🤝｜tech-support

[ornate elk]

Here:
https://www.tomshardware.com/pc-components/gpus/stable-diffusion-benchmarks#section-stable-diffusion-512x512-performance

But it was before we got zluda
So its based on Linux performance

[supple glen]

yo can someone tell me if i can use this model in Stable diffusion and in which of the model folders would i need to put it ?

[ornate elk]

It can be used in Forge or Comfyui and goes into the models/stable-diffusion or checkpoints folder

[supple glen]

i only have the "sd- webui" folder and there is a model folder in it its not possible to run it there?

or is that what u mean

[ornate elk]

Yea you would need an other webui for flux

Forge Webui or Comfyui

[supple glen]

can u tell me how i can see if the models work for stable diffusion webui ? or for which ones it is ?

since i downloaded a couple of models and iam not sure if they even run with stable diffusion webui and i have no clue how to check

[ornate elk]

If they are 2-6 GB files they will run

[supple glen]

am i not able to run higher models ?

or only those specific ones

if it says SDXL 1.0 on base model , it doesnt matter how much gb it has or ?

[crisp junco]

Hi. I have SD A1111, Forge and ComfyUI portable all installed on my F: drive. Ad I’m having problems in installing Triton and sageattention to make a Hunyuan workflow run, I suspence there is a problem Python related. If I check Python version from the ComfyUI folder i get 3.10.9
If I check from python embeded folder i get 3.12.7
a web guide I followed is asking me to install Triton and sageattention in my venv directory, but I don’t have any in ComfyUI. I have it in SD instead.
I’m getting lost. I wouldn’t like to do a fresh install of everything but it looks like I have at least two Python aavailable in my System. Is this the real problem?
Thanks a lot

[brittle lantern]

Is there a Zluda one

[ornate elk]

nope

[acoustic radish]

hey guys, I'm building an AI model that can generate headshot based on flux, lora, and controlnet. Currently we are training and building the model. Is there anyone who is interested in this kind of project or has done something similar before? We can connect to collab or smth. Btw we are a dev agency and we wanna build this project to put in our portfolio, but we are also trying to monetize from it by creating tiktok videos and launching on other social medias.

[chilly knot]

in console i see s/it how i change it to it/s ?

is very slow?

i have 1050 ti gpu/
its ok result for this card?

[fair oxide]

is it possible to make gore in ai?

[ornate elk]

Which webui do you use?

[fair oxide]

bro my prompts not showing me what i want

[chilly knot]

standart webui for automatic

[ornate elk]

with that gpu you should try out Forge webui, it will give you a bit more speed but it/s only with 1.5 based models

[chilly knot]

s /it its standart for all models?

how you think 2.5 second /it its ok or something wrong

[ornate elk]

it/s is iterations per second (fast) and s/it is seconds per iteration (slow)

[ornate elk]

it depends on the model, resolution, lora, extension etc

[chilly knot]

1.5 models

without lora

only promt

512 on 512

[ornate elk]

sounds slow then

whats in your webui-user.bat?

[chilly knot]

set COMMANDLINE_ARGS=--theme=dark --xformers --lowvram --no-half-vae

[ornate elk]

you can try change --lowvram to --medvram

[chilly knot]

for my gpu you think what normal speed

[ornate elk]

idk

but with forge webui you will be faster

[chilly knot]

1 min -1 image

29/29 [01:07<00:00, 2.32s/it]

[winter sphinx]

I am running SD3.5_large-Q8_0 on 16GB of ram in ComfyUI and it works great by itself.
However now I am tryinng to run this workflow with the canny control net and it seems like I am running out of Vram.

Do I need more Vram if I use canny with Q8?

[balmy monolith]

I have a bunch of images that I want to train SDXL with, they are 2400x1600, what size do they need resizing to?

is there a calculator for this?

[balmy monolith]

i trained a lora with these images and the results are severely mutated

[strong depot]

running from command line, now getting this

but its a 4090

[ruby yoke]

but that's a comfyUI error. click show report, please, copy the error log to clip board, and paste it here.

[strong depot]

now it wont start because it cant find torth

maybe the portable is the way to go on windows

[ruby yoke]

portable is what I use. but since you've had so many problems, may i suggest you clean everything up and then watch scott's comfyUI tutorials https://www.youtube.com/watch?v=AbB33AxrcZo&list=PLIF38owJLhR1EGDY4kOnsEnMyolZgza1x to get comfortable with comfy itself. and then move on to trying more complicated stuff?

