500. I have formers show up as n/a in my web ui

[ornate elk]

Okay, yea with that card it's slow :/ on Linux it would be better.

Xformers is sadly not available for amd

[kindred tinsel]

Not neccearily on Linux, on comfy it’s much better either way ;)))). I think it must use some diferent technology or optimization

[karmic crown]

why would image preview make things slower? i am confused

[kindred tinsel]

Generally because it has to render the image with each iteration that passes

You wouldn’t notice in your Mac since the chip is very powerful on its own, but for mid GPUs is very noticeable the effect it has, since removing that unnecessary output makes generation move faster, you just have to know what you are prompting

[karmic crown]

🤣

[kindred tinsel]

So that’s every two iterations I guess, still

If you do 30 if will generate 15 previous

[karmic crown]

but since i using lcm that is about 4-5 images

[radiant stratus]

is there a way to check what an AI sees an area as so I can find keywords to add as negative prompts for things that it's improperly interpereting as being there?

[kindred tinsel]

I didn’t know there was an option to do such a thing but I will change that and medvram, give it full power for single image generation and create 30 batches of 1

[karmic crown]

i will do the testing with 2, 5 and 10, and see if there is a noticable difference

[kindred tinsel]

Well ok the webui you can select latent noise or latent nothing or fill. If you go there you will find that these options either (generate stuff up) or (use the stuff there as a base). You gotta know what you want rather than what the Ai see, if you want for example a late black control tower

Then prompt that and see what it throws at you, Sd1.5 understand some basic human language in normal English so you don’t have to write keywords much more than explain your idea

[ornate elk]

It uses a lot of resources to generate these preview images + showing it in the browser

[ornate elk]

Set it to 6 if you use 30 steps. For example. Or even higher if you want

[karmic crown]

i do not expect any dramatic time save if i revert it to 10 on my m1, but as i wrote i will test with 2, 5 and 10, and see what happes

[kindred tinsel]

It’s an M1

[karmic crown]

i usually use lcm with 10 steps 🙂

[kindred tinsel]

What do you think will happen? :( it’s a powerful chip

Let me test it out with this 2019 piece of antiquity in half an hour and compare results

But let’s keep it at normal sampler, dpm ++ 2M I will use for all the tests one image, 600*900 (average res for my GPU)

[karmic crown]

use 512x512 for 1.5 test

[kindred tinsel]

512*512? Ewww small but okay

[karmic crown]

that is default for 1.5

[kindred tinsel]

Sure sure I’ll use that

[ornate elk]

Yea on Linux it can use rocm and comfyui uses other performance enhancements

[kindred tinsel]

It doesn’t support my graphics card tho, I looked the page up and my model doesn’t show up all I see is a 5500 XM so I’m like , at the end of the line for this GPU

That I bought 6 months ago

[ornate elk]

Yea rdna 1 is done. It won't benefit from anything anymore.

It can work with rocm but only on torch 13, if supported

[karmic crown]

preview set to 2

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3841688468, Size: 512x512, Model hash: ec41bd2a82, Model: photon_v1, Version: v1.7.0

Time taken: 35.3 sec.

[steep ivy]

That's my cmd output when I start the webuiuser.bat @ornate elk

[ornate elk]

Okay looks good. And with which settings do you get the low speed?

[karmic crown]

with 5
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1403469286, Size: 512x512, Model hash: ec41bd2a82, Model: photon_v1, Version: v1.7.0

Time taken: 41.2 sec.

hahaha

with 10

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1816563794, Size: 512x512, Model hash: ec41bd2a82, Model: photon_v1, Version: v1.7.0

Time taken: 43.9 sec.

[steep ivy]

Default, haven't changed anything there. I am using the dreamshaper XL model and xformers (someone suggested I do) so my pytorch is in version ~1.3

[kindred tinsel]

So in your computer the relationship is inversely proportional @karmic crown ?

You increase are jump before image and it increases your render time

[radiant stratus]

So I ran
git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git but I think I made the mistake of doing so before adding python to path, where does the folder that command creates live so I can delete it and rerun it?

(I think that was my mistake because webui-user.bat isn't bringing me to anything)

alternatively the issue may be that I'm an AMD gpu

[karmic crown]

something i eating my cpu, it is is on 44%, i guess i have too many opened windows and tabs in brave

git is not realated to python

[kindred tinsel]

In the webui I wiuls recommend pulling the fork specially made for AMD

There is a link in there for it i think

[karmic crown]

git cloned a1111 to folder from where you run it

[radiant stratus]

I was looking here and I didn't see an AMD specific link

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

[kindred tinsel]

Let me get the fork for u

[radiant stratus]

where is my old copy to delete it?

