🤝｜tech-support

[ornate elk]

Its in the venv

[meager prawn]

i still get the same error after i uninstalled all and only have 3.10.11

i mean this error

[ornate elk]

You need to delete the venv folder again roo

[meager prawn]

ok

[ornate elk]

And delete the .zluda folder

In the stable-diffusion-webui folder

And then you need to reboot the PC too

[meager prawn]

ok i did the 2 steps you told me
should i start webui-user.bat with the arguments --use-zluda

ok i restard now

[ornate elk]

With the ones from my guide

--use-zluda --skip-ort --update-check

You can also add --medvram-sdxl

[meager prawn]

ok wait i add these

[ornate elk]

Afk

[meager prawn]

i get the same error

this is the full log

[median geode]

no xformer?

and also fail to load model

[ornate elk]

Have you replaced the Library files again?

[ornate elk]

What's your GPU?

[median geode]

4060ti 16g

[meager prawn]

Oh no sry I did not think about that

so it is now generating but then stops at 18 of 20 steps and is stuck for like 3 seconds

is there a way to fix it or is it just normal

do i need to enable something in the ui settings
i mean something like enable olive or is it just something not as important

[median geode]

hello cs1o, the models can be loaded now but still no xformers,

[ornate elk]

Nope. Its a visual bug for 1.5 models

Try an sdxl model

[meager prawn]

ah ok thank you

[ornate elk]

Also whitelist the webui in any browser adblocker

[meager prawn]

is there any more optimization?

[meager prawn]

so ublock origin is blocking some things?

[ornate elk]

There is an extension called Tiled VAE for easier upscaling

[meager prawn]

ah ok thank you

do you have any more tips?

[median geode]

Does “no module xformers “ matter ?

[ornate elk]

If you only have 16gb of ram you should increase the Windows Pagefile

[ornate elk]

Yes

[meager prawn]

I have 32

[ornate elk]

@median geode You need to add --xformers to the webui-user.bat

[agile dragon]

besides a slower output generation, having a "low" (8GB) amount of VRAM won't have any other side-effects, right ? (e.g., won't generate less qualitative images) ?

[ornate elk]

Nope

[agile dragon]

@ornate elk "nope" as in "correct" ? 😄

[meager prawn]

does the a1111 support xl modells because it is starting an 10 gig download

[ornate elk]

Yes

Thats only the first time for the clip model

[meager prawn]

ah ok thank you

do i need to download 10 gigs for other models as well?

[ornate elk]

Nope

[meager prawn]

ah ok thank you for all the help

would you say if the nvidia 5060 is going for aroung 400 euros it is good for ai?

do they use less vram because it is native cuda?

[ornate elk]

A 5060 isnt released yet

[meager prawn]

ik but if they are released

[ornate elk]

If it has 12gb or more it would be a good card for ai

[meager prawn]

and if it has only 8 gigs then should i go for something older from nvidia or is amd and the rocm on windows with newer cards worth it

[ornate elk]

For Ai stuff better go for nvidia.
But if you like Linux or focus more on gaming and price, AMD is a good choice

[meager prawn]

I could dual boot Linux

Could I then get close to Nvidia ai performance or is it just that amd is better for Linux bc or drivers

[ornate elk]

Yea AMD on Linux can use ROCm drivers natively so much better performance

On windows we can only use DirectML or Zluda but we could also install WSL2 to use Rocm but its more complex

[meager prawn]

Can it run something like the things zluda can not run?

[meager prawn]

On Linux

Wait can the Rx 6600 use rocm under Linux?

[ornate elk]

Yes

[meager prawn]

So if I use an 1 TB SSD I could run SD much faster than with zluda?

Does it use less vram somehow bc I think I ran into lack of memory when using SD xl

[ornate elk]

Maybe 15 - 20% faster

[ornate elk]

Zluda uses less memory than rocm

[meager prawn]

So sd xl would be slower?

[ornate elk]

Nope

But it would use more ram and pagefile maybe

How fast is sdxl now?

And at which resolution do you use it?

[meager prawn]

I did it at like 512x512 and I think it was around 40 sek

But I can check tomorrow because I have tomorrow my exams so I am need to sleep

But thank you for all the help today

[tough karma]

Hello everyone!

[ornate elk]

Okay, no Problem and good luck!

[delicate citrus]

guys, someone knows a way to use a monitor + monitor ultrawide at same time on a table? some structure, support; can be fixed on the table or wall too. If have a link of the support or a picture, video how it works will be great too

[warped ginkgo]

I don't think I've changed any settings

With the same prompt, same settings, same scale, some months ago as far as I can tell

Mangled merge v3 is the model

[distant plover]

How slow can you go? First generation after loading, if I change anything, and every so often, it takes 10+ minutes to generate an image. Then it will be fast, around 9 seconds for an image, but do a slow one every now and then. No errors. What could be the issue?

[tight depot]

Is there a badhandsv4 equivalent for sdxl that people like to use?

[distant plover]

No need...it does hands pretty good, and it's easy to fix with inpaint if the hands are bad

[tight depot]

i do large batch images and have to delete quite a few for bad hands, but ill try without negative prompts

[distant plover]

yes, it does make many with bad hands

But I think it gets enough with good hands that you have some to choose. Also, if you find one you like except for the hands, inpaint to fix it

[tight depot]

will check that out thanks

[median geode]

i have add one, but doesn’t work

[autumn stag]

ia am using flux redux tool and when i load the vision clip checkpoint i get this error :

i dont know what means

[woeful lion]

im trying to run webui-user but i get this error

i dont even have python 313 installed anymore, i downgraded because torch was failing to install

how do i make it use python310 instead of python313

[vocal burrow]

1/ don t put spaces in your folder names
2/ delete your venv folder and try again

[woeful lion]

thx man, trying it rn

[autumn stag]

i updated comfyui and it works

[woeful lion]

'"git"' is not recognized as an internal or external command,
operable program or batch file.
Traceback (most recent call last):
  File "D:\StableDiffusion\stable-diffusion-webui-master\stable-diffusion-webui-master\launch.py", line 48, in <module>
    main()
  File "D:\StableDiffusion\stable-diffusion-webui-master\stable-diffusion-webui-master\launch.py", line 39, in main
    prepare_environment()
  File "D:\StableDiffusion\stable-diffusion-webui-master\stable-diffusion-webui-master\modules\launch_utils.py", line 411, in prepare_environment
    git_clone(assets_repo, repo_dir('stable-diffusion-webui-assets'), "assets", assets_commit_hash)
  File "D:\StableDiffusion\stable-diffusion-webui-master\stable-diffusion-webui-master\modules\launch_utils.py", line 192, in git_clone
    run(f'"{git}" clone --config core.filemode=false "{url}" "{dir}"', f"Cloning {name} into {dir}...", f"Couldn't clone {name}", live=True)
  File "D:\StableDiffusion\stable-diffusion-webui-master\stable-diffusion-webui-master\modules\launch_utils.py", line 116, in run
    raise RuntimeError("\n".join(error_bits))
RuntimeError: Couldn't clone assets.
Command: "git" clone --config core.filemode=false "https://github.com/AUTOMATIC1111/stable-diffusion-webui-assets.git" "D:\StableDiffusion\stable-diffusion-webui-master\stable-diffusion-webui-master\repositories\stable-diffusion-webui-assets"
Error code: 1
Press any key to continue . . .

do i have to install git

gonna try and see ig

[neon orchid]

Guys, I have a new PC.
I have 12gb vram - Nvidia
Is there a guide for stable diffusion?

[paper dagger]

go to huggingface and look at the stable-diffusion 3.5 model.

youll need to use git lfs i think as well.

[ornate elk]

Make sure you always start the webui-user.bat and not the webui.bat

[ornate elk]

In the pinned messages of this channel

[median geode]

yes i am

[median geode]

but no use

[median geode]

is it bcz torch version requirements don't match?

[neon orchid]

this?

The latest Stable Diffusion install Guides for Nvidia and AMD:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[ornate elk]

Maybe

You can share the cmd log

[ornate elk]

Yes

[neon orchid]

ohhh oksuy

okay*

So... is Zluda, right? o.o

[ornate elk]

Zluda is for AMD

[neon orchid]

oh

[median geode]

this one?

[ornate elk]

@neon orchid You need to check the top guides for Nvidia

[neon orchid]

oh yeah

my bad

ty 🙂

[ornate elk]

No problem 🙂

[ornate elk]

Yea if you have a good internet connecting you could try delete the venv folder and relaunch

[neon orchid]

a question

Compared to Zluda.
What is the difference with [NVIDIA] Automatic1111 Webui (Stable-Diffusion-Webui)?

curiosity more than anything o.o

[median geode]

still the same

[median geode]

do we have any other methods?