[strong depot]

ive used comfy before

just trying to specifically do this video one

[strong depot]

I think I got it

[wicked widget]

is SDXL lora training still inefficient on 12gb or is there ways now

[ruby yoke]

okay, cause i run SVD inside comfy, not from the command line

[crisp junco]

I would like to do a fresh installation of ComfyUI. Would you suggest the portable version or is there some other more efficient way?

[keen thicket]

Hi I've got a problem. Sometimes when I try to generate images I get this error, can you help me please?

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

[crisp junco]

As far as I can understand , It looks like some settings need to know if you are using the CPU or if you have a CUDA compatibile graphic card. Sorry I’m not a tech expert

[worthy plaza]

Hi, everyone!

Has anybody faced with situation when you have paid next new credits for your account for API functionality, but API return "- 0.5 credits " is accessible all time and don't allow to get access?
trying to write to official support but there isn't any answer more than 15 days....unfortunately

please ,help to clarify what's up/ Thank you.

[terse siren]

Is anybody else having issues with SDForge?

[ornate elk]

What issues do you encounter?

[terse siren]

Right now its not working like...AT ALL

Either with LORA or without

It just hangs on "StateDict Keys: {'unet': 1680, 'vae': 248, 'text_encoder': 196, 'text_encoder_2': 518, 'ignore': 0}"

[ornate elk]

Which model do you try?

[terse siren]

Right now Im experimenting with RealVisXL V5.0 which mind you worked totally fine until this morning

even when I type in simple prompts like "a cute dog in a field" it just completley hangs

command prompt shows no real sign of life

Somethings broken

[ornate elk]

Have you restarted?

Also make sure the webui is whitelisted in any browser adblocker

Also disable vpns

[terse siren]

restarted my PC and still nothing

It wont go past StateDict Keys: {'unet': 1680, 'vae': 248, 'text_encoder': 196, 'text_encoder_2': 518, 'ignore': 0}

I have no idea what the hell is going on

It worked yesterday now it wont work

[terse siren]

Is anyone else having an issue with Forge?

[ornate elk]

do you have a vae or clip encoder loaded?

Forge works fine for me

[balmy monolith]

training SDXL with koyha, I have images with aspect of 3:2, but of resolution 2000x3000, can I use these images or should they be scaled down?
Can anyone offer advice with this please, i'm having issue and google does not provide a suitable answer.

[terse siren]

No

[short grotto]

Hello. I would like to ask for support in regards to installing Stablediffusion.

I've got a new PC but put my old SSD in it which already had SD installed. Should I remove it and do a clean install? What's the recommended version these days?

I'm on Nvidia GPU

[ornate elk]

hey, yea better do a clean install

whats your exact gpu?

[short grotto]

RTX 4060 Ti~! I switched from.. 980 Ti.

This is a huge technological leap for me and I'm only learning all the new functions.

[ornate elk]

oh nice card
then i would recommend starting with Forge Webui as you can also use Flux with it

All installation Guides are here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[gentle schooner]

guys what's the difference of impact between "model weight" and "clip weight" on the lora node

[short grotto]

Thanks.

[short grotto]

This will be helpful

[low flare]

Hi, I have bought an OmnioBook Ultra with "Ryzen 9 AI 365", one of the new CPUs with NPU und iGPU with >50 TOPs. Thought I could generate images with this Laptop, but I got stuck.

I wanna run SD with Linux, but my first tries (Comfy with torch/CPU) failed.

Are there any recommendations for image generation with this setup? Would like to use 3.5 medium, but maybe have to use older models?

Thanks for your advice. Don't know what to do next and where to start for AMD-CPU as the guide only has a link for Intel CPU

[terse siren]

Fucking thing worked for two seconds when I tried again and now its not working AGAIN

Should I just backup all my models and LORAs and just do a fresh install of SD Forge

[balmy monolith]

I have a separate directory for all my models, independent of what UI (A1111, Swarm, Comfy etc) then I use Simlink to make a link to the models directory, then if I ever need to reinstall a UI, it doesn't delete any models

You can also set the directories manually, but Simlink's a lot easier

[ornate elk]

maybe do a clean install, and share more infos when you get an error again

[terse siren]

I wish I was getting errors

The thing is literally hanging

sometimes the command prompt is just blank

completley blank

[ornate elk]

The cmd got blank?

