🤝｜tech-support

[ornate elk]

maybe the background

[hot thorn]

and those mediapipes?

[ornate elk]

just other models for the same task, faces, eyes

other quality

[hot thorn]

so just using face and hand should be good enough?

[ornate elk]

hand wont work that great but try it out

face works very good

[hot thorn]

is there a difference between 1st and 2nd face option?

[ornate elk]

8n is better as its larger

[hot thorn]

and just to clarify, it improves clarity or changes the eyes to something else? because if my characters are up close their eyes are great, so i know the trained data is good, its just that eyes from slightly further away get blurry

[ornate elk]

check this helpful guide i found rn: (a bit old)
https://civitai.com/articles/2034/after-detailer-adetailer-stable-diffusion-guide-by-thinkdiffusion

[ornate elk]

what adetailer does is the same as when you inpaint a face manually in img2img

adetailer auto detects the face and then uses inpaint to make the face highres

it all depends on the denois value how much it will change

[hot thorn]

well gonna try it anyways, better to try it out than spend hours or days trying to make it work with random ideas

[ornate elk]

sure adetailer is a real help if the character is further away but also medium closeups can be improved

[hot thorn]

do i do anything with these?

[ornate elk]

nope, only on the inpainting part if you want a lower denois value

[hot thorn]

i gotta say, did it with 1 image that already had ok quality face, and it just turned it from 7/10 into a 12

it did add about 1 min to generation time tho

my images usually take 5-6 min

[ornate elk]

oh why that xD

[ornate elk]

yea its awesome

[hot thorn]

i dont have an amazing pc. like 16gb ram and gtx 1070

[ornate elk]

ah okay yea GTX with 8gb vram and sdxl/illustrious/pony is slow, but i thought its more like 2-3 mins

what are your steps? and do you use hires fix too?

[hot thorn]

with my settings not so fast

pls dont diss my upscaler, i like the results on this one

[ornate elk]

ahh okay, i would go down with the resolution, maybe to 768x1024

then upscale by 2

[hot thorn]

when i upscale too much i get different problems, think they start from 1.6 or 1.7

like complete body deformations, double, triple navels

mangled torso

[ornate elk]

get yourself the tiled vae extension, it helps a lot when upscaling

[hot thorn]

are u just gonna keep changing my entire worldview on generation? xD

[ornate elk]

ah that could be the upscaler xD

[hot thorn]

so far i've just been trying and trying to make my images better with better prompts and some changes here and there with noise, upscalers, and numbers

[ornate elk]

my old PC has a GTX1080, i know the tricks 😛

[hot thorn]

im completely fcked when it comes to complicated options

[hot thorn]

how exactly does this work? and do u think it'd fix my upscaler problems?

or is it because im using the model from screenshot?

latent antialiased i mean

[ornate elk]

mostly yes, would go for more stable latent upscalers like Ltent bicubic or latent nearest exact

tiled vae will speed up the upscaling progress by tiling the vae at the end step

[hot thorn]

ye ive tried all of them, but the results are either not to my liking or worse

[ornate elk]

feel free to share an image

[hot thorn]

guess i will dedicate my entire day today to trying things out

my images are not the kind to share in public places xD

[ornate elk]

then i would recommend using the X/Y/Z script xD

you can compare everything with everything

[hot thorn]

what is this and how do i use it

[ornate elk]

okay then in DM if you like

[upbeat moth]

@ornate elk Hi whenever you’re done and have some time I would appreciate your help with the issue I had earlier regards my video url

Or is there a way to ping the devs here

[ornate elk]

sry cant help with that

[upbeat moth]

Okay is there any other dev here

Don’t tell me I have to email their support

[ornate elk]

if its related to their paid api then yep

[upbeat moth]

It is related to their paid api 💔💔💔💔💔💔

[ornate elk]

its at the bottom of the webui
you can enable it and select what you want to compare, for exmaple checkpoints or hires upscaler etc

even loras and their values

[hot thorn]

tho i assume it will generate several pics which will take a while?

[ornate elk]

yep if you compare 2 checkpoints at 15 and 30 steps you will get 4 images as example

[upbeat moth]

But I am getting the video url successfully, the issue is I am trying my image-to-video endpoint on postman but the video url I am getting is over 1 834 000 characters, like it contains the entire video encoded as a one big data url and obviously postman is going to crash because it can’t handle all of that. I have rewritten my code to upload the storage on AWS and supabase and also added this to my environment variables “MAX_UPLOAD_SIZE= 1024 * 1024 * 50” to be able to handle more but so far, nothing is working. Even when I go out of postman to try it on browser directly it crashes too. I do think that this is an easy fix with a quick answer just don’t want to wait 2-3 days to be able to move on. They should consider maybe hiring few more devs to this server

[hot thorn]

and oof, checked body adetailer, its bad af

[ornate elk]

this is a community server. not an official way to contact stability ai or the devs

[upbeat moth]

“Tech support” 🤝

But okay noted, thanks. I’ll email them

[hot thorn]

@ornate elk is model mixer any good?

also, i dont have anything that has XYZ in it

[upper linden]

oh thank you!

[ornate elk]

havent used it

[hot thorn]

i downloaded it and restarted SD, is there an option to turn it on? or is it a passive extension?

[ornate elk]

you should have a Tiled VAE button in txt2img

oh wait

your on reforge

then you can delete the tiled vae extension.

its already integrated

sry

[hot thorn]

only tile i got is this

[ornate elk]

ah thats also not the correct one

thats it

in reforge and forge

[hot thorn]

should i uninstall the one i just got?

its not even on my installed extension list xD, and gone from available

also what is this?

never noticed it before

[ornate elk]

thats tiled diffusion/vae

it has 2 names

so delete the multidiffusion folder in extensions and your good

[hot thorn]

is it gonna cause problems if left alone? its not active atm

[main nest]

Hey CS1o, how are you doing ?

[ornate elk]

yea better delete if not needed

[ornate elk]

hey im good thx, and you?

[main nest]

Aways good, can I ask you for some help

[ornate elk]

sure

[main nest]

So before I was doing some stuff with SD but I pre-installed my PC and Im fresh start with it without a lot of memory what I was doing, I have some screenshots so now im trying to get what I once had and this screenshot is from start up for Facefusion, can you guide me how I can install FF once again

[ornate elk]

are you on nvidia or AMD ?

[main nest]

nvidia 4070 Super

[ornate elk]

i wrote a startup bat file for facefusion, i can send it real quick.
Just put it into the folder of facefusion before launching it

oh wait

it was for roop

[hot thorn]

@ornate elk where do i find this xyz stuff? i dont see it anywhere

[ornate elk]

here

at the bottom of txt2img

[hot thorn]

and how do i work it to compare upscalers?

