🤝｜tech-support

[acoustic drum]

Where do i find this check

[sturdy ginkgo]

in LMStudio right

[acoustic drum]

Cpu still jumping to wild percentages and still slow

[acoustic drum]

I switched this on

But at least its really uncensored now

[sturdy ginkgo]

Would wait with mixtal till a preset is available or ask on their discord

[acoustic drum]

Ill just fiddle around with the settings

Butbstill, if anyone here could find an answer to error „Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0“ id be gratefull

[sturdy ginkgo]

try a different model and see if the same error shows as mixtral seems to cause lots of errors currently, it outputs but has issues

[acoustic drum]

Yea gotta find a smaller model first, if size is really the problem, but im thinking it cant be, because it runs slow but fine in lmstudio

[torpid ravine]

what are the differences between these models?

[sturdy ginkgo]

8S is the best for ai generation .. the others are more for real time surveilance

[torpid ravine]

ah, thanks!

so the others are just faster versions?

[sturdy ginkgo]

not faster just trained on way more detail making them pickup stuff even with tons of noise or with very little amount of pixels

these are object detection models not really for fixing images

Segm models takes a frame and creates a grid basically (segments) and coordinates where a object is on that grid

[acoustic drum]

Any idea why in lmstudio the model stops writing before the message is done, like mid text

[sturdy ginkgo]

increase the context tokens (some models have a max amount (Phi-2 for instance only 2048 (short 1 or two lines of output) but Llama 32768 (full paragraphs)

[burnt badger]

Is there a limit to how many Loras you can apply to an image at once?

[fresh temple]

Could someone please help me. I'm completely stuck!

I have 100 Low-Res images of construction workers. I have these in a folder, and using them in img2img as Input Directory in batch.

Using Dynamic Prompting, my prompt is:
{white|black|asian} construction worker

I am hoping to get 100 high-res images of different races of construction workers. However, it is selects one race and sticks to it, so get 100 Asian construction workers, instead of 100 random ones of the {white|black|asian}.

The combinatorial generation is DISABLED.
It works perfectly well if I am using multiple batch size or count using one image. But what I need is ONE image with random race, for EACH low-res image in a folder.

[sturdy ginkgo]

less is better 🙂 .. don't go over 6 or so or you overcook the output

[burnt badger]

I've got 7, lol, seems to be working though

[torpid ravine]

what segments can i pick from this?

[fresh temple]

Could someone help me please, I am trying to use Dynamic Prompt but it is not working correctly when using Input Directory.

[dark owl]

Is that the new extension? The old one stopped working.

[fresh temple]

Yes, the extension works perfectly fine when I am doing them one by one. Or Txt2imgs

[dark owl]

Oh then you have the bug

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

[fresh temple]

Ah i am not sure if it is a bug or I need to change some setting in Dynamic Prompt options

[torpid ravine]

what is the best way to correct skin color differences when inpainting?

[rotund lava]

img2img with low denoise strenght

[torpid ravine]

do you know any loras that could help too?

[rotund lava]

no,u need to do an img2img to even out the whole pic

[torpid ravine]

would the best way to preserve the image be with controlnet?

[rotund lava]

no need to use controlnet because you are using a low denoising strenght

[jaunty sphinx]

how do i solve this error bruvs? FileNotFoundError: [WinError 2] The system cannot find the file specified: 'D:\Stable Diffusion Open Vino Version\stable-diffusion-webui\models\Stable-diffusion\majicmixRealistic_v7'

[solemn moon]

i am here

any idea why i might be getting consistant horizontal lines

[mental locust]

Hi, I can't seem to Google the solution to this.

[ornate elk]

Try to put the model directly into models/stable-diffusion

[ornate elk]

What are your settings?

[solemn moon]

Steps: 30, Sampler: DPM++ 2S a Karras, CFG scale: 18.5, Seed: 2009121479, Size: 1152x1152, Model hash: 9aba26abdf, Denoising strength: 0.72, Mask blur: 0, ControlNet 0: "Module: lineart_standard (from white bg & black line), Model: control_v11p_sd15_lineart_fp16 [5c23b17d], Weight: 1, Resize Mode: Just Resize, Low Vram: True, Processor Res: 2048, Guidance Start: 0, Guidance End: 1, Pixel Perfect: False, Control Mode: Balanced, Save Detected Map: True", Version: v1.7.0-RC-75-gaeaf1c51

just thinking it might be the pre processor resolution lemme try that so long

[ornate elk]

Have you tried without controlnet?

[gritty ridge]

If I import and train own model like Lora etc.. trained on very specific architecture related black and white images, small number- with parameter 1 for relatedness to model at the end of the prompt- how is it possible to have in output images in color

[gilded river]

nvm figured it out

[gritty ridge]

I had similar problem with / worse than plague / plagiarism but better

[gilded river]

My guess is that the base model was still trained on colored images

[solemn moon]

but then i have no powers ill try it though, thnx

[hallow latch]

Is it possible to save the generated images to a Onedrive folder?

[ornate elk]

yes in the settings you can define a path where the outputs go into

[hallow latch]

How does it work with 2-auth and etc.. ?

[ornate elk]

? normaly if you have a onedrive folder on windows its just a folder where everything you drop in gets uploaded automaticly

[hallow latch]

Well, if I need to login on my onedrive, I have to accept on my phone and so on.

[ornate elk]

yea but on windows you should stay logged in then

[hallow latch]

Well, I am not at home and was thinking if I could set up so I get these photos on a online drive so I can view em 😄

[ornate elk]

oh well, then you should set it up when your at home

what also works is to install RustDesk on Phone and PC. Then you can do everything on your PC at the Phone

[hallow latch]

I tought that a http link would work but I could set it up 🙂 I just needed the path

like C:\Users\Name\Foldername

To my onedrive

[gritty ridge]

Any idea of stable diffusion failing img2img and inpaint… other text to image is not problematic

[ornate elk]

what do you mean by failing?

[gritty ridge]

Well grnerate is. Switched off and active as an option after 2s

Like doing nothing

[ornate elk]

what does the cmd shows?

[gritty ridge]

Connection from remotehost closed

Connection reset error

Also it is failing finding my named trained LORA - am I suppose to move the json output from trained model in LORA folder

[ornate elk]

What's your GPU ? And what's inside your webui-user.bat?

[gritty ridge]

CUDA

[ornate elk]

Thats not a GPU.
You can check it in the task manager under performance GPU

[gritty ridge]

I have two Intel HD Graphics 4600 GPU

And NIVIDA Quadro

It’s quite old.. but SD is installed somehow

[ornate elk]

Okay how much vram does the quadro have?

[gritty ridge]

16GB

[ornate elk]

Vram not RAM

If you can tell me the GPU name I can check.

@gritty ridge
In any way edit the webui-user.bat by right clicking it.
Then add --xformers --no-half-vae
To the line COMMANDLINE_ARGS=
Save and relaunch

[gritty ridge]

Video memory is 2GB GDDR5

[ornate elk]

Okay then add --lowvram
to the other commands

[gritty ridge]

GPU NIVIDA QUADRO K1100M

[ornate elk]

After you edited you can try img2img again and inpaint with "only masked"

[worldly berry]

Hi there

Is anyone having this issue where the image generation just hangs at 98% or less? I have no idea how to even follow up with this all i know is that i happens randomly and i can't see any error on the console.

torchvision is at 0.15.1+cu118
version: v1.7.0-14-g79548ab4  •  python: 3.10.6  •  torch: 2.0.1+cu118  •  xformers: N/A  •  gradio: 3.41.2  •  checkpoint: a8e08b582e

[gritty ridge]

Thanks so so seems in painting finally finally works

[ornate elk]

Hey what's your GPU and what's inside your webui-user.bat?

[sturdy ginkgo]

Hey guys .. for ComfyUI on a RTX 3060 what cross-attention works better?

[proper pike]

Does anyone have a recent tutorial on best settings to train a 1.5 model? I cannot get it working for the life of me. https://pastebin.pl/view/d724456b

[gritty ridge]

Anyone faced JSON files as an output in supposed to be trained model folder? How am I supposed to load it in SD as a model?

[ornate elk]

A json file is not a file you can use as lora or model

[gritty ridge]

Ok hm I found only that trying to train through dreembooth

[ornate elk]

Hmm

[sly locust]

I try to make that worklows works for sdxl controlnet + animatediff

but I get an error when Im in Ksampler box

any idea how to fix it and why ?

[wind sun]

What do you guys think about this laptop for creating high resolution images with SD

[burnt badger]

I think for 4k you could get a desktop that has a 4090.

You're basically hamstringing yourself if you go laptop for that much money and its not a macbook pro.

[sonic ferry]

Is there something else besides --listen to like use ComfyUI from a phone? When I loaded up IP:8188 from my phone I expected to see what I had loaded on desktop.

[wind sun]

I don’t really care about the money, i just need to be laptop because i am travelling a lot

[white thorn]

hey guys, what are the arguments you run with 4080 16Gb gpu ?

[gritty ridge]

Any idea why it keeps generating json files as a training model and how to turn them into training files

I use dreambooth / Lora

[mighty rivet]

is there any way that i can request my generated image to be a very specific color code?

[jaunty sphinx]

i put the model directly into stable diffusion it still doesn't work

[ornate elk]

16gb vram is pretty high for an nvidia gpu

thats a +

price is a -

[burnt badger]

If money is no issue, you should try to get more VRAm on the video card. at least 12gb, 16 if you can find it.

[white thorn]

na , I have a 4080 with 16Gb , my question is what arguments you think is best optimized

for now I tried --opt-sdp-no-mem-attention

[burnt badger]

Can someone explain why this is happening and how do I correct it?