Wtf

[terse siren]

I dont know what the the hell is going on how the fuck can this thing work perfectly fine one day and completley broken the next?

[terse siren]

when I click run.bat sometimes literally nothing happens

[ornate elk]

Then open up a cmd and run
SFC /Scannow

That checks for system errors

[gleaming prism]

i am trying to find what model this generator uses they say stable diffusion V1 model but it can't do furries so it can't be
https://perchance.org/ai-furry-generator

[terse siren]

I just want to know what could cause it to work one moment and break the next

[balmy monolith]

there should be a log directory, that'll most likely have the raw details of why it hung

[terse siren]

where is it

[balmy monolith]

i don't use SD Forge

do a file search for the word log, that might locate it

[terse siren]

yeah big help there

[supple fable]

Does anyone know how to get clipskip back in the automatic ui? I typed in the thing in settings and it's still not there

[open tree]

Hi there! I found this server recommended on a forum. I recently installed Automatic 1111 on my new computer, but after adding the Adetailer extension, it doesn’t show up in the Text2Image interface. I’ve tried restarting the UI and my computer, but it’s still not working. Any help would be appreciated!

[terse siren]

Now after an entire day of aggrivation it seems to work fine now

[brittle lantern]

tryna run a model i imported on macbook auto1111 and i get getting this when i click on the new model (i did the install from the pinned message)

[balmy monolith]

is this a mac version you are tryi9ng to run?

[brittle lantern]

yes the fork thats pinned in this server

ill try reinstalling and if it fixes very well if not ill wait for help

[balmy monolith]

are you trying to use a lora with you model?

[brittle lantern]

no

[balmy monolith]

is this the model?
noobaiXL_EasyNegative
https://civitai.com/models/923983/noobaixleasynegative

[brittle lantern]

https://civitai.com/models/833294?modelVersionId=1140829

[balmy monolith]

have you loaded it into A1111 as an SDXL model?

[brittle lantern]

ive reinstalled seemed to fix the issue

ill keep messing around

[balmy monolith]

i've still not managed to get lora's working myself!

SD1.5 + lora = 90% fail, at least they generate mutants!

[brittle lantern]

i wish i could help i normally use a dedicated app for gens on mac only this model i use on Auto1111 cuz it doesnt on that applications

[balmy monolith]

it bugs me that the online apps works great with them, but mine - just mutants

[strong depot]

Anyone know if theres a better sampler to use instead of euler with ltxv text to video?

[ornate elk]

Can you show the full cmd log?

[ornate elk]

You need to add Clip_stop_at_last_layer in the Settings, User interface, Quicksettings

Don't forget to hit apply after that and reload ui

[keen raptor]

@ornate elk does no-half-vae really make a big difference or i dont need it

[vocal burrow]

do you get nan errors? then you need it. It's that simple

[keen raptor]

i dont get errors but it just takes forever to show the final image like it gets stuck at 100% done

[vocal burrow]

You should try to install multidiffusion-upscaler-for-automatic1111 extension (available in the default extension list) and enable the TiledVAE features (cf screenshot)
https://media.discordapp.net/attachments/1002602742667280404/1111097170150563881/image.png?width=810&height=228

(just tick the Enabled Tiled VAE)

[keen raptor]

o

does it make it faster

[dense flint]

Generation speed is abnormally slow when using Lora, I am using WebUI Reforge, does anyone else have the same issue?

[supple vessel]

I am trying to deploy the Stable Diffusion WebUI in a local server with 4 GPU for students to play with but i have no idea on how to work on it. I have search through the internet but didn't get any practical tutorial. Do anyone have any suggestions or recommended tutorials?

[balmy monolith]

i'm using koyha to train a lora. can I stop or pause training and continue it another time from where it left off?

[lusty thunder]

Wondering if someone here knows much about the tile feature? Like i want to, given a prompt, generate a tile pattern. e.g. Zebra stripes, but it should be seemless at the edges so could wrap around an object. I selected tile-texture but its not seemless. any ideas how to generate a seemless wrap around texture?

const finalPrompt = `A seamless tileable pattern of ${prompt}. This image should tile perfectly from edge to edge with no visible seams.`;

    const response = await fetch('https://api.stability.ai/v1/generation/stable-diffusion-xl-1024-v1-0/text-to-image', {
      method: 'POST',
      headers: {
        Authorization: `Bearer ${token}`,
        'Content-Type': 'application/json',
        Accept: 'application/json',
      },
      body: JSON.stringify({
        width: 1024,
        height: 1024,
        steps: 30,
        cfg_scale: 7,
        samples: 1,
        style_preset: 'tile-texture',
        seed: 0,
        text_prompts: [
          {
            text: finalPrompt,
            weight: 1,
          },
          {
            text: 'seams, edges, borders, words, letters, numbers, nsfw',
            weight: -1,
          },
        ],
      }),
    });

[ornate elk]

Hey for that I would recommend using forge webui.
You find a guide in the pinned messages of this channel

[open tree]

sure

[open tree]

Despite restarting the console or the PC, Adetailer still does not appear in my UI.

