🤝｜tech-support

[ruby knot]

I saw one for a1111 but I don’t know if it’s any good or if my pc specs are good enough

[snow breach]

@ornate elk have you got a good tut for using SD3.5 in comfy? I've got it running, but I need a simple workflow that uses lora for 3.5 and need to know which model/encoder to use for my 4080 super

[balmy monolith]

the huggingface model page has a link to sd3.5 examples for comfy

https://comfyanonymous.github.io/ComfyUI_examples/sd3/

just save picture to you pc and drag it into comfyui, the workflow is included in the picture itself

the models it uses in each node will be shown, you can then find and download each of the models it uses.

[ornate elk]

yep try the comfyui example link from above

but for loras i dont have a workflow with 3.5 rn

[balmy monolith]

can you just insert the load lora node after the load model, like in sd1.5 & sdxl?

i'm just guessing, i haven't used sd3.5

using Fluxgym, can i continue training a lora from where i left off from a different session?

is it the** --resume** option in advanced settings perhaps?

seems the option to resume is rather complex

[snow breach]

Fresh install of comfy:

when attempting to gen sd 3.5

[ornate elk]

Which one?

[snow breach]

Alright, I re-installed comfy again this time with the portable windows package and I can generate sd3 images... for the moment.

[fair oxide]

what base model is automatic1111?

i have it working but can’t get models to work for some reason.

[onyx solar]

Your question is flawed. A1111 is not a "model". It's software that uses models/checkpoints to infer.

[ruby knot]

I forget its name as it was a while ago

[ruby knot]

I tried install a bunch of extensions on a1111 and it’s been stuck for ages without opening a1111 gui?

That’s as far as it gets even if I close it and retry.

[noble idol]

what do i do if my wildcards just flat out arent working

ive been using them forever never had any issues until today

i tried reinstalling it but it still doesnt work

ive looked it up, but all ive found is people with the same issue and no answers

[ruby yoke]

you go back to an earlier version of comfy

[noble idol]

i dont use comfy

[noble idol]

figured it out

[lament stream]

I have trained some Loras with Fluxgym standart settings.
The samples looks very good so far, but when I use it in webui of SD,
I cant get even close to the trained Lora. I tryed with different
models and different settings in schedule type, sampling
method and percent of Lora. How Fluxgym generate this samples?

[ornate elk]

You installed dreambooth. That breaks the webui.
Delete that extension and then relaunch

[lament stream]

no, dreambooth isnt installed

[ornate elk]

The message wasn't for you

[loud pond]

hello, i am going to upgrade my graphics card from an amd to a navidia, would i have to change my stable diffusion type?

[ornate elk]

Hey, yes, you can follow the nvidia guides from my install guides

Its the first link of the pinned messages

[loud pond]

so i would have to re-install it?

currently i have the one for amd

[loud pond]

i also have a bunch of wildcards, checkpoints, lora's can i just more them over as well?

[ornate elk]

Yep reinstall is needed

[ornate elk]

Yep

[loud pond]

im going from an amd radeon 480 to nvidia 1080, do you know what things i will be able to do now? will i be able to generate bigger images?

from my understanding i will be able to do a lot more with 8gb of vram on nvidia

right?

[ornate elk]

Thats a good card. You can then use sdxl and pony models easily and also upscale images to larger resolutions

And yea native higher resolution is possible

I had a 1080 until last year. Even Cyberpunk 2077 runs on 60fps with medium to high settings

With that card you better use Forge Webui or Auto1111

Forge is a bit faster so I would try that first

[loud pond]

alright! thanks a lot, will i also be able to get comfy ui?

[ornate elk]

yes

[dapper timber]

NotImplementedError: Cannot copy out of meta tensor; no data!

am getting this error

[calm moth]

I'm trying to use Ultimate SD Upscale in ReForge. It keeps drawing the complete prompt in every single tile. How do I fix this?

[ornate elk]

Reduce the denois to 0.3

[manic aurora]

hi guys, anyone faced this problem? i got the models but still cant make it run

[charred rover]

I m sorry I want to double check, do you think he s right?

[ornate elk]

Nope, just restart the PC.
And then get CrystalDiskInfo to check your Drives

[charred rover]

okay thanks! it also cant find Numpy

[ornate elk]

To fix system errors open up a cmd as admin and run
SFC /scannow

[charred rover]

thanks

[ornate elk]

Delete the venv folder and relaunch the webui-user.bat.
But yea after you did the other steps

[charred rover]

it says Disk C is 67% healthy..

[ornate elk]

is it a hdd or ssd? m2 or sata?

[charred rover]

SSD NVM Express 1.2

oh sorry

i m blind it s disk E SSD NVM Express 1.2

[ornate elk]

is it an older ssd?

[charred rover]

yes its 960

[ornate elk]

ah okay, than its normal

if nothing is red on the drives your fine

[charred rover]

okay thank you again

[ornate elk]

no problem

[crystal nacelle]

I’ve been using automatic1111 , SD1.5 with zluda for months and loving it. I want to move to Forge UI, Flux and still keep fast, AMD optimised ZLUDA structure. Can anyone help, is this even possible?

[ornate elk]

Hey you can just follow my Forge with ZLUDA guide.
Its in the link of the pinned messages

[crystal nacelle]

Thank you. Do I then need to instal FLUX separately? This quick list-flux tutorial seems to do that. But it has 2 different tutorials, not too sure which to use https://github.com/lshqqytiger/stable-diffusion-webui-amdgpu-forge?tab=readme-ov-file

[crystal nacelle]

am getting a runtime error when trying to run flux, can anyone help plz. heresa screenshot

[tawny stag]

Hey all, I'm really close to finishing this workflow, but I'm running up against some limitations of Comfy that I don't know how to solve.

The basic idea is this: I want to set the workflow to constantly run in the background and use recorded audio of my D&D game as the input. The transcript is picked up using whisper, then it's fed into a LLM to create a new image every ten minutes or so, which is placed in a folder that has a slideshow running for my players and I to see as we play.

Each individual step works great. I'm able to use the audio recorder in web tools to record the audio on a loop with a set duration and interval between recordings. When that audio is fed into whisper, it picks it up and the LLM summarizes it into an image prompt. Go team. The image generation works fine beyond that point.

The problem is with the scheduling of the prompt / queue itself. I can set comfy to run multiple queue's, and I've been experimenting with Jovimetrix delay nodes to try to space out the prompts, but it seems like the audio->whisper connection wont work correctly unless the new job is scheduled whenever a new audio loop begins, which may be a limitation of the audio recording node.

Any ideas for how I might solve the problem, or any other ways you suggest I might approach it? Thanks so much for your time 😃

[ornate elk]

Nf4 won't work with AMD on windows.
You need to use gguf models

Like flux Q4 gguf

[tawny stag]

Nevermind, I ended up just editing the audio recorder to add a new widget and optionally kick-off a new run at the end of each recording. That seemed to solve my problem.

[fossil sentinel]

ello, was wondering if anyone could help me get caught up with some AI stuff, heard stable diffusion 2 or something was coming out and i was using forge and not sure how out of date mine is, so was just wondering if i could get some help with setting it back up again
and was wondering if anyone could help me train an AI for a character

[uncut jewel]

Did anyone had this issue recently on a simple flux ksampler on Comfyui ?