[median geode]

finally find out what' s goin' on. my old webui folder was still in bin, so the --xformers --no-half-vae didn't work out in editnote💀 after i deleted old webui totally, everything was fine

[median geode]

can u help me check out if the cmd goes well now?

[mortal compass]

Looks healthy to me

[median geode]

thx fr

[mortal compass]

Is anyone here experienced with pytorch and cuda? I recently upgraded GPU from 3070 ti to 4070 ti super, and I was told that removing my old 2.1.2+cu118 versions and installing newer ones might increase my computing efficiency. After doing some googling, I'm not sure if that's true or not, so I'm a bit hesitant.

[neon orchid]

hey guys

a question

i try to use

"Pony Diffusion"

but in all draws...

This comes out, what could be happening?

[mortal compass]

Alright, thanks:) I'm currently asking chatgpt how to uninstall and install them again, should get it right

[mortal compass]

What are your parameters

[neon orchid]

what

[neon orchid]

parameters?

what is that? o.o

[mortal compass]

Parameters are the different sections in your txt2img UI which control the output, for example Steps, CFG, which checkpoint you have loaded, and so on

[neon orchid]

this?

sorry i'm very bad in this xD

[mortal compass]

It's okay

Yes exactly. Send this again but the full webUI screen

[neon orchid]

ohh okay

my bad

sorry reset PC

idk why fail 😦

[mortal compass]

Have you tried with other Samplers like Euler A?

[neon orchid]

mmm nope

Nothing

[mortal compass]

@vocal burrow mods ples

[mortal compass]

it looks like a vae issue to me. Put this in your vae folder, refresh your vae dropdown and try that one
https://civitai.com/models/296576/sdxl-vae

[neon orchid]

kk

i try

[neon orchid]

Ohhh ty bro!

[mortal compass]

Got a better output now?

[neon orchid]

yeah

0 static 🙂

[mortal compass]

When images gets distorted in that kinda way, like static, extreme miscoloring, looking like mosiac or things like that, it's usually a vae issue

Anytime, happy generating

[ornate elk]

Looks good now

[mortal compass]

Ah shit, this was more confusing than I hoped for. Chatgpt recommends me to install Cuda 12.1, but the newest version is 12.4, so why not that one? When I checked cmd nvidia-smi it even said it supports cuda 12.7 o.O wat

[ornate elk]

Wrong VAE

[mortal compass]

We solved all that already 👌

[ornate elk]

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

It should download the 121

You don't need to install any cuda version by yourself

Only make sure the nvidia driver is updated

[mortal compass]

Oh so I need to do it with my comfyui venv activated?

[ornate elk]

Comfyui venv uses 2.3.1

So nothing to do

If you installed it recently

[mortal compass]

Awe okay so then it might not solve my problem after all 😩 I'm troubleshooting cause I think I have RAM leakage when I'm using more powerful models like flux or image2vid. It stays on 70-80% usage after one generation and stays that way until I close cmd

[ornate elk]

If you use Flux make sure to use the T5 GGUF text encoder.
It reduces RAM usage by 10gb

[ornate elk]

Sometimes you also need to reinstall the Nvidia driver because torch might think you still have only 8gb vram

After a GPU switch

[mortal compass]

I'm using google_t5-v1_1-xxl_encoderonly-fp8_e4m3fn currently, so I'll definitely try the T5 GGUF

I take it that its the 8-bit model

Appreciate it man, crossing my fingers

[meager prawn]

so i have 32 gigs of ram but python is using only around 5,2 gigs and my whole system uses 15,1 gigs
my vram is at 8 at all times when i use --medvram-sdx or what it is called

=--use-zluda --update-check --skip-ort --medvram-sdxl these are the arguments

[untold mauve]

hi, I reinstalled zluda yesterday with the rx 6600 and it worked well, but compared to the version I downloaded in June it is much slower to generate images and sometimes it stops at 99% and no longer works... is this normal for this version? before it took me 5 minutes for an image, now it takes like 8

[mortal compass]

Let me see if I got this correct; by "T5 GGUF text encoder" you mean that I should use t5-v1_1-xxl-encoder-Q8_0, which is put in the clip folder, and is used in the Clip node? Or so I thought, but on huggingface they say that I should put it in models/unet. But neither of those folders are picking up the text encoder, only my clip-l file. However, it is showing up in my model node when its put in the unet folder. I believe I must have missed a detail or two xd

[meager prawn]

oh and when i use hiresfix the image in sd 1.5 takes around 2 minutes

[iron lantern]

any clue on what does the score_8 etc and score_9 mean when prompting thanks

[ornate elk]

Whats in your webui-user.bat?

[mortal compass]

They are basically "quality triggers". You use those only with pony models

[ornate elk]

Goes into clip but you need the gguf extension

For triple gguf clip

[iron lantern]

what are they used for need to learn why they are there

[ornate elk]

Seems normal

[untold mauve]

--use-zluda --update-check --skip-ort

[ancient thistle]

hello there, usually i use a Stability Photoshop plugin to upscale layers, but now I am receiving the error "The specified engine ( ID esrgan-v1-x2plus ) was not found"

• usign photoshop 26.1, 64 Mb RAM, 3070 RTX, any clues ? txs

[ornate elk]

Add --medvram-sdxl for better sdxl and pony speed

[mortal compass]

Tried out the GGUF loader; the loader itself worked, but I'm still running into same issue. When I start the upscale generation, the system ram spikes to 87% and stays there until I exit cmd. Even when I close everything else and I'm not generating, it stays on 87%. That's why I thought it was about ram leakage.

Also, got an error that I just now noticed

[untold mauve]

but i don't use sdxl or pony, i just use 1.5

[ornate elk]

you cant upscale flux, for that you would need a 4090 or better i guess xD

maybe it only can work with tiled upscale from the ultimate upscaler node

[ornate elk]

oh okay. but you can still add it, so you can use them too

[untold mauve]

mh but i still don't understand if 8 minutes for an image is normal

[ornate elk]

nope, what are your txt2img settings?

[mortal compass]

Jesus okay xd I have upscaled with flux before though, but I guess this workflow is just too heavy then. That's too bad cause the image output is absolutely amazing:(

[untold mauve]

vae 84000 and clip skip 2, with this settings in old zluda was soo faster

adetailer on or off is same speed

[ornate elk]

set hires steps to 10

and make sure wallpaper engine or any other game isnt running in the background

[untold mauve]

isn't there much difference between hires 10 and 20?

[ornate elk]

nope

[untold mauve]

okay thanks, I'll see how it goes

sorry for my english

[ornate elk]

your english is good

[maiden copper]

😔 what to do?

[ornate elk]

whats your gpu?

and what model did you tried?

[maiden copper]

gtx 1650

[ornate elk]

then you need to edit the webui-user.bat
at the line Commandline_ARGS= you need to add: --xformers --medvram --no-half
then save and rrelaunch

[maiden copper]

okay thanks

[maiden copper]

is it over for me?

or is there any other way i can do it

[ornate elk]

what resolution and model do you try to use?

[ornate elk]

ah you try to use pony

[maiden copper]

😔

[ornate elk]

with your 4gb vram gpu its not gonna work in auto1111

but you can install forge webui

there it can work

[maiden copper]

I’m very new to this, I thought it looked cool so I downloaded the model

[maiden copper]

What’s that?

[ornate elk]

its an alternative version of auto1111 with a better optimisation for lower end gpus

it looks the same

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

[maiden copper]

Ohh

Install process the same?

[ornate elk]

its easier, you just need to download the zip

[maiden copper]

Oh thanks

[signal sigil]

Hey man. I was wondering if you could answer me this qustion?
Is using the "flip argument" for flipping images when training a LoRa on people bad to do?
I've heard from somewhere that the Flip Argument has negative impacts on the results, when training on humans. Is this true?
If so i just wont use it when training on people. Thx

[ornate elk]

hey, sry cant help much when it comes to training

[signal sigil]

Np, np! I just didn't want to experiment with the settings, and commit to training a LoRa that would likely come out badly. Then wondered if anyone hear knew the answer to it. But it's ok m8. Thx for the reply!

[ornate elk]

np, but i heard too that flipping images isnt as good as using more images

[signal sigil]

Yeah. Same here. At the end of the day i just need better Data i think. Haha xD

[rain hare]

how to change the type of sliders???