[main nest]

So about the FF what should I do

[ornate elk]

ill rewrite the launch file then i can send it

give me a few mins to test

[main nest]

sure, thank you

[hot thorn]

got it, will try and see how things go

[main nest]

would you mind tagging me here when you are done

[upper linden]

quick question i installed the zulda for comfyui and get thhis error RuntimeError: CUDA error: CUBLAS_STATUS_NOT_SUPPORTED when calling cublasSgemm( handle, opa, opb, m, n, k, &alpha, a, lda, b, ldb, &beta, c, ldc)

[ornate elk]

done xD

[ornate elk]

you have to replace the 3 renamed dll files in torch/lib

the guide states which ones and how to

[ornate elk]

dont forget to enable Never OOM for tiled vae

[hot thorn]

did

[hot thorn]

question. If i run 3 loras (1 for character, 2 for other details), and my character's clothes or other details are being changed because of those 2 other loras, is there something I can do to force character's lora to be more important or something like that?

[ornate elk]

yes you can adjust the loras strenght

adjust the number from :1 to 0.5 for the 2 detail ones

[hot thorn]

also, i run
prompt
break
detail lora
break
character lora, would simply putting character above detail loras affect it?

[ornate elk]

yep

what comes first gets more prio

[upper linden]

i replaced the files and now it crashed on startup

[ornate elk]

then something in the path settings isnt correct

can you screenshot them?

[upper linden]

sure

should be like this right?

[main nest]

and where should I put this

[ornate elk]

Yep like here

[ornate elk]

After you git cloned FaceFusion
In that folder

[upper linden]

odd idk why its crashing then

[ornate elk]

Have you removed the czluda from the user variables here?

Because it should only be in environmental variables path

[upper linden]

no, im using automatic111 asweel idk if that breaks it

[ornate elk]

Nope

It wont

[upper linden]

okay i did that

still crashes

[ornate elk]

What's your python version?

[upper linden]

3.10.2 i think

[ornate elk]

Ah I see here the two python paths are also in user variable but should be placed where you added the other 2 entrys

[ornate elk]

You need 3.10.11 64bit
I doubt its 3.10.2

[upper linden]

then its probably that

otherwise it wouldn't open like before

[upper linden]

yeah i removed it in the user variables and now its in system variables

but stilll crashed

[main nest]

let me just say, I don't have anything on this PC except the SD installation which I made from your guidde

[ornate elk]

delete the venv folder and relaunch

[ornate elk]

ah okay

then go in the ai folder and open the cmd there

by clicking in the explorer bar and type cmd and hit neter

then copy and paste this command:
git clone https://github.com/facefusion/facefusion.git
then put the .bat file i send you in that facefusion folder

and launch

[main nest]

should I be only in the Ai folder or the SD folder

[ornate elk]

Ai folder

[upper linden]

still crashed

[ornate elk]

Open up cmd in the comfyui folder and run the bat trough the cmd

Then send the output

[upper linden]

Prestartup times for custom nodes:
1.7 seconds: F:\SD-Zluda\ComfyUI\custom_nodes\ComfyUI-Manager

Traceback (most recent call last):
File "F:\SD-Zluda\ComfyUI\main.py", line 134, in <module>
import comfy.utils
File "F:\SD-Zluda\ComfyUI\comfy\utils.py", line 20, in <module>
import torch
File "F:\SD-Zluda\ComfyUI\venv\lib\site-packages\torch_init_.py", line 143, in <module>
raise err
OSError: [WinError 126] The specified module could not be found. Error loading "F:\SD-Zluda\ComfyUI\venv\lib\site-packages\torch\lib\caffe2_nvrtc.dll" or one of its dependencies.
Deactivating virtual environment...
F:\SD-Zluda\ComfyUI>

[ornate elk]

Hmm okay

Can you show me the contents of your c:\zluda folder?

Also is it C:\ZLUDA and not C:\zluda ?

[ornate elk]

Okay open up a cmd and type
Python --version
Ans then
Where python

[upper linden]

Python 3.10.6

[ornate elk]

Screenshot please the
Where python

[upper linden]

C:\Users\crazy\AppData\Local\Programs\Python\Python310\python.exe
C:\Users\crazy\AppData\Local\Microsoft\WindowsApps\python.exe

[ornate elk]

Uninstall every python version you have

Then install python 3.10.11 64bit
Then delete the venv folder and relaunch

And check add python to path when Installing

[main nest]

yup, all good, thanks

btw which is the best face swapping method rn with SD

[ornate elk]

For realism, Reactor

[main nest]

and do you mind sharing a guide how to install it 😄

[ornate elk]

Go into the extensions folder of SD

There open up a cmd like you did before

Then copy ans paste this:
Git clone https://github.com/Gourieff/sd-webui-reactor-sfw.git

[main nest]

done ,thanks

[ornate elk]

Then relaunch two times

Then it should be installed

[upper linden]

done

[ornate elk]

And you have a Reactor tab in txt2img

[upper linden]

still doesnt work

[ornate elk]

Okay, do you have time for discord screenshare in 30 mins? XD

[upper linden]

sure

[main nest]

I have another friend which wish to work with me on my projects and he has top tier PC with 4090 Super so is there any tricks for realism with high settings

[upper linden]

also now my automattic111 isnt working

Failed to load ZLUDA: Could not find module 'C:\ZLUDA\nvrtc64_112_0.dll' (or one of its dependencies). Try using the full path with constructor syntax.

it was working perfectly fine

now the paths broke it

welp i have work in 30

nope

it got worse

because my automatic111 is nowgiving me the same error

[ornate elk]

Delete the venv of auto1111 and it will work again

Thats needed after updating Python

Also delete the .zluda folder inside the stable-diffusion-webui folder

Then relaunch

Also reboot the PC

[upper linden]

where is that?

[ornate elk]

In the stable-diffusion-webui-amdgpu folder

There is a .zluda folder and a venv folder

[upper linden]

its not here

[ornate elk]

Okay then delete the venv

And relaunch

[upper linden]

okay comfy opened

now i replace the file right?

[ornate elk]

Yes

The three renamed dlls

[upper linden]

crashed

god damn it

[fathom flame]

Why am I not able to create images

[upper linden]

OSError: [WinError 126] The specified module could not be found. Error loading "F:\SD-Zluda\ComfyUI\venv\lib\site-packages\torch\lib\caffe2_nvrtc.dll" or one of its dependencies.

same error

[ornate elk]

You can't do this for free here

[fathom flame]

I paid for Nitro

[ornate elk]

Nitro is discord not stable-diffusion

[fathom flame]

Oh

[ornate elk]

For stable diffusion check #artisan-faq

Or install it localy for free

If you have the hardware

[fathom flame]

Ok.. thanks

[upper linden]

it has to be the paths or somethhing

[ornate elk]

Yep

[fathom flame]

Yes . I have a mac 64 gb M3

[ornate elk]

Ah then check the pinned messages in this channel. It has a link to the local webui setup for MACs

[upper linden]

damn

[ornate elk]

I be at the PC in 3 mins

[upper linden]

i have work in 30

so i have to go

[upper linden]

this is in automatic 111 ZLUDA support: experimental
Failed to load ZLUDA: Could not find module 'C:\ZLUDA\nvrtc64_112_0.dll' (or one of its dependencies). Try using the full path with constructor syntax.

