🤝｜tech-support

[crude trellis]

time to test

[cosmic furnace]

Any idea what'r this errors ive got?)

[crude trellis]

also

about this VAE thing

is this something I can get on civitai or hugging face like models and loras?

[ornate elk]

Not until I tested it myself xD sry. But SD.Next uses diffusers Backend so maybe that has some issues

[ornate elk]

Yes exactly
A .safetensor file mostly or older .ckpt

[cosmic furnace]

dang... )

[cosmic furnace]

y btw they was recomending that this diffusers backend works best with zluda

[ornate elk]

Yes but with that ControlNet won't work

Gn 😴

[crude trellis]

gn

[cosmic furnace]

hmmmm do i need to turn on some of theese?

[cosmic furnace]

oh. gn 🙂 . c ya tommorow i guess)

[crude trellis]

😭 😭

its still not working

[unique bear]

what do you connect to the input on the model area

[quick forge]

when I load/change a checkpoint in A1111 it has a little timer like 1.03s/0.3s
What is that timer and is there a way to make loading faster?

[cosmic furnace]

some models settings too much long promt-loras & all shit can make it use too much vram

it says so even when restarting all over?

what resolution & where trying to make?

[covert cape]

I'm trying to port the SDXL LCM from torch to jax and am kinda confused on how to do this well

Currently I'm able to run the txt2img example from the HF in colab

https://huggingface.co/docs/diffusers/using-diffusers/inference_with_lcm?fbclid=IwAR0-saTN5eb9UfpdlK2mYLiXklTkqtKquHVUz-XFgToRnRTJr5FzeonaRpY

and I can replace the diffusers.LCMScheduler with the code from github and that works

but that seems unwieldy for testing out changes on a function by function basis

is there a better approach to doing this?

[viscid gate]

how can i use hires fix with img2img\

[sand dawn]

txt2img > HR fix is basically the same as txt2img > img2img.
The key difference is that the image remains in latent space while switching modes for HR fix.
If you want to do a series of img2img, use a script like loopback scaler

[quick forge]

what does this mean "A tensor with all NaNs was produced in VAE.
Web UI will now convert VAE into 32-bit float and retry.
To disable this behavior, disable the 'Automatically revert VAE to 32-bit floats' setting.
To always start with 32-bit VAE, use --no-half-vae commandline flag.
"

[sand dawn]

use --no-half-vae

[swift kayak]

when using SDXL turbo, why do my images look discolored like this?

those are my parameters

automatic VAE vs. none does not fix it either

[fair oxide]

use this vae for sdxl https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main

[swift kayak]

ahhh thank you very much

[winged hamlet]

Question and help appticiated.

How to create images of 2 person(1 man +1 woman) in same place with different angle of photos?

Imagine one couple in a beach. They want to have a few pictures with different poses, but same clothes, same hair, same style.

So how is the walkthrough of this problem?

My trials, are with
first training Dreambooth,(1 man) ,
second training dreambooth (1 woman) , trying to merge safenetors to one file.

and to generate with controlnet. But it is total fail . .safenetors does not always give same people face, light/clothes/hair meanwhile, only andom things.

Any suggestion, or experience, really appreciated,

[mellow apex]

anyone who has a good idea on how i can inpaint a png car that i found into a landscape scene i created? all inpainting guides ive seen is just talkign about replacing faces on anime waifus

[prisma quartz]

how to do this sequence xyz but on ControlNet Settings?

[vivid viper]

my google chrome or edge does not have an Install from URL in the Extensions tab so how do I install this

https://github.com/Scholar01/sd-webui-mov2mov?tab=readme-ov-file#installation

[maiden haven]

you have got to be kidding me

do you think this is a fucking chrome extension?? it's talking about your extensions tab in the fucking stable-diffusion web ui.. christ

is there a private SD discord that isn't full of people whose first fucking computer was an iphone? i cannot believe how untechnical and trash at python people are in here

im leaving msg me if there is one appreciate it

[unique bear]

Error occurred when executing KSampler:

mat1 and mat2 shapes cannot be multiplied (2x768 and 1280x5120)

what does this mean

[formal trout]

Is there a way to get more pixels here? I'm inpainting an upscaled image and it needs more context sometimes

I could resize the image, inpaint, merge, but lot of work

I'm trying the thing to work with it can see thou

[ornate elk]

Can you show the settings?

[nocturne topaz]

using a 1.5 control net or lora with a sdxl model??

[unique bear]

stable cascade

[pearl bramble]

Hello, Im not sure if it was asked here before, but Im running Stable Diffusion locally using Python and Im getting an import error with triton module.
Ive searched on Google and the best answer was to ignore it, and I am, my images are generating fine, but I would like to understand better this issue.

Thanks in advance!

[ornate elk]

Hey, what webui are you using, and do you have a full screenshot of that error?

[pearl bramble]

Im not using webui, Im developing a tool for a software called Nuke to generate these images using the SD lib.
This is the error:

[manic snow]

This may sound dumb, and i may be overlooking something here, but have you tried simply installing the module via pip?

[pearl bramble]

yeah, I tried by pip, Anaconda etc but nothing works 😦

[manic snow]

That's strange

[pearl bramble]

maybe this does not exist for Windows as Ive read in another forum

[ornate elk]

whats your python version?

[ornate elk]

i think triton isnt used for image generation

[pearl bramble]

2.7 and 3.10

[ornate elk]

ok, yea seems like its not for windows

found these info sites for it:
https://openai.com/research/triton
https://www.nvidia.com/de-de/ai-data-science/products/triton-inference-server/

[pearl bramble]

hmmmmmm thats good!!

[ornate elk]

and the github of it:
https://github.com/openai/triton
and the docs:
https://triton-lang.org/main/index.html

[pearl bramble]

awesome! Im saving all these links, thanks a lot man!! 🙏

[ornate elk]

no problem 🙂

[severe kindle]

turbo models need a lower CFG to work, try using 1.
CFG 8

CFG 1

[urban bloom]

I got this using Forge and SDXL on a 4GB VRAM:

I thought it should work

[ornate elk]

make sure your gpu driver is updated

[urban bloom]

ok

[ornate elk]

wait sdxl

that could be to much for a 4gb gpu
it can work with 4gb but only with lowvram and not recommended

[urban bloom]

Game ready or studio driver?

[ornate elk]

game ready

[urban bloom]

yes well I heard it could work, so I will try

ok

[quick forge]

If I have Stable Diffusion (A1111/ComfyUI/Foooocus) but puit the models/embeddings etc on a separate HDD, would that cause any form of slow down or performance issues?

[vocal burrow]

If by HDD you litteraly mean Hard Drive Disk, not any sort of ssd/emmc/m.2. Then yes.
Storing models/embeddings/loras/etc on hard drive (different or not) will induce slow down during the loading of said models/embeddings/lora. If you can, store everything on some fast device instead.
If all you have is HDDs, will it get even worse by splitting your install on multiple disks. No (or it will be negligeable)

[quick forge]

I meant to include that those UI's/SD is running on an SSD ....an NVME

[vocal burrow]

then it should be all good.

[quick forge]

but I was considering putting the models and such on a separate HDD

[urban bloom]

it worked

[vocal burrow]

If we want to nitpick, it can depend of how much pci express lanes you have available for your disks, what cpu you have, how they re connected.... In reality I doubt it will have any significant impact.

[urban bloom]

SDXL on 4GB VRAM worked

[quick forge]

well this whole system is basically going to be just for stable diffusion

[vocal burrow]

then it will be slow....

hdd = slow loading

it s as simple as that

the larger the file the longer it will take.

[quick forge]

whenever I'm done fucking around with everything I'm going to probably install the basic version of Ubuntu and run it as a headless server of sorts

[vocal burrow]

models takes anything from 2 to 16gb so it s a lot of things to read and transfer to gpu

[quick forge]

yeah I've noticed lol. I've tried to generate some stuff only for it to hang at 95% because my video card is only 16GB

[vocal burrow]

what gpu ?

[quick forge]

RX 6800

[vocal burrow]

it should be enough for most things. You ll get better performances switching to linux indeed. Or you can try the zluda install (check pins)

[quick forge]

im on linux

I was trying to generate a 1024x1024 image, using hi-res fix to get to 2048 and it didn't get past 95%

[vocal burrow]

Might be because of the VAE stage hitting your vram extra hard at the end, try using the Tiled VAE feature of Multidiffusion https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111?tab=readme-ov-file#tiled-vae

[quick forge]

it did mention something about NaNs being generated and converting or using floating point 32 VAE...... I don't have the exact message handy right now. I may have posted it here though yesterday

[vocal burrow]

what VAE was it ?

[quick forge]

Id have to recreate the problem to actually give you information that actually makes sense

actually....

A tensor with all NaNs was produced in VAE.
Web UI will now convert VAE into 32-bit float and retry.
To disable this behavior, disable the 'Automatically revert VAE to 32-bit floats' setting.
To always start with 32-bit VAE, use --no-half-vae commandline flag.

[vocal burrow]

yes but with what VAE / model ?