[ornate elk]

following the pinned install guide in this channel

[ornate elk]

move forge to an other folder that is not in OneDrive Sync

[ornate elk]

whats your webui?

[rain hare]

automatic1111

[ornate elk]

Automatic1111 webui, but if you have a gpu with 4-8gb vram you may want Forge wbeui

[ornate elk]

can you show your versions, at the bottom of the webui?

[rain hare]

this?

[ornate elk]

yes, your torch version and xformers is outdated. but that doesnt affect how the sliders look.

for the sliders make sure you have whitleisted the webui in any browser adblocker

[rain hare]

I whitelist the site, but nothing changed

[ornate elk]

okay, which browser do you use?

and other extensions like dark reader installed that can affect it?

[rain hare]

Opera GX

[ornate elk]

i mean browser extension

[ornate elk]

what you can try is to open up a cmd and type
pip cache purge
then delete the venv folder.
then relaunch the webui-user.bat
it will recreate a clean venv with updated torch and xformers.

If that doesnt help you can try an other Browser like Firefox

[chilly knot]

CS1o, remember I talked about cuda running out of memory - I tried deleting the venv folder (doesn't help)
After one generation, the GPU memory is completely occupied and is not freed, what else can I try?

[ornate elk]

what was your gpu again?

and which webui

[chilly knot]

1050 ti🙂

set PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.6,max_split_size_mb:32
set COMMANDLINE_ARGS=--theme=dark --xformers --lowvram --no-half-vae

[ornate elk]

with forge webui or auto1111 ?

ah i guess a1111

With your card you should try using Forge webui. as its better optimised

[chilly knot]

it works before with medvram

i think about forge now

how you think space on diskdrive not very important?

i have 8 gb free

system drive

[ornate elk]

You should have 15gb free on C at least

[chilly knot]

I'm very interested in what memory consumption looks like normally after generation, should it take 3.8 out of 4 gigabytes in idle time?

or normally the memory should return to its original state

[charred rover]

Please help, I`ve been trying for a hour.

To get the new LTXVideo in ComfyUI working

I ve downloaded the two files and the workflow, and got 3 nodes missing:

LTXVConditioning

EmptyLTXVLatentVideo

LTXVScheduler

'Install missing nodes' did nothing.

Installed LTXVideo custom node - didn`t help.

Error:

"[Inspire Pack] IPAdapterPlus is not installed.

[ERROR] An error occurred while retrieving information for the 'LTXVCLIPModelLoader' node."

But I do have the Inspire pack and everything is up to date.

Cloned the folder to text_encoders folder. Doesn`t work.

removed symbolic link to loras folder, added to config:

{
"Comfy.NodeLibrary.Bookmarks": [],
"Comfy.InvertMenuScrolling": true,
"pysssss.SnapToGrid": true,
"Comfy.UseNewMenu": "Top",
"Comfy.MenuPosition": false,
"Comfy.Queue.ImageFit": "cover",
"Comfy.NodeSuggestions.number": 40,
"models": {
"loras": "C:/Users/HYPERPC/AppData/Roaming/krita/ai_diffusion/server/ComfyUI/models/loras",
"text_encoders": "D:/Comfy/ComfyUI/models/text_encoders",
"motion_modules": "D:/Comfy/ComfyUI/models/motion_modules"
}
}

basically tried every idea that DeepSeek had

(I do have Inspire pack)

[ruby yoke]

did you read comfy's blog post, download his example workflows AND also follow the instructions on the announcement in his discord for what else to isntall, where to get it from, and where to put it?

[charred rover]

I didnt see any discord posts, which discord do you mean? But I saw long form instructions that also included cloning the PixArt-XL-2-1024-MS folder to text_encoders for img2video, this didnt help too

[ruby yoke]

comfy's discord.

[charred rover]

the only difference was on github they said:
'If you don’t have it already downloaded you can download the t5xxl_fp16.safetensors file and put it in your ComfyUI/models/text_encoders/ folder.'
and on discord - to 'clip' directory.

this didnt fix it for me.

[charred rover]

actually the symbolic link error still persists although I removed it from run_nvidia_gpu.bat

when I press Update all

so I have a lower version even though it says Everything up to date

oh not everything but all extensions

[main rune]

Hey @ornate elk, I've decided to move to SwarmUI
Thanks for your previous help, and hopefully my experience with installing SwarmUI goes better than with Forge

[main rune]

But if you're here later, can you tell me if SwarmUI installs Comfy backend by itself or do I need to install it too later?
Want clarification since this showed up

[fair oxide]

how do i open stable diffusion

i opened it with webui-user.bat yesterday but now it doesnt work

venv "D:\StableDiffusion\stable-diffusion-webui-master\stable-diffusion-webui-master\venv\Scripts\Python.exe"
fatal: not a git repository (or any of the parent directories): .git
fatal: not a git repository (or any of the parent directories): .git
Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr 5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Version: 1.10.1
Commit hash: <none>

[main rune]

maybe try rebuilding the venv?

[ornate elk]

Hey, if your on AMD you need to install the Comfyui Backend (with Zluda) yourself.
If your on Nvidia Swarm does it for you.

[ornate elk]

Try run the update bat and then the run bat

[ornate elk]

You installed it the wrong way by downloading it as zip.
You need to follow the install guide from the pinned messages

[maiden copper]

which bat

[neon orchid]

guys

I'm using VAE for Pony Diffusion but...
it keeps giving me erroneous eyes 😦
Does anyone know what could be happening?

[main rune]

try a higher resolution

[neon orchid]

1024x1024

i use

[main rune]

huh

any loras?

[neon orchid]

Yeah

but is character Lora

[main rune]

hmmm, I'm honestly still new-ish myself, maybe try messing with the strength of the lora or CFG

or a different VAE

[median geode]

hello guys, why i cannot find my embeddings and lora in webui, i've put them in the correct files already😭

a lot of embeddings but only one can be seen

[pearl musk]

Hi people, I am using sd-webui-reforge, there is an checkbox, located in Settings-Stable Diffusion, called "Enable quantization in K samplers for sharper and cleaner results." What does this do? In some models I've noticed a drop in quality, rather than improvement, and documentation online does not seem to answer what does this parameter do.

[ornate elk]

Can you show an example

[paper crag]

this is just something minor, but for some reason whenever i type a name in lora search bar, it also showcases models that don't have in name what i type

example, i type 'style'

it also showcases lora models that don't have 'style' in their name

any suggestion how to fix this?

been looking through settings, but can't seem to find something that can resolve this

[vast crest]

hello,i found a good tutorial on how to create a Lora using comyui.but unfortunately i always get errors.i share the error with you,i hope you can help me

[fair oxide]

Relay your issues here for instant assistance

discordapp.com/invite\54SP96qQnB
@vast crest
@vast crest

[ornate elk]

hey, best is to create Loras with extra standalone Tools that are made for that purpos.
Like Khoya_ss or OneTrainer

[ornate elk]

because you have an sdxl model loaded and the not showing embeddings will only work with 1.5

[vast crest]

Hi, I've never heard of these 2 programs. For a beginner, which of the 2 you wrote me do you recommend? If so, please share a tutorial 🙂 thx

[ornate elk]

whats your gpu again?

for beginners both are okay
LoraGuide for OneTrainer:

https://civitai.com/articles/4789/lazy-lora-making-with-onetrainer-and-ai-generation

[vast crest]

rtx 4070 oc 12gb

[ornate elk]

thats good

also stay away from these ticket support scammers

they arent related to this discord

[fair oxide]

anyone here can train lora?

[chilly knot]

cs10 CUDNN greatly affects memory consumption?

its for very good gpu only?

[ornate elk]

Yep it won't help your card

[chilly knot]

if i install it -problem or I can leave it and it won't affect much

[ornate elk]

It could be incompatible

Just use the latest nvidia Driver

[chilly knot]

in new Nvidia driver hew function- System Memory Fallback, i can use ram if no enough vram

i think it help me

[ornate elk]

Yes

That can help

Using fallback

[chilly knot]

is manual how off this function 🙂

i get screen from it

for me i need enable this

but I thought that this functionality was enabled by default and wondered why such an error when there is not enough RAM🙂

[ornate elk]

normaly it should be enabled by default

[chilly knot]

this function enabled in 546.01

my version 530

i hope my gpu support it

[halcyon vessel]

SD 3.5 doesn't load on my automatic1111 directml setup, whereas 1.5 and 2.1 worked (altough with poor results):

File "C:\Users\RainI\Desktop\Stable-Diffusion-DirectML\venv\lib\site-packages\torch\nn\modules\module.py", line 2041, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for LatentDiffusion:
size mismatch for first_stage_model.encoder.conv_out.weight: copying a param with shape torch.Size([32, 512, 3, 3]) from checkpoint, the shape in current model is torch.Size([8, 512, 3, 3]).
size mismatch for first_stage_model.encoder.conv_out.bias: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([8]).
size mismatch for first_stage_model.decoder.conv_in.weight: copying a param with shape torch.Size([512, 16, 3, 3]) from checkpoint, the shape in current model is torch.Size([512, 4, 3, 3]).