[ornate elk]

remove that from user variables

remove C:\ZLUDA here too

thats also user variables

copy these two and add them also into path of environment variables:

then remove them from user variables

C:\ZLUDA should be only added one time and thats in the Environment variables in Path. (You have it there with the %HIP_PATH%bin)
so remove it everywhere else

[knotty turret]

@ornate elk can u help having problems to update comfyui to use some nodes but

i use comfyui zluda

cant even check missing nodes now

[ornate elk]

what feature do you need?
close comfyui and go into its folder.
there click into the explorer bar and type cmd and hit enter

there type
git pull

[knotty turret]

the problem started after i did that

git pull

[knotty turret]

WanIMAGE

People say i need to have updated comfyui to use that nodes

[ornate elk]

can you give me the workflow so i can test

the thing is comfyui cant be updated normaly as the AMD Version is a fork that needs to be merged with the main comfyui repo manually by someone i know. He does that every month at least once

[knotty turret]

send u

i think need make clean install again but dam takes alot time and days to make it work🙃

[ornate elk]

fixed by deleting the venv and repaste the zluda dlls

[tiny gorge]

yo how do i change my comfy ui to use my 32gb vram instead of 8?

[civic fractal]

Anyone know how to fix this error with ComfyUI-MultiGPU node? "ComfyGGUF: "UnetLoaderGGUF DisTorch Multi GPU Unknown model architecture!"
I have updated Comfy, every node and reinstalled still won't work with this Wan workflow

[civic fractal]

Seems scammy

Sad I clicked that link lol should probably change my password now

[ruby yoke]

are you using a model that is fp8 by chance?

[unreal hatch]

Hello, I need a bit help. I am trying to generate some logos (i.e a 404 logo for a website, the page not found thing) but for the heck of me all the results are terrible.

I got up SDXL and SD 3.5. For SDXL I even got some LoRa specifically made for logos like the first from civit.ai (https://civitai.com/search/models?sortBy=models_v9&query=logo) but nothing. All i get is scrawny stuff

any idea why?

[vocal burrow]

You re using an SDXL model with an output resolution of 512x512 (then upscaled but it doesnt matter here).
That s your problem. sdxl is made for 1024x1024. You can usually straight about 30% off the training resolution in one or both direction but here you re halfing it. Therefore it can t apply the patterns it learned.

SD1.4, SD1.5, resolution : 512x512 (there is also a SD2 targeting 512x512 but ... don t use it)
SD2, SD2.1 resolution : 768x768
SDXL, SD3, SD3.5 resolution : 1024x1024

[main nest]

any idea while installing facefusion with this

[ornate elk]

what have you done lol xD
sd-webui-reactor is an extension for stable diffusion not its own programm

[main nest]

wait, im stupid

😄

[ornate elk]

you mixed something up it seems.
Facefusion is its own tool to do faceswap (standalone porgramm only for faceswap)
Stable-diffusion-webui is a tool for generating images, (it needs the Reactor extension to do faceswap like facefusion)

[main nest]

yea yea, I got it

I got messed up with all the messages 😄

[ornate elk]

delete the reactor folder and git clone it into extensions folder of the webui

[main nest]

now everything should be okay

my bad

[ornate elk]

no problem ^^

[main nest]

huh

im doing this on my friend PC now, that's why im doing it a second time FYI

[ornate elk]

ah okay

seems like ffmpeg is not installed

[main nest]

any idea how we fix this

[ornate elk]

https://www.youtube.com/watch?v=JR36oH35Fgg

A PC restart is maybe required after adding the path

[main nest]

okay so we added the ffmpeg, everything is fine, we have the FF now but when I try to open the webui-user.bat for the SD its giving me only this and nothing more is loading

and ofc the second I send the message, it appeard with a lot of downloads happenming and stuff

nvm

[ornate elk]

looks okay

reactor needs insightface to work

[main nest]

how do I install it tho

[ornate elk]

thats stated here:
https://github.com/Gourieff/sd-webui-reactor-sfw?tab=readme-ov-file#viii-for-windows-users-if-you-still-cannot-build-insightface-for-some-reasons-or-just-dont-want-to-install-visual-studio-or-vs-c-build-tools---do-the-following

scam

[main nest]

at this point I don't remember anything, Im complete bot

[main nest]

figured

[ornate elk]

how to?
how do I ?
where can i ask for support ?
support?
??
help

(dont mind me triggering the scam bots)

[main nest]

hahahaha

[ornate elk]

hey, i got ReForge working with Zluda, will add it to my Guides later

[main nest]

I need protobuf 2.20.0 not 6.30.2 ?

[ornate elk]

try this:

[main nest]

in the same cmd

[ornate elk]

there are also other ways to install it:
https://github.com/Gourieff/sd-webui-reactor-sfw?tab=readme-ov-file#installation

[placid pier]

Hey guys, I never used inpaint on A1111, does someone know an easy workflow?

[ornate elk]

hey, always set the inpaint area to "only masked" then set the resolution to a sqaure based on the model version you use, sdxl, 1.5.
Then play around with the denois. start at 0.5 and see how it goes

[main nest]

okay here is the thing, we have working reactor but when i face swap it changes the person from the original picture

here I tried to put a random girls face over megan fox

CS1o I think you need to trigger some bot spams 😄

[ornate elk]

its not for img2img

its either for txt2img or extras tab

if you already have a image you want to swap with an other face then use Extras tab
if you want to create a person with the source face then use txt2img

[main nest]

hm that's new

thats on the extras tab

[ornate elk]

hmm thats strange

if you want to replace faces on existing images better just use facefusion

thats perfect for that

[main nest]

yea actually thats a nice idea

I usually use the facefusion for video face swaps

but for some reason i never thought of just image swaping

[ornate elk]

yea it can do boath pretty good

for faceswap also enable the upscaler and face enhancer

[main nest]

I enabled the face enhancer but I fail to find upscaler

hmm does the facefusion use more CPU than GPU

and one more thing, I will need controlnet

or of corse if there is a new tool for form copying

[random night]

Wao, that's great, i'm looking forward to it

[ornate elk]

FaceFusion does mostly use CPU I guess

[main nest]

Do you mind guiding me trough the process of installing controlnet

[ornate elk]

Go to the extensions tab, click on available, click on "load from"
In the list search for
sd-webui-controlnet
Install it.
Restart

Then you need the correct controlnet models for the base models

[main nest]

now I found those ControlNet models, but how do I add them https://github.com/lllyasviel/ControlNet?tab=readme-ov-file#production-ready-pretrained-models