Did saw a similar error on xflux Githube page, but nobody resolved it yet

[crystal nacelle]

okay awesome. downloaded it and will try now. ive been runnig these commandlines since day 1 SD1.5, do i still need them?
"set COMMANDLINE_ARGS= --use-zluda --medvram-sdxl --update-check --skip-ort "

[ornate elk]

you need the ones from my forge guide

not these (these are for auto1111)

Also whats your gpu?

[crystal nacelle]

RX7600 babyyyy

[crystal nacelle]

im seeing only 1 command line on that guide:

--use-zluda

[ornate elk]

nice one

forge should already set the right stuff automaticly

so only zluda arg is needed

[crystal nacelle]

okay, thank you CS1o.

zluda arg?

[ornate elk]

--use-zluda

[crystal nacelle]

awesome. it worked. first render super slow. but the second render is also mad slow, the cmd window is telling me that it doesnt have enough vram

how has my 8gb gpu only got 1GB free...what can i do to better optimize this? maybe try a different model?

[river narwhal]

I am trying to train a LORA in kohya_ss in SDXL and when i use the SDXL in pretrained model it gives me this error but if i have the exact same parameters but change it to sd1.5 instead as pretrained model it works wonderful. Why is this?

[ornate elk]

which model do you use?

do you have the 7600xt with 8 or 16gb vram?

[crystal nacelle]

i googled the one you suggested and downloaded the first you i found lol. i found another article that gave some more advice '

anyway, this is my model and setup:

flux1-dev-q4

[ornate elk]

dont use the t5 fp16 encoder, better use the t5 q8 encoder:
https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/tree/main

and then try a resolution of 1024x1024

[crystal nacelle]

CUDA out of memory :(. i swear its like this everytime i try instal some new StableDs. gotta set aside a whole day to troubleshoot

[terse siren]

Ok so now controlnet for me is completley broken for no reason now

[crystal nacelle]

it keeps telling its tryna set aside like 1.5GB of but im rocking 8GB gpu and 16GB system ram, whys it crying?

[terse siren]

every time I try to generate ANYTHING with Controlnet I keep getting "press any key to continue"

[crystal nacelle]

which ui you using? send screenshots of your UI and the CMD things for better help

[terse siren]

Im using forge

[Unload] Trying to free 7974.00 MB for cuda:0 with 1 models keep loaded ... Current free memory is 7054.35 MB ... Done.
[Unload] Trying to free 7974.00 MB for cuda:0 with 1 models keep loaded ... Current free memory is 7053.13 MB ... Done.
[Unload] Trying to free 7974.00 MB for cuda:0 with 1 models keep loaded ... Current free memory is 7051.32 MB ... Done.
2024-12-23 09:23:46,907 - ControlNet - INFO - ControlNet Method canny patched.
[Unload] Trying to free 17442.12 MB for cuda:0 with 0 models keep loaded ... Current free memory is 7050.04 MB ... Unload model JointTextEncoder Done.
[Memory Management] Target: KModel, Free GPU: 7050.04 MB, Model Require: 4897.05 MB, Previously Loaded: 0.00 MB, Inference Require: 7974.00 MB, Remaining: -5821.01 MB, CPU Swap Loaded (blocked method): 4897.05 MB, GPU Loaded: 0.00 MB
[Memory Management] Target: ControlNet, Free GPU: 7050.04 MB, Model Require: 2386.12 MB, Previously Loaded: 0.00 MB, Inference Require: 7974.00 MB, Remaining: -3310.08 MB, CPU Swap Loaded (blocked method): 2386.12 MB, GPU Loaded: 0.00 MB
Moving model(s) has taken 8.23 seconds
Press any key to continue . . .

And this is whats in the command box every time it happens

no error message or indication anything went wrong

generating a regular pic with no control net works fine

[ornate elk]

Idk if flux is able to work on 8gb with zluda

But 16gb ram is to less. You have to increase the Windows Pagefile

[ornate elk]

whats your gpu?

and do you tr to use sdxl or 1.5 controlnet?

[terse siren]

sdxl

[terse siren]

NVidia GeForce RTX 2080 super

[terse hedge]

Why do I only get results with so much noise when I use a v-prediction model (NoobAI v-prd-1.0)?

Using AMD webui combined with zluda, following all instructions outlined in the civitai page (1024x1024, 4.5 cfg, euler, etc)

all other models, as well as noobai's EPS version, work fine

[ornate elk]

Can you share your txt2img settings?

[terse hedge]

I'm starting to think this webui version doesn't support v-prediction at all? I'd rather not switch to forge

[ornate elk]

Illust models work on a1111 and forge

Which VAE did you used?

[terse hedge]

How can I check?

[ornate elk]

For 8gb vram you need to use --medvram-sdxl in the webui-user.bat

[ornate elk]

Go into settings, user interface, Quicksettings, there add sd_vae
Then hit apply and reload ui

Then you get a dropdown menu

[terse hedge]

As I've seen, Forge has a "advanced model sampling" feature in settings which allows switching eps and v-pred at will, I don't see anything resembling that in my app

[ornate elk]

I guess Auto1111 doesn't have that

[ornate elk]

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

[terse hedge]

The bottom file in there? And do I just put it in the VAE folder in models?

[ornate elk]

yep

[terse hedge]

It doesn't seem to work

Same settings as before

[ornate elk]

okay, then just dont use the v predict model

maybe it works in forge

[terse hedge]

If that's the only option I'll switch to forge

I don't want to miss out on v-pred, it's much better in terms of quality and performance

[ornate elk]

did oyu used this model?
https://civitai.com/models/835655/illustrious-xl-personal-merge-noob-v-pred05-test-merge-updated

[terse hedge]

This one
https://civitai.com/models/833294/noobai-xl-nai-xl

[ornate elk]

ah okay, that model states which webuis to use and how

in the description

[terse hedge]

Yes, it explicitly states it supports A1111

[ornate elk]

nope only on the dev branch

[terse hedge]

Yes that

I'm using the amd one

With zluda

[ornate elk]

in dev branch mode?

[terse hedge]

So I'm guessing this could be it

No

[ornate elk]

yea thats the issue then

i would go for forge webui with zluda then

[terse hedge]

No guarantee it would work there though

Also I need extensions like controlnet

[ornate elk]

controlnet is integrated in forge

[terse hedge]

I'll go on and install it then, this one? https://github.com/lshqqytiger/stable-diffusion-webui-amdgpu-forge

[ornate elk]

yep, here is my install Guide:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[terse hedge]

Oh that's yours!!

It's incredible, I used it so many times

Thank you so much

I saw it somewhere on reddit

[ornate elk]

Ah no problem 🙂 and thanks! glad it helped

[terse hedge]

Finished installation, works crazy good, thanks again

[ornate elk]

Perfect!

[crystal nacelle]

okay, ill considor upgrading my system ram. whats a safe number to increase system page file to? if all fails i think ill use SD1.5 till i can upgrade my system

[terse siren]

That was never a problem before

[shrewd lagoon]

Hello there,

I recently received the controlnet update on the stability.ai newsletter: https://stability.ai/news/sd3-5-large-controlnets?_hsenc=p2ANqtz-8OzxQv3dxyvrylJFgbMACqnb9mCWJAay52ZhplMYygbuAQmZbYGNtMI5SsNWrTrq3oPv9OzG3zMMz_MTy6_wWuXDCSJ-FPfffvDy5MFtX3dRniaXU&_hsmi=100944169.