[kindred tinsel]

https://github.com/lshqqytiger/stable-diffusion-webui-directml

[kindred tinsel]

That should be in the folder you git pulled

[karmic crown]

try to search stable-diffusion-webui from windows explorer

[kindred tinsel]

Okay first result :
Preview at 2
DPM++ 2M Karras took for a 512 x 512 image 1 minute with 23 seconds at 3.66s per iteration

[karmic crown]

"Windows Explorer has a new name ... It's called File Explorer in Windows 10" 🤣

way to go microsoft 🙂

[kindred tinsel]

Don’t hate on daddy Microsoft

Same image with live preview set at 5 took 1.14 seconds with a 3.53 s per iteration average

[karmic crown]

i do not hate ms, i love what they are doing with open source, i just do not like windows 🙂

i love vscode

[kindred tinsel]

Why tho?

You love macOS or whatever it is called?

[fair oxide]

hey @karmic crown what sd 1.5 anime model do you usually use that is good in following prompts?

[karmic crown]

https://civitai.com/models/6755/cetus-mix is fine

[kindred tinsel]

Thats such a hard question, they do and do not at the the same time lmao

[kindred tinsel]

Now you are just flexing

I hit a solid ceiling, if I have live previews set at every 10 step it defaults to 1 min 14 seconds to generate the image

I’ll try without preview

[ornate elk]

Then your pytorch version is old.
Can you show the bottom of your webui?

[faint shuttle]

is there any way to make comfyui running on linux using Rocm 4.0, as I believe this is the latest supported version for my rx 580 ??

i saw a tutorial on how to install it, but the guy was using rocm 5.7

[steep ivy]

seems like it's 2+ (I assumed it wasn't because I heard xformers doesn't support that version so in it's installation it deletes the new version and installs an older one)

[karmic crown]

preview 1
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 180721704, Size: 512x512, Model hash: ec41bd2a82, Model: photon_v1, Version: v1.7.0
Time taken: 36.3 sec.

preview 10
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 2539402334, Size: 512x512, Model hash: ec41bd2a82, Model: photon_v1, Version: v1.7.0
Time taken: 36.6 sec.

basically no difference

[ornate elk]

Looks good 👍

[steep ivy]

Just reinstalled both, it deleted xformers cuz they're incompatible and now the Loras and embeddings don't show up, should I reinstall xformers and torch in lower version?

[ornate elk]

No, the loras show up if you switch back to an 1.5 model and then restart

Xformers is compatible with torch so dont delete it xD

[crimson leaf]

is it easy to install comfyUI?

[ornate elk]

Its downloading a zip file and running a .bat

[crimson leaf]

on linux .sh?

[steep ivy]

Sorry for the lack of knowledge, how do I do that?

[ornate elk]

Select an other Checkpoint that isn't an sdxl one then close the cmd and restart

[steep ivy]

wait so the XL stands for stable diffusion XL and not a bigger model?

[ornate elk]

Nope for Linux check this:
https://github.com/comfyanonymous/ComfyUI#manual-install-windows-linux

[ornate elk]

Yes

Sdxl models are bigger and slower than 1.5

[steep ivy]

well I assumed it would be slower but not ~5 it/s to 1

[ornate elk]

You should be able to generate a sdxl 1024x1024, 20 steps in nearly under a minute

[karmic crown]

i am getting almost the same times with any setting of preview, thats why i wrote that i am confused, but ok, it make sense that it will slow down things on some machines, i will keep mine on 2

[steep ivy]

Well that's not really fast or what I need (512x512 is alright as well) so i'll stick to 1.5

[ornate elk]

Mac is different than dedicated GPUs, so I'll gues your not affected

[crimson leaf]

i smell many errors already, alr thanks

[karmic crown]

i will test on 1660 when i find time

[ornate elk]

Yea 1.5 and upscaling will work better for you

[fair oxide]

any thoughts on MeinaMix?

[steep ivy]

lets say I want to make things faster, is the bottleneck my GPU (2080 super) or something else (i5-8400, 16GB DDR4)?

[karmic crown]

your gpu, cpu is not too important

cpu is important if you are using cpu only, or have some "cpu+gpu" combination, like my m1 mac

with another gpu, you could have much faster generation

for example 3070

[steep ivy]

What's the most relevant feature of a GPU? just his generation or something like VRAM or Cude cores..

[karmic crown]

"ghost in a shell" and "death note" are my kind of anime, so - not really 🤷‍♂️

[karmic crown]

vram for sure, but more cuda cores can help too

[kindred tinsel]

did I hear ghost in the shell 🤤 ?

[karmic crown]

if you plan to buy a new card set a budget, find what you can buy for that money and simply ask what is the best card for sd of those you can afford

[karmic crown]

then you are probably born after it was released 🤣

[kindred tinsel]

you meanie

but yes

I liked it tho :cc

[karmic crown]

https://www.imdb.com/title/tt0113568/

1995

i have it on vcd, i cant remember what resolution it was, but i guess 640x480 🤣

[kindred tinsel]

man I was born 9 years later chillll, I saw it and liked it thats all

[karmic crown]

https://www.andreszsogon.com/fix-stable-diffusion-automatic1111-webui-on-geforce-gtx-16xx/

set PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:512
set COMMANDLINE_ARGS=--xformers --no-half --opt-sdp-attention --medvram

With those parameters you can generate a 512×512 image in about 15 seconds using the DPM++ 2M Karras model.