[ornate elk]

Thats not the full CMD log

[open tree]

I reinstalled Automatic, but it didn’t work.

[open tree]

Here is the restarted CMD log.

[open tree]

Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.10.1
Commit hash: 82a973c04367123ae98bd9abdf80d9eda9b910e2
Installing requirements
Launching Web UI with arguments:
D:\Stable\system\python\lib\site-packages\timm\models\layers_init_.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers
warnings.warn(f"Importing from {name} is deprecated, please import via timm.layers", FutureWarning)
no module 'xformers'. Processing without...
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.
[-] ADetailer initialized. version: 24.11.1, num models: 10
Loading weights [6ce0161689] from D:\Stable\webui\models\Stable-diffusion\v1-5-pruned-emaonly.safetensors
Creating model from config: D:\Stable\webui\configs\v1-inference.yaml
D:\Stable\system\python\lib\site-packages\huggingface_hub\file_download.py:797: FutureWarning: resume_download is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use force_download=True.
warnings.warn(
Running on local URL: http://127.0.0.1:7860

To create a public link, set share=True in launch().
Startup time: 8.4s (prepare environment: 3.0s, import torch: 2.1s, import gradio: 0.5s, setup paths: 0.4s, initialize shared: 0.1s, other imports: 0.3s, load scripts: 1.1s, create ui: 0.4s, gradio launch: 0.4s).
Applying attention optimization: Doggettx... done.
Model loaded in 2.2s (load weights from disk: 0.2s, create model: 0.4s, apply weights to model: 1.4s, calculate empty prompt: 0.1s).

[open tree]

But Adetailer still doesn’t show up.

[supple vessel]

Thank you for your suggestion! But for me, I am facing the difficulties in deploying the SD model in the server with multiple GPU for multi-students. I need to build from sketch🥲 Can you give me some ideas?

[ornate elk]

Can you reinstall following the guide from the pinned messages?

Because your folder name isnt correct and your using an older python version

Also no launch args used

[ornate elk]

Each GPU can run their own instance. Then you only need to share the different ports

[open tree]

I followed the steps provided on GitHub

[open tree]

The installer downloaded its own version of Python.

[ornate elk]

Yea but its not the recommended way

Its just for the lazy people xD

[open tree]

So, should I download the latest version of Python and follow these instructions?

[ornate elk]

No, follow my guide instructions:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

With Python 3.10.11 64bit

[open tree]

I followed the installation exactly as indicated in the guide, but after installing the Adetailer extension and restarting the UI, it still doesn’t show up.

[ornate elk]

okay, make sure the webui is whitelisted in any browser adblocker

[open tree]

Thank you so much for the help; Adetailer is showing up now!

[ornate elk]

perfect np 🙂

[fair oxide]

how do i make videos with stable diffusion? i specifically have an image i want to make into a video

[bright hill]

how can I pin a batch file to a win10 taskbar?

It seems to have exceptions for .bat and shortcuts

nvm

For anyone's future reference

Create a shortcut to your batch file.

Go to your shortcut properties and change the target to: cmd.exe /C "path-to-your-batch".

Drag and drop the shortcut to the taskbar.

Has to be dragged and dropped

Not right clicked

Very strange

Okay, so I have a weird set of problems

I recently moved my entire stable diffusion folder to another drive entirely

Whenever I start it up, it will ask which browser I want to use

Forget I already set a default last time

And then open a version of firefox that has none of my extensions or default dark mode

[rugged latch]

kinda sounds like a windows bug

when u start stable, by running the batch, it should only start in terminal, then u open the browser and go to http://127.0.0.1:7860

which then opens the ui

to make a links opens in certain browsers in windows u go to settings, type browser, and there will be an option for defaults, go there and select each option u see to open to a specified browser manually for each drop down combo-box

note:, some programs will still call and open legacy browser in windows, this is a function in some programs and cant be changd , example, Jriver Media Center 23

[iron lantern]

is there a way i can see the prompts from a image i made or is there a history log in the program files

[balmy monolith]

i think it's dependent on what gui you are using

[iron lantern]

forge ui

[marsh rain]

So I keep getting a bug where any time I try to import anything to the sketch tab, the image lags and is unusable until it cashes. Judging by leaving my cursor in a fixed position on the image, it feels like the image someone keeps trying to get larger and larger infinitely, without any input on my part. There doesn't seem to be an error log.