Will we have it integrated in the playground notebook ?

[potent oyster]

im getting these weird artifacts from inpainting

my values

it works perfectly sometimes, can someone tell me which values to tweak?

[ornate elk]

16000 min and 24000max for C drive

[open wharf]

i have Python 3.10.9 installed on my computer but whenever im trying to run the Invoke AI installation.
its tells me that no Pythin enviroment detected, any ideas as to how do i solve this?

[ornate elk]

Hey, just run the python installer again and then click modify, next and check "add Python to environment variable's"
Then delete the venv folder if invoke has one and relaunch

[open wharf]

hey thanks for the help!
i did as you said and its seem to work however now i get this when im trying to install
any idea what should i do?

[ornate elk]

Hmm did you downloaded the installer today?

[open wharf]

no, about a week ago

[ornate elk]

Then redownload the installer and try again

Thats what the error messages says

[open wharf]

allright

[chrome rampart]

Wishing the support team and all others Merry Christmas. Thank you for all the help and guidance

[fair oxide]

What i need to downald to make ai work(python and what cuz i forgot)

[ornate elk]

Python 3.10.11 64bit and Git 64bit

[fair oxide]

Still its not working

[ornate elk]

Delete the venv folder and relaunch

[fair oxide]

Where is venv folder?

[fair oxide]

I mean check thhis

[ornate elk]

Run the python installer again and click
Modify, next, check "add python to environment variable's"
Then relaunch

Then it should work

[fair oxide]

Kk

Ye now its working thx

[fair oxide]

it is not working XD

[ornate elk]

Oh you installed it wrong (downloaded as zip and on Desktop)

You need to follow the install guide from the pinned messages

[fair oxide]

There is nothing about torch

[ornate elk]

Why should there be something about torch?

[fair oxide]

Idk

[ornate elk]

It gets automatically installed when launched

[fair oxide]

I made everything that tutorial said

[ornate elk]

Any cmd error?

[fair oxide]

When i click user.bat it says couldn't install torch

[ornate elk]

Can you show me the full CMD log?

[fair oxide]

here

[ornate elk]

It shows you have Python 3.13 installed which is not compatible with any webui

You have to uninstall python 3.13

And then delete the venv folder

Then relaunch

[fair oxide]

what is venv folder

[ornate elk]

Its a folder inside the stable-diffusion-webui-directml folder

[fair oxide]

which version of python do i have to install?

[ornate elk]

3.10.11 64bit

[fair oxide]

Kk thx

[ornate elk]

Np

[lunar perch]

hi what is the installation commands for Stable Diffusion Forge webui on linux?

I remember it was liek 2 lines, really simple, but cant seemt o fidn the right ones

[lunar perch]

eh firgured it out if anyone else was wondierng

#!/bin/bash
cd ~/stable-diffusion/stable-diffusion-webui-forge
source ../venv/bin/activate
python launch.py --theme=dark

[broken shale]

can anyone help me? im using ForgeUI and keep getting the error "AssertionError: You do not have CLIP state dict!", i have them on

[ruby yoke]

you might read through this https://github.com/lllyasviel/stable-diffusion-webui-forge/issues/1075

[broken shale]

this is for a flux model

im using SD 3.5 medium

[half sphinx]

I made a Lora using NoobAi as the base on Civitai, and the sample images came out normal. Yet when Im trying to use it on Reforge, the images come out like in the second image for some reason. I have no clue what Im missing at this point

[ornate elk]

SD3.5 isnt supported in Forge currently.
Only on dev branch

[wicked widget]

for whatever reason loras i train also do not work on reforge. they work on comfy and a1111 and forge just not..reforge. no clue why. try it in a different ui

[broken shale]

oh alr thanks

[hot thorn]

hi, need some help, is there a way, to write a line of prompt with several options/things, and let the SD choose one of these at random every time a picture is generated?
lets say i want to make a very long prompt, and generate 100 images, how do I make it choose these things without breaking the result completely?

I asked in general, but figured its proper to do it here

[wicked widget]

if i understand correctly you want one long prompt that will interchange specific parts?
Dynamic prompt with wildcards

[hot thorn]

to somehow paint the picture a bit better for this case, lets say i want 10 groups of lets say 10 words each

and make prompt choose one from each group

[wicked widget]

dynamic prompt with wildcarfds

[hot thorn]

the way i did it, it sometimes just take 4 from 1, 1 from other, 7 from another

[wicked widget]

youll want to make wildcards of the prompts

[hot thorn]

and break s it all

whats a wildcard

[wicked widget]

if i understand correctly

[hot thorn]

been using SD for 2 weeks only

[wicked widget]

https://www.youtube.com/watch?v=bQK5diN59NA

[hot thorn]

will wath it

different question, since this wildcard stuff is too big for what i want to do. is there a command or some sort of bracket i can use to choose 1-2 or from from a list of lets say 5 words?

[fierce moat]

@ornate elk bro curious, did update torch and cuda improve generation speed ?

[hot thorn]

if i got several groups of words with (word:1.3) (same number value), will it potentially break my image?

[ornate elk]

Yes it can

But cuda shouldn't be the latest version. As most ai stuff needs 12.1 or 12.4

For AMD Zluda 11.8 is used

[fierce moat]

Did you how can I load text encoder: t5xxl_fp8_e4m3fn in this file: https://github.com/KONAKONA666/LTX-Video/blob/main/inference.py
text_encoder part

[light glacier]

Does anyone use SD.Next? How do you set the scheduler to "SGM Uniform"?!

[ornate elk]

Nope but maybe in Comfyui as it can use LTX video I guess

[fierce moat]

It does, but this repo claim can be faster using 8 bit kernels, I tried everything tho, ran out of luck I guess

https://github.com/KONAKONA666/LTX-Video

Also there is one user claimed they managed to port this to AMD GPUs https://np.reddit.com/user/kejos92/comments/1hjkkmx/ltxv_inference_on_amd_gpus/

[calm moth]

Can someone help me with img2img upscaling and Ultimate SD Upscaling in Forge? Whenever I run it, it creates two images, one .jpg and one .png, which are slightly different.

[ornate elk]

Its a setting that saves a jpg image if the output exceeds 2mb in size or larger

[elder mauve]

where am i supposed to attach the clips?

[ornate elk]

you need a clip loader

[elder mauve]

oh

where ?

is there any example workflow for sd3 lora?

[ornate elk]

i only have a basic 3.5 with gguf models, but no lora

[open wharf]

It worked, Now two more questions.
If im gonna copy the InvokeAI folder to another drive will it cost problem?
second :

i recieved this error when tried to install SD3.5 Large
what am i soppouse to do here?

[ornate elk]

it can makes problems moving the folder. better to reinstall it on the drive where you want it to be in the end.
For SD3.5 the application needs a User Token from Huggingface to download the model for you.

You need to create an account on Huggingface to be able to get a User Token

[open wharf]

ohh ok

[ornate elk]

Its only needed for the SD base models.

[open wharf]

is there an optiont o just download the model and put int he Invoke AI folder?

or something like this?

