🤝｜tech-support

[meager carbon]

i mean a 4090 bumps to 450-500w sometimes, and if you have an intel, i could see peak going to 1000w

[median current]

AM5, but I may start looking at power use anyways. Concerned about a faulty psu. Full build is below:

AMD 7950x CPU
Asus ROG STRIX B650E-I motherboard
2 x 48gig corsair 6000mhz, CL30
Nvidia 4090 FE 24gig
Corsair SF1000L psu

[meager carbon]

hmm

[median current]

Only other thing i could think might be DRAM instability, but memtest showed no errors..

[meager carbon]

could also be gpu issue

try downclocking it

[median current]

Yeah good idea. Will try that tonight.

Crashed my pc last night and basically had to take it apart and cycle through. Gpu fan was spinning up but unit was not lighting up. Blank screen. Not posting. Mobo indicator first showed red (cpu) until i took ddr5 out and reseated it. Then mobo showed yellow (dram) until i took 1 of the 2 ddr5 out. Then mobo showed white/green which is vga issues. So i tried flashing bios and finally got things to boot up fine. Rebuilt everything and system works again, but I keep getting gpu/vram maxing out.

It'll either stutter and white screen then crash the .bat or eventually crash entirely.

From process standpoint, it appears to be tied to repeated marigold depth estimation and that being used as a depth controlnet for 4 batch image generation.

Which really shouldn't be that intense, but vram seems to start stacking until it fills up and gpu maxes out even on basic generation.

[ornate elk]

Yes at the PC in 15mins maybe

[meager carbon]

cool cool

[ornate elk]

Yes you can move and image into the lora folder and rename the image to the Lora's name

[hard quiver]

i see. thanks.

[ornate elk]

You can also generate an image then click on the edit button on the lora and click "set preview"

It will use the generated image

[hard quiver]

can't seem to find this edit button on lora.

only these 3 buttons

[ornate elk]

The right one is the edit button

[timber oak]

hello my friends i need help

[meager carbon]

would you like me to uninstall everything and start from scratch?

[timber oak]

my friend is trying to install sd on his laptop

he has an intel cpu with integrated graphics and an rtx 3070

i already equipped the correct gpu in the user bat

updated graphics to new

so cuda is updated

but i still get "runtime error: torch not able to find gpu"

i reinstalled venv several times

i tried removing the test

nothing happens

git pull
@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--precision full --no-half
set CUDA_VISIBLE_DEVICES=1

call webui.bat

GPU 1 is the 3070 and GPU 0 is the intel graphics

i also installed fast api into venv>Scripts

updated pip

We have been sitting here for 2 hours..... We are desperate and broken.... Our success is at your mercy oh great gods of Stable Diffusion🫂

[ornate elk]

You could try reinstall the HIP SDK and then download the zluda files again

[meager carbon]

okay, ill try that

should i uninstall the hip, or just run the installer again?

[ornate elk]

Hey, remove --precision full and --no-half
Then instead add: --xformers --medvram-sdxl --no-half-vae
Then delete the venv folder.
Then open up a CMD and type
pip cache purge
Then hit enter and close the CMD.
Then launch the webui-user.bat

[ornate elk]

Try if you canoverwrite it

[meager carbon]

this is the correct one, right?

and everything look good here?

ill wipe the venv folder just in case.

:/

@ornate elk any idea?

[meager carbon]

zluda works just fine in blender

[ornate elk]

Oh in Blender it works?

[meager carbon]

mhm

[ornate elk]

Then run pip cache purge in a CMD.
Then git clone the directml repo again

[meager carbon]

sure

should i try elsewhere?

other than my user directory?

[ornate elk]

Make a new folder on C and clone it inside that

[meager carbon]

just on the bare drive?

okay.

[gray drift]

Images coming out incorrectly, how to fix this?

dreamshaperXL v21Turbo

[meager carbon]

yeah i just saw haha

sorry

[gray drift]

Thanks xd

[hard quiver]

hi CS1o

do u know what should I do in automate 1.1.1.1 if my goal is for the character to change clothes in random while im making 100 batches of a prompt

should i write

wear shirt, coat, jacket, full sleves t-shirt

or

wear shirt, wear coat, wear jacket, wear full sleves t-shirt

or

wear shirt or coat or jacket or full sleves t-shirt

[carmine agate]

Cfg is way too high

[gray drift]

I'm sorta new to this so Idk why the image came out like this

[gray drift]

What should it be instead?

[carmine agate]

1 to 2 max

[gray drift]

Okay, I'll try it now thanks

[carmine agate]

The model page from civitai should give you good settings

[gray drift]

Thanks! Have you tried this model before?

[carmine agate]

Yup

[meager carbon]

still, no dice.

[gray drift]

I changed the CFG

[meager carbon]

have you tried other models?

or is this your first one

[gray drift]

I tried realism model before, this is a new model I'm trying today

[meager carbon]

is this specific one giving you issues?

[gray drift]

I think so

[gray drift]

I changed it to epicrealism model it came out the same 🤔

[meager carbon]

bump the cfg scale to 8ish

[carmine agate]

Oh I may have been thinking og lightning. Sorry. Did you check the model page on civitai though?

[meager carbon]

and sampling to 20

[gray drift]

Okay, thanks

[carmine agate]

Lcm, turbo, lightning. They all run together in my memory at times. Sorry for the misleading

[gray drift]

Oh

[meager carbon]

i mean looks better

go to #📝｜prompting-help

[carmine agate]

https://civitai.com/models/112902/dreamshaper-xl

[gray drift]

Yeah, thanks for help Xd

[meager carbon]

im p sure it's not a technical issue

[gray drift]

This is the one I downloaded I think

[timber oak]

it didnt work sadly

maybe we did sth wrong

we will try again

[daring rampart]

Hi

[ornate elk]

Checkout my install guide in the pinned messages of this channel

[hard quiver]

do u know how to generate random clothes and stuff if your making a batch of 100 or so

in automatic 1111

[hard quiver]

. this

[ornate elk]

With Wildcard extension

[hard quiver]

wait

[hard quiver]

this one?

[meager carbon]

any idea...?

[hard quiver]

installed it. now what.

https://tenor.com/view/cat-leaving-leave-oops-bye-gif-17519616

jus looked it up

is there an easier way to do that. like in the prompt itself.

[ashen knot]

does anyone know how to use .png bones/annotation file in openpose ?, or it just can't

[carmine agate]

Right, well it states "You can use the Turbo version (not Lightning) as a non-Turbo model with DPM++ 2M SDE Karras / Euler at cfg 6 and 20-40 steps." 😄

[ornate elk]

Nope everything looks correct

[meager carbon]

ill literally try anything

anything that comes to mind

@ornate elk

I tried following this... no dice.

[ornate elk]

Yea because to get that to work you need to run
Pip cache purge
In a CMD
And then delete the venv folder

[ornate elk]

Select None as preprocessor and openpose as model and it will work

[ashen knot]

it works thanks

[meager carbon]

that worked, but zluda does not... should i try sd.next?

[ornate elk]

Yes, that's a good idea

But I have a feeling it won't work too

But worth a try

[meager carbon]

i meannnnn

[ornate elk]

It works 😮

[meager carbon]

ehhhhhh

[quiet swift]

Hello, I try to upload Deforum in a new SD that I have just installed today. But when I go to extension tab, then Available, then Load from, nothing appears. I can only read : URLError: <urlopen error [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: unable to get local issuer certificate (_ssl.c:997)> If someone had an idea, thank you very much

[ornate elk]

i mean it looks better than on the directml

[meager carbon]

once in ui it doesn't use gpu

haha

im reinstalling it though

since i did fuck up my first time

whyyyyyyyy

[hard quiver]

if this is not ticked. I can't see most of my loras. but if I do click it. I can see most of them though still no point. cuz they aren't working for me. why are most of my loras incompatible and what can I do for them to be compatible.

[dusk shale]

oh it is not the 840000-ema-pruned, thx gonna check it out

[meager carbon]

aren't loras version based?

so like some are for xl, others are for 1.5...

[hard quiver]

what do u mean?

[meager carbon]

.

[hard quiver]

i actually don't know what that means.

sry

[meager carbon]

ehm

like

there are different sd models

sd 1.5, sdxl, sd turbo...

[hard quiver]

https://tenor.com/view/nervous-waiting-daunting-no-spongebob-squarepants-gif-5281390

[meager carbon]

and loras are designed for them in specific. so only some loras can work with some models

[hard quiver]

ooh. how do I check which one I have

[ornate elk]

have you tried to delete the venv run pip cache purge and relaunch again?

[meager carbon]

yep

[ornate elk]

for sdnext

[meager carbon]

mhm

ill try once again though!

since my time is worth nothing

[meager carbon]

look at the lora page, it will say there

[ornate elk]

i wasted 2 hours installing RAM into my rig to find out it wont even boot, today

[meager carbon]

OOF

[ornate elk]

my evening is so wasted xD

but now i know, dont buy an additional dualkit of DDR5 6000MHz Ram because Mainboards cant even boot on 5400MHz with 2 dualkits
So back to only 1 dualkit

[hard quiver]

couldn't find it.

[meager carbon]

@hard quiver

[hard quiver]

oh no. I know that

[meager carbon]

well... does it match!