[ornate elk]

Here are the ones for sdxl based models:
https://civitai.com/models/136070?modelVersionId=267493

Drop them into models/controlnet folder

[main nest]

well how can I use those I've sent you

As there is two of them that will be great help to my work

[ornate elk]

These are old. Dont use them

For 1.5 based models use these
https://civitai.com/models/38784/controlnet-11-models

[main nest]

ooh so in this file, all of the old ones are included

and I can still use them but it's just a newer version sort of

[ornate elk]

Its not one file

These are multiple

[main nest]

yea

cool

[ornate elk]

It all depends on what's your image model

If its sdxl based you also need the sdxl controlnet models

If its 1.5 based you need the other ones

[main nest]

Hm I got it

thanks for clearing out the things

[main nest]

hmm

that's what happens when I try controlnet openpose, reactorr with a face and text2img

my guess it's from the controlnet for some reason, cuz when I turn it off but keep the reactor on its generating

those are my options for the controlnet

[ornate elk]

Mat error means the controlnet model is not for the same version as the image model

Sdxl models need sdxl controlnet, and 1.5 models need the controlnet11 models

[main nest]

ah okay

[knotty turret]

@ornate elk

there anything wrong here?

[ornate elk]

Yes cfg needs to be a fixed number like 1 or 2 or 3 and not 2.99

[knotty turret]

already tried with 3 same error

hold on let me restart

[ornate elk]

Is it possible to just enter 3 ?

And not 3.00

[finite zenith]

How do you edit launch arguments in the new comfyui application or exe version

[finite zenith]

This comfyui desktop is so frustrating to use

"This action is not allowed with this security level configuration."

wtf

[lucid yoke]

Looks like a windows thing rather

What are you trying to do?

[finite zenith]

I'm trying to install sageattention cos for some reason WanVideoModelLoader requires it "Can't import SageAttention: No module named 'sageattention'"

Then I'm gonna install sageattention via github link or repository and it doesn't allow me

"This action is not allowed with this security level configuration."

I am the admin of my windows pc

Then for whatever reason comfyUI installation has to break into parts, 1 in C:\ComfyUI, the others in Appdata or wherever it is

Through its shortcut I found it "C:\Users\AppData\Local\Programs@comfyorgcomfyui-electron

But it is in ComfyUI.exe, how do you edit it

[ornate elk]

If you get a security warning you have to go into ComfyUI\user\default\ComfyUI-Manager and there edit the config.ini.
Change the Security Level to Weak, then save and try install the stuff again.
After that set the Security Level back to normal.

You need Comfyui Manager installed before

[finite zenith]

There is no config.ini there

I think I have comfyui manager installed doesn't it comes with the installer

[ornate elk]

Oh you used the comfyui exe installer

Do you have a second comfyui folder not in appdata?

Or maybe its in resources/custom_nodes

[finite zenith]

The second folder is in C:\ComfyUI

[ornate elk]

Is there a custom nodes folder?

[finite zenith]

Yea but both C:\ComfyUI and ComfyUI\user\default\ComfyUI-Manager has custom_nodes

The C:\ComfyUI has most of the wan or the custom nodes I installed

[upper linden]

okay now what

same error

also if your down for a call

i would appreciate it

[tribal anvil]

How do I generate larger images?

[ornate elk]

Not today sry

[upper linden]

damn so what do i do

crashed after i did the edit

[knotty turret]

only that way

[azure magnet]

Hi, do you know someone that could help me to fix comfyui, even paid?

[ornate elk]

move the python pathst to the top

[upper linden]

okay i did that

[upper linden]

still gives the same error

OSError: [WinError 126] The specified module could not be found. Error loading "F:\SD-Zuldav2\SD-Zluda\stable-diffusion-webui-amdgpu\venv\lib\site-packages\torch\lib\nvrtc64_112_0.dll" or one of its dependencies.

[ornate elk]

yea we need to completely check the steps in discord sometime

something is broken

hard to guess

[upper linden]

damn

hopefully tmr

[upper linden]

i can be on around 12pm est

hopefully thats okay

[ornate elk]

idk if im on there but @ me then

[upper linden]

okay

appreciate all the help so far

so thank you

[ornate elk]

no problem,
i also had one case who had the same error and we couldnt fix it.
https://github.com/CS1o/Stable-Diffusion-Info/issues/14
But we didnt screenshared, so maybe we can find a fix.

Also make sure you have HIP SDK 6.2 and not 6.1 installed

[wet cedar]

Does Stable Diffusion handle Image to Video prompts, especially hyper realistic ones? If so, would someone be kind enough to walk me through how to set it up? I have stable diffusion ready?

[upper linden]

i have bothh

how do i uninstal one version?

[clear pawn]

could someone help me out with getting my lora to work in comfyui on runpod. I made it using fluxgym on runpod, downloaded it, its approx 79mb. When I upload it to the lora folder in jupyter lab and try to use it in comfy, I get an error code saying its not encoded in utf-8 or may be corrupt?? Not sure why this is the case as it just came fresh out of the output folder in fluxgym.

[ornate elk]

Right click and select uninstall for each 6.1 version

[magic hearth]

where can i post pictures of sand

[random night]

did you use amuse? how does it do compare to forge zluda?

[supple stag]

hello im trying to downloada this webui and im having alot of difficulties i did skip gpu check and when i tried to launch it said dll failed to load orjson failed to import something like that and when i didnt skip the gpu it says fail to load

i have windows 7

and im on laptop

[ornate elk]

I didnt used it, but it should be a bit faster than zluda as it uses onnx model conversion.
Which means that not every model is compatible. Every model and loras need to be converted to onnx.
Extensions won't work.
So its not a good alternative.
Maybe only if you want to try flux.

[ornate elk]

Hey what's your GPU?
And why Win7 🥲

[supple stag]

my gpu is something like intel hd4400 or smth like that

its pretty garbage

and win7 is because i couldnt seem to be able to upgrade the os lol

i use a 2004 laptop

[ornate elk]

Ah okay, what's in your webui-user.bat?

[supple stag]

you mean in the edit?

[ornate elk]

For CPU only mode you need:
--skip-torch-cuda-test --use-cpu all --precision full --no-half

[supple stag]

rn i just want the program to work

iv been at it like 5 hours

[ornate elk]

Remove --xformers

And add the ones from above

Then delete the venv folder

Then relaunch

[supple stag]

i insstalled manually the torch vision and tortch it failed to install when i didnt automaticly

il do what u say now

also what you mean add the ones above

sorry to bother bro im at it for hours

[ornate elk]

You need these

[ornate elk]

Torch+cuda won't work and xformers won't work as you dont have an nvidia GPU in that laptop

[supple stag]

need what

[ornate elk]

These args

Add them to the commandline_args=

[supple stag]

sry didnt notice

and delete vevn?