[quick forge]

SDXL base 1.0 i think

[vocal burrow]

if it was the SDXL one, try using the sdxl-vae-fp16-fix instead

yup try uisng the fp16 fixed one then

https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main

[quick forge]

I have it. I as mainly just trying to stress test my card a bit

after installing a bunch of fans to keep the case and card cool

aside from A1111, ComfyUI and Fooocus what other UI's are there?

[karmic crown]

forge (modified a1111)

https://github.com/lllyasviel/stable-diffusion-webui-forge

[quick forge]

im going to be looking into that one probably tonight

somewhat tangential here, but is there anyway to setup a credits/user system so I could allow users access to my setup

[karmic crown]

if you install it over your existing A1111, remove controlnet extension first (it is integrated in this)

[quick forge]

good to know

[karmic crown]

Forge backend removes all WebUI's codes related to resource management and reworked everything. All previous CMD flags like medvram, lowvram, medvram-sdxl, precision full, no half, no half vae, attention_xxx, upcast unet, ... are all REMOVED. Adding these flags will not cause error but they will not do anything now. We highly encourage Forge users to remove all cmd flags and let Forge to decide how to load models.

keep this in mind too

[quick forge]

I haven't used any command flags while loading that im aware of.

[karmic crown]

check your webui-user.bat

[quick forge]

I deleted that

[karmic crown]

than your webui is not optimized for your card at all

and it probably works much slower than it could with right params

[quick forge]

im on linux with an amd

[karmic crown]

in that case check webui-user.sh

if we are talking about a1111

[quick forge]

ya i can do that after work. but nothing is optimized at the moment. I'm still fucking about with this that or the other

[vocal burrow]

• Fooocus
• auto1111
• auto1111 directml fork
• SDNext
• SDWebui forge
• invokeAI
• DreamStudio
• StableSwarm
• EasyDiffusion
• Sygil
• TemplateAI
• DreamLikeArt
• ComfyUI
• TensorPlay
• NMKD Stable Diffusion GUI
• Stable Diffusion GRisk GUI
• diffusionbee.com
• stabilityMatrix
• and probably a bunch of other I don't know about yet

[manic snow]

stability matrix isn't a ui on its own, it's an installer/launcher for many different ui's, kind of like pinokio. it has a built-in model installer as well. i love using it

[patent meadow]

Idont understand what the error messages want me to do

I tried try fix and try update

[ornate elk]

Shark

And I thougt stability matrix is just an wrapper for auto1111

[karmic crown]

it can install a1111, comfy and few others, and also has some features that a1111 plugins like civitai browser plus has

it might be interesting for people who install sd for the first time

but for users who use a1111 with plugins like civitai browser, model organizer and know how to use git, it is practicaly bloatware

[manic snow]

nope, you can get most of the popular webuis there! it even has its own little comfyui interface

[brazen flare]

Hi guys ! ^^ I am under the Forge webui but I can't find the "restore faces" setting next to the "hires.fix". Do you know where can I find it ? 🙏

[vocal burrow]

check the settings page

[brazen flare]

supernice !! Thanks ^^

[dense adder]

how do I use DAT upscalers?

[wraith bronze]

I have one of my GPUs dedicated to SD, is there any NVIDIA driver settings I should use that can improve performance since other stuff like gaming or any other task that isn't SD is not done on this GPU?

I have a 3090, and at the moment while generating 8 images it gives me 15.43s/it

[rose ferry]

I am confused with "diffusion_pytorch_model.safetensors" where does that safetensor should reside in?

[wraith bronze]

A very quick Google show its a VAE so try "\webui\models\VAE"

[vocal burrow]

which one ?

[rose ferry]

What do you mean?

[vocal burrow]

diffusion_pytorch_model.safetensors is a very generic name

it doesn t mean anything

[rose ferry]

mmm okay let me find one.

Ah, the VAE stuff. https://huggingface.co/stabilityai/sdxl-vae/tree/main

So the question is that diffusion_pytorch_model.safetensors should be in VAE?

v2-1_768-ema-pruned.ckpt should be in VAE as well?

[wraith bronze]

ckpt should be a checkpoint, that one would go into "stable-diffusion" under "models"

[rose ferry]

not make sense because vae-ft-mse-840000-ema-pruned.ckpt is in VAE but apparently, ckpt in model and safetensor in vae.

[fair oxide]

just be careful about ckpt, those file types can carry malicious codes, most of the models are now in safetensors at civitai

[rose ferry]

I am using huggingface but thank you, moving to https://huggingface.co/stabilityai/stable-diffusion-2-1/blob/main/v2-1_768-ema-pruned.safetensors

[wraith bronze]

Safetensor is not specific to vae, it can be lora, checkpoint, etc. But .ckpt should always be a checkpoint as far as I am aware.

[vocal burrow]

then you don t need anything else expect sdxl_vae.safetensors

[wraith bronze]

I also only use .safetensors, anything I use is rarely a ckpt file

[rose ferry]

okay, what are diffusion_pytorch_model.safetensors do?

[wraith bronze]

I might be way off, but tensors are just generic datastructures, safetensor is a format to store those generic structures in a safe way that is also fast.
While ckpt can contain malicious code (bad), safetensors cannot (good).

I don't use huggingface so I don't know how they tag various things, but if you use civitai you will always know if you look at a checkpoint, lora, lycoris, vae etc and that will help you know where to store the file you download (that is 99.999% of the time a .safetensor file).

[rose ferry]

Roger

Since I do some help from you. Could you answer my question at https://discord.com/channels/1002292111942635562/1176560067961696276

[wraith bronze]

Oh I haven't tried making video stuff in SD yet.

[vocal burrow]

short version, it s training stuff

you don t need it

[wraith bronze]

So I have no idea how any of that works. Well, I did try it, but it didn't work out very well. What I put in was not what came out

By the looks of it, you know more about the video part than I do 😛

[rose ferry]

Thanks!! 😄

[sharp ether]

Hey yall, i'm super really ultra new to stable diffusion and i'm a little lost. I'm watching tons of videos on where to start, sd 101, automatic1111, comfyui, fooocus, and well, did i mentioned i'm new to this hehehe, so, could anyone maybe give some directions on where i could start, which platform to use. My main goal is to learn how and to create an virtual influencer and give life to it. In advance, tyvm for any help. Cheers! (trying to ask for help on this channel since in general it went downhill)

[fair oxide]

if you are going to create virtual influencer model for instagram that will require some advanced deployment, while I would normally suggest starting out with A1111 webui for new comers, but in your case you could start off with comfyui webui which will give you more control with advanced stuff.

[sharp ether]

nice, i got comfyui already working, tried to with a1111 but got lost in the middle and a bunch of errors. so i'll try to start watching some vids on comfy, tyvm

[fair oxide]

automatic1111 is easy to install btw, if you are interested i can give you a quick link

[ornate elk]

Would still suggest auto1111, for starters, for any tech issues here is the right spot to post errors

[sharp ether]

ah, ok, will try to install a1111 again, any suggestions?

[ornate elk]

Also in the pinned messages of this channels you find detailed install guides for auto1111

Depending on your GPU

[sharp ether]

will look for it, ty @ornate elk ❤️ loved the avatar, huuuge fan of one piece

[sharp ether]

i think it will run easy, maybe i'm wrong hahaha butt i'll check it out

[ornate elk]

i mean there are guides for both nvidia and amd ^^

if your gpu has more than 4gb vram, no doubt it will work

[sharp ether]

64 ram / 24 vram

[fair oxide]

the key differences between Automatic1111 and ComfyUi is that Automatic1111 is not as strictly techy as ComfyUI

[sharp ether]

got it, someone was saying that with pinokio it would be easyer

[ornate elk]

you dont need an installer, best is to install it manually ^^ less error sources / dependencies

with my guide its done in 10 mins

[sharp ether]

ok, is there a waay to open the message where you teach how to do it (instal a1111) in a new tab?

[fair oxide]

#🤝｜tech-support message

[ornate elk]

you can also always click on the Pin Needle icon on the top right

[sharp ether]

i went with the pin on top, but i meant like, opening in a new tab so i can still see what you guys are saying or in another channel

sorry for the noob questions haha

[ornate elk]

oh thats not possible in discord :/, but you could copy the whole text into a text file

[sharp ether]

yeah, decided to do that, much easier heheeh, tnks @ornate elk

what would be in your opinion, the best way to train a lora model?

with images i generated

[ornate elk]

khoya_ss

its a seperate tool

made for lora training

[trim sable]

Hi guys, Can I use stability video for commercial use ?

[ornate elk]

@sharp ether Lora training options:
https://github.com/bmaltais/kohya_ss
https://github.com/Akegarasu/lora-scripts
https://github.com/derrian-distro/LoRA_Easy_Training_Scripts

[ornate elk]

watch out for * ´´ characters, these are from text formating and can be ignored

[sharp ether]

you mean the markdown style?

k, will watch out for it 🙂

mind if i ask you some other doubts?

[ornate elk]

no, feel free to ask

[sharp ether]

well, i'm doing some of my own clothing designs, so, with that in mind, remembering that i want to have an virtual influencer made by me as well. If i decided to make the V.Influ use my clothes, the clothes should be trained as lora models as well? (i was thinking on that path) but i might be wrong so decided to ask 🙂

[ornate elk]

yea with loras that is possible

[sharp ether]

humm nice, so i'm not all wrong in my thiking, are there any other ways?