[hard quiver]

but how do I find out what sd version I have installed

[carmine canyon]

In ComfyUI, I'm trying to use WD 14 tagger, However I'm running into this error

2024-04-08 18:06:56.8751234 [E:onnxruntime:Default, provider_bridge_ort.cc:1351 onnxruntime::TryGetProviderInfo_CUDA] D:\a\_work\1\s\onnxruntime\core\session\provider_bridge_ort.cc:1131 onnxruntime::ProviderLibrary::Get [ONNXRuntimeError] : 1 : FAIL : LoadLibrary failed with error 126 "" when trying to load "D:\AI\ComfyUI_windows_portable\python_embeded\Lib\site-packages\onnxruntime\capi\onnxruntime_providers_cuda.dll"

RuntimeError: D:\a\_work\1\s\onnxruntime\python\onnxruntime_pybind_state.cc:636 onnxruntime::python::CreateExecutionProviderInstance CUDA_PATH is set but CUDA wasn't able to be loaded. Please install the correct version of CUDA and cuDNN as mentioned in the GPU requirements page (https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html#requirements), make sure they're in the PATH, and that your GPU is supported.

here are the logs of me trying to install onnxruntime and test if my installation is correct (at the end of the log you can see there are no errors because python doesn't throw an assertation error, indicating python can find my cuda installation):
https://pastebin.com/dYbnf85b

These are the sources i was using:
https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html#requirements
https://huggingface.co/docs/optimum/main/en/onnxruntime/usage_guides/gpu#checking-the-cuda-installation-is-successful

[ornate elk]

Models that are 2-4gb are 1.5 based models.
6gb files are mostly sdxl models

The Lora's only show up when a compatible model is selected

[hard quiver]

its around 9.5 gb

[ornate elk]

What's the name of it?

[hard quiver]

how do I check that?

[ornate elk]

In the models/stable-diffusion folder

There are the models stored

[frail verge]

Any idea why AUTO1111 forgets my settings in ui-config.json? 🤔

It keeps resetting my Sampling Method and Sampling Steps to DPM++ 2M and 20 even though I set them to Euler a and 30 respectively

[lament basin]

did anyone had the issue when using the API that frequently the generated pictures will be 0kb? what is causing this?

[hard quiver]

oh. I see

[cyan valve]

Hi guys, I’m new to SD,I have installed automatic 1111 on my local but how can i check my version is sd1.5 or sdxl ?

[silk cairn]

Hi guys, can some1 point me to a simple LoRA guide for beginners. I installed LoRA Easy Training Scripts but I have no clue how to use it

[solemn moon]

Intro to LoRA’s by Laura 😝

https://youtu.be/ZHVdNeHZPdc?si=Lv1hKFnHcUO2Vwe0

[eternal holly]

can i training a checkpoint model base on models from civitai? are there any requirements for model file? like pured?

[mossy aurora]

When I use SD Forge, after selecting the model in the controlnet, an error is displayed. Why?

Maybe I should ask another way: I installed SD1.8 and SD forge at the same time, but I don't know how to install forge's controlnet correctly and how to use preprocessor and preprocessor model correctly in forge?

[sinful halo]

Depends on what checkpoint you use, some checkpoints aka models are sd 1.5 and some are sdxl, search for sdxl and filter by checkpoint on civitai for sdxl models

Remember that text embeddings and control net models are different from sd 1.5 to sdxl

Idk about loras

[mossy aurora]

When starting, the version will be displayed in the terminal.

[ornate elk]

Check the models page on civitai.
Usually models with 2-4gb are 1.5 based and models with 6gb are sdxl based

[ornate elk]

Python 3.11 is not supported. Also wrong installation method if this is auto1111

[sinful halo]

Man, sorry to bother you again, it's only that even if I use a sdxl model that support certain resolution, 1344x768 landscape in this case, I keep getting multiple characters even if I use a prompt like this; ((1girl:1.5)), sometimes even with vertical resolutions ;-; do you think my prompt is the problem ?
I used every negative word I could've imagined too

[ornate elk]

Hey, do you have an example?

[sinful halo]

Like what ? A image ? Im on mobile now sadly, ;-; but imagine a lot of the same character randomly put in the same image, sometimes the aí even divide them using black frames like a image grid

I'll send some images when I get home

[gritty ridge]

Zero123 is adopted for MAC?

[honest tendon]

How do I actually get good images

this rn is the best pic of a fox i got

[ornate elk]

Yes, then we can take a look.
Normaly sdxl shouldn't create doublicates at that resolution

[ornate elk]

You need to download other checkpoints (models) from civitai.com
These models are trained on higher quality images and will give better results.
For example download dreamshaper v8 from here and put it into the models/stable-diffusion folder.
Then reload the checkpoints on the top right and select it there.
You will automatically get a better fox.
Here is the model:
https://civitai.com/models/4384/dreamshaper
Then work on your prompt. Add some negative tags like, blurry deformed, mutated, onto the negative prompt field.

[honest tendon]

this is the best

i got

[ornate elk]

Yea the 1.5 EMA pruned default is not a good model

[honest tendon]

is it this

[ornate elk]

This is the refresh/reload button

[honest tendon]

im dumb i just realised

FINALLY

A good fox

[honest tendon]

Can I download like multiple models and combine them all

or is that not a thing

also what are workflows used for

[vocal burrow]

Yes you can merge multiple models into one. (it takes some skill to do correctly).
No you can't use multiple models at the same time.

[honest tendon]

Is there a tutorial for it, sounds fun, then again chaotic

probably better just using it individually

[vocal burrow]

Gonna need a more precise questions to answer that.... Asking "what workflows" in A111 is like asking '"what workflows" in photoshop.

[honest tendon]

my bad

[vocal burrow]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#checkpoint-merger

[honest tendon]

thank you

[ornate elk]

You can download multiple models and switch between them.
But Lora's can be used together with models

To get a specific style or character

[slate hatch]

Traceback (most recent call last):
File "C:\Users\august\Desktop\stable-diffusion-webui\venv\lib\site-packages\gradio\routes.py", line 488, in run_predict
output = await app.get_blocks().process_api(
File "C:\Users\august\Desktop\stable-diffusion-webui\venv\lib\site-packages\gradio\blocks.py", line 1431, in process_api
result = await self.call_function(
File "C:\Users\august\Desktop\stable-diffusion-webui\venv\lib\site-packages\gradio\blocks.py", line 1103, in call_function
prediction = await anyio.to_thread.run_sync(
File "C:\Users\august\Desktop\stable-diffusion-webui\venv\lib\site-packages\anyio\to_thread.py", line 33, in run_sync
return await get_asynclib().run_sync_in_worker_thread(
File "C:\Users\august\Desktop\stable-diffusion-webui\venv\lib\site-packages\anyio_backends_asyncio.py", line 877, in run_sync_in_worker_thread
return await future
File "C:\Users\august\Desktop\stable-diffusion-webui\venv\lib\site-packages\anyio_backends_asyncio.py", line 807, in run
result = context.run(func, *args)
File "C:\Users\august\Desktop\stable-diffusion-webui\venv\lib\site-packages\gradio\utils.py", line 707, in wrapper
response = f(args, **kwargs)
File "C:\Users\august\Desktop\stable-diffusion-webui\modules\call_queue.py", line 95, in f
mem_stats = {k: -(v//-(10241024)) for k, v in shared.mem_mon.stop().items()}
File "C:\Users\august\Desktop\stable-diffusion-webui\modules\memmon.py", line 92, in stop
return self.read()
File "C:\Users\august\Desktop\stable-diffusion-webui\modules\memmon.py", line 77, in read
free, total = self.cuda_mem_get_info()
File "C:\Users\august\Desktop\stable-diffusion-webui\modules\memmon.py", line 34, in cuda_mem_get_info
return torch.cuda.mem_get_info(index)
File "C:\Users\august\Desktop\stable-diffusion-webui\venv\lib\site-packages\torch\cuda\memory.py", line 618, in mem_get_info
return torch.cuda.cudart().cudaMemGetInfo(device)
RuntimeError: CUDA error: device-side assert triggered
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

[honest tendon]

do i put these 2 into the Model/Stable diffusion model at the same time

[slate hatch]

vae goes in your vae folder, full model (if it has pruned in username should include vae if not) put in stable diffuson model folder

[ornate elk]

Make sure your nvidia driver is updated

[vocal burrow]

VAE quick guide :
1/ What is a VAE.
It's a part of the stable-diffusion pipeline that encode/decode information from tokens to latent space and from latent spaces to pixels. Aka it transforms math statistics in pictures
2/ Where do I put my VAE ?

• VAE with .vae.pt, .vae.ckpt, .vae.safetensors extensions go into the models\Stable-diffusion folder
• VAE with .pt, .ckpt, .safetensors go into models\VAE
  3/ How do I use my VAE ? Three possibilities :
• Either you name it similar to another one of your model (eg : Anything-V3.0.safetensors + Anything-V3.0.vae.pt), by doing that it should automatically load the VAE when you load the associated model.
• You manually load your VAE by going to Settings -> Stable-Diffusion -> sd_vae and selecting your VAE
• You add an easily accessible VAE dropdown at the top of your page to quickly switch back VAE by adding sd_vae to your Settings -> User Interface -> Quicksettings list

so tldr, yes get both

[vocal burrow]

(the fact that the model is pruned does not mean it necessarily embed the correct VAE)

[ancient plover]

Hey, I just downloaded the 'rich' extension through pip for proper syntax highlighting / text color for python in the console, (Windows 11, zluda stable diffusion webui). I want to auto apply it for stable diffusion, so its easier to work around errors. Normally, you need to run something like "from rich import print" every time you start a new prompt or instance of python before it will apply... I want it to auto enable for all stable diffusion stuff, does anyone know if it would work to add the line to my webui-user.bat file? So it would look like this: if not, is there another way?