[ornate elk]

The venv folder yes

And then relaunch the webui-user.bat

[supple stag]

one second

[magic hearth]

No that's important if you delete it crashes and you need to reinstall

[supple stag]

but he said delete

[supple stag]

i can just unbin for recycler ig care to expand?

[magic hearth]

my bank account is empty

[supple stag]

its deleting like 8 gigs

[ornate elk]

Yes it needs to get recreated

[magic hearth]

eat them

[ornate elk]

@magic hearth stop trolling pls

[supple stag]

these support bots are scaring me bro

[magic hearth]

🧌 🧌 get trolled

🧌 🧌 🧌

oh no theres 3

whatever will we do

[supple stag]

its now downloading 2.5gig

[ornate elk]

Thats the torch version for CPU mode

[supple stag]

says itl take 15 min

[ornate elk]

Ok

[supple stag]

last time i downloaded this manually

cuz it took pretty long time and i thought its odd

[ornate elk]

The download? Its probably capped

At a low rate

[supple stag]

your right but im not talking about the download im talking about the installation

it showed me 0 seconds left for like 20 min

maybe i should ignore this warning

shouldnt

[ornate elk]

Thats a connection error but let it load

[supple stag]

ah its good then its connected

u got an idea what was the import orjson about

[ornate elk]

Nope

[supple stag]

i tried looking on the wed maybe someone had that issue i couldnt find this juts simillar stuff

web

wouldnt it have been faster if i manually installed this

cuz i have the file installed already

nvm i think i deleted it

[ornate elk]

nope

[supple stag]

is it supposed to be on the last second more then 10 seconds

its still on 1 second

but il assumes its installing

[ornate elk]

yep just wait

[supple stag]

i didnt have hot water so im back early

[supple stag]

what now

[ornate elk]

show your webui-user.bat

[supple stag]

ok

[ornate elk]

looks correct

[supple stag]

should i run again

[ornate elk]

and you launched the webui-user.bat and not the webui.bat right?

[supple stag]

yeah

[ornate elk]

relaunch it

[ornate elk]

hm strange

[supple stag]

maybe the skip torch i didnt put correctly

[ornate elk]

its correct

[supple stag]

it dosnt seem to recognise it

[ornate elk]

but i checked and it seems that Python 3.10 does not support Windows 7
But 3.10 is required for stable diffusion

[supple stag]

this is forkked i think

[ornate elk]

?

[supple stag]

i downloaded this python ver from a github that said this is 'forked' for windows 7

whatever this means

i can also use 3.8 offical no?

[ornate elk]

nope

3.10.6 is the lowest version needed

if your Laptop has 4gb RAM and a CPU with 2 or more Cores you should upgrade it to Windows10

[supple stag]

it has 3ram

lol

[ornate elk]

hmm no idea how good that would work

Win7 will run smoother on that than Win10 for sure. And Stable diffusion would be really slow anyway on that machine

so you may should checkout other alternatives

[quiet sand]

how do i use stable diffusion im so confused

[supple stag]

i dont mind it being pretty slow

[ornate elk]

you can install it localy if you have a decent gpu

[quiet sand]

i have uhm a GTX1650 can u help me sir at VC?

[supple stag]

il try to see if i can update then

what is tiny 10

should i download this

[ornate elk]

hey, checkout the install guides (first link of the pinned messages of this channel)
Go for Forge Webui or ReForge for Nvidia

[ornate elk]

Tiny10 is a smaller Version of Win10, its optimised for low end PCs

A lot of stuff is removed to make it compact

[supple stag]

will it work with stable

[quiet sand]

can you join me at diffuser lounge im new at this kind of stuff sorry

[ornate elk]

it should work with stable

but like i said it will be very slow

[supple stag]

so this or the regular windows 10

i was born slow dont worry

[quiet sand]

well that's a no i think thanks anyways im really interested into ai prompt but its too complicated for me

[ornate elk]

i can help you if its to complicated

but the instructions should be really simple

Forge is just a download

[upper linden]

done

[upper linden]

also im ready to stream if you want

[ornate elk]

okay

have you restarted the PC after uninstalling all 6.1 stuff?

[upper linden]

yes

[ornate elk]

and when you relaunch comfyui you still get the error?

[upper linden]

yes

also the same for automatic111

[ornate elk]

okay can you screenshare ?

[upper linden]

sure

do i call

[brazen sun]

sorry to ask again but even activating tiled vae leave me with being at 93-95% for longer than any other part, is this normal?

[brazen sun]

any way to generally speed it up? using amd rx 6600 and a1111 with zluda

[random night]

when do you update the guide btw

[ornate elk]

its up!

finished it a few mins ago and retested

[random night]

thanks man

i'll test it right away

[ornate elk]

let me know how it goes

[knotty turret]

@ornate elk how do i set "Attention optimization method" to "Dynamic Attention BMM"

cant find this config

[ornate elk]

where?

[knotty turret]

stable diffusion

see someone said that makes zluda works faster

idk never see this before not 100$ sure

" After the UI starts, head on over to the System Tab (Standard UI) or the Settings Tab (Modern UI), then the Compute Settings category.
Set "Attention optimization method" to "Dynamic Attention BMM", then click Apply settings "

cant find this Compute Settings

[random night]

it works

though i didn't try any extensions to see if the one works in forge can work in reforge or not

[magic hearth]

it doesnt i tested

[random night]

hmm, can we see your cmd log?

[magic hearth]

i ate it i didnt have lunch i was hungry im so sorry please forgive me

[random night]

...um, no need to say sorry, just take your time

[magic hearth]

tastes like mango tho quite nice

[random night]

now i wonder what you ate

[magic hearth]

cmd log all of it

[random night]

...ok, if you're some super high end AI or virus, i might believe your story

[magic hearth]

no like physically i just printed it made a sandwich together with the computer that shit rocks

[random night]

i have no idea what you're talking about anymore

[brazen sun]

also is this normal?

[magic hearth]

nothing is normal everything is dead everything is gone

its the bees

THE BEEES

[brazen sun]

ah, I see, thank you

[random night]

i think he's high as fuck right now

[brazen sun]

let him BEE high

[magic hearth]

high on fax machines

cuz im real

theres probably a better word

[ornate elk]

adetailer, booru tag and sd-forge-couple worked

[random night]

any luck with inpaint anything?

[ornate elk]

havent tried

[ornate elk]

thats for sdnext i think

not for auto

[brazen sun]

trying amdgpu forge because i heard its faster but I am getting this error, any ideas?

[ornate elk]

the dev broke it 8 hours ago, it needs an update to be fixed

[brazen sun]

ª, so not a me thing? good, thanks! saved me a while of having to go around looking for a fix

[ornate elk]

yep, if you delete it and try tomorrow evening it should work again, the dev fixes bugs quickly

[brazen sun]

great, thank you! also, is it actually faster than regular a1111? curious

[ornate elk]

could be a bit faster

[brazen sun]

oh okay, thanks!