Total progress: 100%|██████████████████████| 25/25 [00:16<00:00, 1.71it/s]
Time taken: 15.81s
Torch active/reserved: 4010/5244 MiB, Sys VRAM: 6144/6144 MiB (100.0%)

it takes much more on my son's PC with the current settings, i need to try this as soon as i come home

[radiant stratus]

damnit

[fair oxide]

ok. i tested the meina mix v11, didnt seem like it follows prompt too well

[radiant stratus]

this was with scaling set to 1 (so no upscaling), just trying to inpaint

do I just straight up not have enough memory with a RX5700?

[karmic crown]

it doesnt work like that

you need to collapse that to disable it

1 will still do upscaling (but to same resolution)

also if you put 0 for steps it will use the same number of steps you had for image generation

for example if you had 20 steps, 0 would add 20 steps for hiresfix

[ornate elk]

What was your settings

For inpainting you have to select "inpaint only masked"

[karmic crown]

i am too tired, so i am sorry if i misunderstood something

[polar cove]

What cross attention optimizer should I use for an A6000? I am running on runpod and get about 1 iteration/s with an a6000 for SDXL

[ornate elk]

--xformers if you can, if not then --op-sdp-attention

[white thorn]

im sorry for that really late answer I wasnt home. im using only -opt-sub-quad-attention --api --theme dark . Only way to make upscaling is u sing medvram-sdxl

[kindred tinsel]

Sad I don’t have xformers

[kindred tinsel]

Wait? I iteration per second on an A6000?!!! Yeah I’m not buying that card then lmao

[ornate elk]

you should use --medvram --opt-sub-quad-attention --opt-split-attention --no-half-vae --upcast-sampling

[mighty rivet]

for automatic1111 i cant seem to find the API parameter for this in IMG2IMG, inpainting

[mighty rivet]

correct me if im wrong, is it "denoising_strength"

[ornate elk]

nope

[mighty rivet]

oh then could u help me with it?

[onyx ledge]

For whatever reason I'm getting this error trying to boot Stable Diffusion now.

[ornate elk]

your torch version is to old for xformers.
whats your webui-version?
start the webui-user.bat and make a screenshot of the cmd#

[ornate elk]

the paramter would be mask i guess

[mighty rivet]

the mask i beleive is the mask image

the string parameter is the encoded image

[onyx ledge]

Ah, thanks. I did have to do some reinstallations lately. Here's the cmd.

[ornate elk]

ohh okay

[ornate elk]

you need to delete the venv folder and then rerun the webui-user.bat
if that doesnt work,
open up a cmd and type: pip cache purge
then delete the venv folder again

[onyx ledge]

Hmmh. Here's what the webui-user.bat cmd says upon doing all that.

[ornate elk]

oh your python version is to high

if you dont need python 3.12 for any other stuff, you should uninstall it

SD works only with python 3.10.6 up to 3.10.13, latest version on windows with installer is 3.10.11
After uninstalling Python 3.12, delete the venv folder before launching the webui-user.bat

[onyx ledge]

Ah alright, I'll look into that. Thanks.

[crimson leaf]

is comfyUI automatically in --medvram mode? I heard this somewhere, but idk if that's right

also, is xformers included in there?

[tawdry wind]

I have the same error

[tawdry wind]

I have exactly the version you mentioned. But still same error.

[ornate elk]

in your screenshot its 3.12

delete the venv folder after uninstalling the 3.12 python

[wind sun]

does somebody have any idea ?

[ornate elk]

yes, i had the same problem an hour ago:
check this:
https://github.com/Gourieff/sd-webui-reactor#installation

oh your using roop

here is the fix for roop:

[wind sun]

welp xD

that is the aftermath of pip install insightface==0.7.3

[ornate elk]

xD

rip

i would recommend using Reactor and not Roop, as roop didnt saw an update since a few months

[wind sun]

i downloaded it from the extension tab in the stable diffusion but after this i got the restart msg, restarted it from the browser with apply and restart UI but nothing happend and my console is now getting spamed with root msgs

[ornate elk]

roop or reactor?

[wind sun]

reactor

[ornate elk]

because for reactor you need to uninstall roop and dlete the venv folder

[wind sun]

oh

[jaunty sphinx]

I'm not sure how it works, but for openvino it seems to work well for stable diffusion models 1.5 and bugs out for sdxl and stable diffusion 2.1

[wind sun]

how do i uninstall it tho 😄

[ornate elk]

delete the roop folder in the extensions folder

[jaunty sphinx]

Fooocus collab works with sdxl models, but not 1.5 and 2.1 for some reason

[ornate elk]

@wind sun delete the reacotor folder too

then read this:
https://github.com/Gourieff/sd-webui-reactor#installation
Read 3. and 4. carefully

[jaunty sphinx]

I think automatic1111 can do all models just like comfyui

and maybe sd.next

rip open vino d;

[wind sun]

welp

[ornate elk]

did you installed the Visual studio stuff ?