[slate hatch]

updated 😦

still get the issue

[honest tendon]

thank you

[slate hatch]

@ornate elk works rn

but ill try with a more extending prompt

if it bricks gg

(advanced prompt)

[vocal burrow]

60 s/it OO wtf, what s your gpu ?

[slate hatch]

i might need an upgrade XD

[vocal burrow]

maybe... depends what gpu your toaster is equiped with

[slate hatch]

2060... (at the time my toaster was considered a good pc) :*(

[vocal burrow]

ok those are NOT 2060 number

something is wrong with your install

[slate hatch]

is 2060 good or bad?

XD

prob something wrong with my install

[vocal burrow]

you should get something like 5it/s on some basic 1.5 512x512 test

not 60s/it

[slate hatch]

is 60s/it bad XD

[vocal burrow]

ooooh you re using animatediff

nevermind

moving forward

[slate hatch]

yeah....

im tryna get into animation

for stable cause im working on a yt video

and tryna test some stuff out

[vocal burrow]

good luck

[slate hatch]

and temporal kit + ebsynth

is too hard

😦 my prompts always = blurry

might just be skill issue

[slate hatch]

is animatediff not the way to go?

[vocal burrow]

there are many workflows for video. I m not an expert on that subject, animatediff is a good point of entry.

It s just that generating animations/movies is way more taxing than creating still pictures

[honest tendon]

So small file goes to models\Stable-diffusion folder, and big file in models/VAE

[vocal burrow]

no

[honest tendon]

but

[vocal burrow]

big file is a stable diffusion model so it goes in models/stable-diffusion

[vocal burrow]

yes but it s not a VAE

[honest tendon]

OH

[vocal burrow]

only the small file is a VAE

[honest tendon]

makes sense

my bad thank you

[keen edge]

Hey, I still have the issue with NetControll not saying ticked when hitting apply, which then prevents me from selecting model in openpose. On recommendation of someone here, I excluded site from AdBlock but it didn't helped

[digital dew]

Hello! I have a small question about VAE.

For some reason, when generating during the entire process, ''Livepreview'' appears normal, but after completion to 100%, the final file comes out like on the screenshot.
This only happens when using VAE 840000 and PonyV6 checkpoint, but after searching in CIVITAI many people have used Pony with VAE840000 and for some reason they don't have a similar problem and I don't understand what could be wrong in my case. Has anyone encountered something like this?
p.s. There is a ''built-in'' VAE in PonyV6, but it seems that on CIVITAI, if I'm not mistaken, they used 840000 instead of it.

[ornate elk]

84000 VAE is an 1.5 VAE and won't work for sdxl based models

You need an sdxl based vae

[honest tendon]

Whats the maximum resolution I can process without crashing

[digital dew]

got it.
Are there any good similar VAE like 840000 but for SDXL?

[gritty ridge]

Where to get new Lora configuration file?

[vocal burrow]

depends of your config and settings. In theory stupidly high if you know what you re doing

[meager carbon]

Infinite in theory

[trim raven]

Hello, how do I properly uninstall an extension that has been installed from github?

[vocal burrow]

you delete its folder in the extensions folder.

[trim raven]

there's nothing to delete in the extension tab?

[ornate elk]

Yes the official sdxl VAE and the sdxl VAE fp16

[hard quiver]

so if im using batch count to generate 100 pics the quality of those pics will not deteriorate afaik though the downside is I can only get 100 pics at a time. so my question is if I use batch size along with batch count to get more than 100 pics. or will the quality worsen then?

[ornate elk]

The quality won't be affected.
Batch count = the amount of images generated one by one
Batch Size = the amount of images generated at the same time (requires more vram)

[hard quiver]

can u elaborate on what u mean by at the same time

cuz I was hoping that batch size works in a a way so that as soon as the first 100 batch count is generated. will move to the second batch count of 100 and so on, that is if the batch size is set to 2 and batch count to 100

so I'll be getting 200 images total.

whats making me confused is what u mean by at the same time. cuz how is quality not getting compromised if ur getting pics at the same time.

[ornate elk]

Stable diffusion loads the model into the vram and creates an image by denoising random noise.
This is a vram heavy task but doesn't have anything to do with the quality.

If you have enough GPU vram you can set the batch size to a higher value.

What affects the quality is:
The model
The Prompt
The resolution
The right amount of steps
A good cfg value
The sampler
Upscaling
Extensions
Lora's
Embeddings

[hard quiver]

oh. so its not like gpu vram will be divied b/w 100 pics or something

xd

cuz thats what batch size sounded like until now

i have a 3080ti

and 32 gigs ddr5 ram

about time I check how many its im getting

[vocal burrow]

nope

[ornate elk]

No the vram will just get filled up more

[hard quiver]

ur supped to see them through the command window thing right?

[ornate elk]

Yes

[hard quiver]

hoping its somewhere along 6-7 atleast

i'll try batch count 100 and batch size 2 and lets see

[ornate elk]

Out of vram

What settings did you used?

[quiet swift]

Hello, I have installed Deforum in SD. SD does work. But when I try to generate videos with Deforum, I always have this error message : URLError: <urlopen error [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: unable to get local issuer certificate (_ssl.c:997)> . I also have exactly the same message when I go to extension tab, then Available, then Load from. If someone had an idea, thank you very much

[vapid sluice]

Is using Pinokio impossible if there’s spaces in my user file name and I can’t change it ?

[digital dew]

thanks!

[vocal burrow]

what country are you from ?

[honest tendon]

I think i maxed out something but I forgot

Oh i maxed out upscale

[quiet swift]

I'm from France. I did use Deforum without trouble since some months.

[vocal burrow]

no vpn ?

[vocal burrow]

no vpn ? @quiet swift

[quiet swift]

I do not know what is a VPN. I'm on Mac by the way

[vocal burrow]

ok, can we get a full log ?

(copy paste everything from the console into a file and drop said file here)

[quiet swift]

do you speak to me ? 🙂

[vocal burrow]

yes

[quiet swift]

ok, thank you but I do not know what is a "full log" and what do you intend by "from the console". I can give you the settings of my video if this is what you ask ?

[vocal burrow]

the white text from the black window thingy that shows up when you launch webui.sh

[quiet swift]

ok 🙂 I give you a captur of my screen Terminal when I launch webui. thank you if you can help 🙂

[vocal burrow]

ok but one including the error

and please copy paste the text into a file and drop the file in here it will make both your and my life easier when the log gets big

[fallow arch]

some body konw this problem? please

3 error

[quiet swift]

ok, sorry. Here is a file with the Terminal Text with the error, which is always the same. It is also the same error message when I go in SD to "extension" tab, then "Available", then "Load from"

[ornate elk]

What webui did you installed?

It doesn't look like the normal automatic1111

[hard quiver]

can someone tell me how to utilize clip skip in stable diffusion

I can't even seem to find it.

[ornate elk]

Settings, User Interface, Quicksettings,
there add Clip stop at last layer

Then apply and reload

[hard quiver]

thanks. worked like a charm.

[fallow arch]

I checked this error on Google, but there is no solution for this error.

[fallow arch]

NORMAL WEBUI

BASIC

[ornate elk]

Can you link me from where you got it?

[fallow arch]

wait a minute

[ornate elk]

Because the basic webui doesn't have Aki-v4 in its name

[fallow arch]

https://pan.quark.cn/s/38230018837e#/list/share

here

I come from CHN

[ornate elk]

That's a Chinese fork of the automatic1111 webui

Or a repack maybe

[fallow arch]

This is the stable diffusion starter modified by our Chinese people.

[fallow arch]

yes

This version often causes various problems

[fallow arch]

This fork no python

Only this interface is also the launcher

Moreover, our network often has problems, even if the VPN is disconnected, the problem persists.

[hard quiver]

does the preferred weight in lora metadata has any importance? should I leave it be to 0? or change it to something?

[ornate elk]

Ah okay is it based on the latest version of automatic1111?

[ornate elk]

The text2video extension might be outdated

[fallow arch]

LET ME SEE

[fallow arch]

HERE

[fallow arch]

OH,LET ME CHECK

Is my error report the same as theirs?

its his error

[ornate elk]

yes its the same

[fallow arch]

OK I try

[hard quiver]

are these speeds good for a 3080ti?

[fallow arch]

here rename ok？

Is it the modelscope in this folder?

[ornate elk]

its probably the modelscope inside the venv folder

[fallow arch]

venv>

?

[ornate elk]

do you have a venv folder?

[fallow arch]

NO

[fallow arch]

HAVE

I found it

[vocal burrow]

@karmic crown you ve seen something like that before ?

[fallow arch]

and then？

[vocal burrow]

@quiet swift your macbook is updated ? its hour/time is set correctly ? internet connetion is stable ?

[ornate elk]

i thought maybe there is a second modelscope folder inthere

but you should check the whole webui folder for them

[hard quiver]

why is there such a difference in speeds?

[vocal burrow]

assuming you re using hires.fix / upscale then the resolution

first pass is low res, second pass is high res

[fallow arch]

OK

[quiet swift]

yes my macbook is updated and hours and time are correct. And no problem with Internet.