[serene badge]

Hi everyone, i hope this is the right channel to ask. I am using AUTOMATIC1111 Web UI, and loaded the checkpoint and yaml for stable diffusion 2.1. It wouldn't generate images with the error NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.Time taken: 1.5 sec.

I tried to enable "Upcast cross attention layer to float32" in the settings like it was suggested and now the image generated are pixelated blue dots just like shown. I am i doing something wrong? Is anyone able to help me out please?
Thank you in advance

[sharp ether]

tyvm

[sharp ether]

was also wondering about legal rights with the images created, are they allowed to be used like i was saying, for commercial purposes ?

[ornate elk]

that depends on the model used, some have information about usage rights at their page

[sharp ether]

will check on that as well

[fair oxide]

given how much you want to push your skill level you could build upon the base model and training your own loras then the rights to commercial use case depends on you.

[sharp ether]

yeah, that's what i'm aiming for, having my own loras trained, with images that i've created.

[fair oxide]

good plan, lots of guides on civitai page

[sharp ether]

yeah, i noticed that and i've noticed that thay have a lora training of their own, inside civitai, but i'm not sure which one sould be the best, kohya, civitai's, any inputs on that?

[fair oxide]

go with the links @ornate elk provided

[sharp ether]

already saved them and already wathing a video on thoery and practice of lora training with kohya. Tks guys, you rock 🙂

[fair oxide]

np, good luck with your venture

[cosmic furnace]

@ornate elk ive been told that zluda now is bad with img2img =\ . yet again rly weird stuff... control net implemented in sd next worked while img2img not... also it managed to generate about 3 times a very fast & big res image and then blum- OS restart & no more generating anything besides txt2img . Oh and it was saying smthing about deep cache full =\

[wraith bronze]

@ornate elk do you have any recommendation on what to enable/disable if I have a NVIDIA RTX 3060 GPU dedicated for SD and want to try and improve performance? Both in SD itself and any driver settings that might help ^^

[ornate elk]

your controlent worked? how?
for me img2img works, but controlnet not, did you changed diffusers?

[ornate elk]

12 or 8gb vram?

[wraith bronze]

12

I use --xformers but that is about it on what I have figured out might help performance

[ornate elk]

yes thats good

[sweet basin]

hwo to fix this jsonmerge

[vocal burrow]

delete venv folder and try again

[mental ocean]

Wondering if anyone would have any tips to resolve this: I’ve been playing with SD using Automatic1111 for about a week now. Today I noticed that it has started slowing way down when generating a batch from the “Prompts from file or textbox”. I did a reboot of my system and the first prompt generated at 19s/it, but the next one slowed way down to around 127s/it. All prompts are using the same model, sampler, negative prompt, steps, cfg, dimensions. The only thing that changes is the positive prompt. I am using an older gpu with only 4GB vram (GeForce GTX 1050, Intel i5-8300, 16GB RAM) but I’ve not had this issue until today. Other than setting directories, I’m only using —opt-sdp-attention in my command line args

[sweet basin]

what for zluda is

[vocal burrow]

what's the prompt ? using any lora embedding, etc ? Also you probably want to use --xformers --medvram instead of --opt-sdp-attention

[vocal burrow]

sorry, what what ?

[sweet basin]

zluda

[vocal burrow]

yeah I got that part. But the sentence does not make any sense.

"what is zluda ?"

[sweet basin]

?

[vocal burrow]

I don't understand your sentence.

There is either not enough words, or too many.

[sweet basin]

"what for"

what is

what is zluda

what does it do

do I need it

..

[vocal burrow]

zluda is a "library" translating cuda (so nvidia specific) instructions into something that AMD can understand.

So far if you wanted to use a1111 (or sdnext or a bunch of other UIs) on windows your only solution was to use directml. Now you can also use zluda instead, it's faster but more experimental.

[sweet basin]

still no idea what it is

[vocal burrow]

your car either run on diesel, gasoline, electricity.

a1111 either run on cuda, zluda or directml

[sweet basin]

so its not important

[vocal burrow]

it's different solutions.

for AMD on windows, zluda is faster but more experimental

[sweet basin]

you mean generating images faster?

[vocal burrow]

But if you have troubles understanding what cuda is, you might want to stick to directml and simple solutions.

[sweet basin]

I have no idea what directml is

neither cuda

I press generate and dont need to know anything else

does it affect image speed progress?

[vocal burrow]

like I said twice, yes it's faster.

But it's experimental so you're more likely to run into errors.

[sweet basin]

your talk like nasa scientist so its hard to understand

[vocal burrow]

Then ask what part you don't understand.

Because people going "I press generate, I don't want to learn anything" is not really encouraging anyone to put efforts into their explanations.

[sweet basin]

"what for zluda is" generates faster for amd

thats it

also deleting venv folder fixed my problem

classic

[notice] A new release of pip available: 22.2.1 -> 24.0
[notice] To update, run: E:\StableAI\stable-diffusion-webui-directml\venv\Scripts\python.exe -m pip install --upgrade pip

[ornate elk]

What's your GPU?

[sweet basin]

where to update this

[sweet basin]

6700xt 16gb

[ornate elk]

Okay and your webui works again now?

[sweet basin]

yeah

[ornate elk]

You dont need to update pip

So that can be ignored

[mental ocean]

No Lora's, but using textual inversions. Example below. If I paste this same prompt 3x in the "prompt from..." text box, the first one ran at around 19.5, then the second took a couple minutes to start and dropped to 135.2.

--sd_model "Standard\dreamshaper_8.safetensors" --sampler_name "DPM++ SDE Karras" --prompt "cinematic photo a centered model, upper body, fashion model, kkw-ph1" --negative_prompt "kkw-ph1-neg, BadNegAnatomyV1-neg, easynegative, negative_hand-neg, UnrealisticDream, ng_deepnegative_v1_75t" --seed -1 --batch_size 4 --steps 20 --cfg_scale 7 --width 512 --height 768

[vocal burrow]

that's ... unusual. What's the full log after a clean sdwebui restart + 2 generation ?

[rough canyon]

Any got any idea why this happens. I made an image using webuit-forge, but want to fix the hands using controlnet. So I open the image in the inpaint tab, but when I open the controlnet 0 option, I can select everything I need except there's no way to upload an image or scribble the mask.

normally, where you see the options "enable, pixel perfect, etc" there should be an option to upload an image no?

[manic snow]

you're in img2img, so you need to check "upload independent control image"

[rough canyon]

oh I see, stupid me 🙂

[mental ocean]

Looks like the culprit was ReActor being turned on and set to use CUDA. Turning it off, or changing to CPU appears to have fixed the issue.
https://github.com/Gourieff/sd-webui-reactor/issues/307

[vocal burrow]

glad you've worked it out :p

[vivid viper]

idk what I am doing wrong and it google is not helping

[ornate elk]

"no Names found" can be ignored

[vivid viper]

i am runing it on amd

[ornate elk]

yes me too

i get this message too

but it can be ignored as it doesnt break anything

[vivid viper]

okay

but when I use it it takes a long time tp load any way to speed it up

[ornate elk]

do you use the commandline args from my amd install guide?

[vivid viper]

yes

[ornate elk]

whats your gpu?

[vivid viper]

but let me try again

Msi Radeon RX 5700 XT GAMING X

[ornate elk]

ah okay

[vivid viper]

let me unstalll the git and python and try again

anthing I should chnage in the git install

[ornate elk]

nothing to change

but you can disable the explorere integration

[vivid viper]

okay will that change any thing

[vivid viper]

is this how to add the mutiple commands

[ornate elk]

nope

they go all in one line

also dont use these

[vivid viper]

okay which ones should I use

[ornate elk]

you need:
--use-directml --skip-ort --medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half --upcast-sampling

[vivid viper]

okay thnaks

[ornate elk]

if you get out of vram errors you need to change --medvram to --lowvram

[vivid viper]

do I space them out

[ornate elk]

just paste them after the =

all in the same line

[vivid viper]

any thing i need to put for python or git and venv_dir

[ornate elk]

nope

[ornate elk]

and make a space between each arg

[vivid viper]

okay

[ornate elk]

best is to just copy and paste this line:
--use-directml --skip-ort --medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half --upcast-sampling

[vivid viper]

finally got it to load an image

okay

[crude trellis]

@ornate elk hey, I just downloaded some VAE, gonna test those now

in which location should I place them?

[ornate elk]

hey alright,
it goes into models/vae folder

[crude trellis]

ok

gonna try generating something now, let's see if i'll still get those vram problems

[cosmic furnace]

in sd next yes...implemented one o.0 erm... yes as recomended to optimize) they implemented an remade control in next sd, which works on diffusers)

i wonder is this succes imgs 2-3 times are random or kinda not... because very hard to genrate again when erorrs-blackout resets starts...

[vivid viper]

@ornate elk finally got it working now thank you so much

[ornate elk]

Perfect, no problem 🙂

[loud needle]

Hey guys. Please could someone give me some tips to make an animation that stays true to an input image for at least 30 seconds. I keep getting results where the image slowly turns into a static colour with no detail anywhere

There's an example of the start frame, a mid frame and the last frame

[ornate elk]

Okay, and what tool are you using?