[ruby yoke]

are you at the latest level of auto1111?

[halcyon vessel]

very unlikely, i set this up like 6 months ago using the directml fork so i'll have to try and update

[neon orchid]

for example...

[oblique walrus]

automatic1111 afaik only supports sd3. not 3.5 or flux. and barely 3

[ornate elk]

What's your GPU?

For AMD you should checkout my AMD install Guides

[oblique walrus]

probably won't help since they're trying to use 3.5

[ornate elk]

Yep xD

[chilly knot]

cs10 CUDA System Memory Fallback in new nvidia driver help!
now this stop cuda out of memory😎

i can generate like before without this error

[halcyon vessel]

I have a radeon RX 5700, i dreaded updating the repo because setting this up last time was a nightmare and i've yet to get it to work, right now it's just erroring out with "[F1123 21:17:49.000000000 dml_util.cc:118] Invalid or unsupported data type BFloat16." followed by a crash

trying to load the older SD models gives "RuntimeError: resource deadlock would occur: resource deadlock would occur"

[ornate elk]

Ah then your best of using the zluda version of Auto1111. Or Forge Webui with Zluda. It has a much better performance than using directml

But SD 3.5 only works in comfyui right now

[main rune]

comfy backend in swarm is not loading right, could this be the issue? I'm using zluda

[tulip mesa]

Anyone else had experience of getting the same output image all the time in OneTrainer? It doesn't seem like it wants to train (using the SDXL pre-config

Like it doesn't train on my concept images at all

[main rune]

have not tried training yet so I don't have any idea

[ornate elk]

Yes maybe

My swarm guide covers the correct setup

You probably installed swarm with directml before

[main rune]

never

kinda baffles me tbh

[ornate elk]

You can delete it and rerun it

Just select the nvidia install and not AMD

For swarm

[ornate elk]

Also does comfyui work ?

[main rune]

did succesfully load, didn't try it though

I've copied a model into the checkpoint to test it, it's currently running but as usual will probably take time as a first run

[main rune]

ah man, I got the numpy error

WOOO

[main rune]

@ornate elk it seems to be working now, I'm just not sure how to see if it was using the comfy backend, but that I can probably experiment with myself

[main rune]

worked! thanks CS1o

[astral falcon]

I'm trying to train an embedding on my art, but I don't think it's processing correctly...it's 73% done and the images it's generating look like this:

Is that normal? Also all the images in the folder are just straight grey

[tiny garden]

I got A1111 back running, but now I'm lacking all those preview images for the LoRA's I have... How do I get them again from CivitAI?

[ornate elk]

With the civitai Browser+ extension

[carmine ibex]

"GroundingDINO install failed. Will fall back to local groundingdino this time. Please permanently switch to local groundingdino on Settings/Segment Anything or submit an issue" how can I install GroundingDINO?

[carmine ibex]

Even when I switched the settings it told me to, I'm getting OOM. This should not be happening; I'm trying to generate a mask on a 1536x1536 sized image. I even did it before but idk why it insta OOMs me

[pastel kettle]

how to fix?

using sdxl

[ornate elk]

Auto1111 or Forge Webui?

[dense seal]

Does anyone have any idea why whenever I select any “checkpoint” type XL or other than SD1.5, it just... crashes? I've just done the installation with this guide an hour ago or so) : https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides#amd-automatic1111-with-zluda (RX 7900 XT & Ryzen 7 7700X)

[ornate elk]

hey, can you shwo the full cmd log?

and how much normal RAM do you have?

[earnest jackal]

how do I get clip skip?

[ornate elk]

Go into Settings, User interface, Quicksettings, and there add Clip_stop_at_last_layer
Then hit apply and reload ui.

[earnest jackal]

got it thank you

[hidden matrix]

is anyone available to help me create a lora ive tried every guide and spent nearly all day on forums and chat gpt to try to fix errors but i cant get it to work

[cedar oxide]

which model? FLUX or SD? please let me know

[hidden matrix]

SD

@cedar oxide

[cedar oxide]

if so, what is the problem?

SD version?

[hidden matrix]

so im using this guide https://colab.research.google.com/github/Linaqruf/kohya-trainer/blob/bc0892647cb17492a106ad1d05716e091eda13f6/kohya-LoRA-dreambooth.ipynb?authuser=1#scrollTo=PtjZHdUcayeJ but once i reach the end of 5.3 it just gives me an error

sd verison v1.10.1

[cedar oxide]

Use this github
https://github.com/cloneofsimo/lora

[loud cipher]

I have a 5700XT. I downloaded ZLUDA using this guide https://github.com/vladmandic/automatic/wiki/ZLUDA#install-zluda and when I generated an image, it took over 5min to even initiate, it didn't even start generating an image so I gave up. I know AMD's especially older GPU's have it hard but is there any alternative for my GPU on windows? Maybe I used the wrong guide/method? Maybe there's a different Zluda guide. I'd appreciate any help. Thank you

[loud cipher]

Any help from anyone would be appreciated. Thank you

[ornate elk]

Hey, yea you used the wrong guide and webui

Best is that you install Automatic1111 or Forge Webui with ZLUDA for that card

Using my Guides

https://github.com/CS1o/Stable-Diffusion-Info

And make sure you get the Rocm HIP SDK 6.1

And uninstall the 5.7 if you have it

[loud cipher]

Thank you so much!!

[ornate elk]

No problem, if any step isnt clear feel free to ask

[neon orchid]

Guys, does anyone know a way to make the AI ​​make two different characters in the same image?

[ornate elk]

Yes thats possible with the Regional prompter extension

[neon orchid]

question and can that extension be used with two Loras?

o.o

[ornate elk]

Yes

Ive used it with 4 different chars too

[neon orchid]

ohhh

Sorry @ornate elk But... could you... show me an example? 😦

I'm looking for videos or guides but I can't figure out how to correctly implement the parameters for "Regional prompt"

😦

[vocal burrow]

Some references, those are not the same extension. But the idea remains, one "common prompt" + one prompt for each region. Each prompt being seperated by some specifc word in capslock. :

• https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111?tab=readme-ov-file#regional-prompt-control
• https://github.com/hako-mikan/sd-webui-regional-prompter
• < https://stable-diffusion-art.com/regional-prompter/>

[uncut jewel]

Heyo is there a way to found backup workflow from a few day session on ComfyUI ?
I did create on an experimented with only preview image nodes instead of save images, so i can't get my hand back into the workflow i created 😅

[pastel kettle]

forge

[slow adder]

what is faster, forge or comfyui?

[vocal burrow]

depends how you use them.

Comfyui is more complex, allows you to do more and is usually slightly faster... But doing so it also provides you the tools to shoot yourself in the foot by creating non optimal workflow.

[agile creek]

Quick question: I am trying to use sd3 api for image-to-image, but no matter what I send in the "image" parameter, I always get the error message: "image: input not instance of Blob" I am using PHP, and for the image parameter I have tried file_get_contents, fread... even tested with images generated by SD itself... Do you know what I might be doing wrong? Thanks a lot! (PD. text-to-image works fine)

[ornate elk]

Then that means you need to use a clip text encoder for the model you tried to use

Flux and SD3 require clip encoders like clip l and t5

[pastel kettle]

I cant find the clip encoder for SDXL

[trim bloom]

Hey everyone!

I recently decided to try to train my own LoRA using kohya-ss for this, but I encountered a problem that when training starts, my CPU is automatically selected as the accelerator device. And I can’t change it.
My configuration: GTX 1050ti 4gb, AMD Ryzen 5 5600x

Can someone help?

Thanks

[teal arch]

is it possible to finetune stable audio open on 1 3090?

[ornate elk]

Sdxl doesn't need an extra one

Make sure to select xl at the top

[pastel kettle]

does pony need one

[ornate elk]

Nope

Only flux and sd 3 3.5

[dense seal]

Sorry for the late answer !
In total i have 32 Gb DDR5-5200, i can definitely buy new ram sticks if it's not enough.
Just did some retesting ; here is "the average" model i took from civitai. The GPU is absolutely shredding it. After rebooting my PC, it seems that it has fixed the XL issues !
BUT, whenever i take a model or anything, it keeps doing the same thing ;

Failed to create model quickly; will retry using slow method.
Applying attention optimization: Doggettx... done.
Model loaded in 15.7s (create model: 6.3s, apply weights to model: 9.0s, move model to device: 0.3s).