[fallow arch]

Too much

@ornate elk

may I should delete which folder

[vocal burrow]

try to open a console in stable-diffusion-webui folder and run venv\scripts\pip install --upgrade certifi

[quiet swift]

ok. thank you. To be sure that I understand, I write what you say ("venv/scripts...") in my Terminal after having launch again SD in terminal ?

[vocal burrow]

first you navigate to your stable-diffusion-master folder then you run the command

[hard quiver]

ok. but how do I know the generating power in general thats being utilized

[vocal burrow]

all of it

[hard quiver]

like i have a 3080ti. I wanna benchmark that with other gpus like 4090. lets say

[vocal burrow]

it always use as much resources as possible

[hard quiver]

to see how slow mine is

[ornate elk]

Use a 1.5 2gb Model.
Resolution 512x512
30 steps.
And for the prompt only
Cat in positive and
Blurry in the negative prompt
Without upscaling or anything

then you get it/s to compare with others, when they use the same settings

[vocal burrow]

Just pick a prompt and its settings, stick to it and compare its results ? do a simple "cat" promp test at 512x512, 30 steps, Euler A, no upscale and voila

[quiet swift]

sorry but I don't understand where I must write your command. I have tried to write it in Terminal (which is the application where you launch SD on mac) but nothing happens.

[ornate elk]

idk which one to rename

[vocal burrow]

open a terminal,

cd /Users/arnaudbigeard/stable-diffusion-webui
./venv/scripts/pip install --upgrade certifi

[quiet swift]

ok, thanks. I did write your command. Terminal reply is : "zsh: no such file or directory: ./venv/scripts/pip
arnaudbigeard@MacBook-Pro-2 stable-diffusion-webui % "

[vocal burrow]

commuting back home I ll check once I get there

[quiet swift]

ok, thanks

[hard quiver]

does it make a difference if I use prompt with or without gap

i.e cat, mouse, dog
i.e cat,mouse,dog

notice one has gap after comma. other doesn't.

[hard quiver]

@ornate elk sry for tag. can u tell me this one last thing.

is there a universal weight I need to set for lora. or should I jus apply lora and let it remain 0

[fallow arch]

IDK?

HHH

maybe here?

he said text2video

[upbeat atlas]

@ornate elk hi, I was trying to re install Zluda so I took a look at the pinned message. If I remember correctly, we had to copy a couple files in the Zluda folder, rename them and paste them in specific folders, do we no longer need to do that?

[thorn thorn]

I'm new to using python to load models. I'm trying to load both stabilityai/stable-diffusion-xl-base-1.0 and the stable diffusion 1.5, but they seem to be too big for my 6gb of vram.

from diffusers import StableDiffusionXLPipeline
import torch

pipe = StableDiffusionXLPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16, variant="fp16", low_cpu_mem_usage=True, use_safetensors=True,
)
pipe = pipe.to("cuda")

Is there anything I can do to make it load (partially?). or should I use some smaller model? How do ComfyUI and Automatic11111 do this (they work for me)?

    pipe = pipe.to("cuda")
               ^^^^^^^^^^^^^^^
...
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 20.00 MiB. GPU 0 has a total capacity of 5.77 GiB of which 42.31 MiB is free

[ornate elk]

hey, thats no longer needed

as this step is automaticly done by the webui

[ornate elk]

every lora is different, the prefered start value should be 1

[hard quiver]

so change all of these to 1 instead of 0 and see what happens. right?

ok

[hard quiver]

also this

[ornate elk]

normaly dont change anything if its not the default value

lora weights have to be changed in the prompt

[hard quiver]

every lora Ive installed had this value to 0 by default.

[hard quiver]

i see

[ornate elk]

you have to try a lora and play around with the weight, and if it needs a specific weight then you can edit the slider

every lora starts with 1 and the strenght goes from -2 to 2

[hard quiver]

alright. I get it now. makes sense

[hard quiver]

and can u tell about this? @ornate elk

some prompts have gaps. others don't

[ornate elk]

i normaly make spaces after a ,

thats how it was since the beginning

[hard quiver]

ah ok

thanks

[flat hemlock]

Hi, can you give me some advice ? In "My Machines", my machine is marked FAILED, and I am unable to start it. Was Ok yesterday.

I am new to this, so any help is appreciated. Thanks.

[vocal burrow]

open a terminal,

cd /Users/arnaudbigeard/stable-diffusion-webui
./venv/Scripts/pip install --upgrade certifi

[vocal burrow]

we have NO IDEA what you're talking about.
what is "My machines" ?
"unable to start it", what is "it" ?

[upbeat atlas]

@ornate elk hi, so I finally fixed the crashing problem. I was messing around with the arguments and I realized I get no crashes when I remove --medvram-sdxl . I still get the same exact speed of generation only now I don't get any crashes. If anyone else has a similar problem they could try this perhaps.

[ornate elk]

okay thats interesting, thx for letting me know

[flat hemlock]

Sorry, perhaps I am not in the right place. I have a problem starting a machine in Think Diffusion. I can start a new machine, but cannot start the one I created yesterday.

[vocal burrow]

Not necessarily the wrong place but you have to give the full details when askin for help. And at the same time you'll probably get better chances asking in their discord. There's probably few people here using thinkdiffusion, we mostly run SD locally.

[flat hemlock]

ok thanks

[dim sedge]

hey can someone help me? ive used loras before and now i wanted to try SDXL Loras. so i installed forge webui (similar to A1111) and SDXL. I think the SDXL seems to work now, i use the model from civitai and the refiner. But now i cant get the lora i want to use to work. (this one: https://civitai.com/models/129556/ps1-graphics-sdxl) The image seems unchanged in style. is it because of the refiner? Do I need to use comfyUI for this?

[chilly knot]

pickles imports detected how is dangerous?

[bitter zenith]

what?

[sinful halo]

@ornate elk Hey man. here are the examples you asked for

and both of my prompts

sometimes i get good ones, one in fifth lol (grid image to show how many are the bad ones)

Model; Animagine SDXL
Sampling; DDIM - 35 steps
No hi res fix at 1344x768 at both vertical and horizontal

i noticed that tiny dupes were caused by the word "doll", but the big ones i have no idea, idk even if this is a prompt problem ... i dont wanna change the model tho, i really liked the good results i got

[ornate elk]

dont use 1girl:1.5

1.5 is to forced

[sinful halo]

ok, ill try

[ornate elk]

also i would lower the lora doll joints maybe

[sinful halo]

ok. good idea

[near acorn]

for w/e reason my art coems otu a bit crystalzied

any idea why?

loosk about fien when loadign but then when tis done its all like that

[ornate elk]

What settings did you used?

[near acorn]

is that wym or

[ornate elk]

Set the upscaler to Resrgan4x+ Anime6b and Denois to 0.5

And hires steps to 10

[near acorn]

well it does this issue even without hiresing

still does with all that too

[ornate elk]

Try DPM++ 2M Karas as sampler

[onyx solar]

Should be using easynegativev2 with that model as well, to help with outputs

[near acorn]

whats that

[onyx solar]

Read the model page:
https://civitai.com/models/4468/counterfeit-v30

When you download models, you need to read. Often-times, there is important information to help with output.

[near acorn]

im not sure wich oen to get

[onyx solar]

the pt since you're using A1111

But as was also suggested, you should adjust your sampler to match what is in the examples, too.

[near acorn]

YEAH BUT THATS NOT THE PROBLEM

oops capslock

it isnt the problem too

cuz that was just a random test

[onyx solar]

Samplers matter.

[near acorn]

every imagie seems to come out that way

[onyx solar]

Change away from the SDE sampler.

You are using Heun

[near acorn]

i normaly litrealy use 2m karas

[onyx solar]

Better.

[near acorn]

well thats already a good step

ty

[severe kindle]

Pickle files can contain runnable code , that executes when the model is loaded, they should really be avoided at all cost IMO unless you totally trust the source.
It could be safe or you could load it and runs a background task to delete your files, or install a backdoor.

[chilly knot]

its audio model

not recomended use it?

[timber oak]

@ornate elk well we tried your pinned guide

my guy

it fucking worked like a charm

my guy, you helped me out a lot several times

love u bro 🫂

[fair oxide]

Does anyone know how to change stable diffusion relying on cpu?
I have an installation problem saying that torch cannont use gpu

[limber pier]

prolly the wrong version of torch

[remote ivy]

Anyone else find that forge uses really high amounts of ram, or is it something wrong with my setup?

[fair oxide]

how could i fix this

[limber pier]

does it say what version you're running in the console

[fair oxide]

how can i check for the version for torch

it just says this

[limber pier]

K so it's not the CPU version then

Looks like you either don't have a GPU, it has failed/died, or it's not detecting it

[ornate elk]

Hey, what's your GPU?

[sinful halo]

so, i let it overnight and sadly same results, 1 in 50 is a good solo girl, the rest are multiple portraits or somethin

new prompt, i added a lot of negative words and some sd xl negative embbedings to it

[honest tendon]

How do you use Lora

nvm

just found the tutorial

[severe kindle]

I wouldn't use in on my local computer unless you trust the user, perhaps contact the author a see if they can update thre code and convert the moel to safetensors.
Its about time the AI world got serious about security.