[wind sun]

do i have to delete also the controlnet stuff

[ornate elk]

nope

[wind sun]

i have it from before

when i was installing the auto1111 from other tutorials

[ornate elk]

lol which guide recommended vs studio for that xD

open up a cmd and type
pip cache purge
then delete the venv folder again

[karmic crown]

i would always use venv\bin\pip cache purge
since in case there is more than one python, it might clean the wrong cache
but in "normal" cases, i agree with you 🙂

i have at least 5 versions on python installed 🙂

[wind sun]

im just deleted my now using vsc and downloaded the linked one

doing all steps like in the link just to be sure there are no freaking errors this time

[ornate elk]

also let it load until it says to restart in the cmd, after clicking install in extensions

[wind sun]

the sebastian kamph one xD

idk to be honest, i buyed the course from one site for 250$ or smth like this tbh i dont really remember and it was zip file with other folders in there with tutorials for everything

[reef smelt]

I'm trying to use Composable Lora and I have Composable LoRA with step enabled
My prompt is ((best quality)), ((masterpiece)), (detailed), high quality, 2girls, <lora:rem_v1:0.6>#cmd\(cooldown\(0.6\)\), rem, <lora:Tohru-10:0.6>#cmd\(cooldown\(0.6\)\), tohru, highly detailed skin, absurdres, HDR but I don't get any result
cmd output is normal like it worked, have tried reloading webui and changing the prompt, It works when I only have one #cmd\(cooldown\(0.6\)\) but not when I have two

it strangely worked before

my goal is to use two loras at the same time

[karmic crown]

you can use more loras at the same time, that is not an issue, but i have no idea what this is #cmd(cooldown(0.6))

[reef smelt]

Well the result is an issue

putting two loras fries the result

[karmic crown]

((best quality)), ((masterpiece)), (detailed), high quality, 2girls, lora:rem_v1:0.6, rem,
lora:Tohru-10:0.6, tohru, highly detailed skin, absurdres, HDR

this should work without a problem

but i have no idea what that thing is

[ornate elk]

it will mix up the two lora characters

[reef smelt]

not just that

[ornate elk]

im using Tiled DIffusion Prompt Control for getting 2 chars with Loras, in one image

[reef smelt]

I guess it did work yet not really

[wind sun]

hm my pc is dying, even my music is lagging xD

[reef smelt]

people say latent couple is good with composable lora

[ornate elk]

let it run xD, it will stop when its done

[ornate elk]

i tried it and havent much success with it

[karmic crown]

i have never tried to generate more than one character and i never used more than one "face" lora, but i mixed 2 style loras, locon, lcm... so you can use more than 1 lora is bascially correct answer

[wind sun]

we just hope my screen dosent turn blue xD

[karmic crown]

but i guess your aproach is the best for two different characters

[reef smelt]

hmm

[karmic crown]

its black with white letters on mac 🙂

[ornate elk]

yes, but it was never the question if he can use two loras at the same time, you can use as much as your vram can handle or the image xD
but generating 2 chars with 2 differents looks in the same image with 2 loras will result in a mix of them. like in the top image

latent couple is an extension that tries to fix that issue

[karmic crown]

i am still waiting for an explanation what #cmd(cooldown(0.6)) is

🤷🏻‍♂️

[reef smelt]

uhh

[ornate elk]

xD yea that makes no sense for me too. never saw it

[wind sun]

so my auto1111 is reinstalling again from scratch now right, cuz its downloading a lot of stuff again

nvm

[karmic crown]

i thought that civitai browser plus was my favorite extension, i just changed my mind it is model organizer, this thing is a life saver

[reef smelt]

this, lets you alter the weight through out the generation

[ornate elk]

its installing the venv stuff again

[wind sun]

right, its done and now all i have to do is install reactor again then restart and hope for the best

[ornate elk]

yep, restart only if it says so in the cmd

[karmic crown]

please send me where you saw that, since that \) doesnt look good to me

[wind sun]

yup, doing it

finally

i love you ❤️

[ornate elk]

no problem 🙂 have fun

[reef smelt]

https://www.youtube.com/watch?v=kfoA0xWv-0Y&t=320s

learned that I need [] on the outside

[ornate elk]

and you should use xformers btw, whats your gpu?

[reef smelt]

cs1o are commas necessary or does spaces do the same thing?

[ornate elk]

commas should be used

[reef smelt]

o okay

[karmic crown]

this make sense, since all brackets are escaped

[wind sun]

oh one more thing, where is the enable button

[reef smelt]

I didn't notice thr [] so probably that

[wind sun]

gtx 3050

[ornate elk]

with 4 or 8gb vram?