[loud needle]

ah sorry. deforum

i can post the settings file if that helps

[ornate elk]

Didn't used deforum :/ can't help much with it

[crude trellis]

where do I setup VAE settings or smt?

[loud needle]

nws dude

[ornate elk]

Go into Settings, User Interface, Quicksettings, there add sd_vae and then hit apply and reload ui

Then you get a VAE dropdown

[serene badge]

does anyone know what problem i could have, if the final result comes out like it has processed with 11 instagram filters? It happens on both SD 1.5 and 2.1, does anybody know how could i fix this please?

[ornate elk]

Hey, yes, your cfg is to high, set it to 7
Then set hires steps to 10, set the upscaler to latent bicubic, the denois to 0.5 and the upscale by to 2

Also 2.1 was trained at a 768x768 resolution

1.5 at 512x512

And dont enable Refiner

@crude trellis did it work?

[crude trellis]

trying right now

😭 right when it was almost done

[ornate elk]

What settings?

Resolution etc?

Did you used the tiled vae extension?

[crude trellis]

I used these VAE I found on civitai

settings

[ornate elk]

Hires steps 0 ....

You need to set it to 10

Otherwise it will go out of memory

On AMD directml

[crude trellis]

ok, sorry

is there something else I should change?

[ornate elk]

NP, you need to restart SD before you try again

[serene badge]

a bit better but it still does the same 😭

[ornate elk]

Also make sure nothing heavy is running in the background

[serene badge]

thank you anyway!

[ornate elk]

Can you select an 1.5 model?

[serene badge]

yup

[crude trellis]

are the VAE I downloaded fine?

[serene badge]

should i do one with 1.5?

[ornate elk]

Yes the one looks okay

[serene badge]

same problem lol

[ornate elk]

Okay try Resrgan4x+ as upscaler

Let me ban that spammer real quick

Ez

[serene badge]

way better!

thank you!

are there any rules about upscalers or is it just trial and error?

[ornate elk]

The problem was that the latent upscalers dont work good with the DPM++ 2M Karras sampler

If you use DPM++ SDE sampler it should work with latent bicubic

Or Euler a

[ornate elk]

Now you should checkout Civitai.com for better Models

1.5 and 2.1 are pretty old and not as good as community models

[crude trellis]

is 12gb not enough? 😟

again

[serene badge]

i did download juggernaut, but it gave me the same problem so i switched to sd again

[ornate elk]

Its enough, but you mostly need the tiled vae extension to prevent that error

Then the upscale should work

[crude trellis]

where can I get it?

[ornate elk]

Click on extension tab. Click on available, then click on Load from.
You'll get a list.
Search for Tiled.
Install Tiles diffusion and then restart

[crude trellis]

ok

[ornate elk]

Then enable only tiled vae

Not tiled diffusion

[crude trellis]

this one right?

[ornate elk]

Yss

[crude trellis]

ok

[crude trellis]

where do I enable?

[ornate elk]

Click on tiled vae. Then enable and set encoder to 1024 and Decoder to 128

[crude trellis]

It seems it worked

thank you!!!

also, I saw some people saying that apparently this would help on AMD cards

is that true?

its rocm

[ornate elk]

Rocm currently is only supported on Linux

Zluda would be your best bet now on Windows. Until Rocm support enters pytorch on windows

[crude trellis]

its not avaiable on windows yet?

[ornate elk]

Not fully supported by pytorch so no webui can use it on Windows

[crude trellis]

hmmm

do you think zluda is worth installing in my case?

as I have a amd gpu

[ornate elk]

Zluda is only for AMD GPUs

[crude trellis]

oh

[ornate elk]

The point of it is to let AMD run nvidia stuff faster

Its cuda emulation

I think it could be very worth for your to try it out

[crude trellis]

should I install it them? does this have any bad sides?

[ornate elk]

Its experimental and not everything that works on Directml works on zluda.
For example sdxl models or IP-Adapter won't work for now

But if your using 1.5 models mostly then you have nothing to lose.

You can generate at higher resolution without crashes

Upscaling without issues

[crude trellis]

sounds like a win win then

[ornate elk]

Yea you can also link the models from directml to the Zluda installation

I have both installed too

[crude trellis]

I might try install zluda on another stable diffusion installation just in case

[ornate elk]

Yes that's what my guide suggests to

I have a directml and a zluda webui

Seperated

Let me know if you try it and get any questions.
But I'm off for now its late here

There is also the SD.next webui which supports zluda too (guide for that in their github wiki)

[crude trellis]

ok

thank you

[ornate elk]

No problem, my guide is in the Pinned Messages of this channel

[hollow schooner]

havent touched stable diffusion in some time

[hollow schooner]

amd

[viscid gate]

how can i generate multiple images of a character with retaining the same face

[old parrot]

Any good plugins to save states of generation parameters for SD Forge?
The one I used by ilian6806 doesn't work with Forge

[hollow schooner]

nevermind i figured it out

[eternal wren]

Hey all i’m trying to install SD but I keep getting error code 1 and runtime error torch could not be installed. any answers?

venv "C:\Users\19096\Desktop\sd 2.0 ;3\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.7.0
Commit hash: cf2772fab0af5573da775e7437e6acdca424f26e
Installing torch and torchvision
C:\Users\19096\Desktop\sd 2.0 ;3\stable-diffusion-webui\venv\Scripts\python.exe: No module named pip
Traceback (most recent call last):
File "C:\Users\19096\Desktop\sd 2.0 ;3\stable-diffusion-webui\launch.py", line 48, in <module>
main()
File "C:\Users\19096\Desktop\sd 2.0 ;3\stable-diffusion-webui\launch.py", line 39, in main
prepare_environment()
File "C:\Users\19096\Desktop\sd 2.0 ;3\stable-diffusion-webui\modules\launch_utils.py", line 378, in prepare_environment
run(f'"{python}" -m {torch_command}', "Installing torch and torchvision", "Couldn't install torch", live=True)
File "C:\Users\19096\Desktop\sd 2.0 ;3\stable-diffusion-webui\modules\launch_utils.py", line 116, in run
raise RuntimeError("\n".join(error_bits))
RuntimeError: Couldn't install torch.
Command: "C:\Users\19096\Desktop\sd 2.0 ;3\stable-diffusion-webui\venv\Scripts\python.exe" -m pip install torch==2.0.1 torchvision==0.15.2 --extra-index-url https://download.pytorch.org/whl/cu118
Error code: 1
Press any key to continue . . .

[formal vortex]

hi

when I am creating an image using sd, showing permission error, how can I fix

[pseudo nebula]

hey uh, the function to proprocess the images by making a text file with their descriptions is gone

i'm checking the extras where i saw one github thing say it was moved to, but there is no such feature there.

it lets you interrogate all of the images in the directory, but there is no output of this

ey did u find a solution

[ornate elk]

Hey, remove the . and spaces from the folder name sd 2.0
Then delete the venv folder and relaunch the webui-user.bat

[ornate elk]

Can you show a screenshot of the complete cmd with error?

[ornate elk]

Hopefully you didn't added --skip-torch-cuda-test, it would use your CPU instead of GPU.
You need --use-directml

[upbeat atlas]

Quick question, would this work with "AMD software: Adrenaline edition" as well?

[ornate elk]

Yes its for adrenalin edition. Zluda is a zip file. And the HIP SDK are modules that doesn't change the adrenalin driver

[hollow schooner]

Oh nah apparently i just used the wrong file/cmd to open the webui😭

[upbeat atlas]

Dude, I have no idea how, but I finally got it to work properly. It's SO FREAKING GOOD. Thanks a lot, it was definitely worth the hassle. 🫡❤️

[ornate elk]

Hey, glad it worked ^^ what was your GPU and what's your it/s on a 512x512, 20 steps, image?

[upbeat atlas]

My GPU is AMD "RADEON RX 6900 XT"
For 512×512 20 steps it gets 6 it/sec which is just crazy to me.
Back in directml version I used to get like 5 it/sec for 344 × 512 images, here I get 7.8, and I get 4 it/sec for 512 × 768 images.
The best part is that it uses so much less VRAM now. In directml I barely could do 1.6 upscale in hires and even then it would gimme a not enough vram error here and there, now I do 1.8 EASILY, for some reason the hires is on an entirely different level compared to everything else.

[ornate elk]

Awesome!
Yes the vram usage is really good now. Hires fix, + using loras + controlnet just works without any errors

Only, depth and IP-Adapter controlnet doesn't work and will crash the driver xD

[upbeat atlas]

Yeah that's absolutely fine I never use those anyway. Thanks again, I'm finally out of the hell hole that was directml 🗿🤍

[zealous leaf]

in REST API, it's clearly defined what kind of possible error messages we can get, I can't find those anywhere for gRPC thought. There is only this piece about error in docs if artifact.finish_reason == generation.FILTER: for safety check.

How can I get and handle rate limit, unauthorized and other errors in gRPC?

[night narwhal]

Hello, Is there a SD1.5 script to train a controlnet for inpainting pipeline?

[umbral plover]

What would be better 3060 12gb or 4060 8gb, i got a chance to buy the second one for almost same price as used 3060 but i think those 4gbs of vram would make a big difference?