[ornate elk]

Make sure the webui is updated and that its whitelisted in any browser adblocker

[marsh rain]

I did a fresh reinstall and ran the updater. I don't use any adblockers.

[ornate elk]

Hmm okay and how large is the image?

[marsh rain]

1024x1024

[ornate elk]

Hmm seems normal

[ornate elk]

Can you try import it in img2img first and then click on sketch?

[balmy monolith]

I tried training a lora.
I have a simple SDXL workflow in comfy, when I generate an image with only the base model, it turns out fine; 4 fingers & thumb, realistic eyes, correct number of limbs etc, but when I use my lora, it returns a mutant

the likeness is there, but it's still mutant

[marsh rain]

I did do this before. Causes the same issue.

[iron lantern]

So I looked it up .... where is the history tab in forge ui

They said I can find old prompts in history tab

[celest delta]

Question, why is doing 2 batch count works, but doing 20 makes the webui completely stuck at the first image? Why does the later use more Vram and get stuck? Shouldn't they be generated one by one? Why does the amount matter?

[dusky bone]

How to make Forge UI update regulary and what settings should I put for a 6gb vram

[ornate elk]

Add
Git pull
To the webui-user.bat

And you dont need to make any changes

[dusky bone]

k, so I don't need xxformers?

[balmy monolith]

i was reading up on xformers just a couple of hours ago, apparently you shouldn't need them, as the other way that it does the calculations is theoretically faster

"The Xformers library is an optional way to speed up image generation and reduce VRAM usage. However, older versions of Xformers below 0.0.20 may produce non-deterministic results"

so maybe you could install them

Here's a post on how to install them, if you feel the need https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/9802

[dusky bone]

ty

[ornate elk]

For forge its mostly not needed

For auto1111 its recommended

[clear elm]

Hello, can anyone point me to correct instructions to install SD on remote Ubuntu VM? Ran through a few YT videos and guides but they don't seem to work.

[hallow latch]

It's not an much of tech-support but what controlnet do I need to get to make it work with Pony checkpoints.. Is it SDXL?

[balmy monolith]

Try this one, I've used it with SDXL
https://huggingface.co/xinsir/controlnet-union-sdxl-1.0/blob/main/diffusion_pytorch_model_promax.safetensors

[balmy monolith]

Controlnet Union XL

[hallow latch]

Is this for every controlnet?

[balmy monolith]

yes, it contains many

[marsh rain]

So after browsing some, I did come to a solution that had me "resize to 100%" (i was on 100%, but clicking down window size then up window size) seems to have PARTLY fixed the issue. It's now at the point above, where it is crashing when imported from image2image to sketch, but importing directly to sketch "seems" to be working fine as a short term fix.

[lethal oracle]

i have some questions guys

[dense tapir]

How do I move my SD onto another drive?

Like the whole thing

[primal wigeon]

Can a1111 tutorials apply to forge ui since it's the same interface

[primal wigeon]

ok so i'm trying to mess around with pony diffusion and i cannot for the life of me understand how the high res fix is supposed to help anything

look at this picture. good, right?

here's hi-res fix applied

she looks like a mandela catalogue alternate

[wicked widget]

lower denoise

hires is a second pass, high denoise will change too much, lower denoise will improve the image

[primal wigeon]

where's denoise

[keen thicket]

Hi I've got a problem. Sometimes when I try to generate images I get this error, can you help me please?

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

[ornate elk]

hey when do you get that error=

most of the time its caused by using a lora or controlnet that isnt made for the model

[keen thicket]

Sometimes when I use some models that other times work without problems. I am not using loras.. And I don't know what is a controlnet so I don't think I am using one

[ornate elk]

okay whats your gpu? and which webui do you use?