[wind sun]

em one sec

[karmic crown]

be sure to have both [ and ] around and that each bracket is escaped, so \( and \)

[ornate elk]

the button should be inside the dropdown

[reef smelt]

hmm

[wind sun]

rtx

[karmic crown]

but discord like to "eat" \ sometimes

[wind sun]

but idk about the gb

[reef smelt]

((best quality)), ((masterpiece)), (detailed), high quality, 2girls, standing up, [<lora:rem_v1:0.6>#cmd\(cooldown\(0.5\)\)], rem, [<lora:Tohru-10:0.6>#cmd\(cooldown\(0.5\)\)], tohru"

[wind sun]

i will search the box rn xd

[karmic crown]

that looks better

[reef smelt]

it might not be updated I think

[ornate elk]

You can check in the taskmanager

Under performance, gpu

[wind sun]

8

[ornate elk]

okay to increase the performance for SD you have to edit the webui-user.bat
at the line Commandline_ARGS=
you add: --xformers --medvram-sdxl --no-half-vae
that will save you a lof of vram, and speeds up the image generation a lot

also needed for Controlnet and upscaling

[wind sun]

will it effect the quality by any chance i mean lower it

[ornate elk]

nope

not at all, its only for performance tuning

[wind sun]

im even thing of clocking my gpu cuz of lack of fps in the games i play but then im thinking there will apear a bottleneck

[reef smelt]

what's the cpu?

[reef smelt]

you can try https://pc-builds.com/bottleneck-calculator/

tho..

hmm

can't test for overclocking...

doesn't really help unfortunately

[wind sun]

it appears as 11% bottleneck but im not sure if this is good or bad thing xD

[reef smelt]

lol think it's good

[wind sun]

im not into overclocking xd

[reef smelt]

duno if it's possible to overclock? nvidia made a special brand that can overclock even though all gpu's have the ability

you can give it more power I think

mayb

[wind sun]

one sec

ooooh i have a clock app for cpu not gpu

idk if i can clock the gpu with nvidia

[ornate elk]

overclocking the gpu can break SD

[wind sun]

that was hard but perfect timing no

[uneven delta]

can i dowload stable diffusion to run in my pc

?

[ornate elk]

yes, whats your gpu?

[uneven delta]

nvidia 3070ti rtx

[ornate elk]

make sure to select graphic intense task in this calculator

[wind sun]

like this ?

[ornate elk]

that will work good, ill give you my install guide

[uneven delta]

thanks

[wind sun]

let me guess, i have to restart it

[ornate elk]

Here is a quick Guide to install Automatic1111 Webui for Nvidia GPUs (Stable Diffusion)

You need to install Git 64bit: https://git-scm.com/download/win
and Python 3.10.11, 64bit: (any python above or 3.10.6) No 3.11, 3.12 https://www.python.org/downloads/release/python-31011/
And check "add python to path" when installing Python.

1. Make a new Folder on your drive (not on Desktop, downloads, documents, Programms, Onedrive) and name it Ai for example:
   C:\Ai\
2. You go into the folder you created in this case Ai, then click in the File Explorer bar (not searchbar) and type cmd then press enter.
3. Then you copy and paste this command:
   git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
4. Press enter and after its done you can close the cmd and launch the webui-user.bat inside the stable-diffusion-webui folder.
   After the Installation that can take a while. Youll get an URL http://127.0.0.1:7860
   Thats the webui you open in Browser.

[wind sun]

now is perfect

[ornate elk]

no need for OC

[wind sun]

but if i want to upgrade my pc, what do i buy better gpu or cpu ?

[ornate elk]

you input the new gpu in the calculator again and check. a 5-8% difference is okay for the cpu

you mostly want a better gpu

[karmic crown]

i gave you a complete script for nvidia, that even sets --xformers in webui-user.bat
just add last sentence and will do the same thing with two clicks 🤣

[ornate elk]

your right xD but i forgot to save it 🥲

[wind sun]

the stable deffusion is working with gpu so its better for me to upgrade my gpu not only for games but for the generating too righ

[ornate elk]

yes, more vram is the key for SD

[karmic crown]

#🤝｜tech-support message

save it 🙂

[wind sun]

im strongly trying to resist the urge of buying a nasa pc with 40's series gpu but im almost to the M4 GTS so its hard xD

[wind sun]

can't find it

[uneven delta]

which one is .bat?

[wind sun]

the one u selected

[uneven delta]

a ok xd

[wind sun]

right ?

[uneven delta]

yes its dowloaing now

[karmic crown]

@ornate elk i have an idea for my nvidia guide 🤣

[ornate elk]

is your webui updated?

[karmic crown]

1. open notepad
2. copy paste this script

@echo off
winget install -e --id Python.Python.3.10
winget install -e --id Git.Git
winget install -e --id mbuilov.sed
mkdir c:\ai\
cd c:\ai\
git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui
cd stable-diffusion-webui
sed -i "s/set COMMANDLINE_ARGS=/set COMMANDLINE_ARGS=--xformers/" webui-user.bat
call webui-user.bat

3. save as install.bat and run it
4. ask @ornate elk for aditional optimizations for your nVidia

🤣

[ornate elk]

haha 😄 perfect

[wind sun]

the pip need a update

[karmic crown]

not really

anything above 20 is fine

[uneven delta]

is it possible to make 3d from text?