[brazen sun]

and sorry to ask this again but is it normal that the last step takes longer than the rest of them (finally checked properly the console and its a step thing, like it takes a lot longer to go from step 19 to 20 that from 1 to 19)

from what I have worked with computers it makes sense but I still I rather ask

nvm just reread this, sorry

[ornate elk]

np

[ornate elk]

At which resolution did you generated?

[brazen sun]

512

1:1

[ornate elk]

Ahh okay with a 1.5 bases model?

[brazen sun]

yup

[ornate elk]

Ah okay. With 1.5 models the ui is bugged as it processes faster than it shows

And then it stops

[brazen sun]

ohh okay

then again, not a problem per se, just taking its normal time

[ornate elk]

Yep

If you try a sdxl model it should proceed normaly and then take a bit longer at the last step

[brazen sun]

okay okay, thanks!

[ornate elk]

NP, if you try a 768x1024 with an sdxl based model with 30 steps, Euler a sampler.
Let me know how long it takes

[brazen sun]

ill test it later and report to ya, although from previous tests, it does take quite a longer while on sdxl

[ornate elk]

Yea sdxl is heavier and takes a bit longer

[azure magnet]

@ornate elk could you help me with this ?

[ornate elk]

Looks okay

[lucid yoke]

yeah it looks installed

[azure magnet]

what? but say "failed"?

[lucid yoke]

thats trying to update the comfyUI it self

not the manager

manager is fine

do you happen to be on AMD?

[azure magnet]

No, Nvidia

[lucid yoke]

hmm no idea why the comfy update failed since we see no logs of that but Manager seems to be working fine

[azure magnet]

Is the last one version i think, but how i can see if it is true?

[lucid yoke]

i couldent tell you sorry

[azure magnet]

Do you know someone can help me?

[lucid yoke]

No, maybe someone will see it later and help ya

[thin whale]

hello, please, I need your help, I already have some knowledge of stable diffusion and sites where you can work with AI, I also made 100 lora models from those sites, but I would like to create more detailed LORA or FLUX models. could someone help me? I also installed the KOHYA program for creating lora models, but it doesn't seem to work, and the tutorials from YouTube didn't help me much either.

[ornate elk]

The update failed. Maybe its already uptodatw

[azure magnet]

Yes, i downloaded the last version, but always said "comfy already upgrade"

[ornate elk]

Okay and does it opens in the browser?

[random night]

hmm, seem like reforge controlnet multi-input is suck just like one in forge

[neon cedar]

I don't have C:\Program Files\AMD\ROCm\6.2\bin\rocblas folder, do I need to create it or something?

[random night]

have you install hip sdk yet?

[neon cedar]

oh lol, i have to read more carefully, thanks

[ornate elk]

Let me know if you have any questions about the setup

[neon cedar]

I had to add "--no-half" to the webui-user.bat file (got it from googling the error) because i got an error when trying to create an image, it's working now

is that normal? why did i get that error

[ornate elk]

What's your GPU?

[neon cedar]

amd rx 6650 xt

[neon cedar]

I keep getting "Error: The 'insightface==0.7.3' distribution was not found and is required by the application" when trying to install reactor and it's not showing up in extension panel

[short rose]

i literally have flash attention installed

[ornate elk]

That doesnt need --no-half

Remove it and make sure your using a model and not a lora as checkpoint.
--no-half slows down the generation speed

[ornate elk]

Thats not how you install insightface.
Delete the insightface folder from the extensions folder

Then read this:
https://github.com/Gourieff/sd-webui-reactor-sfw?tab=readme-ov-file#installation

[ornate elk]

What's your GPU? And which webui do you use?

[short rose]

its stable audio and 3070 ti

[ornate elk]

And which tool do you use for stable audio?

[short rose]

what like toolkit?

sorry im new to this stuff

[neon cedar]

i did the "If you use AUTOMATIC1111 SD WebUI or SD WebUI Forge:" steps, should i do another?

[ornate elk]

No thats fine. But if you want to use the reactor extension you need to follow the guide I linked. It won't work without installing insightface correctly

[ornate elk]

For stable audio you should use ComfyUI, it can handle stable audio and is easier to setup

[ornate elk]

Can you remove --no-half and test again?
Also what size does your models file have?

This is #🤝｜tech-support

[umbral relic]

Oh yes, sorry...

[supple stag]

can i use google collab or smth?

my laptop cant install windows 10 the setup says

[neon cedar]

I'm not sure what I'm doing wrong, i'm pretty sure I did these first steps , also removed --no-half, these steps don't even mention insightface, the model i'm using is 2gb storage space.. idk if this is what you meant

[high flame]

Is there anyone Expert at Stable diffusion who could help me with the perfect setup i need.

i think this shall give me better results consistency then Leonardo

[zenith sail]

anyone have any idea why comfy is still missing this node?
https://github.com/ZHO-ZHO-ZHO/ComfyUI-BiRefNet-ZHO

Even after manually installing.

[obsidian lily]

anyone know what my issue is?

[ornate elk]

Okay sounds good

[ornate elk]

Hey, you downloaded the webui as zip and didnt followed the guide from the pinned messages here.
If you do, it most likely will work.

[neon cedar]

but reactor still doesn't show up, and i get errors when trying to generate without no-half

[ornate elk]

can you screenshot your webui when you get the error?

and the cmd log

[dense spoke]

On pinokio on my Mac it says "ModuleNotFound Error: No module named 'mmgp'" on wan2.1

[ornate elk]

its using cpu mode

open up a cmd inside the stable-diffusion-webui-amdgpu folder and then run git pull
there was a bug yesterday that broke everything

then delete the venv folder and relaunch

[warped canyon]

anyone knows why my img2img sd upscale is making images like this?

[ornate elk]

which webui?, which model and vae?

[warped canyon]

reforge, noobaiXLNAIXL_vpred10, aaaanimesdxlvae_v2

[ornate elk]

denois and sampler? sd upscale script or ultimate upscale or resize ?

[warped canyon]

0.15, euler,

[warped canyon]

sd upscale script

it always worked but last few days has been like this

[ornate elk]

hmmm strange

does normal sdxl models work?

or non vpred models?

[warped canyon]

lemme test

yeah those work

[ornate elk]

so its vpred

maybe try without a vae

[warped canyon]

already did tho

i think that this is not affecting img2img

cause i change to eps and get the same image

[ornate elk]

and you have applied and reloaded between eps and vpred?

[warped canyon]

ah i tried sending an eps pic to img2img and use vpred settings there, and it gets noise, so i guess its not that either...