[keen thicket]

I have got an rtx 4080, but I also have a ryzen 7600X that has an integrated gpu, maybe it's also using that integrated? I am using A1111 stable diffusion

[ornate elk]

you can deactivate the igpu from the ryzen in the Windows Device Manager but normally it shouldnt switch to that if your on an nvidia gpu

[keen thicket]

Understood, do you have other ideas?

[ornate elk]

adding --device-id 1 --xformers in the webui-user.bat

should help too

xformers increases the performance

[keen thicket]

Xformers?

[ornate elk]

you should either have --xformers OR --opt-sdp-attention in your webui-user.bat commandline_args= line for the best performance

[keen thicket]

Understood, I'll try later and I'll let you know thanks!

[primal wigeon]

where is denoise

[brittle lantern]

whats some good budget GPUs for well SD that can also run flux, dont mind used or refurb either (idk where else to ask)

[ornate elk]

at the hires fix

[ornate elk]

any nvidia with a lot of vram

4060ti with 16gb vram

[noble fable]

Hello, Im pretty new to comfyui and stable diffussion.(ai in general) Can i use any checkpoint with any lora or do i have to match for example a flux lora with a flux checkpoint? 🙂

[ornate elk]

hey, yep thats how it works!

lora and model needs to match the base

[noble fable]

Okey must it match exakt version? like sd 1.5 vs 3.5? 🙂

[ornate elk]

yep

[noble fable]

Alright thanks! 🙂

[noble fable]

Another question, How do i use a checkpoint merge? Do i use it with a base model or can i use it stand alone? 🙂

[brittle lantern]

So a merge model in short is like the creator combined to models u use it like normal put it in ur checkpoints or model folder and use it

[brittle lantern]

Alr

[noble fable]

Thanks 🙂 Im trying to play around with Loras and try to create a image of a Male orc using this lora https://civitai.com/models/19597/orc-diffusion and using SD 1.5 https://huggingface.co/pt-sk/stable-diffusion-1.5/blob/main/v1-5-pruned-emaonly.safetensors
but im getting this error,
mat1 and mat2 shapes cannot be multiplied (256x4096 and 768x320)
Do you have any idea what im doing wrong?
Adding my comfyui workflow

[brittle lantern]

@ornate elk ur turn when ur available

Idk this

I use mac specific software

[brittle lantern]

I also recommend going on civit AI and filtering stable diffusion 1.5 checkpoints by year or month they are better than base 1.5

Also civit AI has a lot of NSFW so if u want u can have them but set it to blur in settings so if u want to see a specific one and it’s blurred u can rather than having all gone or all showing without blur

[noble fable]

Thanks will testa a newer one 🙂 Yeah i saw the NSFW stuff 😄

[brittle lantern]

I personally work with both more NSFW than SFW but this server is SFW so won’t be sharing any more info u can DM if interested but ask before u do

[noble fable]

Maybe you have a recommendation on a good model for fantasy (dnd) style. Like orcs and stuff

[brittle lantern]

But Yh so many great SDXL and 1.5 models both have already peaked in quality because how long they have been out they had enough time to be trained etc.

[brittle lantern]

I’m on mobile rn bout to watch but when I go upstairs I can try finding some for u

[noble fable]

I appreciate it!

[brittle lantern]

For Loras some have keywords that’s needed to be put in to the positive prompt for it to actually take effect so when u look for them check in the description or the side box info if it has any clip skip or keyword and weight requirements

[noble fable]

I will thanks!

[brittle lantern]

No problem

Good luck on your Stable Diffusion adventure

I use 3 different UIs

Draw things app on Mac then Automatic1111 on Mac and comfyUI on Mac

[noble fable]

thanks, I can't seem to get any model on SD 1.5 working, Guessing im not using the correct workflow or Clips for it, I did get Flux.1 and SD 3.5 working

[brittle lantern]

What GPU u rocking?

[noble fable]

A lowly 1050ti 😄 A potato i know, But i don't want to invest into anything big yet
It has worked, Just takes a bit of time 😛

[brittle lantern]

Ah Yh fair I wouldn’t do 3.5 or flux on that I think maybe SDXL is a stretch but doable

[noble fable]

I do have a AMD 5700XT on my gaming rig, But does stable diffusion support amd?

[ornate elk]

Yes

[ornate elk]

Wrong workflow then. But with a 1050ti I wouldn't recommend using comfyui.
Forge Webui will give that card the best performance

Interesting that you got flux and sd3.5 working xD

I guess you used the smallest gguf models of it?