[ornate elk]

you can manually download it

and put it into the invoke models folder

[open wharf]

ok thanks!

[full scaffold]

Merry christmas everyone, I'm in need for some help

When trying to generate images in Automatic1111, I get these results:

Best I've managed is to get a silohuette but still made with that kind of noise

[ornate elk]

hey, which model did you used?

[full scaffold]

Thank you for answering so quick CS1o! I've been using sd3.5_large.safetensors

[ornate elk]

np, thats the issue then. Because Auto1111 currently doesnt support SD 3.5

[full scaffold]

Oooooh, that makes sense lol

What should I use instead of A1111?

[fierce moat]

comfy ui the master

[ornate elk]

SD 3.5 currently only works in ComfyUi

[full scaffold]

Seems like it's chartflow time then, thank you very much!

[ornate elk]

For Flux you could also use Forge Webui, they work on SD 3.5 support too but not ready

[fierce moat]

flux better than 3.5 tho in my opinion

also faster

[full scaffold]

Is Flux another big model? Sorry for not being much up to date, last time I used this was about a year and a half ago and things seem to have changed FAST

[ornate elk]

yep Flux is also a big model like SD 3.5

its the more used Model compared to SD 3.5 i guess

[fierce moat]

its ranked better than 3.5

[full scaffold]

I'll download both and try them out! Thank you very much for your help! Hope to get some nice results soon!

[fierce moat]

best open source text 2 image model

[full scaffold]

Oh, so I should really try Flux

[ornate elk]

yea its cool but requires a lot of resources. Better you have a 12gb nvidia gpu or more vram and 32gb RAM

[full scaffold]

I've got right now 64GB Ram and a 3090 24GB

It should be ok then, right?

[ornate elk]

thats perfect

[full scaffold]

Let's go with that option then!

[ornate elk]

also dont go for the full flux model (its 23gb) and it will still overflow into your Pagefile. So better use the smaller variants FP8 or Q8 for example

much more fun

[full scaffold]

I was about to download the model, so that's a good call! What would be the difference?

[ornate elk]

a bit of quality

but much faster

[supple dawn]

GPU 0

[ornate elk]

What does it shows in the right top corner?

[full scaffold]

Thanks! Flux seems to be working great now!

[ornate elk]

Nice no problem!

[ornate elk]

Okay with that GPU you should use the Forge Webui.
It gives your GPU the best performance

[supple dawn]

not strong enough for stable diffusion?

[ornate elk]

It is strong enough

You said you want it faster

Forge Webui is the only way to get the best performance on that card

[supple dawn]

I'm looking at forge webui now

[supple dawn]

then I have to delete what I have now and download it the forge webui, right?

[ornate elk]

You dont need to delete the other webui but you can

[wicked widget]

3.5 is nice for artsy, flux is nice for specifics

i like making meme posters with flux

[limber dune]

Guys, please help me. I basically decided to download stable diffusion for sillytavern, the problem is that it doesn’t connect me, if I try to connect me says this: "Could not validate SD WebUI API: SD WebUI returned an error." Doing some research I found that just add "-api" to the sh file of webui-user, when I tried it, nothing changed, I also tried to add "-allow-code --xformers -enable-insecure-extension-access -api --cors-allow-origins=", but it didn’t work. After however I realized by looking around that I had to add these commands to the batch file, not to the sh, but it is here that the problem is. When I try to start it from there, it sends me an error message: "ERROR: Could not find a version that satisfies the requirement torch==2.1.2 (from versions: 2.2.0, 2.2.0+cu121, 2.2.1, 2.2.1+cu121, 2.2.2, 2.2.2+cu121, 2.3.0, 2.3.0+cu121, 2.3.1, 2.3.1+cu121, 2.4.0, 2.4.0+cu121, 2.4.1, 2.4.1+cu121, 2.5.0, 2.5.0+cu121, 2.5.1, 2.5.1+cu121)
ERROR: No matching distribution found for torch==2.1.2"
Basically says that I don’t have the right version of torch, even though I’ve tried everything. I tried to download python again, to download newer versions of pytorch, but nothing. The error message also tells me: "Looking in indexes: https://pypi.org/simple, https://download.pytorch.org/whl/cu121", I tried to open both links, I clicked on "torch" and chose a newer version but I download a file of which I don’t know what to do, You know how to help me?

[ruby yoke]

what's sillytavern?

[wicked widget]

llm frontend

[open wharf]

hey, so i downloaded some SD 1.5 models, i copy pasted them to the "Models" folder, but it dosen't seem like invoke is recognizing them,
uner the "Models" Tab" he saids "No Matching Models" and in the Generation tab the "Model" dropdown is grayed out.

[ornate elk]

Wrong Python version installed.
You need 3.10.11 64bit
Uninstall all other versions and delete the venv folder.
Then relaunch

[ornate elk]

How big are the model files?

[supple dawn]

I downloaded the forge webui and still very slow and sometimes it doesn't generate the image

what kind of video card should I buy for to work well with stable diffusion?

[supple dawn]

but my video card is good, i don't know why is so slow

[ornate elk]

Can you show the Settings you used? Also try with 1.5 models

A GPU with 4gb vram is just not as good as newer GPUs

[ornate elk]

Any nvidia GPU with a lot of vram.
12gb vram is a good start point

[supple dawn]

When i generate the image i opened this

Memoria 92%

Why GPU 0%? Sometimes go up to 10%

[supple dawn]

it takes about 7-10 minutes to load the image that’s normal?

[ornate elk]

Ah I see the problem your HDD is maxed out.
Did you installed to Webui on the HDD or the SSD?
No matter what, you have to set the Window pagefile to the C drive and increase it.

That should fix the lag and the HDD usage should be 0

[supple dawn]

I downloaded Webui on the E drive that’s the HDD.

so then I have to install on C drive and it will be fine?

[limber dune]

ok, i'll try it now

[limber dune]

where can i find the venv folder?

[chrome stream]

Does anyone know how to fix this issue? i had stable diffusion on my pc before and it got deleted but i dont knbow if this issue relates to that

[limber dune]

BRO YOU'RE A LIFE SAVER

[limber dune]

YOU HAVE NO IDEA WHAT I DID ALL NIGHT YESTERDAY TO SOLVE THIS PROBLEM

[limber dune]

ty

[terse siren]

How do you do this?

[gentle dock]

Hi, I trained a Lora on Forge, is there a way to use it on A1111? It works perfectly on Forge.