[ornate elk]

thats okay

[wind sun]

then whats the problem

[ornate elk]

did you tried if it works?

also check the bottom of your webui for the versions

[ornate elk]

not with the local webui, stability release txt23d a week ago so we need to wait

[void kelp]

Hi guys, I'm new to the server and I need a lot of help. Who can I ask to?

[wind sun]

the version is fine i think cuz there are no errors but its not generating anything so imma go to sleep now and maybe later i will get on again and try and fix this shi

[ornate elk]

For technical questions you can ask here

[karmic crown]

for general questions https://discord.com/channels/1002292111942635562/1002292112739549196 or https://discord.com/channels/1002292111942635562/1004159122335354970

for prompting help https://discord.com/channels/1002292111942635562/1011743094309396631

[void kelp]

Thank you! I've been having this problem for days, but I just can't seem to fix it. On kohya, trying to train the lora, if I enter a custom checkpoint, it always gives me the following error. Are there solutions? Thanks in advance>> (I'm executing this on colab pro)
[Dataset 0]
loading image sizes.
100% 25/25 [00:00<00:00, 580.90it/s]
make buckets
number of images (including repeats) / 各bucketの画像枚数（繰り返し回数を含む）
bucket 0: resolution (512, 512), count: 500
mean ar error (without repeats): 0.0
noise_offset is set to 0.0357 / noise_offsetが0.0357に設定されました
preparing accelerator
loading model for process 0/1
load StableDiffusion checkpoint: /content/drive/MyDrive/CheckPoint/realisticStockPhoto_v10.safetensors
building U-Net
loading U-Net from checkpoint
U-Net: <All keys matched successfully>
building text encoders
Traceback (most recent call last):
File "/usr/local/bin/accelerate", line 8, in <module>
sys.exit(main())
File "/usr/local/lib/python3.10/dist-packages/accelerate/commands/accelerate_cli.py", line 47, in main
args.func(args)
File "/usr/local/lib/python3.10/dist-packages/accelerate/commands/launch.py", line 1017, in launch_command
simple_launcher(args)
File "/usr/local/lib/python3.10/dist-packages/accelerate/commands/launch.py", line 637, in simple_launcher
raise subprocess.CalledProcessError(returncode=process.returncode, cmd=cmd)

subprocess.CalledProcessError: Command '['/usr/bin/python3', './sdxl_train_network.py', '--enable_bucket', '--pretrained_model_name_or_path=/content/drive/MyDrive/CheckPoint/realisticStockPhoto_v10.safetensors', '--train_data_dir=/content/drive/MyDrive/Elon_Musk/image', '--resolution=512,512', '--output_dir=/content/drive/MyDrive/Elon_Musk/model', '--logging_dir=/content/drive/MyDrive/Elon_Musk/log', '--network_alpha=1', '--save_model_as=safetensors', '--network_module=networks.lora', '--text_encoder_lr=5e-05', '--unet_lr=0.0001', '--network_dim=8', '--output_name=Elon Musk', '--lr_scheduler_num_cycles=1', '--no_half_vae', '--learning_rate=0.0001', '--lr_scheduler=cosine', '--lr_warmup_steps=50', '--train_batch_size=1', '--max_train_steps=500', '--save_every_n_epochs=1', '--mixed_precision=fp16', '--save_precision=fp16', '--cache_latents', '--optimizer_type=AdamW8bit', '--max_data_loader_n_workers=0', '--bucket_reso_steps=64', '--mem_eff_attn', '--xformers']' died with <Signals.SIGKILL: 9>.

[karmic crown]

for colab is the best to ask the author directly, since most of use here use local installs like a1111 or comfy or both

[void kelp]

In your opinion, what error is this? And then should I ask colab or a1111?

[uneven delta]

is there a way to improve the quality of the images?

[ornate elk]

use community made models from civitai.com
use good prompts, and negatives
then later upscaling can help a lot

[uneven delta]

thx

[void kelp]

Sorry if I'm insistent, but who should I contact for this problem? I'm really desperate

[karmic crown]

as i already wrote, the person who wrote a script for colab

where did you get it?

it is not that we do not want to help, we just do not use it, so we have no idea how to help you

i never saw anyone who knows how to answer this kind of question in the last 30+ days

[uneven delta]

what is CFG scale for

[karmic crown]

long answer
https://decentralizedcreator.com/cfg-scale-in-stable-diffusion-and-how-to-use-it/

short answer
The CFG scale adjusts how much the image looks closer to the prompt and/ or input image.

[void kelp]

Thanks so much for the reply first of all. I got the script from camenduru and don't worry, you are very kind!

[uneven delta]

how do i dowload civitai models to the stable diffusion ia generator thing

[ornate elk]

download the .safetensor file (2gb or more) then put that into the models/stable-diffusion folder

[karmic crown]

or install civitai browser plus plugin and it will download it and put it in a right folder for you

[urban merlin]

does anyone here have any experience with OpenVINO stable diffusion?

[livid jetty]

"LayerNormKernelImpl" not implemented for 'Half' i keep getting this message

[karmic crown]

you probably need to add --no-half to your command line params, what is your gpu?