[ornate elk]

ahh okay, you only need --xformers --no-half-vae

[ornate elk]

you need to edit the webui-user.bat,
at the line commandline_args=
you add: --xformers --no-half-vae

[white thorn]

I saw that xformers was for old gpu

[worldly berry]

Nvidia 12 GB , 3060 currently my web user bat file is empty...

wait i have this line
set PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:32

[ornate elk]

nope, thats not true every gpu can use it

and it uses less vram than --opt-sdp-attention

[ornate elk]

remove that line, not needed for your gpu.
at the commandline_args=
add: --xformers --no-half-vae
then save and restart

[burnt badger]

Okay, I'll do that, what is forcing it to half? is it a failure somewhere in the system?

[white thorn]

you think I should be able to hires x2 an SDXL 1216 x 832 ? with my 16gb vram

[ornate elk]

its for sdxl and anime vae files. they need that

[ornate elk]

yes should work

[white thorn]

yeah cuz its not for now I will try Xformers

[ornate elk]

hmm you may need tiled diffusion extension, than it will work for sure

[kindred tinsel]

Hello !!! Does anyone know why my stable diffusion is throwin this error:

NotInplementedError: Cannot Copy out of meta tensor: no data!

[white thorn]

yeah even with xformers OutOfMemoryError: CUDA out of memory. Tried to allocate 1.93 GiB (GPU 0; 15.68 GiB total capacity; 12.96 GiB already allocated; 1.48 GiB free; 13.35 GiB reserved in total by PyTorch)

[ornate elk]

yea then its to much and only work with tiled vae

only activate tiled vae

[white thorn]

im trying ok

what is diffusion for ?

[ornate elk]

its for generating different subjects in parts of the image

[white thorn]

Ah ! thats why I had 2 girl lol

it work when I use small value need to find a sweet spot I guess

[ornate elk]

yes, you have to test around. also can be slow

you could also upscale in img2img with the sd upscale script

[white thorn]

yeah someone just told me its bad and hires is much better

[white thorn]

my max value seems to be around encoder 2352 and decoder 128

[ornate elk]

thats pretty good

[ornate elk]

hires is good to use for the first gen, then later upscale in img2img to even higher res

[white thorn]

isnt it a bit pixelated for a 2432 1664 image?

[sturdy ginkgo]

it isn't vector graphics so no matter the resolution you will get some pixelation

zoomed in 600x and all i can see is ghosting artifacting not really pixelation

[kindred tinsel]

Hi there I am getting this error or warning in ComfyUi but I don’t understand what it means t.frac.out…. Should I change somr setting in the launch for the ui? I only have —directml in there

[sturdy ginkgo]

are you using an AMD gpu

[kindred tinsel]

Yep! RX 5500 XT

Sadly that’s all I have for now

[sturdy ginkgo]

try a SD 1.5 model instead of SDXL

seems SDXL uses too much VRAM .. also disable any controlnet nodes if trying SDXL

[kindred tinsel]

I’m not using sdxl at all

I don’t even know how to use it :) I only work with 1.5

[sturdy ginkgo]

ok

is it this exact same issue: https://www.reddit.com/r/StableDiffusion/comments/18kq51m/6700xt_getting_5773sit_then_running_out_of_memory/

[kindred tinsel]

Yeah it is but the task manager reports GPU being used not cpu as some comments on the post suggest

I used the same fork even

[sturdy ginkgo]

Might have to go the Linux Subsystem for windows route or run on Linux

[worldly berry]

dammm i've been dealing with this for a long time, never thought about the bat file, it works even better now, can you share why this two parameters?

[unreal marsh]

does anyone know how to fix error 128 when installing stable diffusion?

its very annoying issue :((

[ornate elk]

nice, np.
--xformers is a general optimization tool, it uses less vram and boost the performance. its good for every nvidia gpu.
--no-half-vae is for compatibility with vae files of sdxl and other 1.5 vaes

[ornate elk]

can you show a full cmd screenshot?

[ornate elk]

have you installed directml inside comfyui?

[kindred tinsel]

How do you do that?. Or do you mean the first steps when you use pip to do so because I think I did

[ornate elk]

the pip install torch-directml

[kindred tinsel]

Yeah I think I did that, how can I check for it

[ornate elk]

you can run it with the command

then it should be used

[kindred tinsel]

Yeah I always do that to run it

That’s why the error is weird and isn’t always there

[kindred tinsel]

Ummmm this thing keeps going up in seconds for this extension, it did this last time too, literally killed my entire webui with one clikc.

Is there a way to stop it or to terminate the install and revert back?

halp please is at 844 seconds and counting :c

[ornate elk]

check the cmd

[kindred tinsel]

it got frozen but its back to normal now

[ornate elk]

this extension is very heavy

so the long load is normal

if you ever uninstall it, you also need to delete the venv folder

[kindred tinsel]

I might not worry about that too much, if I ever uninstall it it will be with the entire webui, aka once I replicate webui in comfy with workflows

[burnt badger]

Is there a syntax that lets me turn a prompt word into a nonfunctional, but lets me retain it in the prompt for activating later. Similar to the // function in C?

[worldly berry]

while the performance in general is like 5 times faster, it seems that from time to time this weird thing happens, like halting at 93% 98% 😦
I mean at least now restart is even faster but im still wondering what the heck... Could it be that the RAM hit a ceiling and something got lost in the way?

[ornate elk]

Does it completly freeze? Or just takes a bit longer at the end?

The last step is usually "applying the VAE"
So it depends what VAE your using.
If you use for example the 84000-mse VAE for 1.5 models or the sdxl vae you can remove the --no-half-vae from the webui-user.bat

But If it completly freezes than you should check that your webui Version is updated.
You can check that at the bottom of the webui in browser.

It should say webui version 1.7.0, torch 2.0.1 and xformers 0.0.20

[fair oxide]

👉 👈

any idea if there's a fix for this?

[mental locust]

Load Clip Vision SD1.5\model.safetensors ... ?

Anyone know how to find that file?

Both of those actually, I can't find either

From this tutorial https://www.youtube.com/watch?v=wFahkr-b7HI

[worldly berry]

the render is halt but the page seems operational, its gray like if it was really waiting for something.

[topaz maple]

Is it possible to select different VAE for hires.fix in A1111?

[formal trout]

I don't understand what to do, but it would be even nicer to use my Automatic1111 installation as server, is that possible?

[worthy bluff]

hey all, i use Stable diffusion with the Auto 1111 thing and today their was an update and i have and AMD GPU, it was working fine till today

now i get
RuntimeError: "LayerNormKernelImpl" not implemented for 'Half'

Time taken: 0.0 sec.

i think ive fixed it by adding to cmd args --lowvram --precision full --no-half --skip-torch-cuda-test

the lowvram sucks its slower than before but it works

seems it wont work without it

okay maybe its not fixed, every image done is just a blur

[kindred tinsel]

Does anyone know why after say 20 generations I start getting black images?

[inland zephyr]

What do you use as a refiner with an XL model that doesn't come with one? The same model again, or none?

[formal trout]

finally running Comfy, any flags I need? In Automatic1111 I use --medvram-sdx --opt-sub-quad-attention --opt-sdp-attention --no-half-vae

RX 6700, Ubuntu

[dark owl]

gpu is running out of memory for larger batches

or you have a node set to stop rendering after so many frames

[fickle lantern]

Hi guys, i am newbie here. Can anyone help some question here? Appreciated it

I have been trying to recreating videos of girl making dances. The videos are getting from youtube and tiktok and trying to recreate and become my own to avoid copyright issue. There are two ways as i of found.

1. Ebsynth Utility + img2img (adetailer and controlnet(tile & openpose) - Prompts - CFG 9, Denoising 0.75
2. Mov2mov + Controlnet (tile & openpose) - Prompts - CFG 7, Denoising 0.45

both ways doesnt give good result. The shirt always changing or else the shirt wont change at all (hopefully it will be a different outfit but consistent). The moves and cloths are very flickering. Are there any better way? or need to continue to test out the harmony between the promps the CFG scale and denoising strength?

[rotund lava]

that just normal you could try with temporalnet2 controlnet or use a character/clothes lora so outfit doesnt change much or just fix errors by hand

[fair oxide]

@ornate elk https://huggingface.co/segmind/Segmind-Vega

[haughty pewter]

Suffering with animated diff. The nice image is from my basic workflow without controlnets or animatediff. Then when I intro AnimateDiff or controlnet into the process, all goes to hell, the input image and model are the same. Anyone have ideas? @dark owl

[tulip tide]

hey guys, can anyone help? I keep getting ValueError: '/data/DuaLipa/DL 3.jpg' is not in list when I'm trying to train model in dreambooth, anyone know why?

[formal trout]

I'm getting stutter and crashes

[solemn moon]

turned out to be prepro resolution 😅

[fair oxide]

@ornate elk

[worldly berry]

https://gyazo.com/523eef35b34e3d206dc36f3ac915970c just for context.
https://gyazo.com/6b9d8b916e5ffb20673cedf8143e2437
https://gyazo.com/0bcd634340fce84d3815edb2373eed5f might be a deadlock?

[subtle quarry]

What the hell does this mean?

I have all my stuff on an external drive, there shouldn't be anything ai related left on my main drive.

[ornate elk]

Your GPU driver is not updated.
And also check if your webui is on the latest version

[ornate elk]

Check it you graphic driver is updated

[zenith crescent]

Nothing related to GPU driver. Its a common error. It ususally happens when you load and reload a model multiple times. SImple fix.... just close everything and reload.

it happens a few times to me

[subtle quarry]

he was talking to me-

oh wait

sorry im stupid

[zenith crescent]

[ornate elk]

Lol its not a common error if I never got it with a 1080 over a year

[subtle quarry]

it's 5am and I have no idea how I'm still awake lmao

[ornate elk]

Are you on a laptop quinzel?