[ornate elk]

I think you can you try to remove
"Great composition" and
for sure remove "concept art" as this could trigger it for multiple poses of a person in one image

[fierce elm]

Hey guys. Does anyone know how to use Demofusion in webui? I tried but the output looks missing a big piece and blurry compared to normal output.

is it something wrong because of settings? I kept them default

[sinful halo]

yeah lol, why did i put concept art xD bro, thank you for sharing some brain lol

[chilly knot]

dont understand hhow to download

[honest tendon]

whats the difference

[honest tendon]

you did something wrong

maybe you messed up a path

or simply your computer needs a restart

[vernal cape]

Hey everyone,

I could really use some assistance with something. I'm trying to figure out how to create a new artistic style by combining influences from multiple artists using stable diffusion prompts. I've experimented with various examples, like Artist Name 1 and Artist Name 2, as well as trying to emulate the styles of individual artists like Artist Name 1 and Artist Name 2. However, despite my efforts, I'm not achieving the results I hoped for. Any suggestions or guidance would be greatly appreciated!

Also, does anyone know of any straightforward stable diffusion prompts that I could use to test and experiment with different artist styles?

Thanks in advance for your help!

[visual warren]

bot

[umbral fox]

I'm using mov2mov from Stable Diffusion web ui,
The video is not generated even after more than an hour.
What settings do I need to make to generate videos?
The mov2mov items are still displayed on the screen.

■Environment
Stable Diffusion web ui ver.1.8.0
■How to install
Use the installation URL from github (https://github.com/Scholar01/sd-webui-mov2mov.git) and install from the URL on Extensions on Stable Diffusion.

■Checked "Allow other script to control this extension" in ControlNet

■Things used for operation confirmation
・Sample video of about 10 seconds
・Resize mode: "Crop and resize"
Video output size
Width:
512
Height:512
・Movie FPS:10

■Other concerns
The following message is displayed in red in the Movie Editor item.
[This feature is in veta version!!
It lnly supports Windows!!!
Make sure you have installed the ControlNet and IP-Adapter models.]

The PC I am using is Windows,
ControlNet is installed,
I have already installed IP-Adahttps://huggingface.co/lllyasviel/sd_control_collection/tree/mainpter from the following site, but the error message does not disappear.

・ip-adapter_sd15.pth
・ip-adapter_sd15_plus.pth
・ip-adapter_xl.pth

[severe kindle]

I don't know, I never had to download an lfs file from github, only from hugginface and they have a convenient download button for the items.
What Github repo is this.
You could clone the repo, but that will be slow and could download several times the amout of data than just downloading the model.

[chilly knot]

its safe downolad safetensors version this pickles file?

user made safetensor version

[hard quiver]

whats a pickle file?

[ornate elk]

Mov2mov is broken and needs a manual fix.
Here is a link to the fix:
https://github.com/Scholar01/sd-webui-mov2mov/issues/140#issuecomment-2017033461

[bitter zenith]

what

[hard quiver]

so never to download pickletensors even if the creator is trusted?

[chilly knot]

maybe you can

[hard quiver]

wym?

[chilly knot]

you can if you trust and people not said is bad

[hard quiver]

i see

[chilly knot]

but i dont like any risk trust or not

[umbral fox]

Thank you!!
I will try to fix it in my own environment.

[severe kindle]

From python's own documentation
https://docs.python.org/3/library/pickle.html#

safetensors is a format for tensors that does not execute arbitary code so at the moment safetensors are considered a secure format.

[quiet swift]

Hello, I open a terminal and wrote your command. Terminal reply is : zsh: no such file or directory: ./venv/Scripts/pip
arnaudbigeard@MacBook-Pro-2 stable-diffusion-webui %

[vocal burrow]

ok what's in the /Users/arnaudbigeard/stable-diffusion-webui/venv/Scripts/ folder ?

[quiet swift]

inside the stable-diffusion webui/venv/ I don't have a "scrpit" folder. Inside venv folder, I have 4 folder = bin - include - lib -share. And I have 3 files =dev.txt - pyvenv.cfg - x.txt

[vocal burrow]

inside bin ?

[quiet swift]

Inside bin, there is 52 elements.

[vocal burrow]

is there pip ?

[quiet swift]

there is pip pip3 and pip3.10

[vocal burrow]

ok then

cd /Users/arnaudbigeard/stable-diffusion-webui
./venv/bin/pip install --upgrade certifi

[quiet swift]

Terminal did reply : Requirement already satisfied: certifi in ./venv/lib/python3.10/site-packages (2024.2.2)

[notice] A new release of pip available: 22.3.1 -> 24.0
[notice] To update, run: python3.10 -m pip install --upgrade pip
arnaudbigeard@MacBook-Pro-2 stable-diffusion-webui %

[vocal burrow]

Is there a "Install Certificates.command" file in your /Applications/Python 3.10 folder ?

[quiet swift]

yes, there is this file in Application/Python 3.10 folder

[vocal burrow]

run it

[quiet swift]

when I double click on it (is it "run it"?) there is a window that open in which is written : Last login: Wed Apr 10 10:19:43 on ttys000
/Applications/Python\ 3.10/Install\ Certificates.command ; exit;
arnaudbigeard@MacBook-Pro-2 ~ % /Applications/Python\ 3.10/Install\ Certificates.command ; exit;
-- pip install --upgrade certifi
Collecting certifi
Using cached certifi-2024.2.2-py3-none-any.whl (163 kB)
Installing collected packages: certifi
Successfully installed certifi-2024.2.2

[notice] A new release of pip available: 22.3.1 -> 24.0
[notice] To update, run: pip3 install --upgrade pip
-- removing any existing file or link
-- creating symlink to certifi certificate bundle
-- setting permissions
-- update complete

Saving session...
...copying shared history...
...saving history...truncating history files...
...completed.

[Opération terminée]

[vocal burrow]

try again to run the stuff that gave you error yesterday

[quiet swift]

ok

[quiet swift]

It does work !! :)) Thank you very much. After having generated a video, I went in "extension" tab, then Available, then Load from, just in order to see if the same error message appeared. And the list appears, there is no more error message. Thank you !!! :))

[honest tendon]

where can I save this

[trim raven]

this morning I tried to reinstall my environment but I got this error

[honest tendon]

whats the difference between Pruned and Full?

[merry fractal]

/pic

/pic a cat in a house

[honest tendon]

wtf r u typing

[merry fractal]

pls help me

/pic how to create a picture

[honest tendon]

How do i make it more vibrant

Im trying to make it look like this - from https://civitai.com/images/55758

how do i force my Webui to like use the VAE given

https://huggingface.co/samle/sd-webui-models/blob/23997884993f591d36bed7c645892d3d0828017f/pastel-waifu-diffusion.vae.pt

[ornate elk]

First go into settings, User interface, quicksettings, there add sd_vae
Then hit apply and reload ui

Then download a VAE and drop it into the models/VAE folder

Then you can select it in the dropdown

[vocal burrow]

what is "this" ?

[honest tendon]

Will this still work if the VAE is already in the folder?

[vocal burrow]

pruned <=> trimmed off, it's lighter model. They removed part that of it that are useless 99% of the time. Making it way lighter.

[pastel steeple]

anyone knows how long it takes for a credit purchase to become active? I've bought credits 5 minutes ago but I still get an error about insufficient funds if I try to generate an image via API

[honest tendon]

awww

AHHHH

[pastel steeple]

works now

[ornate elk]

You need to select the vae

In the dropdown

[upbeat gazelle]

I have an issue with 1it/s no matter what in forge and sdxl models on a 4090. I have installed cuda with cudnn update, pytorch, latest gpu drivers etc etc. On win 11. Anyone encountered something similar?

Googled all day and tried whatever fixes I found, no change.

[hard quiver]

what is this supposed to mean lol?

can i use it with anymodel and stuff?

[light lynx]

I get the error:

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

anyone know any fixes?

[ornate elk]

hey, whats in your webui-user.bat?

[ornate elk]

whats your gpu?

[light lynx]

gtx 1080 ti

[hard quiver]

can someone tell me how do I use embeddings?

[light lynx]

you can find youtube tutorials

it's not hard to look for

[brisk sigil]

stick them in embeddings folder and then just call them like a lora in a1111

[hard quiver]

am I supposed to put them in adetails folder

[hard quiver]

alright. thanks

[light lynx]

my bad not ti it's just gtx 1080

[light lynx]

hello?

[brisk sigil]

not properly installed? start over https://stable-diffusion-art.com/install-windows/

[upbeat gazelle]

the default https://pastebin.com/VZ8vsLyV Ive tried with --xformers, but apparently it's not needed for forge

[ornate elk]

okay, make sure you have the latest nvidia driver

[ornate elk]

yea, we should focus first on getting auto1111 to work, then forge will work too

so whats in your auto1111 webui-user.bat?

[light lynx]

I do

[light lynx]

started over like 10 times

[upbeat gazelle]

I don't have it installed right now, hold on.

[ornate elk]

this here is the recommended install guide for auto1111 on nvidia:
#🤝｜tech-support message

check if you have done these steps

[light lynx]

I have

[ornate elk]

okay, can you share a screenshot of the whole cmd after launching it?

[light lynx]

Sure

[ornate elk]

follow my install guide from here:
#🤝｜tech-support message

[brisk sigil]

ewall - try deleting your VENV folder

[light lynx]

Like I said I reinstalled it 10 times

[ornate elk]

okay, open up a cmd and run
pip cache purge
then delete the venv folder and relaunch the webui-user.bat

[light lynx]

even starting from scratch

[ornate elk]

also do you have an intel cpu ?