[limber pier]

Always go for vram

[green ferry]

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument mat2 in method wrapper_CUDA_mm)
hey im getting this error when trying to enable a lora as an additional network. it seems like an easy fix in the code but idk what code to adjust to put the argument there

[ornate elk]

hey, wats ypur gpu and whats inside your webui-user.bat?

[green ferry]

gpu is nvidia and i installed forge's webui with default settings

should i just take forges settings in the webui out?

[ornate elk]

No thats okay

Make sure your GPU driver is updated

Also make sure the model and lora are the same base Version.
1.5 or sdxl.

Dont mix up 1.5 loras with sdxl models

[cosmic furnace]

and some ppl said that img2img works just fine for them on zluda next 😑

[ornate elk]

For me too

But I dont get controlnet to work

[cosmic furnace]

As if smone who was up at that time didnt felt like investigating reason -errors then? o.0

or too often error happen - blah then?)

control net implemented? o.0 as if its random huh...

[ornate elk]

Can you show me an example of the implement setting working?

[cosmic furnace]

well before full reinstall ive managed to do 2-3 img at high res in img2img & then it started to error

also didnt quite sure what to do when "deep cache" is full

[ornate elk]

Is that an error message you get?

[cosmic furnace]

mb shudnt enable that feature for speed up 😄

[ornate elk]

Yes thats probably why img2img doesn't work for you

[cosmic furnace]

yeah last time cudnt generate any image because it was saying deep cache i enabled is full

[cosmic furnace]

not sure maybe it was before enabling it 🤔

[ornate elk]

Try again without it for test

[cosmic furnace]

not one , but few of them whos at the top... managers? lol dunno

said that img2img doesnt work'ish

[river flame]

I want to add Cyberpunk style to Forge, how do I do it?

[cosmic furnace]

and few or more ppl said img2img work with zluda for them -_-

[river flame]

and also. I m getting error 128 trying to install stable diffusion web ui. Added Python PATH in 2 places, it started installing but then error 128 again.

[cosmic furnace]

doesnt seem like path problem

very rude & stupid attempt but oh well 😄

as i understood nsfw is prohibited))

768 in 17 sec... rly slow -_- y ?

mb cuz original was 1600x828 ? and its control net ?

and now 1280-1024 40 steps in 13 sec o.0 but black & white -_-

[cosmic furnace]

about 3its is slow y ? =\

[ornate elk]

can you show the error as screenshot?

[ornate elk]

on that resolution, i think not

[green ferry]

i think you need to take the quotations out

[green ferry]

i updated my drivers for nvidia and the model and lora are the same version but the error happened again. I think I need to add something to the data line but idk which python script to edit

[ornate elk]

Can you show the complete cmd?

[ornate elk]

What model and lora do you use?

[cosmic furnace]

rocm 6.1 for linux already ;O

mb for win soon too?)

[ornate elk]

hopefully 😮

[green ferry]

based64 and the loras are sd 1.5

[river flame]

thank you very much for trying to help, but I think I`ll just give up and work in Forge UI, it seems to be faster on my 1080ti.

[ornate elk]

if default auto1111 works, you should use that

make sure to use --xformers --medvram-sdxl --no-half-vae

[river flame]

where do I put this?

[ornate elk]

in the webui-user.bat

at the line COMMANDLINE_ARGS=

[river flame]

thank you so much

[ornate elk]

np, that will make it much faster and also faster for sdxl

[green ferry]

@ornate elk did you have an idea of which python script to edit? https://stackoverflow.com/questions/59013109/runtimeerror-input-type-torch-floattensor-and-weight-type-torch-cuda-floatte this is an atricle I was reading on it but idk what line to edit to remove the error

[cosmic furnace]

i though control wud be more accurate o.0

[ornate elk]

nope, hav you tried an other lora or model?

[green ferry]

yes its just with the additonal networks extension by kohya

[ornate elk]

So it does work when using the lora normaly?

Because additional network extension isnt needed

I dont know if forge even supports it

[green ferry]

yes it works with the lora normally but i was thinking it was just an easy fix to get it updated for the kohya extension

[green ferry]

yeah i just downloaded normal automatic1111 and the additional networks work:(

[ornate elk]

OK, yea forge don't support every extension for auto1111

[karmic crown]

but also do not need some of them, like controlnet

[chilly knot]

hi

SSL error Decryption failed or bad record mac when i try to use upscale and something downloading
i think is bad connection?

[karmic crown]

forge is made by lllyasviel, so he integrated controlnet and added some crazy things, i would use it right away if it was faster than a1111 on my computer

check it out when you have time

if i remember correctly controlnet and tiledvae are integrated

[glossy spindle]

I am using SD web-ui by A1111 with the Dreambooth extension installed. I don't see an "Input" tab like I see in tutorials. Was the UI changed or should I see an Input module?

[river flame]

does it matter if I install Cascade on Pinokio or on Forge?

[ornate elk]

Yes I already tested it.
In a 1080 you get 3.08 it/s instead of 1.99 it/s on auto1111

But extensions like regional prompt control from Multidiffusion doesn't work :/

[river flame]

can I install SDXL Lightening on Forge?

[ornate elk]

Does forge support it already?

[river flame]

I watch a video How to install Cascade on A1111 and Forge, so I guess yes

[river flame]

nvm, I installed ComfyUI

[solar sundial]

hey guys, so I have xformers installed. but when I run webui.sh it says it cannot find it. Not sure what I'm missing here?

[noble belfry]

yes. i actually just posted about that on another discord. im able to generate 1024x1024 images on a 1060 machine with only 6gb vram in 30 seconds. Its a very good combination.

[river flame]

this is great news cause I m confused with all these nodes

[noble belfry]

haha. well dont delete comfy, its useful to have 🙂

but forge is very easy to install and run from its zip and its a fork of a1111 so its simple to use.

[old parrot]

What would be the reason SD Forge generates every other photo slowly?

First generation takes 30 seconds, second would be 30 mins if I don't interrupt, then the 3rd generation takes 30 seconds. All parameters are the same each time

[noble belfry]

well it does a lot of loading and unloading and moving to optimize time, so maybe if you're switching models or using a refiner

[old parrot]

No refiner or new models

[noble belfry]

sorry time=memory.

any loras?

[solar sundial]

how did you get this running? I'm having a hard time

[ornate elk]

On Linux you need to build xformers yourself

[solar sundial]

oh geez. from source?

[ornate elk]

nope, here is a short guide (at the bottom)
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Xformers#building-xformers-on-linux-from-anonymous-user

[old parrot]

Yes 2 or 3 loras, same ones each time

[noble belfry]

hmm. there is an optimizations tab in settings, but to be perfectly honest im pretty new to forge and wouldn't go telling you to start clicking things i havent tried myself. 🙂

30 minutes is obscene though, it doesnt sound like a slight optimization issue

[old parrot]

I just skip and restart when I notice but it's strange for sure

restarted my PC and it stopped lol

[velvet dagger]

NansException: A tensor with all NaNs was produced in Unet. Use --disable-nan-check commandline argument to disable this check

I get this when I try using a model thats not stable diffusion or dreambooth, do i have to merge the checkpoints or?

[noble belfry]

you wouldnt happen to have your model files on a different disk to the one you're running forge on would you?

[old parrot]

I had the same issue until I switched to SD Forge

[velvet dagger]

is that a new model or a new program?

[old parrot]

New variation of Stable Diffusion

[velvet dagger]

rad

[old parrot]

Prolly is the reason, they're running on my HDD, about time I upgraded this SDD anyway

[noble belfry]

oh mine run on my HDD. but its the same HDD as my OS

[old parrot]

I just realized it only happens when I switch away from the SD tab, hoping it's a setting within Edge

[chilly knot]

when i start generate in extra tab with gfpgan i have error -ValueError: max() arg is an empty sequence

[valid flame]

can anyone help me y getting this error

not open

no india

yes

yes

ru hacker wat 😄

[noble belfry]

jfc dont click random urls from greenleaf users on a really large discord.

(or anyone frankly)

[valid flame]

sure

[valid flame]

just i clicked that link but nthg open anything problem ?

[old parrot]

Apparently if you use Edge browser with Efficiency mode on, tabs go inactive when not focused.
Made SD slow when switching back to Runescape lol

[noble belfry]

aaaaaaaaaaaaaaaaaaah!

that should not effect the actual server, but weird

[solar sundial]

oh boy... still saying no xformers after following the steps...

/stable-diffusion-webui/venv/lib/python3.10/site-packages/torchvision/io/image.py:13: UserWarning: Failed to load image Python extension: '/home/daniel/stable-diffusion-webui/venv/lib/python3.10/site-packages/torchvision/image.so: undefined symbol: _ZN3c1017RegisterOperatorsD1Ev'If you don't plan on using image functionality from torchvision.io, you can ignore this warning. Otherwise, there might be something wrong with your environment. Did you have libjpeg or libpng installed before building torchvision from source?
warn(
No module 'xformers'. Proceeding without it.

and my answer is "yes" to the question it is asking me in the error

[river flame]

Juggernaut XL 9 Lightning (4 steps) doesn`t work for me (Forge). Should I just switch to the normal version?