[subtle quarry]

Yea, with a 1660ti

[ornate elk]

Check in the Nvidia control panel under 3D Settings. That the dropdown is set to "nvidia high performance device"

[subtle quarry]

I believe that's already been done

[zenith crescent]

well i got a 3080ti and i get it sometimes even when i had my drivers updated, however it never happenned to me when i was using the Ui with like 2 or 3 models. However loading and reloading models multiples times like every 3 minutes does cause the issue for me.

[subtle quarry]

currently updating the driver though.

[zenith crescent]

but tbh.... its small error and easily fixable

[ornate elk]

Okay did you used sdxl models too?

[zenith crescent]

I don't use SDXL anymore, i got tired of it.

[subtle quarry]

sdxl doesn't work on my rig, i tried it and it just instantly gives up

[zenith crescent]

i'm more into SD 1.5 models

[ornate elk]

Yea wouldn't recommend them for a 16xx card

[subtle quarry]

I also only have 8gb of ram, so

when it needs to offload, there's not much to offload onto.

[ornate elk]

Oh 8gb ram

Thats on the edge

[zenith crescent]

Yhea....

[ornate elk]

Well yea than that could be indeed a issue caused by loading models

[zenith crescent]

https://tenor.com/view/cat-nod-catnod-yes-vibe-gif-21529329

[ornate elk]

I think we already increased your pagefile. But then restart is the only fix like Joker said

[subtle quarry]

yeah i ain't got much on here.

comfyui works alright

[zenith crescent]

My advice would be to just stick to one model and not change models too much, and if you do want to change models and it causes the error just close it and start again. So it loads the model you wanted to change into at the start.

[subtle quarry]

I forgot how slow my computer was with ai since i've been using my friend's remotely using ngrok (comfyui) so uh...might just have to keep using his until i get my better rig when i move out.

which sucks, cuz i hate using other people's stuff but i guess i gotta suck it up-

booting up a1111 again, it's gonna take a few mins.

[zenith crescent]

have you tried stuff like this?

[ornate elk]

Minutes?

[subtle quarry]

yeah i have medvram on

and yeah, minutes

like a minute or two.

[ornate elk]

Is SD on an SSD?

[subtle quarry]

it has to stabilize since if I try to do things too early it locks up.

It's on a uh

...i dont actually know

hold on

it's an external drive

[zenith crescent]

so... prob USB 2.0 or USB 3.0 at prob... 20 to 300 mbs?

[ornate elk]

Yea thats the bottleneck

[subtle quarry]

it took like 2 minutes to load the model i tried to get

then when i pressed "generate" it kicked me with the error

[unreal marsh]

for some reason it just fixed itself idk why either

[ornate elk]

Okay xD

[subtle quarry]

sometimes i forget xformers takes forever to load in, even on my main drive.

so i have to wait even longer after pressing generate

[zenith crescent]

isn't Xformers..... supposed to only work on 20 series and up? Cause the 10 series doesn't work with it well?

At least from what i heard

[ornate elk]

Nope, xformers works for GPUs starting with GTX 1xxx

[zenith crescent]

Not sure where i heard that from but....🤷‍♂️

[subtle quarry]

yeah, not enough memory error kicked in this time. damn.

guess i'm stuck using my friend's computer.

[ornate elk]

@subtle quarry if you have enough Space on C: you might want to install SD there and only let the models be on the External

Also did you already increased the Pagefile ?

[subtle quarry]

Nah, not a storage issue, it's a ram issue

and i have enough space on C:, i just wanted to move stuff out

[ornate elk]

It would be loading faster if SD is on C.
Also your pagefile is your "virtual ram" that caches files if you dont have enough free RAM left

So increasing the pagefile could remove that error completly

[subtle quarry]

ahhh. kay. What should I increas it to? apparently i don't have one.

must've removed it or something.

[zenith crescent]

in some cases chrome can also take up like 1Gb of RAM so... maybe try to find another UI?

[subtle quarry]

I don't use chrome, I use firefox

[zenith crescent]

or just upgrade the ram

[subtle quarry]

can't upgrade the ram, getting a new rig when i move out in feb

[ornate elk]

Set it to system managed for C. And disable it for other drives.

Then restart the pc
Then test again

[zenith crescent]

Just make sure you think it's capable of doing the Ai stuff

[subtle quarry]

It is, it's a 3060ti, so it should be fine

i would hope it'd be fine

[ornate elk]

Yea thats okay

[zenith crescent]

Just make sure the RAM is not 8gb again😂

[subtle quarry]

kek, yeah it's gonna be like 42 or something

my bf is giving me his old ram and upgrading his, so

and he had 2 16 and 2 8

Unless I can't math, in which case shrug

[zenith crescent]

Mismatched RAM?

[subtle quarry]

Just the capacity.

[ornate elk]

Would go with the 2x16

[subtle quarry]

they're all the same brand i think.

and if not, i won ram in a christmas giveaway so i can just plug that in instead

[zenith crescent]

i think it's... acceptable IF they are at the same mhz speed.... if not then...

[subtle quarry]

They're at the same speed

i dogged him about it.

[zenith crescent]

Well if it's same brand, same speed, just diferent capacity.... i guess it wouldn't hurt to try?

[subtle quarry]

i'll probably fix all this ai stuff later, my brain is kinda too tired to be doing a bunch of stuff.

I mean, it's worked for him for like...a year

with no issues

no bluescreens, no ai crashes

no random hiccups

[ornate elk]

Let me know if the pagefile helped

[zenith crescent]

Well.... i leave him to you @ornate elk

Good luck!

[subtle quarry]

i increased the pagefile size, screw it. I won't do it later-

now just to wait 3 minutes for it to boot

and then another 2 after that for xformers to apply.

[worldly berry]

version: v1.7.0-14-g79548ab4  •  python: 3.10.6  •  torch: 2.0.1+cu118  •  xformers: 0.0.20  •  gradio: 3.41.2  •  checkpoint: d394871e48
I think webui is updated as what the version indicates 1.7.0-xxxxxx
Seems similar to what people report here, getting hanged at 95%
https://www.reddit.com/r/StableDiffusion/comments/15agkxo/out_of_memory_using_a1111_and_sdxl_10/

[ornate elk]

Make sure you restarted the Laptop

[subtle quarry]

yep! I restarted it.

[ornate elk]

Hmm okay, your vram usage is also pretty high, what settings did you used ?

I see you also used controlnet

[subtle quarry]

aaand it bluescreened.

yeah so ai is a definite no on this thing now.

[ornate elk]

Bluescreened wtf

[subtle quarry]

And i know it was from the ai stuff since the err code was "MEMORY MANAGEMENT"

so

[ornate elk]

Are you on win10 or 11 ?

[subtle quarry]

win10

[ornate elk]

You can open up a cmd as administrator and run
SFC /scannow
That will check windows for corrupt files

[subtle quarry]

i've ran this so many times on this computer

and each time it always comes up with something corrupted

starting to think this thing is cursed.

Oh that's a first, nothing came back wrong

[ornate elk]

And you have 15-20gb free space on C right?

[subtle quarry]

i have 166gb free on C

so plenty of free space.

i really dont feel like moving stuff around but if i have to i guess i will

it's unfortunately not as easy as just drag-dropping, i have to reinstall everything...

..oh fun now my speakers aren't working

[ornate elk]

What is this laptop xD

[subtle quarry]

msi gf65 thin

there's an audio driver issue and i dont know how to fix that, so might have to restart the laptop again

there we go ffs. okay so no ai on this computer anymore, that's fun

[ornate elk]

I mean you could disable the pagefile again if you want to.

[subtle quarry]

i'll probably keep it for emulation purposes, as I can kinda get away with switch emulation

[ornate elk]

Okay, then try to install SD on the SSD and test with one model

You dont have to move the models over as you can edit the webui-user.bat with --ckpt-dir "Path/to/Models"

[subtle quarry]

so that's how you do that

okay i had no idea how you changed directories.

[ornate elk]

Yea thats the inbuild way. The other way would be using Symlinks

[subtle quarry]

Does this look about right?
--ckpt-dir "D:\stable-diffusion-webui\models"

[ornate elk]

Yes if that don't work change the \ to /
And make a \ or / after models

[subtle quarry]

ty! I'll report back in a couple mins.

[ornate elk]

Also dont forget to add --xformers --medvram --no-half

[subtle quarry]

forgot --no-half

[ornate elk]

If you had it before

[subtle quarry]

yeah, i had to use it

--no-half-vae worked sometimes, but i had to use --n-half

i am going to curbstomp my computer, it keeps freezing up-

[ornate elk]

Ah yea the initials venv setup takes up a lot of ram

Also if you have loras there is --lora-dir

Also check how big the pagefile is set from the system

If its to small you can increase it to 20000mb

[subtle quarry]

that's what I have it set as

so hopefully that's enough

i'll just do a model stuff first and then if that doesn't crash out, i'll start adding the rest of the arguments

[worldly berry]

some times it comes after a succesive generations, but i am on the range max of 1024 x 1024
would --medvram make difference in this case?

[ornate elk]

If it won't work. Set Min to 20000 and Max size to 320000

[ornate elk]

Without hires fix ?

But with controlnet enabled?

Can you try without controlnet?

[subtle quarry]

yeah i don't think this is gonna work, i'll just use my buddy's.

it's taking over a minute to load in models

like i click on the model, and it's perputually loading.