[light lynx]

yes

[ornate elk]

check in the taskmanager under the performance tab if their is GPU0 and GPU1 or only GPU0

[light lynx]

one second still try the purge thing

mmm

gpu 0

[ornate elk]

okay so only the nvidia gpu is listed, that would be good

[light lynx]

so what does that mean exactly?

[ornate elk]

that does mean it didnt used the intel igpu

[upbeat gazelle]

Installed and ran a very simple prompt at 1024x1024 in a1111, it's at 9it/s, so that's promising. Opening forge again and its installing a bunch of stuff
inside the a1111 .bat:
@echo off

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

call webui.bat

[ornate elk]

looks good, also 9it/s is good if you used an sdxl model (6gb), on 1.5 models it should be much faster on 512x512

[upbeat gazelle]

it was the default downloaded model sadly.. 😩

[ornate elk]

oh then try it at 512x512

because the default is an 1.5 based model

not made for 1024x1024

[upbeat gazelle]

waiting on few installs on forge

[ornate elk]

did you retried after purging the cache and deleting the venv folder?

[light lynx]

yup same thing

[ornate elk]

whats in your webui-user.bat?

[upbeat gazelle]

a vrooming 18-24it/s at 512x512 for default model

[ornate elk]

thats good!

so no issues here

[ornate elk]

at the line COMMANDLINE_ARGS= add:
--xformers --medvram-sdxl --no-half-vae

[light lynx]

Sure

[upbeat gazelle]

I ran a benchmark via System Info extension in forge to get some info, the it/s looks fine there, but I don't know the parameters its using for genning. Maybe any info in this can help https://gyazo.com/7344f00615b28fc950a59e7f5f51dc2b

[ornate elk]

what i also would suggest is to open up a cmd as admin
and run
sfc /scannow
that will try to fix system errors.
after its done restart the PC.

[light lynx]

alrighty

[ornate elk]

looks okay, but no infos on the steps, sampler or resolution

[light lynx]

same issue with the adding commands but I'll try the scan now

It does say their is a pending thing which a pc reboot is needed for

so I'll restart my pc in a minute and ping you when I'm back or just say back

[ornate elk]

oh okay, then reboot and then run the check, also make any windows updates before

[upbeat gazelle]

Yeah. So I gen at 1024x1024, default dpm++ 2m karras, 7cfg, 20steps. sdxl v10 checkpoint. prompt "a chair", 1.45it/s

[ornate elk]

on forge or auto1111 ?

[upbeat gazelle]

forge

[ornate elk]

1.45 its is to slow for sdxl on a 4090

[upbeat gazelle]

https://gyazo.com/fd12615ffcf9727edfd9c6588240f83f

yeah.. 😅

[ornate elk]

oh you made batch size 4

try with 1

[upbeat gazelle]

6its

[ornate elk]

yea thats better

batch size means "generating at the same time, using more vram"
while batch count is "generate each image after image"

[upbeat gazelle]

huh

so.. then I have a normal speed? 😩

[trim raven]

so today, whether from an old installation or from scratch, it's impossible for me to download torch...

[ornate elk]

yea, i guess so

[upbeat gazelle]

Lol, well I guess that's what I get for trying to be efficient. I've always ran 4 batch size. I took over a year break from SD, used 4 batch size back on a 3090 too. I guess i turned it into a muscle memory default

Many thanks for the help! I'll be able to pass on this wisdom now 😎

and just get another 3 4090, so I can run a pseudo 4 batch size that way

[light lynx]

alright

back

[ornate elk]

no problem 🙂 you can still use batch size, but you have to compare what is faster, batch size 4 with sdxl or batch count 4 with sdxl, also check how much vram is used

[light lynx]

I've tried it but it still does not work

same issue

[ornate elk]

hey, open up a cmd and type
pip cache purge
then delete the venv folder and relaunch the webui-user.bat

[upbeat gazelle]

It's just 1second difference of time taken for the gen and ram stays the same. But now I know that it splits the vram for the genning, so big prompts with loras and extensions should run better on batch count 👌

[trim raven]

it's already been done

[ornate elk]

hm what you can try is update to python 3.10.11, delete the venv folder and relaunch

[ornate elk]

you can try update your python to 3.10.11 too then delete the venv folder again

[light lynx]

alrighty

[ornate elk]

@trim raven if the python update wont help, try this command inside the venv
pip install torch==2.2.1 torchvision --index-url https://download.pytorch.org/whl/cu118

[light lynx]

should I do the same if this does not pan out once it's finished?

Also yeah doesn't work

[ornate elk]

nope, you should reinstall your nvidia driver probably

[light lynx]

alright I'll just try that a different time

if that doesn't work I'll just wait for stable diffusion to update and try their newer version

[trim raven]

thank you, solved

[ornate elk]

its not the fault of stable diffusion, as for my GTX1080 it works perfectly

mostly it didnt recognised your gpu or driver

so the clean reinstall could fix that

what you can also try is to make a new folder and git clone the repo again

[sleek zephyr]

Hi, I'm a member of the community with subscription, I use AI for artist projects.

I just now read that the outputs generated on your site are public and you publish them on the bulletin board and I would like to delete my creations, is it possible?

Alternatively, can I delete my profile?

Thanks if you can help me

[sleek zephyr]

On this site https://stability.ai/stable-video

[elder turtle]

Please send that question to the stability.ai support, we can not help you here

[sleek zephyr]

I sent several emails and also wrote in the chat on the site but I never got a response. Others told me to write here :/

[elder turtle]

Open a #1010934719455707218 ticket and ill see what I can do

[sleek zephyr]

ok thanks I'll try now because I've never made a tiket :/

[weary crescent]

Good morning ! Excuse me for disturbing you but I have a little problem with the stable installation.... can someone help me quickly? :3

[ornate elk]

Hey, please follow the install guide of the pinned messages of this channel depending on your gpu

Then make sure you have a good internet connection and that your PC time and date is right.

If your from asia, Russia, or Iran you may need a VPN for the install

[distant warren]

did anyone know why that happened?

[light lynx]

I know, it worked like a dime for me before, I will 100% reinstall my driver though tomorrow

I know it only takes a few seconds but not interesting in stable diffusion right now, I'm gaming 🥹

[ornate elk]

Make sure you whitelist the webui in any adblocker

Also what does the CMD shows?

[distant warren]

cmd did not showing anything

[distant warren]

i just launched webui and then it shows this error immediately

[ornate elk]

did you used the install script for MAC ?

[distant warren]

yes

[ornate elk]

this from here?
https://github.com/viking1304/a1111-setup

[distant warren]

no i used this https://github.com/AUTOMATIC1111/stable-diffusion-webui.git

[ornate elk]

then use the one from my link, it will do everything automaticly and also sets the best parameters needed for mac

[distant warren]

ok! THANKS!!

[ornate elk]

no problem 🙂 read the description on github how it works

[timber coyote]

been trying to install the zluda version on my pc but i get this error

[ornate elk]

seems like a bug with the latest update a few hours ago,
try to run this command inside a cmd inside the stable-diffusion-webui-directml folder
pip install torch==2.2.1 torchvision --index-url https://download.pytorch.org/whl/cu118

[timber coyote]

it's currently downloading torch

[timber coyote]

did it, it said success but still the same

[ornate elk]

go inside the venv

and rerun that command

[ornate elk]

looks okay

[timber coyote]

still the same

[ornate elk]

did you followed my zluda guide?

[timber coyote]

i did

[ornate elk]

okay

[ornate elk]

whats the cmd showing now?

[ornate elk]

oh thats not how you run a command in the venv

@timber coyote you need to run the Activate.bat via a cmd inside the SD-Zluda-webui\stable-diffusion-webui-directml\venv\Scripts

[ornate elk]

that should work

[timber coyote]

it seems to be working
should i do anything for the SD.Next?

[ornate elk]

nope, sd next is not needed

[timber coyote]

it launched in my browser but after generating

[ornate elk]

whats your gpu?

[timber coyote]

7900xtx

[ornate elk]

okay, same as me,
how does your paths looke like? in the environment settings

[ornate elk]

okay looks good

is your driver updated?

[timber coyote]

running at 24.3.1, says up to date

[ornate elk]

which zluda verison did you downloaded?

[timber coyote]

v3.7

[ornate elk]

okay whats your cpu?

[timber coyote]

7950x3d

[ornate elk]

okay you may need to disable the Igpu from the cpu

in the devicemanager

[timber coyote]

the selected one right?

[ornate elk]

yes

[timber coyote]

still the same, also

[ornate elk]

that looks okay

[ornate elk]

when do you get this?

[timber coyote]

after pressing generate

[ornate elk]

also i hope you have skipped the step with the rocm library folder (because for a 7900xtx its not needed)

[timber coyote]

i did skip that

[ornate elk]

try restart the PC

@timber coyote
if it still crashes then do this:
go into the C:\ZLUDA\ Folder.
There copy the cublas.dll and the cusparse.dll and paste them inside it again.
Rename the copys to cublas64_11.dll and cusparse64_11.dll
Copy these two files to stable-diffusion-webui\venv\Lib\site-packages\torch\lib, and overwrite if asked.

then relaunch the webui-user.bat

[timber coyote]

seems to be stuck on this

[timber coyote]

it launched -> generating now

[ornate elk]

perfect, let it run

it can take up 15-40 mins

for the first gen

[timber coyote]

is that for every new model or just the very first gen?