[noble belfry]

what cfg?

[solar sundial]

anyone know what I did wrong wtih this?

cat webui-user.bat
@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --medvram-sdxl --no-half-vae

call webui.bat

[river flame]

1.8

testing CFG 7

[noble belfry]

ive it open rn: these settings working fine as wine

[solar sundial]

Jason Vorhees?

[river flame]

wow it really works! magic, thank you

[noble belfry]

its 4 step, but 6 step and 1.5 seems to give me a good consistent combo

[solar sundial]

I'm also continuing to get this error when loading up the SDXL base safetensors model:

/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/_refs/init.py", line 4755, in empty_like
return torch.empty_permuted(
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 14.00 MiB. GPU 0 has a total capacity of 5.79 GiB of which 13.69 MiB is free. Including non-PyTorch memory, this process has 5.77 GiB memory in use. Of the allocated memory 5.46 GiB is allocated by PyTorch, and 196.17 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

[ornate elk]

with 6gb you need --medvram instead of --medvram-sdxl, but the issue comes from the not working xformers

you need it for sdxl on 6gb

or you could try --opt-sdp-attention

maybe it works with that

[solar sundial]

yeah i followed the link you sent me for xformers

so --medvram --opt-sdp-attention on the commandline args?

[ornate elk]

yes

[solar sundial]

same error on xformers... this is frustrating

is pop_os maybe my issue? Should i change distros?

[chilly knot]

how many vram need for xl

[noble belfry]

absolute floor minimum would be 4gb vram with forge

[river flame]

in Forge, I don`t need to rename .bin to .pth, right?

[noble belfry]

but its like trying to drive to work in a childrens car

[noble belfry]

you shouldnt need to change anything

[chilly knot]

forge what is it

i have 4 gb)

[noble belfry]

it is a version of a1111 that works very well with low end machines

but 4gb will be very very slow even at that

https://github.com/lllyasviel/stable-diffusion-webui-forge

[chilly knot]

a111 take more vram?

[noble belfry]

yeah its less memory efficient. think of webui-forge as an optimized version of a1111

[velvet dagger]

?/?

im getting this in the sd forge i just downloaded

its referring to the word count in my prompt bar

TypeError: 'NoneType' object is not iterable

i also get this error

nvm i think i fixed it

i tried editing the notepad so that it went to the same path as my previous sd with all those models

[noble belfry]

did it work?

wait wait, scuttlecrab. did you by any chance try to load an example image via png info?

sometimes you get that error if the image is referring to settings for things your version of the program doesnt have

[velvet dagger]

it worked! and no i hadnt

i think the issue was that i had no real model place into the forge

but once i linked the pathways it grabbed the models from my other sd models folder and was able to generate @noble belfry

[noble belfry]

nice!

[brittle oak]

Hello everyone I would like to request your help

has anyone experienced this issue using Google Colab...?
does anyone knows how to solve it...?

I got a firts error because of a missing module, and then I tried to install it, but I got a new problem

this was the firt message

[magic hollow]

so i got an amd card.... i was wondering if there was a way to get chatbots and stable diffusion to work locally or what os i might need?

[feral radish]

Hello guys, As a senior full-stack developer, I am currently looking for a job opportunity. My skills are React/Next.js, Vue.js, Angular.js, Node.js, Python/Django, etc. Please don't hesitate to ping me if you need a developer for development. Thanks.

[half plank]

Does anyone here are having this type of problem today with paperspace?

could you help me with that?

Yesterday i was using it without problem, but today its like this, dont know what i did

[fervent sinew]

did you update anything?

if not I would try reinstalling dependencies

random non-obvious errors, from experience, are 9/10 due to some dependency being an incorrect version

[half plank]

haven't updated anything

i dont know if were to download the newest version, can you help me with that?

[half plank]

can you?

[fervent sinew]

you're using collab right

unfortunately I don't know anything about collabs. sorry

[half plank]

collab you mean by paperspace?

i dont know much about SD, a friend of mine showed me this and i felt in love

[half plank]

but just learned how to install it on paperspace and it was working until now 😦

can you explain me a bit about it?

[fervent sinew]

paperspace is a collab

essentially some extra code to make it work thru jupyter or some other version that runs in linux

in other words, I don't know how it works. sorry

I only run local

[half plank]

I understand, my friend said that to run local, we need a strong pc, is it true?

[fervent sinew]

depends what you mean strong

I'd say GTX GPU and up

[half plank]

Sen, thanks for help me, it really works reinstalling dependences

[half plank]

Thank you!

[fervent sinew]

Knew it!

[nocturne topaz]

paperspace can do jupyter notebook and standard linux terminal

paperspace is in a world of hurt at the moment, their systems shat themselves 3 days ago and haven't come right yet

[valid flame]

Y this error already I download all files still getting error

This one not installing

Plz help me guyz anyone

[unique bear]

I get this error when trying to merge models Error occurred when executing CR Apply Model Merge:

Error while deserializing header: MetadataIncompleteBuffer

File "D:\AI Projects\ComfyUI\execution.py", line 152, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
File "D:\AI Projects\ComfyUI\execution.py", line 82, in get_output_data
return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
File "D:\AI Projects\ComfyUI\execution.py", line 75, in map_node_over_list
results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
File "D:\AI Projects\ComfyUI\custom_nodes\ComfyUI_Comfyroll_CustomNodes\nodes\nodes_model_merge.py", line 119, in merge
merge_model = comfy.sd.load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, embedding_directory=folder_paths.get_folder_paths("embeddings"))
File "D:\AI Projects\ComfyUI\comfy\sd.py", line 494, in load_checkpoint_guess_config
sd = comfy.utils.load_torch_file(ckpt_path)
File "D:\AI Projects\ComfyUI\comfy\utils.py", line 13, in load_torch_file
sd = safetensors.torch.load_file(ckpt, device=device.type)
File "C:\Users\BenTs\AppData\Local\Programs\Python\Python310\lib\site-packages\safetensors\torch.py", line 308, in load_file
with safe_open(filename, framework="pt", device=device) as f:

[ornate elk]

Hey, yes thats possible. I have 2 AMD install guides. You can find them in the Pinned Messages of this channel.
For chat take a look at GPT4All or LM Studio or Kobold ccp

[valid flame]

execution.py how to enable

[brave nexus]

hello guys ,I have a small question for you
is there anyway to download old auto1111 web interface(nearing 2020. October )?
I am kind of missing the old version of SD 1.5

[nocturne topaz]

the new version of a1111 will still run all of your sd1.5 models just fine

[nocturne topaz]

python execution.py?

[valid flame]

yes

[nocturne topaz]

one of the models you downloaded is corrupted

[valid flame]

how to find wich one

[nocturne topaz]

which what?

[valid flame]

how to fix that error bro

can u help me

[brave nexus]

it's true that new version of a1111 can run sd 1.5 successfully. but as we all know that ,afte sd2.0 . a1111 interface will generate different pictures compare with old version of a1111. many artist style in sd1.5 are forbidden to use

[obsidian bear]

Hi guys, what are the vram requirements for training the 3.7B stable cascade stage C LoRA? I notice they're quite high. Training on the 1B needs about 15GB but the 3.7B skyrockets to 40

[nocturne topaz]

no, as you have given zero context as to what your system is, what the execution.py even refers to, and how you got an error message/what you need it for?

[nocturne topaz]

it shouldn't, there are compatibility settings in the back that you can change to make sure the old ones match

[nocturne topaz]

3.7B parameters, should be doable in 12gb at batch size 1 if what I read is believed

[obsidian bear]

alright, thank you 🙂

[nocturne topaz]

check in over on #🔧｜finetune

[nocturne topaz]

ah ok.. it looks like the ipadapterApplyFaceID node is missing an input image

'NoneType' object has no attribute usually means missing input

[valid flame]

how to fix this

i am new for comfyui

so help me out

[nocturne topaz]

there will be a node on your screen called IPAdapterApplyFaceID....

[valid flame]

yes

[nocturne topaz]

it will have an input of an image

[valid flame]

that only getting erroe

[nocturne topaz]

is that connected?

show us that particular node without the error message

[valid flame]

connected but loading time ther only stuck and getting this error

[valid flame]

ok

please check

[nocturne topaz]

connect the mask from the image to the attn_mask on the Apply IPAdapter FaceID, if that doesn't work, no idea as it looks like that is the sample workflow provided?!

[valid flame]

yes workflow from youtube

[river flame]

my Forge stopped lauching.

I m getting this

and Press any key

this happened after I tested using refiners

[tacit pier]

Hi Team i am using pinokio and installed Stable Fusion web UI, and i am wondering is there a way i can use Control net? thanks

[nocturne topaz]

you installed the a1111 cascade extension?

[river flame]

no I didn`t install anything

[nocturne topaz]

god knows why it is trying to run the cascade decoder pipeline then

[nocturne topaz]

the "File D:\Forge\webui\extensions..." says you do have it installed...

[river flame]

oO

[nocturne topaz]

right.. so whatever you did is trying to use that extension, but that extension requires a different version of diffusers to the standard forge install

[river flame]

I deleted it, now it s just

[nocturne topaz]

whats the last line say ?? Press any key to continue?