[ornate elk]

Okay, yea thats still because of the slow external drive

You could put your most used models to ssd

[worldly berry]

Sorry yes i had two layers of control net enabled and with out hires fix.

[ornate elk]

Okay then add --medvram-sdxl
To the webui-user.bat

And test again

[chrome rampart]

afternoon all. Any tips for webui for 3060 card with 12Gb memory ?

of course xformers

[ornate elk]

--no-half-vae

If you use sdxl vae or anime vae

[worldly berry]

im on the same spot, what is your issue? is your instance hanging too randomly?

[chrome rampart]

thks cs to make sure for sdxl something with medvram?

[chrome rampart]

nothing yet installing card in few hours here

[ornate elk]

If an 1024x1024 30 steps sdxl image take more than a minute add --medvram-sdxl

[chrome rampart]

clear

first need to replace my power supply so hope o BOOM 🤣

[ornate elk]

GL xD
Also after the GPU install. Check for driver updates and delete the venv

[ornate elk]

Can you try without controlnet, if it hangs there too?

[tulip tide]

does anyone know why this is happening?

[vivid vine]

Hey folks, I'm training a lora with 60 images, 10 epochs training time, 768x768, Network Dim 64, alpha 32, and adamw8bit, 0.0005, 1 batch size, bf16.
I'm running it on a Razer Blade 16 laptop with a 4060 gpu, 2023 model. Does 1.52it/s sound right for a 2950 step training? I know the laptop gpu isn't as powerful as a desktop gpu, but could it really be THAT much less powerful? 4060 on desktop seems to get 15-25it/s from what I see online. Any suggestions or help would be much appreciated!~~~~

[red prawn]

Anyone knows why I keep getting weird deformities on the faces? Seems like it gets worse after the hires.fix

[kindred tinsel]

Weird though, I don’t run batches mostly send image after image. It only happened today which I found concerning but thanks for the info 🙇‍♂️🙇‍♂️

[sonic ferry]

Anyone got a guide on setting up the ControlNet for poses in ComfyUI? I have the manager installed, installed Fannovel16 ComfyUI's ControlNet Auxiliary Preprocessors, but then what?

[safe inlet]

Is there a "better" Nvidia driver version to use? 2080 TI Windows 10

[karmic crown]

has anyone ever managed to generate any SDXL image with 1660 Super?

1.5 models work fine, but whatever i tried sdxl doesnt work

[ornate elk]

have you tried with --lowvram?

[karmic crown]

not really

i had like 40s/it with medvram

so i just removed sdxl models

i am trying to find fastest 1.5 settings that works

for some reason looks like opt-sdp-attention works faster than xformers on 1660 super

[ornate elk]

okay, thats interesting

[karmic crown]

last step takes forever

19 steps are like 1.23it/s and 20th is 1.31s/it
wtf?!

1.5 wit DPM++ 2M Karras

[ornate elk]

hey, looks interesting, but it didnt worked for me, getting Nan errors

[karmic crown]

@ornate elk can you try to guess which crazy combination seams to work the best/fastest 🙈

Emphasizing the word "CRAZY" 🙈

[ornate elk]

no xD i cant even imagine

[karmic crown]

--xformers --opt-sdp-attention --no-half

yes together 🙈

[ornate elk]

for real ? xD

and using one or the other is slower?

[karmic crown]

i see a little boost on it/s
like 1.4x instead of 1.2x

but... i am using 2.1.2+cu118 and 0.0.23post1+cu118

for some reason last step of DPM++ 2M Karras takes much longer, with eulera for example it has te same speed as previous steps

since euler a, does not have a problem with last step, i will do a few tests with euler a and 4 images to see which one will be the fastests, performing second test currently

[ornate elk]

alright

[karmic crown]

hmm, with 4x1 they are all in 1 second frame (1min22-1min23), and xformeres is like 1sec faster

acutally only 0.2 compared to sdp only

blah

i will left xformers and ho-half (that thing must be there) 😦

and i will use my mac for sdxl

1.5 works faster on 1660 than on my m1, but with sdxl my m1 drastically outperforms 1660

[ornate elk]

yea i thought so xD

I wish --opt-sdp-attention would work for AMD cards without errors :/

[karmic crown]

install macos and use my command line 🙂

[azure magnet]

@ornate elk sorry for annoying you, but can you help me to fix it if you know how?

[karmic crown]

did you update your extensions after updateing webui?

[azure magnet]

Idk, do you know how to solve it?

[karmic crown]

go to extensions tab, than click on check for updates

and finally aplly and restart ui

[azure magnet]

I'll try but i don't believe it work

[karmic crown]

this looks like outdated plugin to me

[azure magnet]

No, same error

I'll try to install again two shot

[karmic crown]

last commit was 9 months ago

looks like authors didn't update that plugin

[azure magnet]

Ooooh, you have right and save me, i'll move to regional prompter

Anyway i have even this

[fair oxide]

probably it's not supported on a1111 yet, i tried it too but a1111 wouldn't load the checkp[oint.

[karmic crown]

you do not have styles.csv, which is not important if you do not use them

https://github.com/Douleb/SDXL-A1111-Styles/blob/main/Professional Styles Only %2B Comfyui Portage to A1111 (no GPT) (120%2B)/styles.csv

it looks like this

i do not use that anymore

[azure magnet]

Sorry i don't understand how to download it

[karmic crown]

you can create one yourself, or download any of those from the internet, that is just one of those

it should be in your stable diffusion folder

you can use this to create it yourself

you can download that one i send as an example like this

some others https://www.reddit.com/r/StableDiffusion/comments/11mkefu/is_there_a_collection_of_styles_for_web_ui_i/

but that was too messy for my taste, since i had to take care which one i click first

[azure magnet]

ok i'll check it

[hollow schooner]

It was running perfectly fine before i dunno what’s wrong

[hollow schooner]

better version of what this looks like

[ornate elk]

You have to add
--use-directml
Since the latest update

Then it will work again

[hollow schooner]

okay thanks so much it started working again :)))

[formal trout]

hey @ornate elk you that know this stuff, is there anything to do to make Comfy run smoother here?

[ornate elk]

Nope, comfyui should use the needed stuff automatically

But make sure you installed the directml dependency

[formal trout]

hey what about that? I'm on Ubuntu

[ornate elk]

Oh then not needed

[formal trout]

How sad, hate that stuttering like it is going to crash, and it has crashed when using "live" in Krita mostly, also sometimes just using Comfy

[vital storm]

hey there, ive been trying out turbo on comfy ui, for some reason, it looks like the SamplerCustom node and VAE Decoder node is reloading the model each prompt/config change, is that expected?
also takes about 2.5mins/workflow completion, wasn't it meant to be instant?

im on m2 mac pro, 16gb memory

[sweet slate]

Hey everyone, I was wondering If anyone knew how to avoid getting these artifact appearing on certain checkpoint

[gritty wave]

Anybody have a idea why every image I generate turns out like this?

Windows 10, AMD RX570 GPU with 8 Gb of Vram

It complains about xFormers not being installed when I start it up, I've tried several things to get xFormers to work but haven't been able to.

[signal sigil]

Hey guy's. I was wondering if any of you would know the answer to this, but I'm currently training a loRa with custom captions, How would i describe clothing for each person, when there are two people in a image?
Also if people are interacting in the image, say like shaking hands for example. Do i need to need to add the caption "handshake?" Or is describing interactions not needed?
Thx btw.

[onyx solar]

DreamShaper XL Turbo DPMpp SDE is the model you're using. You need to read this page:
https://civitai.com/models/112902/dreamshaper-xl?modelVersionId=251662

Specifically, look at the "About this version" section for CFG and sampler information.

[gritty wave]

Thanks for the response, I went and made sure I was using the correct sampler and set CFG to 2 with 5 steps, and the result was pretty much the same. I'm currently downloading a different model to test.

[onyx solar]

Did you set your VAE to None?

This has a baked in VAE.

[vital storm]

how do you do that on comfy ui?

[onyx solar]

That depends on your workflow. If it's got a baked VAE, then you can just connect the VAE node from the model loader, assuming you're using a model loader node that has a VAE node connection, to the input connection where you need it.

Comfy being so customizable, it's all dependent upon workflow setup.

[vital storm]

thanks, that's interesting, will have a look

[onyx solar]

I use a VAE switch node to flip between the baked VAE and the proper SDXL VAE when necessary.

[vital storm]

do you have any idea about my issue @onyx solar

the main thing is, it's not "live", I'm wondering if Im don't something wrong, I'm brand new to comfy

[onyx solar]

Really depends on the workflow. If it's reloading it every single time, there's either something in your workflow at or ahead of the model node or you've got some kind of setting that's doing it; though I don't really know what that would be.

The length of time is also dependent upon your setup. Not knowing what hardware you're running on or what your workflow looks like, I can't tell you exactly what the causes of your issues are.

[vital storm]

hardware is m2 Mac Pro 16gb ram. and I downloading the image from comfy example GitHub and dropped it into comfy and got the workflow

[onyx solar]

I'm no help with Mac stuff; couldn't tell you what to do there. I'm sure there are other Mac enthusiasts here that could, though.

What I would suggest as far as workflow is to understand what each node and connector does, then build the simplest workflow you can of your own first, just to understand the process. Depending upon what you downloaded, the workflow could be complicated and have things that impact speed to load or cause additional steps to be taken before image inference.