[ornate elk]

only for the very first gen

and if you update zluda versions

[timber coyote]

it's already done

[ornate elk]

perfect

[timber coyote]

thank you very much for your help

[ornate elk]

no problem 🙂

[light lynx]

oh nvm, reinstalled driver and it still doesn't work

so strange

[potent drift]

can someone help me please? why is this impaint thing NOT IMPAINTING at ALL

[idle phoenix]

this should have been a dragon , but seems like torch is failed how can i fix it ? pip uninistall not working

[ornate elk]

what settings did you used?

[ornate elk]

make sure to selecte Inpaint Area to "only masked" and then set the resolution to 768x768

[light lynx]

still doesn't work btw

uninstalled my driver in device manager rebooted, did a fresh install of the whole thing

still not working

[light lynx]

@ornate elk got anymore suggestions

or should I try a reset of the pc

[ornate elk]

What's your pyhton version?

[bold copper]

Hey folks, installed stable diffusion webui yesterday. Did a few generations that worked well but were EXTREMELY slow, like 70 second per iteration. I'm running an rtx 3070, 48gb of ram and a 3700x processor. I'm not entirely sure what I'm doing wrong, I've updated my drivers and I've ran the userwebui bat code to install xformers.

(Also included a test generation I did with 512x512, 8 interations and some generic peramaters just to test speed. Even that was super slow)

[sinful halo]

idk nothing about anything but use your task manager to see if it is using your cpu

[bold copper]

that's it at full use

[ornate elk]

you need to edit your webui-user.bat
at the line COMMANDLINE_ARGS=
you need to add: --xformers --medvram-sdxl --no-half-vae
then save and relaunch the webui-user.bat

[bold copper]

I'll try that

testing now with a 10 it generation

That's alot better, it's now at 1it/s

I'm using a complex model and I think that's around where it's normal to get

[potent drift]

btw guys any help for not making her in white top all the time?

[bold copper]

that's better, I think?

[bold copper]

is that better?

[ornate elk]

That's the best for 8gb vram cards

[bold copper]

Yeah it's a whole lot better

Cheers

[ornate elk]

Because your using an sdxl model

[bold copper]

Oh, should I be looking at lower end models?

[ornate elk]

You don't need to, but if you want faster speed then try out some 1.5 models, these are usually 2gb in size and much faster

[bold copper]

Oh nice

Okay

I'll see if there's a pony model for 1.5

[ornate elk]

Sdxl models are trained on 1024x1024 resolution, while 1.5 models are trained on 512x512
So use upscaling for 1.5 models to get higher res

[ornate elk]

There is one

[bold copper]

Gotcha

[bold copper]

can you send it to me?

[ornate elk]

Here:
https://civitai.com/models/257749?modelVersionId=297064

[bold copper]

cheers!

[nova ermine]

I'm looking for a checkpoint, lora, etc... which may be able to generate cartoonish facial expressions like this. At the moment I'm using 3DAnimationDiffusion, but I get a lot of struggle on giving my characters expressions like this.

[ornate elk]

White woman triggers the top color

[idle phoenix]

@ornate elknothing just standart settings

[balmy barn]

Hey folks, I just tried to run an image with a new workflow. and I get the following:

Error occurred when executing KSampler SDXL (Eff.):

module 'efficiency-nodes-comfyui.py.smZ_cfg_denoiser' has no attribute 'SDKSampler'

File "C:\ComfyUI\ComfyUI\execution.py", line 151, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
File "C:\ComfyUI\ComfyUI\execution.py", line 81, in get_output_data
return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
File "C:\ComfyUI\ComfyUI\execution.py", line 74, in map_node_over_list
results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
File "C:\ComfyUI\ComfyUI\custom_nodes\efficiency-nodes-comfyui\efficiency_nodes.py", line 2246, in sample_sdxl
return super().sample(sdxl_tuple, noise_seed, steps, cfg, sampler_name, scheduler,
File "C:\ComfyUI\ComfyUI\custom_nodes\efficiency-nodes-comfyui\efficiency_nodes.py", line 713, in sample
samples, images, gifs, preview = process_latent_image(model, seed, steps, cfg, sampler_name, scheduler,
File "C:\ComfyUI\ComfyUI\custom_nodes\efficiency-nodes-comfyui\efficiency_nodes.py", line 483, in process_latent_image
comfy.samplers.KSampler = smZ_cfg_denoiser.SDKSampler

[idle phoenix]

NotImplementedError: No operator found for memory_efficient_attention_forward with inputs: query : shape=(1, 4096, 1, 512) (torch.float16) key : shape=(1, 4096, 1, 512) (torch.float16) value : shape=(1, 4096, 1, 512) (torch.float16) attn_bias : <class 'NoneType'> p : 0.0 cutlassF is not supported because: xFormers wasn't build with CUDA support Operator wasn't built - see python -m xformers.info for more info flshattF is not supported because: xFormers wasn't build with CUDA support max(query.shape[-1] != value.shape[-1]) > 128 Operator wasn't built - see python -m xformers.info for more info tritonflashattF is not supported because: xFormers wasn't build with CUDA support max(query.shape[-1] != value.shape[-1]) > 128 Operator wasn't built - see python -m xformers.info for more info triton is not available requires A100 GPU smallkF is not supported because: xFormers wasn't build with CUDA support dtype=torch.float16 (supported: {torch.float32}) max(query.shape[-1] != value.shape[-1]) > 32 Operator wasn't built - see python -m xformers.info for more info unsupported embed per head: 512
Time taken: 0.7 sec

[ornate elk]

Can you try an other model?
For example dreamshaper v8 from civitai.com

Also what's your gpu and what's inside your webui-user.bat?

[onyx solar]

You need to look backwards at the node connected before that where sdxl_tuple is connected and check that everything there is correct. Also, having "refine_at_step" as a negative number is weird to me...you might need to set that to a normal number. (Though I'd recommend not using Efficiency Nodes at all anyway, but that's just me.)

[balmy barn]

Is there anyway you would be ok with sharing your workflow? I don't really care what workflow I use as long as it is good. and has a spot for loras

It would also be nice if it had a spot for a base checkpoint and refiner checkpoint

[onyx solar]

If you're using a good SDXL model, you don't really need a refiner. You can, but you don't need to.

But you should learn howthe stuff works. Start with something very basic. I have a super simple basic workflow to start people off with. No refiner, but works great with decent models.

You can start here:

Discord is being stupid.

[balmy barn]

I was able to download it.

[onyx solar]

Odd that it's showing me an error on my end, but that's good.

[onyx solar]

So...to add LoRA support, simply stick a "Load LoRA" node in between the "Load Checkpoint" and "Positive" & "Negative" nodes in that flow.

(Note that there are a ton of ways you can do this kind of thing and many little tiny changes will result in different end images.)

[balmy barn]

Does this make sense here?

[onyx solar]

Yeah, that will work.

Once you understand little bits of things, swap orders of things or adjust settings to one extreme in one place to see what will happen. Learn what each setting does, what the vocabulary is, etc. Then you can make much better images and experiment easier since you'll really understand what you're doing. Comfy isn't as difficult as some people think and a super easy workflow like the one I shared can help you learn.

Essentially, that workflow only has 8 nodes that are actually doing anything and technically, one of them is redundant, depending upon the model. (The "Load VAE" node could be taken out if your model has the VAE included.)

I leave it in because some models don't embed the VAE.

[balmy barn]

the issue i see is that there is no option for none. and when you bypass it. it throws an error. For the VAE box

[onyx solar]

You always have to have a VAE; that's what decodes the image from the latent noise. If you have a model with the VAE built in, you simply connect the VAE dot in the Load Model node directly to the VAE dot in the VAE Decoder node.

[balmy barn]

ohh... magic

i like it

[onyx solar]

If you want to get fancy, you can drop in one of several "switch" nodes from people that will allow you to connect 2 or more VAE node sources into one where you pick which one you want. (One of my favorites is from rgthree that automatically chooses the first non-null input.)

Oh...and the other thing to understand is that although workflows tend to feel logically like they should be from left-to-right or top-to-bottom, they actually first look at the end result and work backwards to ensure there's a full path of execution.

[balmy barn]

that kinda makes sense... its like making sure there is a bridge ahead before driving over the water.

[sharp valley]

hello i just wanna ask what are the chances of getting a virus from --disable-safe-unpickle

i trusted the models i downloaded but needed the command just to be capable of running training

[balmy barn]

Not sure which VAE Switch you were referring to but is this essentially what it would look like?

[sharp valley]

my files corrupted so i deleted the whole thing but now im kinda scared of the possibilty of one of them being a virus

[onyx solar]

Comfyroll's VAE switch is just fine and an easy way to implement a switch like that.

[balmy barn]

Did you say the one you use does this automatically? What is the node called? Also I would really like to add a refiner model. How would I do that?
I'm sorry for asking so many questions... Its just I never really find anyone that breaks things down like this for specific things like this.

[onyx solar]

Look for the package from rgthree for the various switches he's got. But beware that he's got several things in his package that are pretty advanced.