[river flame]

yes

[nocturne topaz]

and if you push any key it disappears?

[river flame]

yes

[nocturne topaz]

delete the venv folder, then run webui-user.bat

[river flame]

sorry, what is venv

[nocturne topaz]

Virtual ENVironment

its in the webui folder

its like a sandbox for python

[river flame]

I dont have this folder

[nocturne topaz]

D:\forge\webui\venv?

[nocturne topaz]

cool... run webui-user.bat then

[river flame]

I remember it said the same when everything was working

[nocturne topaz]

whats in your webui-user.bat file?

[river flame]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--theme dark --xformers --medvram-sdxl --no-half-vae

@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%/embeddings ^
@REM --lora-dir %A1111_HOME%/models/Lora

call webui.bat

[river flame]

should I just reinstall Forge?

[nocturne topaz]

get rid of xformers medvram-sdxl and no-half-vae as they aren't used in forge.. if that doesn't work, yeah backup the models folder and reinstall from scratch

[river flame]

ModuleNotFoundError: import of xformers halted; None in sys.modules
Ok reinstalling

[nocturne topaz]

maybe it does need xformers, definitely doesn't need the other two

[river flame]

my 1080ti is not very fast. Can I generate everything in 512x512 and then if I like something, take the seed and generate it in 1024x1024, I will be the same, right?

[wheat sparrow]

Hello 🙂 I wanted to use this https://civitai.com/models/269232 checkpoint. I'm trying to recreate generations but no matter what I provide in the prompt, I get such type of result. Do you guys know what the problem can be? I don't have such results with other checkpoints.

[wheat sparrow]

Nevermind, I've figured it out. I had wrong VAE settings. Take care! 🙂

[indigo anchor]

hello guys...i need some help please...i have rtx 3070/80go vram

i have been using sd 1.5 and it was fine and fast/secondes even with hires fix

but with sdxl is too slow even ehen using --medvram

anythoughts?

[vocal burrow]

Define "too slow"

Sdxl is heavier by design

[indigo anchor]

it takes 3 minutes some time plus for simple promptes .. i use auto 1111 ...but when using comfy ui it is not that bad

so i think it is somethink with SD

[ornate elk]

You need --xformers --medvram-sdxl --no-half-vae

[indigo anchor]

Thank you, i am seeing drastic improvement

[valid flame]

plz someone help to clear this

[short stratus]

it takes time to generate an image due to this msg, I have downloaded some CN models but what is this?

[manic snow]

it's downloading a file that's required to run the script/extension that you're including in your generation. it'll only do it once, depending on your settings, and after it'll continue like normal

[short stratus]

it did that 3 times, and it's still doing it

[manic snow]

Did you change any settings in between? And did you let it finish?

[short stratus]

nope, same, I am just wondering why is it downloading 'nothing' every time (the folder path is empty)

[manic snow]

just give it a while, the files can be over a gigabyte in size, sometimes more

[fierce hollow]

Is it possible to make images on a transparent background? If so, how?

[golden topaz]

hey I want to increase my training and generating time. I am planning to buy some gpu's, cpu's and RAM's. I hace around $100.000 budget. Do you have any hardware recommendations for my request which are to increasing training and generating speed

[solid prism]

hey i wanna know about your project

[solid prism]

could u explain more?

[fierce hollow]

I want to make stickers for Telegram with anime girls (We are discussing, of course) So that there would be a transparent background around the girl. Is it possible to make a transparent background immediately in a neural network?

[solid prism]

just wait for a moment

[fierce hollow]

Any news?..

[ornate elk]

You can use the Rembg extension to remove the background

[fierce hollow]

Where can I find this extension and how can I use it?

[astral copper]

Can i run CCSR in A1111?

[ornate elk]

you can install it via the extensions tab and then use it in the extras tab

[vestal torrent]

Anyone know how long Stability.AI billing takes to respond via email? They won't accept my credit card and I already emailed them

[cosmic furnace]

img2img work with zluda? directml or next?)

[crisp prawn]

trying to fix this:

0.0 seconds (IMPORT FAILED): D:\Ai\ComfyUI\ComfyUI\custom_nodes\comfy_vid2vid

[noble belfry]

is your comfyui up to date?

same question with vid2vid. generally if theres an import error one thing or another is older than it should be and things have been moved around since 🙂

[sweet basin]

can you use zluda and directml at the same time?

I mean install both and use it

[indigo anchor]

hey guys...what version of ipadapter really works for sdxl....because it always get stuck for me

[crisp prawn]

i always run update all in comfy manager

[sullen frost]

Hello can I generate at least 1 image per second with SDXL turbo with a RTX 3050 GPU?

[worn nymph]

ı get this error how can ı fix that

[noble belfry]

yeah dont install into that folder

Literally helping someone else with the same issue in another discord rn 😄

install to C:\ai or even your downloads folder, but a lot of other places cause permission nightmares

[sullen frost]

or do I have to use SD Turbo?

[worn nymph]

ty so much ı am trying rn

[noble belfry]

it should definitely work, i was able to replicate that same error earlier and its definitely an issue with permissions even though it doesnt say so plainly

[ornate elk]

Yes you can (install boath without issues)

[ornate elk]

The sdxl ones from here:
https://huggingface.co/h94/IP-Adapter/tree/main/sdxl_models

[ornate elk]

Hey, no it will take longer than one second

[sullen frost]

If I use this model will it take much longer than sdxl turbo to generate a single image?
https://huggingface.co/latent-consistency/lcm-lora-sdv1-5

[ornate elk]

Thats not a model, its a lora and is for sdxl models to use less steps

Less steps = shorter generation time

[sullen frost]

how do I use a lora?

[hybrid siren]

Apologies if this is the wrong channel to ask this kind of question, but I am trying to use Depth Anything in ComfyUI and I keep getting this error: Any advice here?

[ornate elk]

Here is a quick lora guide:
#🤝｜tech-support message

[sullen frost]

Thank you

[reef smelt]

Hey I'm trying to use SDXL turbo but my generation speed is horrible, between 20 to 50 seconds to generate an image?

I looked at https://www.youtube.com/watch?v=WigWNkxxB_E but they didn't mention any external configurations I had to do

[ornate elk]

hey, whats your gpu and txt2img settings?

[snow lance]

upgrade something and i'm getting a NVML version mismatch now when starting ComfyUI, any ideas?

maybe torch version needs upgrade for 535.161?

[reef smelt]

rtx 2060, 1 step, cfg 1

[ornate elk]

do you use --xformers --medvram ?

[snow lance]

looks like something doesn't support 535.161 that was released last week, not sure how to roll back but i think that's the issue

[ornate elk]

have you updated comfyui too?

[reef smelt]

I didn't try

[snow lance]

few days ago

what's the command to update via command line? just git pull?

yeah, just ran git pull, updated a few things, same error tho, think it's related to the update gfx driver not compat with pytorch

[noble belfry]

roll it back?

[snow lance]

i know how to roll back major versions of nvidia in ubuntu, but the minor version 535-161 to 535-154

[noble belfry]

i can't really help you with linux commands, but im sure others can

[snow lance]

maybe i can get lucky with pytorch nightlies supporting new driver

[noble belfry]

they normally are quite ahead, just not out. and on windows im reliant on binaries 🙂

[snow lance]

will see if i get lucky

[noble belfry]

fingers crossed. 🙂

[snow lance]

thanks 🙂

eep, auto installs the newest cudnn too, remember having an issue with that a few months ago

that worked, yay!

[noble belfry]

excellent! I'm sure we'll get a few others asking the same so ill just point at your chat since you fixed it yourself 😄

(but take the credit) 😄

must be nice being able to easily compile your own binaries. 🙂

i know you can on windows but... you can eat an elephant with a spoon i guess

[fiery island]

Using Stable Swarmui is it possible to do img2img

[old parrot]

https://www.youtube.com/watch?v=A6dQPMy_tHY&ab_channel=OlivioSarikas

Is this still the best upscale method?
img2img SD upscale 2 times, then ESRGAN 4x-UltraSharp 2 times?

[quick forge]

does anybody know where, in ForgeUI config files, that

I can tell it where to look for the models directory.

[snow lance]

do it! 🙂

[noble belfry]

forge\webui\webui-user.bat

@quick forge

for reference mine looks like this

[quick forge]

sorry, on linux

[noble belfry]

`@echo off

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

set A1111_HOME=C:/AI/stable-diffusion-webui

set VENV_DIR=
set COMMANDLINE_ARGS=%COMMANDLINE_ARGS% --ckpt-dir %A1111_HOME%/models/Stable-diffusion --hypernetwork-dir %A1111_HOME%/models/hypernetworks --embeddings-dir %A1111_HOME%/embeddings --lora-dir %A1111_HOME%/models/Lora

call webui.bat
`

ah

[quick forge]

I keep forgetting to specify that

for some reason it doesn't want to use the venv dir i set in webui-user.sh

[pallid kite]

How do I keep the original file name after upscaling through img2img Batch?

[lament wyvern]

Colab links for animate diff and wwarpfusion?

[old parrot]

PNG info send to txt2img doesn't work, terminal says: ValueError: too many values to unpack (expected 2)
Any ideas please?