[vital storm]

that's a great suggestion, will do. I moved from automatic1111, trying to get used to comfy

[onyx solar]

https://tenor.com/bkJ4m.gif

[slim agate]

Hello I'm using this colab to train a SDXL model:
https://colab.research.google.com/github/camenduru/kohya_ss-colab/blob/main/kohya_ss_colab.ipynb

I keep getting this error:

    sys.exit(main())
  File "/usr/local/lib/python3.10/dist-packages/accelerate/commands/accelerate_cli.py", line 47, in main
    args.func(args)
  File "/usr/local/lib/python3.10/dist-packages/accelerate/commands/launch.py", line 1017, in launch_command
    simple_launcher(args)
  File "/usr/local/lib/python3.10/dist-packages/accelerate/commands/launch.py", line 637, in simple_launcher
    raise subprocess.CalledProcessError(returncode=process.returncode, cmd=cmd)
subprocess.CalledProcessError: Command '['/usr/bin/python3', './sdxl_train.py', '--enable_bucket', '--min_bucket_reso=256', '--max_bucket_reso=2048', '--pretrained_model_name_or_path=stabilityai/stable-diffusion-xl-base-1.0', '--train_data_dir=/content/drive/MyDrive/Dave/images', '--resolution=767,768', '--output_dir=/content/drive/MyDrive/Dave/model', '--logging_dir=/content/drive/MyDrive/Dave/log', '--save_model_as=safetensors', '--output_name=Dave', '--lr_scheduler_num_cycles=10', '--max_data_loader_n_workers=0', '--learning_rate_te1=1e-05', '--learning_rate_te2=1e-05', '--learning_rate=1e-05', '--lr_scheduler=cosine', '--lr_warmup_steps=1880', '--train_batch_size=1', '--max_train_steps=18800', '--save_every_n_epochs=1', '--mixed_precision=fp16', '--save_precision=fp16', '--cache_latents', '--optimizer_type=AdamW8bit', '--max_data_loader_n_workers=0', '--bucket_reso_steps=64', '--xformers', '--bucket_no_upscale', '--noise_offset=0.0']' died with <Signals.SIGKILL: 9>. ```

[solid tide]

Is there a reason Lora Training runs at ~1s/it when my ComfyUI can generate 6.37s/it?

Comfy runs 3.11 while Trainer runs 3.10 Python, but I don't believe that could make that drastic of a difference, could it?

[karmic crown]

it depends what you have in your workflow (number of steps, sampler...)
macs with 16GB are relatively slow (I have M1 MacBookPro 16GB)

[green aurora]

Hello, im using dreamstudio but all my 6000 of my images are gone; cant get any response from the team, support is disabled on the website.
is there anything else i can do?

[karmic crown]

Steps: 20, Sampler: DPM++ 2M Karras, Size: 512x768, Model hash: ec41bd2a82, Model: photon_v1, VAE hash: 735e4c3a44, VAE: vae-ft-mse-840000-ema-pruned.safetensors, Version: v1.7.0

Time taken: 1 min. 11.9 sec.

[vital storm]

I think you might be confusing which model I'm referring to, I'm looking at stock SDXL Turbo, and "instant generation"

I'm using the stock workflow (workflow dragged and dropped from the image), which is steps:1, CFG: 1, sampler: Euler A, VAE: whatever came with stock SDXL Turbo

[karmic crown]

sorry, i haven't seen that you mentioned turbo

what dimensions are you generating?

1024x1024? or something else?

[karmic crown]

Steps: 1, Sampler: Euler a, CFG scale: 1, Seed: 372646444, Size: 1024x1024, Model hash: 2e58e3704b, Model: sd_xl_turbo_1.0, Version: v1.7.0

Time taken: 25.6 sec.

Steps: 1, Sampler: Euler a, CFG scale: 1, Seed: 722140019, Size: 512x512, Model hash: 2e58e3704b, Model: sd_xl_turbo_1.0, Version: v1.7.0

Time taken: 9.6 sec.

[karmic crown]

btw. be sure to use 512x512, not just because of speed, but because other sizes like 1024x1024 produce bad images (with default turbo model) i have just tested, and there is also a note on stabilityai page under Limitations - The generated images are of a fixed resolution (512x512 pix)

images are generated faster, but it takes much longed to start generating than with some other models (maybe becaouse i download 13GB version)

[strange thunder]

Could anyone please be able to direct me to the "lora syntax" the LoRa Control - Dynamic Weights Controller extension is referencing? It is here: https://github.com/cheald/sd-webui-loractl

It says to use the existing lora syntax, but I have no idea where I can find that reference sheet

[strange thunder]

Or direct me to a lora scheduler I can use in a1111

[ornate elk]

Your missing a vae for that model

A vae fixes these purple dots and the washed out colors

[opaque mesa]

Hello, I am suddenly having an issue when launching webui-user.bat, I have an infinite loading screen, tried several times, rebooted my PC but it is still on infinite loading, is there a way to fix this please ?

[ornate elk]

Hey, what does the CMD show?

[vital storm]

512x512

[vital storm]

although its much faster than mine, is it not instant for you as well?

[light bone]

run.it does not open when running the bat file how do I solve this problem

[hollow bloom]

I'm sorry that my question is related to the API part, but I can only use the API right now. But I don't know why I can't use version 1.6? Does anyone know why version 1.6 was shot down? About two or three weeks ago, it was working perfectly without any bugs or problems, but it has been a week since it stopped and we have not received any explanation from the team as to what is the reason and why it cannot be used?!!!!!

[opaque mesa]

Doesn't seem to show much, no error message.

[fair oxide]

@ornate elk my pc rebooted due to power fluctuation in the middle of generating an image, after the pc boot up im getting the error in the attached file that's failing to start webui

already deleted the venv folder after purging the cache and updated the webui but that didn't help

[ornate elk]

You installed the webui the wrong way, (by downloading it as zip). Now the webui wont update anymore.
To fix this go into the stable-diffusion-webui folder and click in the File Explorer bar and type cmd and hit enter.

Then use the following commands:
git init git remote add origin https://github.com/AUTOMATIC1111/stable-diffusion-webui.git git fetch --all git reset --hard origin/master git branch --set-upstream-to=origin/master master git pull
Then delete the venv folder.
Then edit the webui-user.bat and add --xformers --medvram --no-half-vae to the Commandline_ARGS=
After that relaunch the webui-user.bat

[ornate elk]

you should install the webui using my guide, and not downloading any zip file for it.
My install guide for nvidia gpus: #🤝｜tech-support message

[ornate elk]

update or install python again, maybe some file got corrupt

also run an sfc /scannow as admin in cmd

[fair oxide]

thanks, doing that now

[chrome rampart]

morning all. i installed new 3060 (YAY) and works amazing can now finally use SDXL and comfy. Just wonder about startup and maybe still can improve. See below

Launching Web UI with arguments: --xformers --autolaunch --no-half-vae
Style database not found: E:\A1111\stable-diffusion-webui\styles.csv
[-] ADetailer initialized. version: 23.11.1, num models: 9
[AddNet] Updating model hashes...
0it [00:00, ?it/s]
[AddNet] Updating model hashes...
0it [00:00, ?it/s]
ControlNet preprocessor location: E:\A1111\stable-diffusion-webui\extensions\sd-webui-controlnet\annotator\downloads
2023-12-29 12:14:02,722 - ControlNet - INFO - ControlNet v1.1.424
2023-12-29 12:14:02,827 - ControlNet - INFO - ControlNet v1.1.424
Loading weights [ef76aa2332] from E:\A1111\stable-diffusion-webui\models\Stable-diffusion\realisticVisionV51_v51VAE.safetensors
add tab
[([button, button, button, button, button], button, gallery), ([button, button, button, button, button], button, gallery)]

[AddNet] Updating model hashes...
100%|███████████████████████████████████████████████████████████████████████████████| 98/98 [00:00<00:00, 11796.63it/s]
Creating model from config: E:\A1111\stable-diffusion-webui\configs\v1-inference.yaml
Applying attention optimization: xformers... done.
Model loaded in 13.4s (load weights from disk: 1.0s, create model: 0.4s, apply weights to model: 11.1s, apply half(): 0.1s, load VAE: 0.3s, calculate empty prompt: 0.4s).
[AddNet] Updating model hashes...
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 98/98 [00:00<00:00, 49932.19it/s]
E:\A1111\stable-diffusion-webui\extensions\sd-webui-additional-networks\scripts\metadata_editor.py:399: GradioDeprecationWarning: The style method is deprecated. Please set these arguments in the constructor instead.
with gr.Row().style(equal_height=False):
E:\A1111\stable-diffusion-webui\extensions\sd-webui-additional-networks\scripts\metadata_editor.py:521: GradioDeprecationWarning: The style method is deprecated. Please set these arguments in the constructor instead.
cover_image = gr.Image(
Running on local URL: http://127.0.0.1:7860

To create a public link, set share=True in launch().
Startup time: 46.3s (prepare environment: 19.2s, import torch: 3.9s, import gradio: 0.9s, setup paths: 1.8s, initialize shared: 0.3s, other imports: 2.0s, setup codeformer: 0.6s, load scripts: 2.8s, create ui: 14.0s, gradio launch: 0.5s).

generate result just to be complete: Startup time: 46.3s (prepare environment: 19.2s, import torch: 3.9s, import gradio: 0.9s, setup paths: 1.8s, initialize shared: 0.3s, other imports: 2.0s, setup codeformer: 0.6s, load scripts: 2.8s, create ui: 14.0s, gradio launch: 0.5s).
100%|██████████████████████████████████████████████████████████████████████████████████| 31/31 [00:14<00:00, 2.08it/s]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 31/31 [00:15<00:00, 2.06it/s]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 31/31 [00:15<00:00, 2.18it/s]

[fair oxide]

uninstalled python and reinstalled it again with all the required options as before, then ran sfc /scannow and got the following message...

Verify and Repair Transaction completed. All files and registry keys listed in this transaction have been successfully repaired.
but running the webui is still giving me the same error as before

[ornate elk]

okay, do you have the full cmd log after starting it?