[near sonnet]

https://your-ctrl.blogspot.com/2024/12/unveiling-googles-willow-chip-quantum.html

[ornate elk]

No problem 🙂

[ornate elk]

Edit the webui-user.bat with notepad and add the command to the commandline_args args line

[ornate elk]

Its better you have it on C. But you also need to change your Windows Pagefile settings

[limber dune]

guys i have another problem, Lora dont want to load

[ornate elk]

Make sure you selected a compatible model

[supple dawn]

I did the page file and HDD really 0% and much faster thanks❤️

my only problem is hardly any space on my C drive😂😭

[ornate elk]

np, you can leave the models on the HDD and link them to the webui by editing the webui-user.bat file

Make a Folder on the HDD and call it SD-Models for example.
Then in this Folder create a new folder and call it Stable-Diffusion and one Folder for** Lora**.
then move the models in the Stable Diffusion folder. And Loras into Lora.
Then in the webui-user.bat of Forge you can edit to make it look like this:

`@echo off

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

@REM Uncomment following code to reference an existing A1111 checkout.
set A1111_HOME=D:/SD-Models
@REM
@REM set VENV_DIR=%A1111_HOME%/venv
set COMMANDLINE_ARGS=%COMMANDLINE_ARGS% ^
--ckpt-dir %A1111_HOME%/models/Stable-diffusion ^
--hypernetwork-dir %A1111_HOME%/models/hypernetworks ^
--embeddings-dir %A1111_HOME%/embeddings ^
--vae-dir %A1111_HOME%/models/vae ^
--controlnet-dir %A1111_HOME%/models/controlnet ^
--lora-dir %A1111_HOME%/models/Lora
git pull

call webui.bat`

[supple dawn]

I will try it thanks

by the way there are difference between Forge Webui and the stable diffusion? in quality

[ornate elk]

Nothing the other webui is called Auto1111

Both webuis can use Stable-diffusion models

The models and Settings make the quality differences

Not the Programm to use them

[supple dawn]

I copied this and edited the webui user.bat as you said and now it won't start

And i maked folders in E too

[ornate elk]

oh its E drive

then you have to change the Letter at the top path

where it shows D:

to E:

[supple dawn]

Ohh

[ornate elk]

also a good tip: uninstall "advanced system care". its noot good and can slow down your PC.

[supple dawn]

succeeded🥳Thank you

[supple dawn]

Okayy

[ornate elk]

perfect, np 🙂

[radiant pagoda]

can someone please send me a download for sd 1.5? i cant seem to find it

[half island]

Is there a way to install support for intel arc along with my cuda gpu?

[limber dune]

I'm continuing to have problem after problem: I finally found a way to get the lora to load in the dedicated section in the UI. The problem is that what I write doesn't follow the lora, it doesn't make the character I want (I'm using an anime lora). You say to check if the model is correct, but I know neither how to do it nor where to look

And of course I write the right command to invoke it, which is the one between “<>”

[ornate elk]

If you have problems with Invoke. I would suggest using Automatic1111 Webui or Forge Webui.
They are more commonly used. Easier to setup and more people that can help

[ornate elk]

Shouldn't be a problem

[warped canyon]

is it possible to separate img2img from inpaint to a different tab?

[limber dune]

im using automatic 1111, without it i cant connect stable diffusion with sillytavern

[half island]

I just don't know how i'd install intel's part with cuda together. Do i have to compile both? Or just add whatever extra intel one needs?

[ornate elk]

Ah okay

[ornate elk]

You mean in the webui ?

Intel needs a different webui

Best would be to use SDnext

[limber dune]

so what can i do?

[half island]

Ah, so this node pack can't do multi "brand" gpu then i guess https://github.com/neuratech-ai/ComfyUI-MultiGPU

[ornate elk]

I guess not for intel

Only if you have multiple nvidia ones

[ornate elk]

What's the lora based on? 1.5 or pony or sdxl

Ans which model do you use

[radiant pagoda]

can you check dms please?

[open wharf]

one of them is arround 19GB the other one is arround 144mb. which now that i think is kind of small isnt it?

[ornate elk]

the 144mb isnt a model

its a lora

and the 19gb one shouldnt be used

both are not SD 1.5 models

[open wharf]

ohh wait i got it mixed up

1.9GB

not 19

[loud pond]

@ornate elk with an GTX 1070 will I be able to run comfy ui?

[ornate elk]

1.9gb should work

[ornate elk]

yes it will work

but not great optimised

basic stuff will work

[loud pond]

oh alright thanks ill install it!

[open wharf]

I created a video Cupture of my computer to demonstrate the entire process, maybe it can help with figuring out what am i doing wrong.

https://youtu.be/c0MOwfrqYoI

[ornate elk]

the model file is corrupt

it says part at the end of the file and you have a file with the same name and its 0kb

so redownload the model

and delete the other broken parts

[vocal sundial]

yo, can anyone help me figure out why my images are coming out like this?

[ornate elk]

Which webui and Settings do you use?

[vocal sundial]

old forge

[ornate elk]

And the model and VAE ?

[vocal sundial]

VAE none

model tested various and it happened on most of them

both Pony and XL

[ornate elk]

Okay and when did it started to happen?

Any new extensions installed recently or any setting changed?

[vocal sundial]

yeah, I got prevent arctifacts to work

they don't give those cloud images anymore

but I don't think it's related

I formatted my PC recently and it started after

[ornate elk]

Hmm okay

I would suggest deleting the ui-config.json and the config.json for a complete reset of the settings. Then relaunch and test.

If that doesn't fix it you should try ReForge or normal Forge

[vocal sundial]

I'll try that

I didn't know reforge was a thing

gonna try it later

[dapper timber]

i have a1111 and 2050, what model should i download?

[vocal sundial]

its still fked

I'll try reforge

[terse siren]

Yeah...that didn't do shit

ControlNet is just completley borked on my PC

I dont know what happened

It was working one day and now it decides it doesnt want to work the next

[little bough]

Hi, is there a way to revert an extension back to a previous version in A1111/Forge like you can with the version of A1111/Forge ? thanks

[wicked widget]

ponyxl or illustriousxl

[vocal burrow]

You can git checkout them at a specific commit

[vocal burrow]

Works fine to me. The "blockiness" and virtual artifacts are due to your prompt and settings, switching to forge won't fix them.

Screenshot the whole webui page so we see your settings

[terse siren]

And it doesnt show any sort of error message in the command prompt whenever I try to make a control net pic

it just does Press any key to continue

So can anyone please try and help me without using the phrases "have you tried turning your computer on and off" or "Works fine for me, you're doing it wrong"

[vocal sundial]

Yeah, I figured out that the checkpoints I was testing required pony tags

That's was dumb of me, thank you.

[unreal surge]

im getting this error when i try to load an xl model how could i fix this? i just updated it

more errors above i didnt see

[potent oyster]

anyone know how i update python packages in comfy?

[loud pond]

@ornate elk thank you so much with the help in installing stable diffusion for nvidia (I can finally generate images in seconds lmao), but quick question in my old stable diffusion for amd I had a bunch of extensions is there anyway I can get those or do i just have to manually install each one?

[loud pond]

i just ended up moving everything from the extensions folder into my new one ( tell me if this was the wrong thing to do tho)

[ornate elk]

Its okay to do that

[loud pond]

@ornate elk please could you help. I am trying to use AnimateDiffusion extension but whenever I try to use it, I just get this error

[loud pond]

giving the same error when generating normal images now

[ornate elk]

Can you show the full cmd log?

[faint tinsel]

Could someone point me to what im doing wrong? I'm trying to use this cross hatch model to turn this food image into the style of this model, but it doesnt seem to generate what i want. This is the model im trying to usehttps://civitai.com/models/211733/crosshatchxl

this is one of the images that i want to change the style of with the model i linked

[neon orchid]

Hi guys, a question.
Does anyone know a way to train a Lora?

[dapper timber]

getting this whenever im trying to generate

[ornate elk]

Which model and webui do you use?

[ornate elk]

With tools like Kohya_ss, OneTrainer etc

[noble idol]

stuck on this screen

restarting doesnt work

[ornate swift]

Try refreshing your page?