Why is it permanently doing the slow method instead of the fast/normal one ?
Otherwise everything seems to be working well !! Sorry for the disruption :)

Oh, just noticed, it does crash time to time (whenever i click on sdxl models or "flux_dev" ?)

[dense seal]

Confirming, ANY SDXL (or anything that is NOT SD1.5) just crashes...

Base model : pony (works), SD 1.5 (works), SDXL 1.0 (crashes), Flux.1 D (crashes).
launching with args : --use-zluda --update-check --skip-ort

[ornate elk]

Auto1111 doesn't support flux, remove it from the models/stable-diffusion folder or install forge webui

Stillt strange that it crashed with sdxl

32gb ram is enough

[dense seal]

oh, alrighty then!
So what about SDXL ?

[ornate elk]

Make sure you don't run Wallpaper engine or any Games in the background

Wallpaper engine causes issues

[dense seal]

I don't, ram usage is usually low and so is my gpu/cpu (usually under 5%)

i don't have wallpaper engine

[ornate elk]

Can you show me the environment path settings ?

[dense seal]

maybe i should add some other args ? What are you using ?

[ornate elk]

Where you added the hip path bin etc

[ornate elk]

A 20gb GPU shouldnt need anymore args.
But you can try add --medvram-sdxl

[ornate elk]

I'm on a 7900XTX and only use the ones from my Guide

What you can try is to update your python to 3.10.11 64bit

Then delete the venv folder

[dense seal]

OMW

[dense seal]

well, i did install it yesterday so isn't Python already the lastest ?

[ornate elk]

Nope

Yours is 3.10.6

[dense seal]

that's what it's been asked in the guide, no? Maybe i was confused

[ornate elk]

My guide only lists 3.10.11

Also open up a cmd as admin and run
SFC /scannow
That will fix windows system issues

[dense seal]

did everything you said, unfortunately any SDXL 1.0 is still crashing

nvm even SD1.5 is crashing now

I think i gotta wipe my windows, i think it's the best solution.
Maybe the many, many installations of pythons and C++ thingies broke everything up.

sfc /scannow won't even fix the things

[ornate elk]

What does it say?

Does it say errors can't be fixed?

[dense seal]

yeah, and now it won't even scan

[ornate elk]

Rip

[dense seal]

guess i'll be formatting everything then. Sorry for the disruption again ! Well at least now i know why everything is laggy/buggy haha

[ornate elk]

Yea NP, system errors like that can cause everything (and we can't know what).

Better clean install then follow the guide again and you should be fine

[granite sky]

Hello,
W11, 4090, 5800X3D, 32Go RAM, 1T SSD M2 + 2T SSD M2.
I'm trying to use AnimateDiff, that has worked for my 1try so I made a proper prompt but doesn't work anymore, I cannot make videos anymore.

I've made the installation following the links in the description here : https://www.youtube.com/watch?v=jOfKHOhb25w&t=1138s&ab_channel=VisionIA
Nvidia installation and the other, both work for pics.

I've installed pinokio to try SD3 ; CUDA and visual studio too in my quest to AI videos trying to use FakeApp (I failed, errors), was struggling to install cudnn_9.5.1_windows, I think i've failed for this last one.
Reading the message in the CMD window it's talking about CUDA and I don't undersatand what, thinking CUDA is not required for my version of SD I've uninstaled it with Visual Studio, AnimateDiff still doesn't work.

Is it a CUDA story?
"D:\UIPICS\A1111\system\python\lib\site-packages\huggingface_hub\file_download.py:797: FutureWarning: resume_download is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use force_download=True.
warnings.warn(" -> I don't understand, is it a python story too? I have python-3.13.0-amd64 installed.
Also, how to "force_download=True"? Where?

Thank you for helping me.

[untold mauve]

when i installed my zluda days ago it was on fresh windows install but still appear "failed to load quickly". i think it's just like this for everyone at this point

[pastel kettle]

ok so it seem slike the problem has to do with the refiner model

[balmy monolith]

Im' using swarmui with flux 1 dev, when generating an image, i'm consistently getting blur

Any ideas please?

[ornate ruin]

Hi CS1o 😅
Like over a week ago I had to factory reset my laptop and was told to just get Swarm.UI
But I really don't like it a whole lot, and even with McMonkey one of their main devs I believe? We couldn't get Face IP Adapter to work properly. Hence why I'd like to come back to a1111 again

In short, how much will you hate me when I soon have to re ask questions like what some of the optimal prompts are for me like -low-vram or so... 😅 Will try to find most stuff from the chat history but I can't assure I'll find everything on my own. lol..

[balmy monolith]

nvm, think i found it, was cfg = 1 fixed it.

not setting it to 1

[cosmic pulsar]

Hello everyone, maybe who knows how to solve this problem?
I tried to install Derrian's LoRA Easy Training Scripts, but it didn't work
When running install.bat throws an error.((
'Unable to copy 'C:\Users\Dasss\AppData\Local\Programs\Python\Python310\venvlauncher.exe''
'Unable to copy 'C:\Users\Dasss\AppData\Local\Programs\Python\Python310\venvwlauncher.exe''

[earnest jackal]

anyone know how to install dora?

ok

[dense seal]

Don't click, it's a scam

[earnest jackal]

ok

[earnest jackal]

did he also tried to get you like that as well?

[inland oracle]

Hi, in civitai there is what called style. How can I use them after I install them? And where I should store them ?

[fair oxide]

how can we fasten the speed of stable diffusion my internet is really fast my pc is Processor: Intel Core i7-11800H
Graphics Card: NVIDIA GeForce RTX 3060 (6GB GDDR6)
RAM: 16GB DDR4 3200 MHz

but it generates very slowly

[light prism]

Hey i have been developing an own application that uses Stable Diffusion for a while and am continuously learning more about Stable Diffusion and was wondering if there are official numbers ob Training dataset size?
I visited the laion website and did not finde deatils on what was actually used in the end.

And if there are no official numbers in what range of numbers are we talking about? Probality more than anyone of use can even download themselves 😄

[rugged latch]

boop

i havea werid error, it doesnt disrupt anything, rather, id say it more erroneous than anyting but, wheni access my stable via the mobile link, and then i close the browser on the phone, my termnial prints this

why it do dat

happens only when u access the sharelink, then on the deivce u accessd from, close the browser

xd

again, everything works fine after and during

im jus curious

[ornate elk]

Yes thats normal in Auto1111

Can be ignored

[ornate elk]

Which webui do you use?

[ornate elk]

I can help you with the Settings sure and for prompts also ask in #📝｜prompting-help

[rugged latch]

q-q

ig nu body knows

YK

big question

arent there devs here?

why non of them seem interested in talking here in tech

like dang, cudos to Cs1o for being like the main, if not only mod interested in halpin ppl

i dont bellieve ive ever seen any of the devs say a word to anyone here

i think i saw yusuke say 1 ting

and it may have been to me, when i said they have the same name as mai bestie, which i thought was pretty cool

[teal mantle]

Black image EVERY TIME

D:\StabilityMatrix\Packages\Stable Diffusion WebUI reForge\modules\processing.py:1036: RuntimeWarning: invalid value encountered in cast
x_sample = x_sample.astype(np.uint8)

Its always this why is this happening

[past vortex]

Just got started getting A1111 running on RX 6700 (ZLUDA). Finally managed to get to image generation, but been walled by this error:

" RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument mat1 in method wrapper_CUDA_addmm) "

Any help would be appreciated. Thanks! 👍🏿

[carmine ibex]

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 1024.00 MiB. GPU 0 has a total capacty of 12.00 GiB of which 0 bytes is free. Of the allocated memory 10.82 GiB is allocated by PyTorch, and 245.07 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation.

I got OOM but with Segment Anything which I was not getting before, even on that same model. Idk why I am getting this

Same image too

[hidden matrix]

is someone available for a discord call to help me fix a lora im trying to create i have a data set already everything is captioned but whenever i try to run the command to train it says it cant detect any images

[ornate elk]

What's your GPU?

[teal mantle]

Solved it by restarting thanks

[ornate elk]

Can you show the txt2img settings you used?

[dense seal]

Quick update: it's the case !
Fresh win install and everything went smoothly 🙂
Guess my old Win was corrupted to the neck. Thanks again for the support !!