[fair oxide]

yes thats what i posted in txt file

before that is the normal version and commit hash

right after that line is the load error in txt

would i need to clear anything inside Appdata?

[fair oxide]

a suggestion ... when you have to post a huge chunk of text its probably better to use the + menu on the left side of chat bar and upload as text file

[chrome rampart]

ok i will thks

[fair oxide]

@ornate elk let me do a quick reboot

[chrome rampart]

ok after tip torqx i now upload file. So i installed my new 3060 and works great can finally use sdxl and comfy now (<1 min). I just wonder if i check cmd (see file) if all is ok or can improve. Thanks for checking and any tips

[fair oxide]

@ornate elk not much luck, still getting the error as before, let me repost the entire console ...

this is the entire log @ornate elk

i wonder if there is anything related to AppData

[fair oxide]

with 3060 for command line arg, you should be fine with just set COMMANDLINE_ARGS=--xformers in your webui-user.bat file

[ornate elk]

hmm, that error doesnt provide any helpful hint.

[chrome rampart]

a ok but wonder more about the other lines with addnet and so on. ALso if the iterations are ok? speed is improved HUGEEEEEEEEEEE sinc i had 1660 lol

[fair oxide]

i have no idea either but anything related to AppData files maybe?

[ornate elk]

remove the additional network addon, its not needed, and move your loras and lycoris to models/lora

[ornate elk]

in appdata is python installed

so if you clean installed python its ok

[fair oxide]

it is i think altho i reintalled python with default settings and selecting path

[chrome rampart]

ok network clear but strange is loras are in models/lora already

[ornate elk]

thats good

you maybe installed it only for lycoris

but not needed anymore

[chrome rampart]

looks like

[fair oxide]

@ornate elk anything related to ui-config.json about that error?

[ornate elk]

you could delete config.json and ui_config.json to reset all settings

[fair oxide]

ok let me try that

[fair oxide]

wow that fixed the glitche, right after i renamed the config.json

thanks @ornate elk 🙂

webui is up on my browser now and active

[ornate elk]

nice, np 🙂 good to know for me aswell

[fair oxide]

but isn that supposed to recrate a new config.json file

[ornate elk]

yes it should, if not you need to use git pull

[fair oxide]

it recreated ui-config.json

[chrome rampart]

thks cs and torrqx

[fair oxide]

ok let me run git pull

i dont see it creating a new config.json in the folder

[ornate elk]

run it again and reload the folder

[fair oxide]

dont see it still, but anything related to cache.json ?

[ornate elk]

idk

[fair oxide]

let me change some settings inside webui and save the changes, if that would have any effect

[ornate elk]

maybe that creates the config

worth a try

[fair oxide]

yes it did

well that was something new, glad we got to sort it out

[lost crescent]

Anybody know why A1111 has all of a sudden got to spend 10 minutes at the start of every session building a sd_xlrefiner.yaml file?!

I tried to change the number of checkpoints loaded at startup to 10 ... but that has not worked.

[ornate elk]

Try load a 1.5 model then restart the webui

[left basin]

Hello guys, I keep getting this message when uploading pictures to my fine tuning model

But couldn't find info regarding eligibility, any chance I can make it work?

[lost crescent]

1.5 model also spends 10 minutes creating a new yaml file!!! 😦

[ornate elk]

Do you have a screenshot of the full cmd log?

[lost crescent]

Where is this LOG to be found?

[ornate elk]

Make a screenshot of the cmd after launching the webui-user.bat

There is no specific log

[opaque mesa]

Thank you very much for your help, but it didn't work for me ! After two computer crashes ( weak PC ) and problem still unsolved, I will just completely reinstall, that will be far easier and time saving haha.

[ornate elk]

Oh okay, but then install it following my guide:
#🤝｜tech-support message

That will make it easier to update and troubleshoot in the future

[karmic crown]

i am using a1111, which is usually slower than comfy, so you should get faster generation times than me. i noticed that generating images is inded much faster than with LCM, but unlike LCM it take a lot of time to even start generating the image

@vital storm btw, which version did you download (13GB or 6GB)?

6GB works faster

i tried both

[vital storm]

6gb, fps16. I downloaded 13gb, will see if that makes a difference

[karmic crown]

loading of 13GB version was like 40seconds

[vital storm]

I just came around LCM, got any guide to get started?

[vital storm]

I'm going through the source code to see why it's "reloading" the model "again" or what's causing the slow init.
the progress bar is frozen for about 50s, with ?s/it, then done all in one a flash.
something is "freezing" it

[karmic crown]

i think i will stick with lcm
check out this
#🏞｜general-with-images message

just 8 steps, no upscale, 1.5 model

[vital storm]

this is really good, will definitely have a look at LCM as well.
dynamically typed languages are really not my thing. idk how people do it. hats off to that 🎩

[karmic crown]

that is a combination of lcm and sdxl renderer

https://civitai.com/models/195519?modelVersionId=225222 LCM 1.5 version (be sure to use the right one)

https://civitai.com/models/171159?modelVersionId=236130 (1.5 lora that make 1.5 looks more like SDXL)

[ornate elk]

Looks more like you installed some extensions that are loading.

Also thats not the full screenshot

From to top to bottom

[ornate elk]

Yea you have many extensions that all require different stuff. That takes so long at startup

If you dont use Kandinsky or prompt gallery or dynamic prompts, or auto-sd paint then remove those extensions and delete the venv folder after that.

[karmic crown]

@ornate elk i now use dynamic prompts yaml instead of styles.csv 🤣

i have much more control of my prompts now

[ornate elk]

Okay

[lost crescent]

This has only begun recently. It never used to wait and formulate a YAML file at all!

Some claim that I should limit the number of checkpoints loaded at startup - i set that to 10 - but it didn't stop the Yamlling

[ornate elk]

Have you tried set it to 1 ?

[lost crescent]

OK, I will try 1 🙂

[lost crescent]

Is there a cache to empty in A1111?

[ornate elk]

No, but you can run
pip cache purge in a cmd to clean temp install files

[wind sun]

hey anyone know how to face swap videos ?

oh dont tell me i need adobe apps

[torn oracle]

Hello,
Do you know if the MacBook Air M2 chip is a good device for image generation with stable XL ?

[karmic crown]

how much RAM do you have?

[torn oracle]

I don't know 😅 but the shop website says :
8-Core CPU
10-Core GPU
8GB Unified Memory
256GB SSD Storage
I don't know if it's useful 😅

[karmic crown]

you will probably be able to use 1.5, but it will be slow as hell

forget SDXL

i have m1 macbookpro 16GB and SDXL is extreamly slow

[torn oracle]

Slow but possible ? With good results even if it's slow ?

[karmic crown]

with m2 with 24GB or more, you can probably use sdxl with more decent speed

[karmic crown]

if you need 20 minutes for 1 image, whats the point?

[torn oracle]

Yeah it's true lol

[karmic crown]

better get m1 macbookpro with 16GB than M2 air with 8GB

for ANY task

[karmic crown]

if you have enough money, buy at least pro version with 16GB of RAM and 512GB SSD
256GB is not enough space, this is my 3rd macbook pro with 512GB SSD, so trust me
also i had 8GB on my first macbook pro (16GB is the mininum that i would recommend to anyone)

you will be able to use sdxl, but it will be slow compared to nvidia, but it will be much faster than with that one you mentioned

[gray leaf]

Run diffusion is good for bad laptops

[torn oracle]

24GB unified memory is good ?

[karmic crown]

yes, but be careful, that says up to
in most cases you need to ask for a specific version and to wait, since they usually do not have those laptops on stock
the have basic versions, but not the "advanced" models
i had luck with this one i have now, since someone order it and then decided not to buy it
so i waited only 2 days instead of 2 weeks 🙂

[torn oracle]

An Apple assistant just made me a shopping cart with the 24GB unified memory, of course I didn't bought it for now lol

[karmic crown]

256GB SSD will be full fast, better get 512GB

the problem with macbooks is that you cannot upgrade memory, and if you change drive it might cause some problems (i saw that)

you cannot change your mind later and say i want to add more mory, since memory is soldered to the motherbord

if you want to save money, buy smaller version

mine is 13.3"

74,05 GB available of 494,38 GB

[tame yoke]

is there a way to have something here by default

[wind sun]

for 30min loading im starting to think there is something wrong

[karmic crown]

just restart ui

[rugged compass]

4080 vs 3090?

sd only a leisure hobby

4080 looks vastly superior for other purposes such as gaming

[tulip saffron]

Hello all. I'm new to AI training. I have a style that I want to use for Stable Diffusion: specifically the color palette and medium (digital painting with subtle brush strokes). Should I train a model and output a checkpoint or should I go for a LoRA? If I train the model on images that have the style I want but they all have cityscapes, for example, can Stable Diffusion then apply this style to a nature landscape or does the model need to be also trained on images with nature? Thank you!

[ornate elk]

Then a 4080 would be the better way, better performance and less power consumption, and longer support

[rugged compass]

Reckon it will be worth it to wait for super release?

As it might bring the price down, and 4070 ti also has 16gb speculated

[true rover]

In what folder should I install LORA?

[wicked flint]

@ornate elk hi there, do u have any guide about:
making model, and generating an image using AMD GPU?