[uneven delta]

i dowload the .safetensor file and i put in the stable diffusion folder but when i open the url i dont see the model

showing up

[karmic crown]

please show me where you put that file

and check the filesize

[karmic crown]

looks ok

[livid jetty]

AMD Ryzen 5 7600 and where is the command line param? Im sorry its my first time using stable diffusion. and the video guide guy wasn,t as thorough as i hoped

[karmic crown]

click on refresh button here

can you see it after refresh?

[uneven delta]

yes yes

thx

[karmic crown]

that is CPU (processor), what is you GPU (graphics card)?

[livid jetty]

oh sorry

[ornate elk]

Edit your webui-user.bat that it looks like this
#🤝｜tech-support message
To use sdxl models

[wind sun]

the version is 1.6.1 of webui bat

[uneven delta]

?

[karmic crown]

go to your sd webui folder and type
git pull

that will update to 1.7.0

[livid jetty]

AMD Radeon RX 7900 XT

[wind sun]

sd webui folder ?

where is it

[karmic crown]

i recommend reinstalling following this guide #🤝｜tech-support message

[livid jetty]

gotchat

im also using some safetensors

just fyi if that is important

[karmic crown]

same folder where you found webui-user.bat

rename the folder

[uneven delta]

why do edge images look way better than mine?

[karmic crown]

so you can move them to new installation

[livid jetty]

al;so should i delete the stable diffusion file?

[karmic crown]

for now rename the folder, so you can take things you need from there

if your net is not fast enough

[hallow latch]

Now I am curious, I just added a PNG image and it's only loading when doing Interrogate CLIP.. Its currently on 300/0.25 sec and the console only says ( image ) ?? Hmm

[livid jetty]

oh its stable diffusion webui

[wind sun]

whats the name of the folder or the file im looking for

[karmic crown]

where is the file that you run to open a1111?

the folder where you installed your a1111

[wind sun]

oh

sd is short for stable diffusion

yup im dumb, ignor this

[karmic crown]

i should be less lazy i guess 🙂

[livid jetty]

what is the last command for using cmd in file explorer, i wanna make sure its everything

[wind sun]

u r fine, im lacking a sleep a lot

[karmic crown]

and write stable-diffusion-webui 🙂

[karmic crown]

in search type cmd and click it when you see it

[livid jetty]

ok

[karmic crown]

"Press Windows+R to open "Run" box. Type "cmd" and then click "OK" to open a regular Command Prompt.

you need normal one

[uneven delta]

doesnt generating images should consume gpu like i can see and my gpu is at 6% usage

[wind sun]

now im 1.7.0, hoping for it to work

[ornate elk]

For the best performance you need to edit your webui-user.bat

[uneven delta]

how

[karmic crown]

update your plugins now

[uneven delta]

aaaaa

[wind sun]

dont be mad but how exactly

[ornate elk]

Right click , edit

[karmic crown]

that is too low, you need better command line params

[karmic crown]

check for updates

i have to find images on the net, since my ui looks diffrent 🙂

[wind sun]

i ckecked

[karmic crown]

my a1111 looks like this

[wind sun]

yo

[hallow latch]

Why can't I use the Interrogate CLIP or Interrogate DeepBooru? I will only load for me? :/

[karmic crown]

@wind sun https://github.com/lobehub/sd-webui-lobe-theme

[hallow latch]

Ohh cool, how do you install that one?

[karmic crown]

go to extensions, and find lobe theme

[hallow latch]

AssertionError: extension access disabled because of command line flags

Damn

[karmic crown]

also, i recommend adding those in user interface part of the settings

to have this

[hallow latch]

Me or the other guy? 😄

[uneven delta]

sorry what u mean better command line params

[ornate elk]

the ones i linked you, --xformers --medvram-sdxl --no-half-vae

in your webui-user.bat

needed for better performance

[karmic crown]

i was talking in general, for easier access to those commands realated to my ui screenshot, sorry if i confused someone

so it was for @hallow latch and @wind sun 🙂

[wind sun]

yay i face swaped finally

its kinda naked one otherwise i was gonna send it for yall xD

[hallow latch]

Hey @ornate elk the man 😄 Got any idea why I cannot use the Interrogate clip function anymore? 😄

It only loads in enternity 😄

[uneven delta]

like this?

[hallow latch]

Yes

[karmic crown]

but i will never understand why people add space between = and -- 🙂

[uneven delta]

is it ok if it say Style database not found: C:\AI\stable-diffusion-webui\styles.csv

[wind sun]

cuz we are not devs and it seems like it need space xD

[karmic crown]

i am an old grumpy dev 🤣

[wind sun]

you and @ornate elk are the best ones

so whatever you say boss

we all love yall for the help

[uneven delta]

ye

[karmic crown]

both will work (space and no space) i just like properly formated code, but in the end it is the same for you as a user

[ornate elk]

That shouldn't be the case

[wind sun]

so can i ask if u know which is the best lora and checkpoint for realistic girls

[ornate elk]

What does the cmd shows?