As far as refinement for SDXL, it's really not used excessively anymore. But ultimately, you can do it several ways. If you wa;nt further examples, you can find a bunch of them here, directly from Comfy:
https://comfyanonymous.github.io/ComfyUI_examples/

That includes several examples for SDXL; with refinement if you insist:
https://comfyanonymous.github.io/ComfyUI_examples/sdxl/

[balmy barn]

For me the refinement is not as important and I agree with you... However my wife uses it and almost leverages the refinement process like a LoRA in itself where it allows the secondary model to change some of the finer details towards the end of the process. For example usually we would use an Anime style Checkpoint for the base and then a detailed checkpoint like Nightvision as the refiner... It makes some amazing results.

[onyx solar]

Instead of treating it like SDXL-specific refinement, you should use an Advanced KSampler node and learn to tune the denoising, passing that latent into another KSampler node for the "refinement" that you are doing now. From there, it's going to be easier to understand how you can make even more improvements by inserting other nodes into that mix.

In fact, you can take a fully denoised image, re-encode it into a latent, tune denoising down to between .5 and .8 to really play with styles.

[balmy barn]

Do you inject new prompts into the new latent space for the refinement? This is where I get confused.

[onyx solar]

You can, but keep in mind that when you use Comfy, it's all about doing whatever you want to do. Sometimes you'll get what you expect, sometimes you'll learn a lesson. Sometimes you'll stumble upon amazing results. But that's what makes it great; you can do just about whatever you want.

Hell, I have a workflow that doesn't have anything to do with image creation or manipulation at all. All it does is read in a directory of files, go through the images in the directory, & create a matching file with all the booru tags for the images.

You can even setup workflows that do complex math and have nothing to do with images at all, if you're so inclined and find the nodes you need.

[balmy barn]

I have so much to learn.

[silk cairn]

Hi guys, can someone send me some basic .toml settings for a standard SD model for LoRA training

[balmy barn]

So I feel like my process is really redundant and I still dont even have upscaling or hires fix I do think I'm starting to understand a bit about how this pasta dish is coming together though... Any thoughts on how I can consolidate or add upscaling and hires fix or implement some of the above things you mentioned?

I think I got the upscaler figured out possibly.

[onyx solar]

If you're patient, you could look into SUPIR. It's my favorite upscaler, but it does take a while to render. (Obviously depending upon hardware, resolution, & settings.)

[balmy barn]

I have a 4090 128gb DDR5 i9 etc... I really cant wait for the 5090 to launch

[onyx solar]

You should look into SUPIR, then. Go find the node pack in the manager, click on the link in the manager to go to the git site and read up on it there.

I suggest keeping the upscaler workflow separate once you figure out how it works, though. That way you can just pass images into it when you want.

[limber pier]

sure hope it has the rumored 32gb

[onyx solar]

I don't even get out of bed for less than 48GB of GDDR7.

[limber pier]

This is the look I have at the thought of that.

[onyx solar]

According to Anandtech, GDDR7 supports non-power-of-two capacities, including 24Gbit & 48Gbit chips. Max spec for GDDR7 is 64Gbit, though.

What's killer is the prefetch per-pin architecture. Each channel supports up to 32x8b per pin!

[limber pier]

damn

yeah if it's a vram upgrade over the 4090 i'll buy it immediately (and keep the 4090)

4090 will remain as an inference card

[limber pier]

awesome

[harsh saffron]

What's up! Im having trouble with the stable diffusion 1.5 my computer is pretty slow but i have a school project to do with this and im running into some issues... I have an intel i7 processor with 16Gb of ram and a small NVIDIA GTX 750Ti... What could i do to make it work? i tried to run with --lowvram in the webui-user thing but that dosent seem to work since it dosent show anything when i try to generate a cat.. Could someone help troubleshoot this please? thanks!

that thing loaded for about 10 mins, before i tried to run it with the lowvram thing...

[limber pier]

Isn't that only 2gb of vram?! 😂

You're gonna need a newer GPU

I'm amazed you can run it at all

[harsh saffron]

idk man its a bad one lol

[limber pier]

Yeahhhh nothing you can do

[harsh saffron]

its running on overclock

[limber pier]

You can't use fancy options to get around the fact your GPU is a decade old with like 1/4th what's generally seen as the minimum amount of vram

[harsh saffron]

okok so basically, my thing ain't gon work? xD

[limber pier]

Yeah you need a better GPU

Vram is the most important part

I don't think you can run it at all with less than 4gb and it'll suck ass at that

[harsh saffron]

damn... allright welp that's bad lmao

[limber pier]

What're you trying to do anywayw

[harsh saffron]

basically, for college i need to make a video of presenting a local stored ai to a certain client.

and i thought that it would work on the school computers, but they dont run on NVIDIA so i thought well mine had one itll work xD

[real jay]

Hi

[harsh saffron]

But allright, thank's for the help anyways, have a good one!

[limber pier]

Yikes

They assigned you a project without providing a system that can run it? That sucks

[harsh saffron]

It does, but hey its Quebec and a public college.. they dont even have the budget to get the IOS program for us to learn the language. We only do Android xD

they expect us to make up for their shitty budget lmao

[harsh saffron]

but hey, would a 1660TI work?

with AMD?

or 2060 and Intel?

[limber pier]

yes

both of those would work

they'd be a bit slow but they will work

[limber pier]

ouch

sad that public education doesn't get funded too well

[harsh saffron]

perfect, ill borrow my friends computer tomorrow and ill make the video there. Can i tag you in here if i ever get any problems?

[limber pier]

you can try, but i'm usually away from here most of thurs and fri due to my work sched

[harsh saffron]

I know, theyre money laudering tf out of us xD

[limber pier]

there's a lot of ppl here who are usually glad to help

[harsh saffron]

Allright well perfect, thanks a lot man, I wouldve wasted so mutch time xD realy thank you!

[limber pier]

also real quick

if teh quality of what you get out of it matters at all

don't use the default model

https://civitai.com/models/46422?modelVersionId=127207 get this

[harsh saffron]

what do you mean by that?

[limber pier]

the whole 1.5ema pruned blah blah safetensors thing at the top

that's the default sd15 model

it's pretty horrendous

you'll download this one and drop it in your stable diffusion folder... models/stable-diffusion

[harsh saffron]

lmao, so this one would be worse quality but would still work?

[limber pier]

refresh with the blue arrow

the edfault one will technically work but it looks like total ass

[harsh saffron]

oh XDD

[harsh saffron]

well my client is a tattoo artist, so i just need black lines on white background basically

client *

[limber pier]

here's a high res one for ya in case you need it

[harsh saffron]

looks trippy af

[harsh saffron]

would it be able to create simple things like an outline of a shark? basically a drawing (tattoo style)

[limber pier]

absolutely

just give it a shot

"tattoo art style, blah blah blah"

i'll warn ya though, if those generations are coming back quickly enough, you start to feel like a rat pushing a lever for cocaine

[harsh saffron]

lmao, allright well ill try that out tomorrow and if i get back with something cool ill send it here lmao, thanks for the tips!

[limber pier]

yup!

no prob

hopefully you get horribly addicted and you join the rest of us in a state of perpetual ruin and gratification

[formal trout]

Hi, I just reinstalled Ubuntu, and now I'm having trouble running Automatic1111

[fallow arch]

bro,its my error, somebody know why error?

[ocean basin]

Is there a tutorial to instal comfyui on windows, amd rx6600 gpu? I have tried zluda a1111, but its still not good enough for my job(clothing)

[dreamy lodge]

Couldn't launch python

exit code: 9009

stderr:
Python was not found; run without arguments to install from the Microsoft Store, or disable this shortcut from Settings > Manage App Execution Aliases.

Launch unsuccessful. Exiting.
Press any key to continue . . .

Keep getting this error when starting webui-user.bat. Ive already installed SD fully and it worked the first time no problem but upon restarting its been doing this

[ornate elk]

What's your python version?

[dreamy lodge]

3.10.6

[ornate elk]

Then run the installer again and click modify, next, and then check "add python to system variables"

[dreamy lodge]

it has been

[ornate elk]

Okay then delete the venv folder and relaunch the webui-user.bat

[dreamy lodge]

Wheres that again?

[ornate elk]

Or install python 3.10.11 64bit

[dreamy lodge]

Also when i type python in cmd it redirects me to MS Store

[ornate elk]

Type in a cmd
where python

[dreamy lodge]

Did i accidentally install MS store version?

[ornate elk]

Maybe

So uninstall every python you find and install 3.10.11 64bit

[dreamy lodge]

that windows python exe is a blank exe. IDK what it is but its locked behind alot of security

[ornate elk]

Try uninstall it via the App section of the system settings

[dreamy lodge]

its not there. Probably just a remnant

[ornate elk]

Ok

[dreamy lodge]

i installed 3.10.11 and its still not working

@ornate elk i deselected the app alias's for python 1 and 3 in settings and python works now

somehow they where asigned to the microsoft page

[ornate elk]

Okay strange but good that it works now

[dreamy lodge]

Completely works now

[ornate elk]

Don't forget to add the performance args to the webui-user.bat

[dreamy lodge]

Can you elaborate on what those are. I havent touched SD in a while

[ornate elk]

What's your gpu?

[dreamy lodge]

nvideia RTX 3050 TI 4GB(Laptop)

[ornate elk]

Ah then you need
--xformers --medvram --no-half-vae
Add these to the line COMMANDLINE_Args=

[dreamy lodge]

thanks!

[ornate elk]

Np, with that you get the best performance and have less vram usage. So you can do higher resolution generations

Also in the SD settings you should activate FP8 support for sdxl models

[dreamy lodge]

k