[sweet slate]

Many thanks man, fixed my problems

[true rover]

@ornate elk what is this

[stark parrot]

hey can anyone tell me why when i install the extension ''replacer'' it will not show in my tabs? it shows that i have it installed in extensions but i only seen the replacer tab one time and it never came back

[lyric vector]

My browser been lagging when ever I switch away from my Webui when I click back its just white and waiting for it to load tried opening a separate window for it but its still lagging when ever i check back Im not currently rendering anything using Opera GX

[formal trout]

I'm getting crashes while trying the img2vid Comfy workflow https://comfyanonymous.github.io/ComfyUI_examples/video/

I'm on Ubuntu, RX 6700

No idea what could be, but Comfy usually make my PC lag/Stutter, and in many opportunities it crashes while performing certaing operations

Tried the regular model and the fp-16

[marsh flume]

So i updated my A1111 ui again and now my "save prompt" button is missing

[ornate elk]

Yes, do you have an AMD GPU?

[ornate elk]

Make sure to not use the Ooera VPN function.
Also you can try an other browser to test.

[ornate elk]

In models/lora

[wicked flint]

yup im using amd gpu

[ornate elk]

Then here is the guide.
Make sure to delete the other webui that was for nvidia and run the command I mention at the top:
#🤝｜tech-support message

[wicked flint]

cool! would be great if its pinned

[karmic crown]

i guess, you can try to disable GPU acceleration in browser

[wicked flint]

if i want to make models, how many images should i have/prep?
is there is any guide for it?

[lyric vector]

Okay I can try those aswell moving to another browser and turning off hardware accel

slowness happens aswell lets say I have a 4k image I upscale I tab out of webui later I'll come back and its going to load abit looking at my ram usage my browser doesnt really take that much like 2gb less i dont have anylimitations on how much cpu/ram it uses

[ornate elk]

I dont think you cant train on AMD rn on Windows

[wicked flint]

i want to make talking head videos using the photos i have, is that possible to achieve without training a model?

i want to turn the image to be cartoonized and then turn it to a talking head videos

[ornate elk]

Okay, idk about the talking thing

[wicked flint]

https://huggingface.co/spaces/vinthony/SadTalker
tried using this and it works really well

with the combination of SD and RVC it should turns out really well

[ornate elk]

okay, then you need to install the SadTalker extension

[wicked flint]

mb sorry i just realized that im asking a wrong question approach x,x
what i need is:
"how to converting the photo that i have to cartoon/anime/other artstyles?"

[karmic crown]

img2img

[wicked flint]

THIS ONEEEEEEEE

[karmic crown]

https://www.youtube.com/watch?v=ZvNR9De_88k

this?

https://www.youtube.com/watch?v=BO4GTG1Gb-4

[wicked flint]

Thankyou, yeah thats what im looking for

[ornate elk]

oh yea, igm2img is the way to go, maybe later use Controlnet

[sturdy ginkgo]

hey guys.. so i am trying to launch comfyUI and i get this error:

Traceback (most recent call last): File "D:\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\main.py", line 220, in <module> loop.run_until_complete(run(server, address=args.listen, port=args.port, verbose=not args.dont_print_server, call_on_start=call_on_start)) File "asyncio\base_events.py", line 653, in run_until_complete File "D:\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\main.py", line 124, in run await asyncio.gather(server.start(address, port, verbose, call_on_start), server.publish_loop()) File "D:\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\server.py", line 617, in start await site.start() File "D:\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\python_embeded\Lib\site-packages\aiohttp\web_runner.py", line 119, in start self._server = await loop.create_server( ^^^^^^^^^^^^^^^^^^^^^^^^^ File "asyncio\base_events.py", line 1525, in create_server OSError: [Errno 10048] error while attempting to bind on address ('127.0.0.1', 8188): only one usage of each socket address (protocol/network address/port) is normally permitted

no other instances of Comfy are active

or anything using that port on my network card

[karmic crown]

the other is probably in "zombie" mode

check processes

https://www.techrepublic.com/article/how-to-use-task-managers-processes-tab-to-troubleshoot-issues-in-windows-10/

[sturdy ginkgo]

okay.. guess the online version from a website also loaded it local 😄

[wicked flint]

yay i made it!
anyone know how to check if im using the latest stable diffusion version?

[karmic crown]

you are not 🙂 you are using 1.6.0 (i see that in console log)

go to your stable directory from cmd, and type
git pull

[wicked flint]

done, what should i do next to check the stable diffusion version?

[karmic crown]

git status

[karmic crown]

good

thats fine

but i think your command line is not as good as it should be

check this #🤝｜tech-support message

use one from step 5

[wary mulch]

Hello mates! Appreciate any help. automatic1111 does not starting ^
AttributeError: partially initialized module 'modules.sysinfo' has no attribute 'format_exception' (most likely due to a circular import)
Press any key to continue . . .

[karmic crown]

what you have above

[wary mulch]

I also had this

but I hadn't changed any directories

[karmic crown]

can you copy the complete output, put in a text file and send text file?

[wary mulch]

The issue started then I turned on both COntrolNets (basic and for animateDiff)

[ornate elk]

lol if you had followed my guide you would have the latest version + the right settings in the webui-user.bat

[karmic crown]

During handling of the above exception is still on top

Basically, you get an errror, than another and another

i need to see what was the original one

[wary mulch]

I'll insert the whole text, 1 sec

[karmic crown]

but
argparse.ArgumentError: argument --controlnet-dir: conflicting option string: --controlnet-dir

[karmic crown]

RuntimeError: Error running install.py for extension F:\stable-diffusion-webui\extensions\sd-webui-roop.
C

this one is the problem

remove roop

[wary mulch]

By just deleting Roop folder ?

[karmic crown]

delete the roop folder in the extensions folder

better use reactor

@ornate elk right?

[ornate elk]

yes exactly, also he may want to update his webui

[wary mulch]

yay, You fixed it, Many Thanks! Going to use reactor, hope i'll be installed with no issues

[karmic crown]

check for update of webui and plugins

as CS1o just wrote

[wary mulch]

There is only for extensions. How may I check it for the whole 1111 ?

got a lot of errors in ui

[karmic crown]

go to stable diffusion folder and type
git status

if you do not see Your branch is up to date with 'origin/master'.
then
git pull first

[wary mulch]

but now it's up to date

[ornate elk]

lycoris extension can be deletet after you moved your lycoris to the models/lora folder

[wicked flint]

u're right the guide i saved on the notepad got mixed @_@
rn im reinstalling it using the proper guide from the capt @ornate elk

[wary mulch]

I'm a bit of noob, Should I type in cmd ?

[karmic crown]

yes, but better rename your current install
and reinstall following CS1o's guide
then move models you need from old to new installation

[wary mulch]

Thanks. do you mean renaming this main folder ?

[karmic crown]

yes

[wary mulch]

Do you mind me asking you to your guide link ?

[karmic crown]

see mine 🙂

#🤝｜tech-support message - guide

click there 🙂

[ornate elk]

thats the one for amd

[karmic crown]

@wary mulch you have amd right?

[ornate elk]

@wary mulch use this if you have an nvidia gpu:
#🤝｜tech-support message

[karmic crown]

i am sorry if i mixed something up

[ornate elk]

np, looks like hes not on AMD

[karmic crown]

but anyway you have both guides now 🙂

[ornate elk]

im rewriting the nvidia guide now for the final Pin

[wary mulch]

yup nvidia. And should I use those args with 3060ti ? What args in addition to xformers? Or neither of it is nessessairy ?

[karmic crown]

my nvidia guide has "ask @ornate elk for right command arguments for your card" as a last step 🤣

[ornate elk]

--xformers --medvram-sdxl --no-half-vae

[wary mulch]

But I'm generating on 1.5 models, sdxl is taking too much time on my 3060ti

anyway thank you guys for reaching back

[karmic crown]

it should not take too much time with those arguments

[ornate elk]

yea should be much faster now

[wicked flint]

@ornate elk its the latest update right?

[ornate elk]

yes

[karmic crown]

wait?

[ornate elk]

hes on AMD

its the latest for directml

[karmic crown]

it should be 2.0.0?

ok

sorry

my mistake

listen to CS1o

[wary mulch]

DOes it better to hold webui on HDD or SSD ? I know ssd is faster but how long it may be alive with a lot of recordings and erasing on it ?

[ornate elk]

SSD is better and you can move the models on HDD if you want to save space and then add:
--ckpt-dir "Path/to/Models"
to your webui-user.bat

[wicked flint]

so sad, so nvidia has the latest version which is 2.0.0
while AMD is 1.7.0? 😢

[karmic crown]

no no

1.7.0 is the latest

i was talking about sometning else

and i was wrong

i was talking about pytorch

[wicked flint]

🙇‍♂️ thankyou @karmic crown and @ornate elk

[karmic crown]

i am using 2.1.2 on mac, nvidia use 2.0.2 and amd use 2.0.0

IndexError: list index out of range

i am trying to use regional propmter, but i have no idea what i am doing wrong

back to basics, with simplest posible promt, without any negative

[wary mulch]

Got new errors that interrupts UI to start. Any thoughs ? After installing reactor and adding those arguments above

[ornate elk]

your running on an old xformers version, and you didnt read the install guide for Reactor on github, as its not plug and play:
https://github.com/Gourieff/sd-webui-reactor#installation

[wary mulch]

I did . Deleting animDiff helped to start.

[ornate elk]

okay, you also installed the visual studio stuff?

[wary mulch]

Yes sure,

[ornate elk]

can you show me the version of the webui in browser, scroll to the bottom

[wary mulch]

then reboot. So now I need to bring animatediff back xD

[ornate elk]

okay, your xformers version is old. 0.0.20 is the latest.
you can use --reinstall-xformers
in the webui-user.bat to upgrade it.
after that remove it again

[wary mulch]

I also did "git pull" in main folder and nothing happened to dowloading

[ornate elk]

yea its updated. 1.7.0 is the latest

[wary mulch]

Now cmd has just avoided this command, or I did it wrong

[ornate elk]

you still need --xformers in there too

--reinstall-xformer --xformers

after its done delete the --reinstall-xformers

[wary mulch]

Thanks you it's updated.