[noble idol]

i did

like 7 times

[dapper timber]

noobai and a1111

[ornate elk]

noobai only works in forge or in auto1111 on dev mode

[ornate elk]

what does the cmd shows?

[dapper timber]

how to get dev mode

[dapper timber]

and what is dev mode

[ornate elk]

by going into the stable-diffusion-webui folder, there opening up a cmd and then type git checkout dev

its the experimental developer branch

more features but buggy

better use forge

[dapper timber]

then noobai will work normally?

[dapper timber]

i would need to uninstall a1111 for it?

[ornate elk]

thats not the full cmd log

but make sure you whitelist the webui in any browser adblocker

[ornate elk]

nope

[noble idol]

ah one sec

how do i

send the whole thing

[ornate elk]

click into the cmd and press ctrl+a

then crtl+c

[noble idol]

kk

just paste it here?

[ornate elk]

whats your gpu?

ah 2060

[noble idol]

yeah

[ornate elk]

remove --skip-torch-cuda-test from the launch args

[noble idol]

alright

relaunching

[ornate elk]

also show your other launch args

there is some issue with controlnet-dir

also delete OneTrainer from the extensions folder. Its not an extensions but a whole seperate tool

after that delete the venv folder

also delete the X-Adapter extension

its for comfyui, not forge

[noble idol]

what will deleting the venv do again

does it yeet anything

i cant remember

[ornate elk]

it resets the core files so that you can use the webui again

no images or models will get removed

[noble idol]

aight

one minute

[ornate elk]

also delete sd-webui-controlnet-main

its the wrong extension

also delete this:
sd-webui-loractl

[noble idol]

wait what is that

i can find everything else but idk what that is

[ornate elk]

also delete this: its not compatible with forge:
sd-webui-regional-prompter

[noble idol]

or where to find it

[noble idol]

oh tragic, is there an alterative then?

i need it

[ornate elk]

yes

[noble idol]

okay i deleted the venv, the control net extension,xadapter,onetrainer,loactl,regional prompter-

did i miss anything

[noble idol]

whats it called

[noble idol]

it does this now

[ornate elk]

https://github.com/Haoming02/sd-forge-couple

[noble idol]

its still not working

im just getting different errors now

[noble idol]

thank you

but yeah now its just doing OTHER stuff aside from starting up

[ornate elk]

typical stability matrix errors. Wouldnt recommend using that tool

try open up a cmd and run
pip cache purge
then delete the venv folder again and relaunch

[noble idol]

alright

[ornate elk]

if that doesnt fix it you have to reinstall forge

[noble idol]

okay so before i go through reinstalling it

[noble idol]

lets make sure i do this correctly

when you say 'open up a cmd'....?

it doesnt let me send the command

[ornate elk]

what do you get?

[ornate elk]

thats not a cmd

[noble idol]

oh shit

what is then LOL

[ornate elk]

click on the windows logo and type cmd

hit enter

[noble idol]

alright

removed some files apparently

alright

venv is gone

trying to relaunch

oh

it needs to download stuff now

@ornate elk

help-

[ornate elk]

Rip

Reinstall needed

[noble idol]

is there an way to do this where i dont lose stuff like my style cards or data from extensions like infinite image gallery?

[ornate elk]

You have to save those before or install forge the normal way and move the stuff over

[noble idol]

before as in, before everything fucked up?

[ornate elk]

Before reinstalling it

[noble idol]

ah bet

now to find where style cards are kept

[ornate elk]

Cards?

Normaly styles are in styles.csv

[noble idol]

idk what theyre called

[ornate elk]

Other stuff is in loras, Embeddings folder

[noble idol]

aight

[ornate elk]

You can save the whole models folder

And Embeddings

And outputs

[noble idol]

alright bet

might have fucked up there but

i got the extensions and the styles.csv

outputs though-

i use something called infinite image gallery, it keeps all my images its an extension

it should still have that data right

[plain pond]

yo mr@ornate elk sorry for bothering.. what happend to my automatic1111?, how to fix it? thanks

[ornate elk]

Idk

If you didn't deletet it then yes

[noble idol]

fair enough

[ornate elk]

Hey, can you show the full cmd log?

[noble idol]

help-

i cant reinstall this using the old method, stability is the only thing that installs everything on my shit wifi-

crazy because swarm still works

loaded up in two seconds

but forge is still dead

[ornate elk]

do you have Git installed?

[noble idol]

gonna have to remind me what that is-

[ornate elk]

remove --disable-nan-check from the webui-user.bat
then delete the venv folder and relaunch
Also whitelist the wbeui in any browser adlocker

[ornate elk]

a programm need to git clone repos

[noble idol]

this?

so i do have it

so why doesnt it work

[ornate elk]

show me the extensions folder

[ornate elk]

why did you copied them over?

why is there stableSwarmUi in ?!

[noble idol]

man i have no clue

you gotta understand

im always awake

i almost never sleep

idk what im doing when tired

[ornate elk]

breaking stuff it seems xD

[noble idol]

yeah-

ive been awake uhh

28 hours atp?

[round spade]

Hey, I have a simple issue...
Auto1111 gives me an error because my python version is too new, how can I downgrade my python?

[plain pond]

alr lemme try wait

[ornate elk]

wow 😮
delete the sd-canvas editor,
the full eyes- detection
the stable-swarmui

then delete the venv folder and relaunch

[ornate elk]

uninstall every python version you have. Then install Python 3.10.11 64bit
then delete the venv folder and relaunch

[noble idol]

aight

sec

[round spade]

I have no idea where the venv folder is, also I'm on Linux...

Ugh... this is so inconvenient...

[noble idol]

i think its somewhere around where your models folder is

i just search for it in the little search bar-

[ornate elk]

Its in the main folder of the webui

[noble idol]

still not loading up

its just on that screen-

[ornate elk]

Then reinstall and dont put every extension back in

[noble idol]

alright

[ornate elk]

There are mostly a lot not compatible or outdated

[noble idol]

[plain pond]

hey thats work mr thankyou very much, but idk somehow I feel like a new problem will arise.. hmm.. ugh.. thankyou btw haha

[ornate elk]

Np 🙂

[noble idol]

you wont believe it

no extensions

still wont get past that screen

its taunting me

[ornate elk]

Can you show me the cmd log?

[noble idol]

yeah

let me open it back up

[ornate elk]

i see no error here

[noble idol]

yeah but

its still not getting past this screen

its nnot doing ANYTHING now beyond putting this screen up then sitting there

[ornate elk]

make sure its whitelisted in any browser adblocker

and which browser do you use?