[ornate elk]

okay, then no clue, vpred is just messy, i also dont understand the hype of it xD

i get much better results without it

[warped canyon]

found the fix

this wasnt on 0

so i changed back to 0

[ornate elk]

ahh

nice

[neon cedar]

I did the steps, already had git pull, then deleted venv, it gave me error when i tried to boot, tried again and it does boot the webui but generating doesn't work, it doesn't even give me error like before, and this is my log

[ornate elk]

looks perfect

it compiles zluda

that will take around 15-30 mins for the first time

after that it will use your gpu to generate images in seconds

[neon cedar]

ok so i just leave the cmd open, what should it say once it's done?

[ornate elk]

yep let it load until its done

if you already pressed on generate you will get an image when its done

[austere pine]

I'm trying to use this model: https://civitai.com/models/811067?modelVersionId=907264. Using the recommended settings for CFG, Sampler and steps

But images are coming out all purple and weird like this

Any clue what I'm doing wrong?

They load normally in the preview but when the final image comes out it's like that

[ornate elk]

which webui do you use?

the model is mixed with vpred so the webui needs to support it.
also someone posted this about it:

[austere pine]

Currently using SwarmUI 0.9.3.1

[azure magnet]

Hi, how can remove all thing that comfyui need to works?

[ornate elk]

Try set the resolution to 832x1216

[neon cedar]

it's generating images now, although there's one bar that has been stuck at 55%

[ornate elk]

Thats okay its the same bar just older

When compilation was in progress

[neon cedar]

I keep getting this error when trying to generate

[ornate elk]

What's in your webui-user.bat?

AMD does not work with --xformers

Remove it from the webui-user.bat and then delete the venv folder and relaunch.

[neon cedar]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--use-zluda --update-check --skip-ort

call webui.bat

i deleted venv like 3 times already, I will try again. It started giving me that error after installing 2 extensions, i deleted them from the folder but it still gives me xformers error

[neon cedar]

deleted venv and launched webui-user.bat and gave me this, it closes when i press a key

[ornate elk]

Relaunch

That error appears sometimes after deleting the venv but shouldn't happen again when relaunching

[flint drift]

hello, can i ask do you have any idea on how to resolve this when i'm generating?

URLError: <urlopen error [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: certificate has expired (_ssl.c:997)>

this is the error message that i'm getting

[ornate elk]

Hey, where and when do you get that?

[flint drift]

i just re-installed stable diffusion the other day and i'm getting this error when generating

[ornate elk]

Can you show me a full cmd log ?

Also make sure your PCs Date and Time are correct.
And if your from Asia you may need to use a VPN for the Installation and change the DNS to install extensions

[ornate elk]

The complete log pls

[ornate elk]

Okay seems like a connection issue

Restart the PC.
Delete the venv folder and then relaunch the webui-user.bat

[flint drift]

okay i'll try

[flint drift]

restarted and it finished reinstalling those files but it's still getting the same error

[ornate elk]

Make sure the webui is whitelisted in any browser adblocker and dont use opera vpn

[flint drift]

yeah i made sure i have no adblocker and i'm not really using vpn now

still the same error

[ornate elk]

Okay try updating your Python to 3.10.11 64bit and then open up a cmd and run
Pip cache purge
Then delete the venv folder and relaunch the webui-user.bat

[flint drift]

tried it still same error sadly

[ornate elk]

Hmm, are you from Asia, Russia or Iran?

Then you would need to use a VPN for the install.

Also are you connect trough WiFi or Ethernet?

Maybe also try change your DNS to 1.1.1.1 or 9.9.9.9

[ruby egret]

Completed all steps for 9070xt per the guide: https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides#amd-automatic1111-with-zluda, got error as shown in picture. gfx 1201 (9070xt) not present in tensile library within AMD ROCm 6.2.4 (shows as 6.2 in file explorer) What's next?

win11, using AMD Adrenalin drivers, not PRO. Up to date.

[ornate elk]

hey, the log shows that its using your AMD CPUs Igpu instead of your 9070xt
Disable that in the Device Manager

Readeon TM Graphics

then relaunch

[ruby egret]

can i use --use gpu 1 or whichever?

forget the specific command

[ornate elk]

you can try --device-id 1

but it may not work

[ruby egret]

that's the one

will try

disabled igpu, restarted OS, tried --device-id 1, got same error.

[ornate elk]

you also need the gfx files for your gpu

[ruby egret]

agreed

[ornate elk]

have you replaced them already?

[ruby egret]

no

[ornate elk]

also python 3.10.11 64bit is recommended

[ornate elk]

then do that first, i will need to edit the guide to specify that step for 9070
also you have to restart the PC after that

[ruby egret]

https://github.com/ROCm/ROCm/issues/4485

not relevant

where is gfx1201 rocm files?

[ornate elk]

https://github.com/likelovewant/ROCmLibs-for-gfx1103-AMD780M-APU/releases/tag/v0.6.2.4

You need the gfx1201 ones

The guide shows what how to proceed with the files

[brazen sun]

been trying amdgpu forge and its quite slower than regular a1111, could there be a reason for this?

[left junco]

Anyone use Forge? I could swear i used to have a directory list at the right of the Lora tab. Was that removed? Was it a plugin? was it SD-WebUI? Anyone know

[brazen sun]

like quite a bit slower

[ornate elk]

The button is hidden on the right top of the lora tab

[lucid yoke]

Did you use zluda?

[brazen sun]

this right?

[ornate elk]

Can you show the cmd log?

[ruby egret]

Got em. I'm not sure what's goin on here, i'm just letting it run

[left junco]

@ornate elk No that's just for the 'buttons' in 3. At one point i had a dir list at the left side.
So 1 is the button you mean. I know that one. I mean 2. I had that at one point as a dir-list.

(This is a mockup obviously 🙂 )

[ornate elk]

Yep looks good, let it run

[ornate elk]

That dir list got removed by an update

[left junco]

Pity

Because with nested dirs the top gets a littlemessy 😛

[ruby egret]

it's been at it like an hour

i have no models in there lol, what model is it compiling

[left junco]

What are you doing? Getting it working on ROCM?

oh nvm AMD 9070XT

Does that work? Coz NVidia cards too expensive for me now:/

[ruby egret]

it seems to be

loading my first model right now

[left junco]

Setting up controlnet.

[ruby egret]

functional yay

thanks, cs1o

[left junco]

Nice, and you have oodles of vram i saw. (I only have 11 so 16 is oodles to me 🙂 )

All preprocessors installed. Now the hard part 😦 Gonna need to find all the controlnet Models and put them in the right folders

[ruby egret]

good lord, 3.4ghz clocks with inpaint

[brazen sun]

I really cant read this, have diferent settings between normal a1111 and forge?

[left junco]

forge and reforge don't do controlnet for flux?
I have to use comfy i guessing?

Are you sure the refiner is for the model type you are using?

[lucid yoke]

you have the flux specific control nets?

[left junco]

I do. is pent an hour getting them all.

[lucid yoke]

hmm im not too sure about forge/reforge

[left junco]

a1111?