[ornate elk]

Ah glad it worked!
And no problem 🙂

[keen raptor]

Can u help me bro

Why the Lora dosent show

i even pressed restart

[cedar oxide]

I can't understand your error. Could you explain more detail please?

in your screenshot, what is the nwsj.... safetensors file?

[keen raptor]

Its a Lora

[cedar oxide]

if so, why did you tell the lora doesnt show?

Really can't understand

And which model did you train?

[keen raptor]

My bad bro

[keen raptor]

I downloaded one from CivitAi

Like

I downloaded a detail tweaker

When I paste in Lora folder

It dosent show/

even after restart

[ornate elk]

Its for flux

[keen raptor]

oh

[ornate elk]

It won't show up if you not select a flux model

[keen raptor]

No wonder

[ornate elk]

Also Auto1111 doesn't support flux

So it will only work on forge webui or comfyui

[keen raptor]

Understandable thanks

[cedar oxide]

Ops, sorry to misunderstanding and at first I understood you trained your lora model and can't show the result lora file.but you downloaded from civitai, lol

[keen raptor]

Hehe

[mortal compass]

Tried to make a simple upscaling workflow for SDXL/Pony. The nodes that are visible are the nodes that I'm using.
However, on the last VAE Encode and KSampler, it takes an unusual amount of time to complete. When the KSampler got to 75%, I received a "torch.PutOfMemory Error", which I find to be very strange.. My GPU is a 4070 ti super, which should be able to handle this with ease. I'll post the log below. Thanks for reading!

[ornate ruin]

A1111 probably won't get any major updates anymore will it? Like SD3.5 support

Or where can I learn more about recent news myself?

[timber oak]

hello there fellow ai enthusiasts

i have an annoying issue

[vocal burrow]

dev branch is still updated, so there s no reason to classify it as abandonned.

[timber oak]

over the course of a few years i have accumulated over 4k loras

some of them sd, others sdxl

and most of them pony

now i wanna clean up and delete all sd 1.5 loras

bc i only use pony models and sdxl works with pony sometimes

i cant find any way to sort the files by version

do i rly have to delete the loras by checking each and everyone of the loras in automatic/comfy and delete them one by one?

or is there a simpler way?

i also dont wanna delete all my loras and redownload bc it already took ages

(in total my lora folder is 456gb big)

(according to tree size viewer)

(help)

[past vortex]

I switched over to SDNEXT and using Deepcache. It's working fine now 🙂

[past vortex]

Do the file names not include SD_1_5 or something?

[timber oak]

nope

only some

but i found this program

stability matrix

its epic

its ui + civitai/huggingface + file explorer all in one

[past vortex]

Do you not have a list of the SD 1.5 LORA names? Feed that list into Python or some automated task thing something and delete it

[timber oak]

i dont rename my loras

i usually keep them like that

but stability matrix can autosort them

[hexed loom]

My ComfyUI crashes after trying to use JuggernautXL :(

[ionic dragon]

Heya!
I'm trying to get into SD and I'm looking for a ... mentor? 😄
Someone willing the help a newb out and answer some question for a little bit of pocket money ? 👀

[ruby knot]

It depends what you need to know? I’ve been messing with it for a bit.

[willow temple]

HI, does anyone know what I could be doing wrong here ... I've never used detail daemon, I usually fiddle with pretty simple work flows just to have fun on occasion. I thought this would be neat I snagged a random image to try it on but my start image is same as end image so when you compare a and b there is just zero changes.

[loud shard]

hello i can't get my depth map to generate correct images and i suspect the controlnet model is picking up the change in brightness too much

the problem is the stripes

i can't make them any better

i already tried bluring it but didn't do anything

[ruby knot]

Is there a way I can copy my Automatic1111 folders onto my other computer and make them work with no internet? I’m asking as it doesn’t have WiFi on the mb for some reason, I can backup the whole folder and copy paste it back to my laptop when extensions brake my install but if I try load it on my pc with better specs it errors and doesn’t load.

[hidden matrix]

im trying to run the training command for a lora but it always gives me the error that there isnt any images in the dataset (there are 61 in total all the same dimensions and all named in numerical order 1.jpg 2.jpg and so on) can someone please help me

[runic stratus]

I have a SD3.5M Diffusers question, if anyone can help, greatly appreciated.

I can only seem to find FlowMatchEulerDiscreteScheduler and FlowMatchHeunDiscreteScheduler for SD3.5M but they both give a brighter result than the Euler from comfy results. Any idea why my comfy results are prettier (deeper, less bright overall and slightly better contrast?
I'm using Euler + simple for both

pipe.scheduler = FlowMatchEulerDiscreteScheduler.from_config(
    pipe.scheduler.config,
    timestep_spacing="trailing",
    shift=3.0
)

[serene comet]

what is the use of xformers ?

do I need it?

[thorny iron]

So umm..

How does one install A1111 with the new 3.5

Or do I use Forge? :<

[honest canyon]

Hello, i am really giving up... i tried a workflow.. at first it was a bit difficult... and then it worked... after a while i gt some problems.. i am really trying to fix these since 4 hrs... but now i am giving up... maybe someone of you can help me... otherwise i will see myself in Photoshop again 😄 My Problem: in the Manager it says missing nodes --> comfyui-art-venture... i uninstall, install, updated.... and its still in the list of "Missing Nodes" ... does someone hav an advise?

[ornate elk]

Is your comfyui and manager node updated too?

You can also manually install the nodes via Git clone

[ruby knot]

Hi I have a RTX 4060ti 16gb and flux took maybe 5 mins to do an image at 768 by 768 and it was just noise, what am I doing wrong? I’m using comfyui

[ruby yoke]

sounds like you have the wrong VAE selected.

[ruby knot]

I have no VAE I don’t normally have to load them on normal sd?

[ruby yoke]

you have to have a VAE - either it's part of the model, or you load one seperate from the model - but it can't give you anything at all without one. What checkpoint are you using?

[ruby knot]

Oh ok thought they where optional my bad, the flux Schnell-fp8 one

[ruby knot]

Also weirdly I’m getting better images with some sd 1.5 models then some sdxl ones.

[vocal burrow]

SD3.5 isn't supported yet in Auto1111

[vocal burrow]

xformers is a library used to speed up generation, use --xformers command line argument to turn it on.

[keen raptor]

Can someone tell me why my images turns out like this when I'm using Juggernaut XL

It loads normal during generating

But when final result

[keen raptor]

It blanks out into this

[ruby yoke]

No, they arent optional. Youll need tobgobdownload the vae for flux

[ornate elk]

Can you show the txt2img settings? And which VAE do you use?

[keen raptor]

Oh.

I DIDNT SETUP VAE

but I used an anime xl model

It worked

[thorny iron]

What about forge

[keen raptor]

It was normal 30 steps and 1024 resolution cfg scale 7

But the image still shows in the generation

Just blanks out in final

Is it cuz of no vae

Could it be the other model had baked in VAe

[ornate elk]

In development

[runic stratus]

Hi everyone! I'm using Stable Diffusion 3.5 with Diffusers, but I'm a bit confused about the scheduler and sampler options. I only seem to find FlowMatchEulerDiscreteScheduler and FlowMatchHeunDiscreteScheduler. SwarmUI and ComfyUI have more choices like SGM, DDM, DPM++, etc. When I try using schedulers from SDXL, I get messed up images.
What are my actual options for schedulers and samplers with SD 3.5 and Diffusers? Thanks!

[thorny iron]

What is your thoughts?

Do you think A1111 or Forge is better at the moment? :<

[ruby knot]

Do you know where it is? I don’t remember seeing one unless missed it?

[untold mauve]

hi, best vae for sdxl? for anime and for realism

[honest canyon]

first of all.. thank you for your effort... but yah i tried now a lot.. i change those folders "segment anything" and "omegaconf" ... i updated my pip... everytime i open this workflow.. it will disappear what i showed before... and in the manager (also updated) ... it says "(IMPORT FAILED) comfyui-art-venture" ... i am lost...

[ornate elk]

For Stability, auto1111, for more features currently Forge

Flux works only on forge and Comfyui

And 3.5 will work next in forge too

[knotty turret]

@ornate elk man i think did big mistake didnt know stable diffusion dont work on AMD gpu

can you help me with it or there no way?

[hot vector]

Hello everyone

I recently explored a home interior design website with AI and came across a feature called "Virtual Staging," which fills an empty room without altering the walls or floor. I attempted a flow where I detect the floor and walls, generate a mask image, and then perform inpainting using depth and Canny ControlNet. However, in my results, the floor and walls still change, despite adjusting the prompt and strength parameters.