[fair oxide]

i have a weird issue. When i was installing locally i installed the wrong version of python. After realising, i uninstalled and got the correct version. But even after adding the new version to my windows PATH, When I open stable diffusion Webui-user the batch file. It opens the CMD and says error the Path for the (old/wrong version) of python doesn't exist.

Even after I put the new python in the scripts folder aswell. The weird part is I can open the Webui (Python File) using Open With and selecting my compatiable Python.

How can I fix?

[ornate elk]

can you switch to an other model?

[ornate elk]

you have to delete the venv folder and relaunch the webui-user.bat,
it will get recreated with the right python version

[ornate elk]

realisticvision is good

[fair oxide]

Just yoink it out and and attempt relaunch?

[wind sun]

i guess civitai

[ornate elk]

Yes

[radiant stratus]

and @karmic crown sorry I went to bed right after this. here is how it was setup

[fair oxide]

Thanks so much ❤️

[hallow latch]

Damn, what a dick 😦

[young bridge]

hey guys

how do I update stable diffusion from 1.6 to the latest version 1.7

[karmic crown]

this looks fine, except for this resize part, but i haven't played much with that
@ornate elk can you check those images?

[ornate elk]

Switch to resize to, then select 512x512
Thats needed for inpainting

[radiant stratus]

hrm but that will lead to scaling the image down and losing information

[ornate elk]

No it wont

[karmic crown]

@radiant stratus he knows better than me 🙂

[radiant stratus]

how so?

[karmic crown]

because you are impaining just a part, not a whole image (i guess)

[ornate elk]

If you inpaint the resolution you set is not the end resolution.
Make sure to select "only mask"

Then the inpainted areas get inpainted with a 512x512 resolution

[karmic crown]

as i thought

[radiant stratus]

ok so just set it to this?

[wind sun]

good night guys, thank you a lot for the help today ! yall real G's

[ornate elk]

Run git pull in a cmd thats in the folder path of stable-diffusion-webui

[young bridge]

i cant use git pull

bcs i got the repo through zip

[ornate elk]

Ohh

Then you need my fix

Give me a sec

[young bridge]

ok

[ornate elk]

Good night !

[karmic crown]

git init is coming 🙂

[ornate elk]

You installed the webui the wrong way,
To fix this go into the stable-diffusion-webui folder and click in the File Explorer bar and type cmd and hit enter.

Then use the following commands:
git init git remote add origin https://github.com/AUTOMATIC1111/stable-diffusion-webui.git git fetch --all git reset --hard origin/master git branch --set-upstream-to=origin/master master git pull
After that relaunch the webui-user.bat

[young bridge]

stable-diffusion-webui 1.6?

[karmic crown]

i was just started to write with the help of his friend "git remote add" 🤷🏻‍♂️

[ornate elk]

You can also rename it if you want to just Stable-diffusion-webui, but do that before the fix

[young bridge]

what does these commands do

[karmic crown]

actually, i should add to my script, since i currently just print error message in that case

[young bridge]

git pull already up to date

that what it says

[karmic crown]

0.0.5 might be released this year after all

[karmic crown]

show me git status please

[young bridge]

how

[karmic crown]

just type
git status in the same folder
and show me the result

[ornate elk]

They connect the folder with the github repo so you can git pull again to update it in the future

[young bridge]

Installing requirements
Launching Web UI with arguments:
No module 'xformers'. Proceeding without it.

1.7.0

why no xformer?

isnt that the command for speeding up

[karmic crown]

that's why i asked you to run git status first

[ornate elk]

You need to edit the webui-user.bat again and there add
--xformers --medvram-sdxl --no-half-vae
To your COMMANDLINE_ARGS

[karmic crown]

to see which files are changed

[young bridge]

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

like that?

[ornate elk]

Then you have the best performance

[young bridge]

yeah

can i rename my folder now?

bcs its still 1.6

[young bridge]

how do i run it

[ornate elk]

You should have done that before the fix. But you can do it now too.
If any error occur you need to do the fix again

[young bridge]

its fine doesnt bother me

what has changed in the new version?

is it faster?

[ornate elk]

On github is a huge changelog to read. But nothing performance wise has changed.

[karmic crown]

just type git status inside your folder, and it will print the status

[young bridge]

yeah i know but i cant understand all changelog its too advanced

[ornate elk]

Mostly Bugfixes and little additions/changes

[young bridge]

On branch master
Your branch is up to date with 'origin/master'.

Changes not staged for commit:
(use "git add <file>..." to update what will be committed)
(use "git restore <file>..." to discard changes in working directory)
modified: webui-user.bat

Untracked files:
(use "git add <file>..." to include in what will be committed)
modules/textual_inversion/preprocess.py

no changes added to commit (use "git add" and/or "git commit -a")

so?

[karmic crown]

this is strange

Untracked files:
(use "git add <file>..." to include in what will be committed)
modules/textual_inversion/preprocess.py

[young bridge]

i dont know what that is but how can i fix it

[karmic crown]

git reset --hard origin/master
should hopefully fix that