[gloomy vapor]

hello, i'm a newbea. we usually use what tools to finish our task?

[ornate elk]

Depends what you try to do, most images tasks are done in a local Stable Diffusion webui

[ornate elk]

Make sure to whitelist the webui in any browser adblocker

[gloomy vapor]

thank you , i installed comfy-ui,and run a sample sucesfully, but i didn't know all the components how it works and if i want to create a customrize pciture how can i do

[ornate elk]

Comfyui can be a bit overwhelming in case of where to start or what you can do.
On the comfyui github you can find a list of example workflows. Best is to try them out and test them.
Or you install auto1111 for getting easier into SD

[gloomy vapor]

got it , i will try .

[verbal mirage]

Does anyone know how to fix this?

[onyx solar]

Yes. You make sure the directory and file exist.

It's literally telling you what and where.

[vale sundial]

Are --medvram-sdxl and --skip-ort the only commandline args I should use once Zluda is installed?

[fiery island]

Need assistance with stable swarmui

[dense copper]

hi guys , im 1st time here , may i ask it is possible for i have tutorial foe <How to use this sever> 😂

[ornate elk]

yes, you normaly dont need more

[fiery island]

May I ask what your vram is when selecting the use of this commandline. I have RTX3070 Ti 8gig just asking for self use

[vale sundial]

I have an AMD RX 6600 8GB. Zluda is the only way I can run SD with tolerable speed

[fiery island]

What's zluda

[vale sundial]

#🤝｜tech-support message

[verbal mirage]

I know that, only thing is. I'm not sure where sd_x1_refomer_1_0.yaml is

[onyx solar]

Well, first off, it's sd_xl_refiner_1.0

[fiery island]

Have you tried Stable Swarmui. For pictures only. I use yammersperfectdesign sdxl model, with 8 step sdxl lightning lora (8 Sampling Steps, CFG - 1) 1024x578 resolution. It reminds me of forge. It's super fast

[cosmic furnace]

@ornate elk so have you made some tests ?)

[river flame]

I want to make a picture from a sketch in Forge. As I understood, this should be done with Scribble controlNet, but it has no models. I googled and found only one, diffusion_pytorch_model.bin, but it gives me "TypeError: 'NoneType' object is not iterable". The cmd says, the shapes cannot be multiplied.

I also get this error a lot when I use the default checkpoint

[fiery island]

https://huggingface.co/lllyasviel/control_v11p_sd15_scribble

No models with that. Hmm let me try

[dense sandal]

I was doing a vid2vid in animateDiff and when it stopped diffusing it went error. I did a couple of times now so I don't know what happened this time:

Also don't understand the error message.

using this workflow

oh nevermind I am stupid it has error because facerestore was enabled and I didn't have any input image.

[ornate elk]

you need --xformers --medvram-sdxl --no-half-vae (when using auto1111)

[river flame]

I also found this https://civitai.com/models/287869/scribbleray-sdxl but don`t know how to use it.

[vocal burrow]

It s a lora. Therefore :

LoRa/Lycoris quick guide :
1/ Where do I put my LoRa : stable-diffusion-webui\models\Lora, Loras are NOT stable-diffusion model but rather patches for said models.
2/ How do I use my LoRa, you can either :

• Invoke it directly in your prompt by using the following syntax <lora:filename:weight> (or <lora:alias:weight> depending of the setting in setting->extra network). eg : <lora:animeTarotCardArtStyleLora_v20Offset:1> for using the Lora animeTarotCardArtStyleLora_v20Offset.safetensors
• Use the UI "Extra Networks" / card system by doing as described in my screenshot. Clicking an associated lora card should automatically insert the lora prompt liked mentioned above. https://cdn.discordapp.com/attachments/1002602742667280404/1191779742534348870/loraHowTo17.jpg
  3/ Be careful, LoRa are trained specifically for some SD version. Lora trained for SD1.5 (and derivatives) models will not load on SD2 models and vice versa (same goes for SDXL). Also even though Lora trained on SD1.5 models should work for every SD1.5 derivatives, they'll probably work best with the model that has been used for its training. Therefore, (by default) only Loras compatible with the model you currently have loaded will show up.
  4/ LoRa, LoHa, LoCon, what are these ? They're all commonly refered as "Lora" but differ slightly technically :
• LoHa : Lora with hadamard decomposition
• LoCon : Lora extended to redisual blocks.
• More explanations : https://stable-diffusion-art.com/lora/, https://www.sdcompendium.com/doku.php?id=lora_0014, https://www.sdcompendium.com/doku.php?id=locon_training_0071, etc

[river flame]

thank you so much

When I start Forge, it says 'CUDA Stream Activated: False' in cmd, is it how it should be or should I do something? I have 1080ti.

[vocal burrow]

it's just a debug log, you can force activate it with --cuda-stream

cf https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/395#discussioncomment-8580475

[river flame]

thanks

[ornate elk]

test of what?

[river flame]

I ve set exactly the same settings as the guy on the video guide and I still cant get anything close to my sketch (his results are perfect): Canny + IP-adapter (for reference). The only difference is he uses A1111 and I use Forge. Can this be the problem?

[river flame]

I ve installed A1111 like him, I m using his old checkpoint, his sampler, everything. Nothing works

[noble belfry]

do you have a trailing slash somewhere?

WAIT

you literally said that haah, sorry

[old parrot]

I whitelisted webui in the adblocker, still getting error ValueError: too many values to unpack (expected 2)

Any more suggestions would be appreciated

[sweet basin]

what to do

zluda error

[quick forge]

does anyone in here use ForgeUI?

[old parrot]

I removed everything

time to try a different browser, Edge is trash

Rather try to fix the issue, works on my other PC thru Edge with no problems

Lol trying to fix an issue instead of a nooby workaround is hardly stubborn.

Icognito didn't work either so

[ornate elk]

ahh okay, thanks i will edit it

[ornate elk]

you didnt replaced the files maybe?

[sweet basin]

what files

[old parrot]

I only tried to prove it wouldn't work

[ornate elk]

the zluda files, like stated in the guide

[sweet basin]

I did everything as you said

but it seems different today

maybe you edited

[ornate elk]

yes i updated it

should be easier now

[sweet basin]

I did every step I think

[ornate elk]

whats your gpu?

[sweet basin]

6800

[ornate elk]

XT ?

[sweet basin]

normal

maybe path part incorrect

[ornate elk]

can you show a screenshot of the paths?

[sweet basin]

top or bottom path

[ornate elk]

bottom

[sweet basin]

wrong then

nah still the same

dlls wasnt replaced

[ornate elk]

does it start now?

[sweet basin]

no

RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check
Press any key to continue . . .

[ornate elk]

can you show the path?

of the system variables you added

also run this in a cmd:
where hipinfo.exe

[sweet basin]

this?

C:\Program Files\AMD\ROCm\5.7\bin\hipInfo.exe

[old parrot]

Reinstalling SD fixed it, thanks anyway

love you

[sweet basin]

this ok?

do I need to remove dlls from zluda folder

with 64 names

will try classic remove venv folder trick

[ornate elk]

no dont remove them

you need to copy and place them inside the torch/lib folder

[sweet basin]

I did

[ornate elk]

np, wow thats a nice increase!

[ornate elk]

ist your gpu driver updated?

you nned at least 24.1.1

[sweet basin]

24.2.1

wait maybe venv folder removal will fix

[ornate elk]

if you do that you then need to copy the dll files over again

dont forget that

[sweet basin]

yeah

oh now it is different

before dll files wasnt there

now it is and it was replaced

looks promising

hmm it was fixed btu got another problem

[deep canyon]

Hi guys

[deep canyon]

I'm trying to take a picture of a car and then sort of have it redone in an art style

does anyone know how one might do this?

[ornate elk]

Thats not a problem

[vocal burrow]

you can ignore the "style database not found" thing. It just means you don't have any "style" registered yet

[sweet basin]

yeah waited for a while and it loaded up

trying to make something nothing happening tho

[vocal burrow]

Can be done with inpainting, pix2pix, ip-adapter, etc. Pick your poison

[sweet basin]

hmm doesnt generate anything

[vocal burrow]

what does the log says ? what settings ?

[ornate elk]

It will take 15-40 minutes for the First image

I will add that to the guide

[sweet basin]

so I need to wait?

[ornate elk]

If you clicked generate and the button is still Gray then yes

[sweet basin]

yes it is

[ornate elk]

Then watch a 30 minutes YouTube video or so ^^ and let it load

Dont interrupt it

[sweet basin]

that will happen only once right ? 😄

[deep canyon]

im a noob

[ornate elk]

Yes once, but if you update Zluda files (the DLLs) then it will load again one time

[sweet basin]

--ckpt-dir "Path/to/models/stable-diffusion" --lora-dir "Path/to/models/lora" --controlnet-dir "..." --embeddings-dir where to add this

COMMANDLINE_ARGS= ?

[vocal burrow]

eg : https://stable-diffusion-art.com/instruct-pix2pix/

[sweet basin]

you should write COMMANDLINE_ARGS= cuz it will be unclear for most of people

[deep canyon]

oh okay

yeah ive only used auto1111 before

[vocal burrow]

there are instructions for auto1111