[noble idol]

how do i do that

ngl some shit called norton

idfk why it pulls that one up

i have other browsers but

[ornate elk]

norton is crapware, remove that

[noble idol]

whatever

[noble idol]

oh i bet

it looks and feels shit but

its never been an issue before

should i try to make it start up on smth else

[ornate elk]

which browser do you use?

remove the norton browser extension

[noble idol]

rn?

opera

crapware uninstalling

didnt even know i had it until my ai started opening on it ngl

so i just kept it

[ornate elk]

its not needed and just causes problems

[noble idol]

fair enough

longest uninstall ever

[ornate elk]

also make sure your not using the opera browser vpn function

[noble idol]

im not

im raw dogging the internet quite frankly

anyways, lets see if it opens in opera

or if i have a 3rd browser-

[ornate elk]

if not try in firefox, its the better browser anyway

[noble idol]

alright

oh i have like 3 OTHER browsers

wtf am i doing

howd those get there

oh crazy

[ornate elk]

if you care about privacy they are all bad xD

[noble idol]

it opened on opera

sweet

[noble idol]

nah, let the world see my search history-

thanks CS

ill be back when i break something again

[ornate elk]

No problem, and keep the extensions to only the needed and compatible ones

[noble idol]

fair enough

[noble idol]

im back

its not saving my defaults-

[ornate elk]

Did you have edited the ui-config.json ?

[noble idol]

i just uhh

use the little defaults tabs

whats that

it saved the prompt this time

not my dimensions

or anything else

[ornate elk]

Ah you used an extension maybe

[noble idol]

no?

well i added one yeah, the one you gave me but

[ornate elk]

Defaults tab?

[noble idol]

yeah

[ornate elk]

Ah okay

[noble idol]

its always been there, its never been an issue either

[ornate elk]

Hmm and they dont get applied?

[noble idol]

nope

it said applied

so i closed out and added the extension

but came back

nothing was like it should have been

so i did it all again, applied, got the same shit

[ornate elk]

which one?

[noble idol]

the regional prompter alternative

[ornate elk]

Ah okay

[noble idol]

[ornate elk]

And no other extension installed?

That maybe alters the settings

[noble idol]

just one but it shouldnt alter any settings

it just saves all my images in a tab

[ornate elk]

Hmm okay

[noble idol]

its just not saving dimensions now

wack

i need it at 1216x832'

i have it memorized but changing it everytime i get on is annoying

whats this-

would this be considered "things i dont need"

[ornate elk]

Yep

Depends on you but always check if the extensions are still getting updated

If they are not Updated for over a year then they most likely won't work and can break your webui

[noble idol]

sad

thats fine

i can likely remove backgrounds better manually anyways

havent seen any sites better at it than me yet at least

[ornate elk]

there is also this tool:
https://fyrean.itch.io/bgbye-background-remover

[noble idol]

oh bet

let me try it

[noble idol]

mmmm

i can say with confidence im still way better

what IT did

what i did

to give it a fair chance i also did a much simpler image, and it did do better

but mine is still cleaner

[ornate elk]

Nice!

[noble idol]

that extension you gave me works perfectly

thank you

[ornate elk]

Nice np

[loud pond]

how would i do this? do i just take a screen shot of the entire image or should i just copy and paste the error?

[ornate elk]

Generate an image. Click into the cmd and press Ctrl+a then Ctrl+c

Then paste it here

[loud pond]

Is this ok?

[ornate elk]

Update your nvidia GPU driver and then delete the venv folder.
Then relaunch and try to generate a normal image

[loud pond]

venv folder?

where would i find this?

[ornate elk]

In the stable-diffusion-webui folder

[loud pond]

Thank you!

ill do it once ive finished updating my drivers

[crystal nacelle]

I have 8GB AMD card and 16GB ram and was wondering if i can install SD3 on forge or automatic 1111? i cant seem to find an install guide for my prefered UI 😦

[clever smelt]

Why not?

[fair oxide]

heelloo!
How can I change my sampler when using cli?
what im currently running for image gen is
python scripts/txt2img.py --prompt "PROMPT" --plms --skip_grid --n_samples 1 --W 1024 --H 576
but I dont see any option for other samplers

or better yet, how can I get GUI instead of using the command line?

I use command line since I couldn't find an interface to use instead with pull request 56

https://github.com/CompVis/stable-diffusion/pull/56

But it's hard to tweak things most the time since I cant really see

Sorry if i'm not the most knowledgeable on this stuff, I'm sorta new to AI as a whole

Im using Arch btw (sorry I kinda have to)

[fair oxide]

Omg my images look so bad 😭

[exotic cloak]

Both automatic1111 and comfyui should Work on arch

[fair oxide]

Yeah but I have an Intel GPU

and i can't get Automatic1111 to use cpu instead

[exotic cloak]

https://github.com/openvinotoolkit/stable-diffusion-webui/wiki/Installation-on-Intel-Silicon

Don't have a Intel gpu but i read openvino everywhere so that looks promising

[lusty frigate]

I have an amd gpu (7900xt), ive recently have been able to download SD and generate images with it, BUT ive got an issue: RuntimeError: Cannot set version_counter for inference tensor

how do i exactly go about this issue/bug?

Ive searched forums and sites, yet i am still confused

Ive got both git and python (3.10.6) installed

i just cant get the extras tab upscaler or SD scaler to work

[loud pond]

yeah i had the same issue when i had a amd gpu your using automatic1111 right?

[lusty frigate]

i cloned the forked version

I followed a video, i got all of it and everything, but i had some fatal erros

[loud pond]

id reccomend following the guide in the pinned comments it should fix a lot of the errors, and when i had the issue with upscaling i had found that apparently it was an issue with amd gpus

this one helped me quite a bit

[lusty frigate]

i was thinking abt doing it through zluda, tried it, dunno if i even did it right

[lusty frigate]

i got all of tht, dunno how im coming up with these fatal errors regarding the repository

dunno if i even pathed git or not, i dont remember it having a path option

[loud pond]

would you mind sharing the whole error?

[lusty frigate]

how would i do tht? copy and paste the entire code? pic?

[loud pond]

yeah just copy and paste the entire code

[loud pond]

how much gb of vram do you have? 20gb?

[lusty frigate]

24gb or it might be 20gb, cant remember

[loud pond]

so just checking. generating an image is fine, its just upscaling?

[lusty frigate]

yeah, cant upscale with certain upscalers in the extras tab

i dont think i can even do img to img scale option either

note: i dont know much abt programming, this is my first time doing this, no prior knowlegde of git, python, etc.

never knew itd be this much more work for amd comapred to an nvidia

[loud pond]

https://github.com/lshqqytiger/stable-diffusion-webui-amdgpu/issues/503

is it the same error as this?

[lusty frigate]

yup

[loud pond]

if it is, i think its just amd in general then.

[lusty frigate]

saw something abt zluda, and git checkout tags/v1.9.3-amd

[loud pond]

im really not sure then, my coding knowledge is quite limited. id reccomend asking "cs1o" once hes online just show him this and he should be able to tell you whether its solvable or not

[lusty frigate]

alrighty then, ty for the help

ill try to troubleshoot this tonight, if i cant then ill mess around with it in the morning

[ornate elk]

Hey, I have the same GPU. Please follow the The Forge with ZLUDA Guide from my Guide list:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[lusty frigate]

uhm im at the part where i have to add C:\ZLUDA folder and %HIP_PATH%bin to my PATH

i dont rlly know HOW to edit it, heard it could break my system or sum...

ik how to get there

[ornate elk]

Ah okay, check that it looks like here:
https://github.com/CS1o/Stable-Diffusion-Info/issues/7#issuecomment-2408541054

If its like in the screenshots, you won't break anything

[ornate elk]

You can install Forge with ZLUDA, you find the guide In the link from the pinned messages.