Can someone help me achieve this feature? Should I train custom models such as text embedding, inpainting, ControlNet, or implement a segmentation model to solve this issue?

Thanks

[vocal burrow]

it does work, check the guide in pinned posts for AMD specific instructions

[knotty turret]

u right THX GOD

[knotty turret]

Rx 6600 use hip sdk 6.1.2 or 5.7?

got confuse on guide

[untold mauve]

6.1.2

[knotty turret]

thx

[ornate elk]

Stable diffusion works on AMD, checkout my Guides 🙂

[knotty turret]

https://tenor.com/view/tom-and-jerry-tom-the-cat-phew-relieved-sigh-gif-17213206

[knotty turret]

just need do 1 generation now

i already running it

but i have no model

need 1.5 base model

and wait this too

[ornate elk]

Here:
https://civitai.com/models/4384/dreamshaper

[knotty turret]

thx

[ornate elk]

Yea first image will take that long

[knotty turret]

have pony too on my old webui

works?

[ornate elk]

Yes

[knotty turret]

or its better use this dreamshapper

[ornate elk]

But use the 1.5 model for the first image gen

[knotty turret]

u have the link?

[ornate elk]

And you need to add --medvram-sdxl to the webui-user.bat

[ornate elk]

The one I send a few seconds ago

[knotty turret]

oh mb

[knotty turret]

have 8gb vram

still need -medvram?

[ornate elk]

Not --medvram but --medvram-sdxl

[knotty turret]

to use pony model

[ornate elk]

So it only uses medvram for pony and sdxl

[knotty turret]

Oh ok thx

can be anything first image

simple prompt?

[ornate elk]

Yep

Doesn't matter

[knotty turret]

should i download

vae too?

[ornate elk]

Just type cat or something xD

[ornate elk]

You can do that yea

But first generate the first image

[knotty turret]

Ok

forget to say i using the forge version

still need --medvramsdxl ?

[ornate elk]

Oh then nope

[knotty turret]

lol didnt see it

hmm...

[ornate elk]

Can be ignored

[knotty turret]

alright

[ornate elk]

SD next is worse to use

[knotty turret]

time to wait

thats it thx again you saved me

[ornate elk]

No problem 🙂

[knotty turret]

And what about comfyUi

Sometimes like to use that one too

[ornate elk]

You can install it via my Guide.
But first make sure forge works

If one webui with ZLUDA works its easier to set the others up too

[knotty turret]

this is normal right?

[knotty turret]

alright

tell u the result later

[ornate elk]

Okay 👍

[fiery ice]

Hello guys, in comfui (flux), would anybody know how to conect unet to modelname? I'm using the node comfy-image-saver:Save Image w/Metadata.

[knotty turret]

i slept lol

it finished

😃

[ornate elk]

Nice!

[knotty turret]

i will try your guide for comfyui later

[ornate elk]

Sure

[ruby yoke]

https://huggingface.co/black-forest-labs/FLUX.1-schnell/tree/main look in the VAE folder on this page

[ruby knot]

Thanks

[knotty turret]

do you know about collab for some reason now i cant connect local

[ornate elk]

What do you mean you can't connect local?

And are you talking about comfyui or forge?

[knotty turret]

oh the msg from the guy got deleted

maybe he could help

[knotty turret]

Google collab

[ornate elk]

Dont click on scammer discord invites

[ornate elk]

Ah nope can't help with collab.

[knotty turret]

ah ok lol

[ornate elk]

And these discord scam fake supports won't either xD be careful.

[knotty turret]

ok

[keen raptor]

@heady granite my Model HAD baked in VaE apparently

but still no work

this is the model

[ornate elk]

whats your gpu?

[keen raptor]

4070

[ornate elk]

make sure you have --xformers --no-half-vae in the webui-user.bat

[keen raptor]

i did

you can see here it shows an image while generating

but now it dosent when done. ‘only for this model’

[ornate elk]

can you show your full cmd log?

[keen raptor]

is it this

[ornate elk]

the cmd log

the black box

with white text

[ornate elk]

looks normal

[keen raptor]

i tried a diff model and it worked’

[ornate elk]

then maybe redownload the jugggernaut model

[keen raptor]

but i feel like it has to be something with the Ui cuz else why would it Preview the generation

if it totally dosent work

[ornate elk]

you can try delete the config.json and the venv folder and then relaunch

that resets everything

[keen raptor]

mayne

im so sad

😩

[iron lantern]

*** Error running after_extra_networks_activate: C:\ai 2\webui_forge_cu121_torch231\webui\extensions\sd-forge-couple\scripts\forge_couple.py
Traceback (most recent call last):
File "C:\ai 2\webui_forge_cu121_torch231\webui\modules\scripts.py", line 876, in after_extra_networks_activate
script.after_extra_networks_activate(p, *script_args, **kwargs)
File "C:\ai 2\webui_forge_cu121_torch231\webui\extensions\sd-forge-couple\scripts\forge_couple.py", line 127, in after_extra_networks_activate
raise RuntimeError(
RuntimeError: [Forge Couple] Not Enough Lines in Prompt... [1 / 2]

what does this mean ?

[ornate elk]

It does mean you have not enough separated lines

You have so split the prompt with Enter key

[main rune]

I can't seem to load GGUF files into comfy

[iron lantern]

wont allow me to do that it just starts making the image

[distant hamlet]

Hey, so I have SD set up with automatic1111 and it works for SD1.5 but 3.5 just produces pixelated noise. What could be causing this? I should have the GPU power, I am running a 4090.

[ornate elk]

Do you have the gguf Node installed?

[ornate elk]

Auto1111 doesn't support 3.5 currently

You would need to use ComfyUI

[main rune]

I thought I did, gotta check

[distant hamlet]

Thanks. I have ComfyUI set up as well but I don't really understand it. When I attempt a prompt it says "no tokenizer" not really sure what this is referring to.

[main rune]

I did have it installed, but apparently it suddenly needed an update for some reason
I did update it before this yet somehow...

[distant hamlet]

Sorry, I just redid it. Giving someone an error off of week old memory isn't good lol.

"CLIPTextEncode

'NoneType' object has no attribute 'tokenize'"

I am aware this is related to the clip files that are available with the model download, but I put them into both clip folders in models, clip and clip vision, and while they show up in the folder pathing in comfyui, I can't drag them into nodes and they don't seem to do anything.

[distant hamlet]

Seriously. I can't figure out what to do with the text encoders. I add them to the models/clip folder, they do show up as text encoders in comfyui. But I can't drag them into the node model and they don't change anything in terms of errors.

[deft hemlock]

should be the same if i have the nan_ problem right ?

return getattr(processing, '__controlnet_original_process_images_inner')(p, *args, **kwargs)
File "F:\ai\sd.webui\webui\modules\processing.py", line 998, in process_images_inner
devices.test_for_nans(samples_ddim, "unet")
File "F:\ai\sd.webui\webui\modules\devices.py", line 265, in test_for_nans
raise NansException(message)
modules.devices.NansException: A tensor with NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.

i tried all of those settings

both in webui user bat

and the settings itself

is the lora gone ? like useless ?

https://tenor.com/view/car-dreak-gif-4225905613113080592

[proven vortex]

yo peeps ❤️
i have this problem with advanced controlnet and doesn't matter what I do I can't fix it, might have missed a solution somewhere but im close to giving up

[remote plover]

I've been running into issues running stable diffusion on my steam deck, I think maybe I'm still missing dependencies I have installed alot of them, I have received mix messages of weather SD works with python 3.11 or if 3.10 is required

[scenic pilot]

Hello there, can someone can help to understand why I getting this issue with Forge, when I generated 2 images + Hirefix, the first one is corrrectly generated at a decent speed then at the second one, I get this message and my pc start to be super slow ( forge stuck at 50% ) :/

[ruby knot]

Is it hard to learn how to use ComfyUI ? It most the time if I try add any extra nodes it doesn’t end up working anymore?

[sharp lichen]

does anyone know what causes a image in flux to look like this in a lora?

[knotty turret]

@ornate elk Tried your guide now for Comfyui but after the dll files idk why not working

[ornate elk]

Have you replaced them ?

[knotty turret]

yep

just detail

This comfyui its on my D: its normal cmd show C: ?

[ornate elk]

Yep

[knotty turret]

tried to run again the bat file with admin

he download files again

[ornate elk]

Looks good

You need to replace the files again

[knotty turret]

ok

[knotty turret]

gonna try again now

[knotty turret]

same error

[ornate elk]

Did it asked to override?