Probably not as forge and reforge are forks

[lucid yoke]

A1111 is outdated compared to forge

[left junco]

Yeah i stopped using a1111. respect to creator though. He was at the forefront

[lucid yoke]

yeah it was peak but it stopped recieving updates

but yeah comfy/SwarmUI (comfy backend) are imo the future as it supports most models but if you use AMD idk how well its gonna work

[left junco]

that's the good thing about Comfy. the module approach allows people to make a puzzle piece and give it to the community.

Comfy is too cluttered. It need better modularization in the GUI.

[knotty turret]

Anyone here know the ~max settings i can use on wan 2.1 with L4 GPU?

[left junco]

Or maybe, that's just what it is.

It's a combination of framesize and number of frames insnt it. And on what model you use 14b or 1.3b iirc

But no, i do not.

[knotty turret]

you talking to me?

[left junco]

Yes. The real answer AI would give is; it depends

[lucid yoke]

thats where SwarmUI comes in. id give it a go

[left junco]

ok. I have comfy running anyway. I assume it calls the comfy api?

[knotty turret]

89 length its good number or too high?

[lucid yoke]

you can mount it as a backend yeah. https://github.com/mcmonkeyprojects/SwarmUI/blob/master/src/BuiltinExtensions/ComfyUIBackend/README.md

[left junco]

If it's too high it will just OOM. Then you lower it. That's how i do it locally. But it takes forever so i sorta stopped video untill i have a 5090. (HAH noever)

[knotty turret]

i just wanna cool setting tbh

dont need to be perfect

[left junco]

McMOnkey my brother from another monkey mother

[knotty turret]

too low?

[left junco]

always 16!

WAN is 16fps

[knotty turret]

Oh ok

so ig 33 length

[left junco]

how mwuch VRAM?

what is width and height?

[knotty turret]

24gb

[lucid yoke]

also: Wan 480p model or 720p?

[knotty turret]

My image is 1080 x 1919

[lucid yoke]

nvm 720p

[knotty turret]

i need resize it?

[left junco]

yeah image size don't matter. Just make sure video size is same aspect ratio

[lucid yoke]

this yeah unless you told it to use initial image size

[knotty turret]

idk how to do it

🙃

wanna see my workflow setting?

[fair oxide]

i am on linux, just download a1111 (nvidia), what things should i do before launching the program, the webui user.sh is more messed than webui user.bat

[knotty turret]

wait so i need resize it?

got confuse

[left junco]

Start image doesn't matter. The video output size is wht matters.

[knotty turret]

you said yeah but then dont matter

[knotty turret]

oh ok

mb

[left junco]

@fair oxide stop and install forge wich is more up to date.

[knotty turret]

so what number should i put here?

to get max possible from image

[left junco]

@knotty turret can you just show the workflow as more large area

[fair oxide]

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

[left junco]

Yes. @fair oxide

The is alsoo re-forge. Though i am nit sure wich is better. I use Forge, it's easy to use compared to ComfyUI

[fair oxide]

one more thing, in linux for the ai i downloaded python3.10 using the tar.gz source, is it alright?

[knotty turret]

@left junco need anything else?

[knotty turret]

Ok

[left junco]

1 is output video width, 2 is output video heieght, 3 i number of frames

[knotty turret]

yeah alr know it but what number is the ideal

[left junco]

you can probably do 120 frames with that

[knotty turret]

hmm ok

[left junco]

Maybe 150. Trial and error. Set it to a value. If it gives OOM error lower it. If not. Increase it untill it does and then you know.

jealous of all that VRAM 🙂

[knotty turret]

ok

what about res?

[left junco]

well, if you increase resiolution you will be able to do less frames.

[knotty turret]

i think its better i look for small images so i can get everthing on it

[left junco]

lower resolutions it can do more frames.

[knotty turret]

dont want get just part of it

[left junco]

Just make your image same dimension

[knotty turret]

or can just take my 1080x1919 image resize to 720x1280

and work with it?

[knotty turret]

you mean like 512x512 768x768?

[left junco]

No, source image same dimension as target video.

[knotty turret]

wait so i can just put 1080x1919?

[left junco]

Put? The source image size doesn't matter, it's about the aspect ratio being the same.

No

[knotty turret]

srry dont speak english too well

i dont think getting what u trying to say lol

[left junco]

You have a start image, yes?

[knotty turret]

Yes (the image i want to make the img2vid)

[left junco]

OKay so the target will be So (1) 1280p and (2) 720p

[knotty turret]

ok will make test

[knotty turret]

always?

1280 x 720?

[left junco]

Depends on your source image but i read WAN prefers it when you use the models height. so 480 or 720 depending on what model you used. You use 720 i think you said

[knotty turret]

because like red(1080x1919) and white(1280x720)

thats why i confuse

looks like gonna miss alot from image

[left junco]

oh your source is portrait?

[knotty turret]

that one yeah

[left junco]

Oh, yes then width is 720 and height 1280 i guess.

[knotty turret]

😭 i dont think portrait its good idea

lol

[left junco]

It doesn't clip it it resizes the source image to the size of the video.

[knotty turret]

ohhh

i dumb

so ig portrait 720x1280

not portrait inverse

😂

[left junco]

Just do it with 4 frames. You will see what it does

[knotty turret]

Ok

[left junco]

SCAM

[knotty turret]

yeah

[ornate elk]

No problem 🙂

[knotty turret]

CS1o

my goat

coding kitten already helped me but

just to confirm

the 720p wan model the max u can do is 720x1280 and 1280x720?

[ornate elk]

Its trained for that yes

[knotty turret]

i using gpu with 24vram now

[ornate elk]

The smaller one is for 480p

[knotty turret]

ok

[knotty turret]

and what about frames

good number?

@left junco forgot to say thx mb

you helped alot

[left junco]

np, good luck making movies

[knotty turret]

it worked fine with 720x1280 for portrait

[wanton wagon]

Is there a darkmode cosmetic version for Linux Automatic1111?

all of the looks in the app are just recolors of the bright version

[left junco]

@wanton wagon Dnno, you could try using a browser plugin.

@lucid yoke It's making 1st image 🙂

[left junco]

VAE mismatch?

@lucid dawn Ask this in Grok maybe:
I'm trying to use a refiner, but it's complaining of mismatched dimensions ?
"shape '[2, -1, 12, 64]' is invalid for input of size 20480"
I'm basically doing
ksampler(model) --latent--> ksampler(faceic_model(refiner)))

I never use SDXL. But you can't use 1.5 VAE on SDXL iirc.

(I never liked it because of the refiner BS. Go figure :P)

lol i think preprocessor and model were not properly selected

That's a police woman 🙂

[ornate elk]

Add --cuda-stream to the launch args and try again

Forge should be much faster with that

[left junco]

controlnet doing nothing it seems. images take forever in flux :/ annoying testing new stuff

[ornate elk]

That means something isnt compatible