[lusty frigate]

i got a question abt tht %HIP_Path%bin

i downloaded an older version

but i also downloaded a newere ver

i deleted the older version, yet tht shortcut is getting an error abt how the older ver being non existant

how do i make it to go to the newer ver?

[ornate elk]

Do you mean the webui, Zluda or hip SDK?

[lusty frigate]

hip SDK

[ornate elk]

Ah just uninstall the hip SDK stuff from the Control panel

Then install the version from my guide

There shouldn't be an shortcut

But feel free to Screenshot

[lusty frigate]

so i just delete anuthing related to the older ver?

[ornate elk]

Uninstall the 5.7 stuff and if it says "couldn't be found" you can hit delete

As its mostly already gone

Then reboot the PC

[lusty frigate]

i can restart after i deleted it all right?

[lusty frigate]

is there anyway u could take a look at my code?

zluda is failing to load?....

wait

i edited the path for zluda and the hip sdk

but its not there?

[ornate elk]

Then add them there again

Also your Python is old and on a custom install path

Better uninstall Python and install the 3.10.11 64bit in its default Path.
Also check "add python to path"
Then restart the PC again

And delete the venv folder before the next relaunch of SD

[lusty frigate]

add python to path, and then just install? no custom at all?

wait does tht mean i gotta redo the edit the system varibales again?

[ornate elk]

If Zluda and hip is not in the path then yes

But if you added them again it should stay there

[lusty frigate]

they are this time

[lusty frigate]

by tht u mean...

tht venv right?

[ornate elk]

Yes

Oh and also delete the .zluda folder at the top

[lusty frigate]

was tht a duplicate or sum?

[ornate elk]

No its for the zluda files

But it mostly contains the wrong 5.7 files

At the relaunch it gets to 6.1

[lusty frigate]

so now i can just laucnh tht webui app?

[ornate elk]

Webui-user.bat

[lusty frigate]

im guessing it ran and still fine for you?

[ornate elk]

It reinstalls the core stuff and then should open up in browser

[lusty frigate]

should i be worried abt some of these minor errors?

[ornate elk]

Looks good.
The warnings are normal for Auto1111. They will be fixed with the next update

[lusty frigate]

failed to create model quickly, etc.?

[ornate elk]

Yep

Can be ignored

The compiling will take a bit

First image gen will take 15-40 minutes.
After that it will take seconds

[lusty frigate]

u can still enhance images right?...

[ornate elk]

Like upscaling?

[lusty frigate]

ye

[ornate elk]

Sure

Everything works

Only Flux and SD3.5 models are not compatible

For Flux you would need to use the Forge Webui with ZLUDA

[lusty frigate]

this is normal im assuming?

[ornate elk]

You can relaunch if its not open in your Browser

[lusty frigate]

stable diffusion? its in my browser atm

[ornate elk]

Ah okay then type Cat or Something and hit generate

Then you have 15-40 minutes time to wait

[lusty frigate]

ahm i guess i gotta wait for it to compile FIRST, then the generation will take place?

[ornate elk]

If you dont hit generate it wont

It would compile and then you would still have to click generate

It will take the same amount of time if you generate or not while its compiling

[lusty frigate]

uhm this is my screen atm

dont mind my prompts and stuff, i honestly dont even know if they work or not

[ornate elk]

Thats not good

You activated hires fix with wrong settings so it will take much longer and can freeze for a bit

Better close and relaunch and just type cat with default settings

[lusty frigate]

hires fix?...

[ornate elk]

Yes hires fix

You enabled it

[lusty frigate]

whts tht? where?

[ornate elk]

Look at your screenshot below the negative prompt

Below sampling method

[lusty frigate]

whts tht supposed to do?

[ornate elk]

Hires fix = upscaling.
But you have to set the Hires steps to 10 and the Denois slider to 0.5 for it to work good

[lusty frigate]

is tht for every model and everything? or a general rule of thumb?

[ornate elk]

A general rule of thumb, for a good result and less waiting time

[lusty frigate]

ive since restarted and just did cat with nothing else

ahm i think i still got the same screen

[ornate elk]

Looks good

[lusty frigate]

still gotta wait for the compile to finish im guessing

[ornate elk]

Yep

[lusty frigate]

then its the actual 15-40 minute generation time?

[ornate elk]

It will be finished when you get the cat image

[ornate elk]

Compiling time

One time only

[lusty frigate]

oh thts the time it takes to compile? thought it was the actual image generation time

[ornate elk]

Nope

Just the compiling time

After that it will take seconds to generate images

Did you used Stable diffusion in the past with directml?

[lusty frigate]

i did prior, worked fine, had errors abt the upscaling part though so thts why i thought i should give zluda a shot

i spent abt 5 hrs trying to firgure out how to download SD directml

i followed a video gudiing me through it all

he did not have errors

i however did

[ornate elk]

Most video guides are outdated OE false.
Wouldn't recommend anyone using directml

With ZLUDA you are 2-4 times faster at generating and you use less vram at the same time.

[lusty frigate]

im far too late for tht, now i know after seeing how much faster zluda was compared to directml

[ornate elk]

You can also train Loras with it

I made a Guide for that too

[lusty frigate]

im not too good in tht department atm so ima leave tht for later

though ive had prior expericne in using SD like ui and features

[ornate elk]

Also important is to disable Wallpaper engine when generating images. If you have that running.

[lusty frigate]

is tht bad?

i havent used in a while anywho, was jus woindering

[ornate elk]

It will slow down the generation and can even cause PC crashes when both used at the same time

Even when its paused

[lusty frigate]

well thts a good note

i have successfully generated a simple cat

upscaling works like a charm now

[ornate elk]

Perfect!

[lusty frigate]

oh my days everything works!

thank the lord for you

[ornate elk]

Also Inpainting works now and faster

Directml had Problems with that

[lusty frigate]

if it wasnt bc of ur help, i probably wouldve given up trying to use this

i hoenstlywas debating on moving back to an nvidia

but now its working fine and fast

[ornate elk]

Yea understandable, its not to easy to get it work but thats why I made those guides

[lusty frigate]

tyvm kind sire, i shall be on my way, i can finally learn to enjoy wut everyones been using this for

i shall be going to bed, tyvm again

ive also gained some basic ish knowledge abt some stuff today going through all this troubleshooting

[ornate elk]

NP, and have fun!
Feel free to ask me for any AMD and SD related question.
There are some tricks to gen faster when using sdxl or pony with upscaling too

[lusty frigate]

most of the models i use arent supported on those ones, but ill be sure to experiment

[ornate elk]

Ah sure give them a try, your not limited on 1.5 models anymore as Zluda uses much less vram

[keen thicket]

Hi! I need a hand. Sometimes Stable Diffusion takes incredibly long time to generate an image. Can anybody help me?

[ornate elk]

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

[gilded willow]

Running forge web UI in Linux mint with a 7900xtx. Image Gen works at reasonable speed but hires fix just hangs pretty much indefinitely, even at only 1.5x upscale.

[keen thicket]

I've got a 4080, I didn't touch the webui-user.bat

[ornate elk]

And which settings did you used?

Also add --xformers to the webui-user.bat commandline_args line

[ornate elk]

Make sure to set the hires steps to 10, then it should work

[gilded willow]

Is this some AMD issue? I don't have trouble running more steps with hiresfix on Nvidia cards with less vram.