[knotty turret]

yes

[ornate elk]

Okay strange

[knotty turret]

😔

[daring night]

Hello, I am new to the stable diffusion. I have email support from their website and haven't received a response yet and it has been a while. Is there anybody from support that can assist me?

[knotty turret]

what should i do

[ornate elk]

Can you delete the venv folder and relaunch the ComfyUI start bat again but without Admin rights

[knotty turret]

ok

[knotty turret]

should replace the files again after this?

[ornate elk]

It should first open up in browser

[knotty turret]

did nothing

the cmd just closed

[knotty turret]

it didnt open the browser

[ornate elk]

Then open up a cmd and run
Where python
And then
Python --version

And screenshot

[ornate elk]

Hmm

Try uninstall python from the ms store

That usually causes the closing cmd

[knotty turret]

never

used ms

ok

let me remove it

i dont have it

[ornate elk]

Hmmm okay

Strange then

You can try to reinstall Python completely

[knotty turret]

🙃

[ornate elk]

Then restart the PC

[knotty turret]

ok

[ornate elk]

Then delete the venv folder and relaunch

[knotty turret]

just delete both?

[knotty turret]

i did what u said

again

[ornate elk]

Have you deleted the venv too?

[knotty turret]

Yep

trying tutorial on youtube

maybe works

[remote plover]

./webui.sh: line 258: bc: command not found.
loading stable diffusion model: RuntimeError
Then
RuntimeError: HIP error: invalid device function
Stable diffusion model failed to load
qt.qpa.plugin: Could not load the Qt platform plug-in "xcb" in "/home/deck/%%%%/python3.11/site-packages/cv2/qt/plugins" even though it was found.
This application failed to start because no QT platform plug-in could be initialized. Reinstalling the application may fix this problem.

[glad fiber]

which os are you using?

[remote plover]

Steam os 3.6.20
Qt version 5.15.12
Kernel version 6.5.0-valve blah (64-bit)
Graphics platform X11

[glad fiber]

could you run echo "$SHELL $0" and send the result please

i think one of the problems is the shell you are using

[remote plover]

Bin/bash /bin/bash

[glad fiber]

idk then

maybe i would try to run pip install -r requirements.txt because the ui may have not installed all dependencies

you could also try to find what program the ui is expecting to have the bc command available

[remote plover]

Okay let's check requirements, I was reading posts made online about Steam Deck not having certain dependency and headers fully installed. But I don't know if I installed the right things.

Just as I thought steam os is archbased and use pacman mainly pip isn't installed

[glad fiber]

@remote plover i forgot to ask, but did you run webui.sh or webui-user.sh?

[remote plover]

Webui.sh

[glad fiber]

the "right" one is webui-user.sh

nvm

i mixed it with windows

looks like that is your solution

[remote plover]

At line 258 it says : if [ $(echo "$libc_vernum < $libc_v234" | bc) -eq 1 ]; then

If that's the answer to the other thing you mentioned

That might be the answer I'll look it up

Can confirm that I have Bc installed. It no long gives that part of the error but everything else remains.

FutureWarning: Importing from time.models.laye4s is deprecated, please import via timm.layers warnings.warn(f"Importing from {name} is deprecated, please import via timm.layers", FutureWarning)
QObject::moveToThread: Current thread (####) is not the objects thread (####).
Cannot move to target thread (####)
qt.qpa.plugin: Could not load the Qt platform plug-in "xcb" in "/home/deck/%%%%/python3.11/site-packages/cv2/qt/plugins" even though it was found.
This application failed to start because no QT platform plug-in could be initialized. Reinstalling the application may fix this problem.
loading stable diffusion model: RuntimeError
RuntimeError: HIP error: invalid device function
Stable diffusion model failed to load

[remote plover]

Webui-amdgpu was recommended on reddit somewhere and it does this

ModuleNotFoundError: No module named '_ctypes'

[distant pivot]

in stability matrix/swarmUI/comfyUI workflow, I can seem to move groups anymore. there was an update or something i guess. anyway, is there an option now that locks them or something?

Never mind, it's something in by browser i guess. cause another one didn't have any trouble.

[knotty turret]

@ornate elk tried yt tutorial to get the comfyui but damm so much harder

[ornate elk]

yea wouldnt recommend that

these other tutorials are bad or complivated

[knotty turret]

maybe try again your guide?

im lost fr

[ornate elk]

you can try

best is if the cmd launches and closes itself again, to make a video of it and pause and screenshot what it shows at the end, then i can help

[knotty turret]

you wanna see the install cmd process?

or just the end

anyway gonna record

[ornate elk]

the end

[ornate ruin]

Does Cache FP16 for Lora actually just use more RAM as it says or does it mean that it uses extra VRAM?

Also - A1111 already supports SD3?

Found a setting called 'enable T5 text encoder' it's under a Stable Diffusion parameter called Stable Diffusion 3. 🤔

[knotty turret]

mb made video but u can just skip

[ornate elk]

Auto1111 Supports SD3 but not 3.5

[ornate ruin]

I thought that's the same.. okay. But the setting needs to be turned on for it to be able to utilize it or does it just boost it for the trade of extra resources?

[knotty turret]

Did u see something wrong cs1?

[ornate elk]

yep it didnt seemed to install torch 2.3.1

[knotty turret]

😲

[ornate elk]

can you check the venv/lib/sitepackages folder for torch 2.3.1 ?

[knotty turret]

Ok

[knotty turret]

some torch folders

[ornate elk]

hmm looks good

[knotty turret]

🫠

[ornate elk]

can you relaunch the cmd (not as admin) and record it?

[knotty turret]

Ok

[knotty turret]

.

[ornate elk]

for SD3 not needed but for flux and sd 3.5 later

[ornate ruin]

Found a setting for Vram limiting. Could it be wise to limit it to 7 when I'm on an 8gb vram device?

[ornate ruin]

Ah okay. Interesting.. thank you

[ornate elk]

ah okay that is before you replaced the zluda files?

[knotty turret]

you mean the 3 files that i have to rename?

[ornate elk]

in auto1111 ? nope in forge maybe

[ornate elk]

copy, rename, and replace

[knotty turret]

i stoped on step 6

[ornate ruin]

Why in forge but not auto1111? Just for understandment,

[knotty turret]

yes that video is before

that

[ornate elk]

in forge the limiter is made better and can help when using flux, but with 8gb vram you dont need to limit it anyway

[ornate elk]

then do the other next steps

[knotty turret]

ok

[fair oxide]

Has anyone else had issues with tokenizers during running sdwebui? I thought I must have done something incorrect, but checked out multiple sources in the event the official instructions were just not updated for Apple Silicon, but everyone has done the exact same I did so I have no idea.

[ornate ruin]

Thank you very much.
Now I've seen so many settings that I didn't even look for lol.

1 Clip skip and VAE to add it to the t2i tab found that one 🙂

And 2ndly an option to change the default option on start-up for parameters like hires fix to already have it active with my preferred settings

[knotty turret]

just try to start now? or do something else

[knotty turret]

asked to replace and did it too

@ornate elk

i installed forge amd on ssd but this comfyui on hdd maybe thats the problem?

zluda folder inside ssd

[ornate elk]

thats not an issue where zluda is

[knotty turret]

😔

[ornate elk]

it closes to fast to see whats the error

[knotty turret]

let me print

[knotty turret]

did two prints one from now and from the video

[ornate elk]

whats your adrenalin driver version?
whats your hip sdk version?

[knotty turret]

Hip 6.1.2

ROCm and hip sdk is the same thing?

[ornate elk]

on windows yes

[knotty turret]

ok so yeah 6.1.2

[ornate elk]

and your auto1111 works with zldua right now?

[knotty turret]

yep let me show u

real quick

[knotty turret]

.

[ornate elk]

okay good

thats forge webui

[knotty turret]

Yep

[ornate elk]

can you delete comfyui and git clone it again into that folder? and record that

i reinstalled comfyui right now without problems

[knotty turret]

Ok

[ornate ruin]

There's no way to find a Loras actual name except for the parameters used in the positive/negative prompts from an images meta data is there?

[ornate elk]

lora hash will be alkso in the output

[knotty turret]

@ornate elk

[ornate elk]

your zluda folder is inside a zluda folder instead of being directly C:\ZLUDA

[knotty turret]

oh

[knotty turret]

put this on C

[knotty turret]

better now?

[ornate elk]

needs to be ZLUDA

not zluda

it has to be the same name as the environment path variable