[wind sun]

What do you guys think about that for SD

[karmic crown]

looks fine to me

since it is a laptop

as you would say "това е добре братко" 🤣

[wind sun]

xDDD

[sturdy ginkgo]

anyone know how to fix OoM errors or size mismatch for proj_in.weight: copying a param with shape torch.Size([1280, 1280]) from checkpoint, the shape in current model is torch.Size([1280, 1664]) errors with Ipadapter in ComfyUi

using the SDXL Ipadapter_plus model

[young bridge]

hey guys

which model is the best one to generate greetings cards on social media

[ornate elk]

thats a question for #📝｜prompting-help

[torn coral]

someone has a extension for faceswapping 1.7 version?

[karmic crown]

https://github.com/Gourieff/sd-webui-reactor

[torn coral]

its installed but doesnt show up?

[karmic crown]

did you click this

[torn coral]

yh

im reinstalling sd

[torn coral]

someone knows why reactor wont show up

[ornate elk]

You may need to delete faceswap

Also you need to follow the install guide for reactor in github

[torn coral]

i did vscode isntalled and everythinf

it worked before

thats why im so confused

[sharp lichen]

@ornate elk Do you take Dms?

[karmic crown]

vscode?

[torn coral]

yh

does it only work with 1.5.0?

[karmic crown]

what vscode has to do with stable diffusion?

i am confused

[torn coral]

brother u need to have it installed

to use the extension

[karmic crown]

if you are talking about VS C++ Build Tools, that is something completely different

[ornate elk]

Yes

[ornate elk]

For reactor both can be used

[karmic crown]

nope

vscode is just an editor

vs c++ build tools = compiler and some other stuff

Visual Studio 2022 = ide + compiler....

3 different things

vscode = Visual Studio Code = my favorite editor = my favorite microsoft product

[ornate elk]

I only say how its told on the guide. I know the differences xD

[karmic crown]

guide is fine, there is no vscode in the guide 🙂

[ornate elk]

Yes right, only visual studio

Oh he said vscode

Yea now I get it xD

[karmic crown]

when someone say "vscode" i instantly turn into Gollum dont touch "my precious"

[ornate elk]

Do you know vscodium?

Microsoft freed version of vscode

[karmic crown]

yes I know, same thing as Chromium vs Chrome

[fallen zephyr]

I received a failure:
Stable diffusion model failed to load

[karmic crown]

it is posible that model is not properly downloaded

[fallen zephyr]

So I may uninstall it and try again

[karmic crown]

sure

[ornate elk]

whats the model size?

[fallen zephyr]

It gives me the failure during the start of webui-user.bat

[ornate elk]

can you post a full screenshot of the cmd?

[fallen zephyr]

I just made a reinstall, if the failure happens again I'll post the full cmd

[fair oxide]

where do I get started downloading stable diffusion? The faq link is confusing

[ornate elk]

for nvidia or amd gpu?

[fair oxide]

nvidia

[ornate elk]

Then here is my install guide:
#🤝｜tech-support message

[fair oxide]

Thanks

I got my old stable diffusion back, but now it won't let me gen images

I keep trying to do hi-res fix, but it keeps running out of space

I don't know why it's doing this, it used to generate images before

now it's giving me a new error

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

man this sucks

[steady mountain]

Anyone in here know why I might get a ./webui.sh: line 256: 5876 Segmentation fault (core dumped) "${python_cmd}" -u "${LAUNCH_SCRIPT}" "$@"
error on my 7900XTX ?

[karmic crown]

post full log or screenshot

[steady mountain]

how much do you want?

here is a shorter one

[karmic crown]

i do not see that you have any params in log

have you change webui-user.sh and added params for your card?

#🤝｜tech-support message <- AMD guide

since you are on linux, you edit webui-user.sh

and you do not need --use-directml
so you should have something like
--medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling

on macs --upcast-sampling is the "magic" that stops crash similar to yours

[steady mountain]

I will try them out

I attempted similar tricks on another install

but we will se

[karmic crown]

i have imac with radeon, but on mac it use metal, so its not the same

but anyway even on my m1 i have a pyton crash without -upcast-sampling

[steady mountain]

/

looks good?

[ornate elk]

the new install or the old?

[fair oxide]

It was the old install

I managed to get back my good version

but it seems to be working now

I changed the .bat file

I changed it from medvram to lowvram

[ornate elk]

whats your gpu?

[fair oxide]

a shitty one XD

it's an nvidia gtx 1650

[ornate elk]

that needs --medvram --xformers --no-half

[karmic crown]

yes

[ornate elk]

i mean you can use --lowvram if you really need to

[fair oxide]

Well, when I had it to medvram, it kept saying I ran out of memory

[ornate elk]

okay, depends what your trying to do

[steady mountain]

Sadly no joy

[fair oxide]

All I'm doing is the hires fix

since my gpu can't do it normally

hell, it couldn't even do it even if I didn't have hires fix checked

[karmic crown]

still crash?

pleease post git status

[steady mountain]

sorry man. too much of a gui stunted noob to understand you

[ornate elk]

what settings, make sure to use hires steps:10 or 15 if not that will cause out of vram

[fair oxide]

damn, even with lowvram it couldn't even generate an image

[steady mountain]

git status?

[fair oxide]

My hires steps are on 0

[ornate elk]

maybe share your settings you used

[karmic crown]

go to your stable diffusion folder from terminal
and type
git status

[fair oxide]

How would I share that?

[ornate elk]

thats bad, and will use the same amount as steps used, not doable for your gpu

[fair oxide]

But I was able to do it perfectly fine last time

It used to generate images with no problem

I could even do hires fix

[fair oxide]

So I don't know what the problem is

[ornate elk]

you can do Win+Shift+S and then mask a screenshot area, and Ctrl+V in this chat

[fiery arch]

Hi guys, does anyone know how to uninstall Stable Diffusion from Mac OS? I wanted to clean it and re-do the installation. Thank you

[fair oxide]

that's not what I meant lol

[karmic crown]

looks fine, so i am confused

[fair oxide]

I meant, where do I share my settings?

Do I just take a pic of stable diffusion?

[steady mountain]

yes, you are not alone

[karmic crown]

please go to your folder where your webui is and type pwd, and post the result

i have a better solution for you, i just need to know where is your installation

[fair oxide]

it fails at the last second

[ornate elk]

if you got an out of vram error you need a clean restart of the webui to clear the vram

before trying again

[karmic crown]

@fiery arch also, which mac do you have?

[fair oxide]

So just exit out of the program and restart?

[fiery arch]

macos 14.1

[karmic crown]

which model? m1, imac, how much ram...

[fiery arch]

macbook pro 16, M1 Pro, 16GB

[karmic crown]

ok, than you need same params as me
just tell me where your current installation is
go to your folder where you run webui
and type pwd

it will give something like this

❯ pwd
/Users/viking/stable-diffusion-webui

[fiery arch]

/Users/hong/stable-diffusion-webui

[karmic crown]

great

[ornate elk]

also use denois 0.5 and try Esrgan4x+Anime6b

[fair oxide]

mkay, thank you

[karmic crown]

https://github.com/viking1304/a1111-setup/releases/tag/0.0.5

download my script and just run
sh a1111-setup.sh -f all

and wait 🙂

[steady mountain]

I am going to try a docker install tomorrow. The weird thing is that I got a version working at first, after a pretty troubled and chaotic late night install. But it failed after a while. I am worried I might have fucked up the original first ROCms install. and that there is some kind of leftovers in the system somehow. Stable diffusion say ROCm 5.42, BUT

I might just try to purge the ubuntu install of all ROCm and redownload

[ornate elk]

also this version of the webui is old. you should try the new install with --medvram --xformers to test if it might be better

[fair oxide]

how do I update it?

[karmic crown]

@fiery arch my script will update your webui to the latest version (1.7.0), restore original webui, and webui-user, remove the old venv, aplly some fixes and reinstall lauch webui which will create new venv for you

it will not touch your models or extensions

[fair oxide]

I was able to generate something this time

woohoo

[fiery arch]

where should I put your folder to?

[ornate elk]

you need to add git pull to the webui-user.bat in an empty line before webui.bat

then delete the venv folder
it will install the updated torch and xformers

[karmic crown]

wherever you want, just do not put it in stable diffusion folder you have

[fair oxide]

okay, I'll try that

thanks

[fiery arch]

i see. so it automatically detect my sd folder and update it?

[karmic crown]

try to remove venv folder and let webui create a new one on start

[karmic crown]

by default is use stable-diffusion-webui inside your home directory, thats why i asked where is yours
there is an option to use other folder, but it is not needed for you

[fiery arch]

installation complete as is?

[karmic crown]

please show me what you have in the bottom of your webui

[steady mountain]

How do I easiest remove the venv folder and let the webui create a new one?

[karmic crown]

not that

user interface of automatic1111

the bottom part with version

[steady mountain]

are you are real viking viking1304? 🙂

[karmic crown]

find your webui folder from file manager and just remove it from there

[karmic crown]

looks good

[fiery arch]

so installation complete?

[karmic crown]

yes

[karmic crown]

try to generate something simple like 20 steps without any upsacale

[steady mountain]

I do not understand what you are saying really. I already know were the folder is, it is a folder in my home directory

[fiery arch]

If I have a ssd with SD installed previously, will your script auto update SD on my Mac or on the SSD?

[karmic crown]

remove venv folder that is inside stable-diffusion-webui folder

[steady mountain]

ah. so I should just delete the venv folder?

[karmic crown]

yes