[noble fable]

flux1-dev-Q4_0.gguf :p

[noble fable]

yeah i thought 1.5 would be easier 😛

[ornate elk]

Yea comfyui normaly starts with the basic 1.5 workflow

Here is a basic lora example:
https://comfyanonymous.github.io/ComfyUI_examples/lora/

[noble fable]

Thats exactly the workflow i have 😮

[ornate elk]

Oh 😮

[noble fable]

Aaah exepect i have a DuelClipLoader

[ornate elk]

Yea that won't work

[noble fable]

Yeah without that it worked!

It only took like 20s lol

Im going to try with my 5700xt, See the difference

[ornate elk]

For the 5700xt you need to follow my AMD Zluda Guides.

They are on the first link in the pinned messages of this channel

[noble fable]

Great thanks!

[ornate elk]

Try Forge with Zluda or Auto1111

If that works you can later install comfyui too

The setup is not as easy as on nvidia

If any step isnt clear, feel free to ask me

[noble fable]

I might take you up on that 😄

[ornate elk]

Sure. I'm back in 8 hours, gn

[noble fable]

gn 🙂

[noble fable]

When you're back, im getting this issue starting webui forge after following the guide
\meson.build:1:0: ERROR: Unable to detect linker for compiler clang -Wl,--version

[mental locust]

Can anyone tell me the best video to video methods?

I know of animatediff, batch image to image, cogvideox, ... I don't know if SVD or LTX does it, I don't think Mochi does it, and mimicmotion isn't what I'm looking for. I'd like to convert video to anime

[ornate elk]

Hey, would need to see the full cmd log

[supple vessel]

I have deploy the SD on my VM server through SSH tunnel and run ./webui.sh. How can I visit the SD URL? I cant visit: http://127.0.0.1:7860

[noble fable]

Sure, Here is the full log 🙂

[ornate elk]

Add --listen to the webui.sh commandline_args

[ornate elk]

You have the wrong python version installed

3.12 is not compatible

You need to uninstall that.
Then install Python 3.10.11 64bit
Then delete the venv folder and relaunch the webui-user.bat

[supple vessel]

Ic. thanks.

[noble fable]

Yeah that worked thanks!

[gentle dock]

Hello I installed ROOP, and now everytime I try to launch SD I get this error.

Is there anything that I'm missing?

[unborn mauve]

It used to work fine but I decided to delete all the models and extensions (without deleting the whole a1111 folder) and git pulled a1111 web ui then installed my models again

it used to work fine the last time I used it a few months ago

[finite zenith]

Is there a difference in detail with different VAE?

Or different VAE only changes color and tone?

[ornate elk]

Roop is deprecated and you need to delete it from the extensions folder.
Then delete the venv folder and relaunch the webui-user.bat to fix your webui.
Then install Reactor extension for Faceswap

[ornate elk]

Your webui isnt updated even you did a git pull

Best would be a clean reinstall using the Guide from the pinned messages

[iron lantern]

possible to remove something small from a image gen ?? like a extra hand without redoing the whole image ??

[wicked widget]

inpainting

[calm cairn]

I installed oldify on Forge and now it doesn't open anymore. I'm used to Forge having problems so I also have Auto1111 installed, and I clicked install on oldify and now it doesn't start either. Anyone know how to fix this?

[ornate elk]

whats oldify?

[calm cairn]

under extensions tab I loaded / populated the list automatically and it was an option about halfway down the page which is designed to recolor old photos called "oldify"

I'm sure there's an error message on cmd when I run it, so I could paste that in if it would help, but I don't know what exactly is undefined or where in the code it lies

[ornate elk]

To fix your webui you need to delete the extension and then the venv folder.
Then relaunch

[calm cairn]

How do I delete the extension? I would normally do it in the browser but I can't get it to load into the browser. Is there a way without launching?

[ornate elk]

delete it from the extensions folder

[bold cliff]

can somebody help me with installing perfect world into my webui?

[ornate elk]

whats that?

[bold cliff]

It's some checkpoint merge from civitai, I have never used any checkpoint mergers and have no idea how they work

[ornate elk]

ah okay, they work like normal checkpoints

just drop it into the models/stable-diffusion folder

[bold cliff]

oh

thanks

[ornate elk]

np

[calm cairn]

so I found the extension folder AND VENV for auto1111 but I only found the extension folder for forge, a search of folder contents + subfolders did not find a file or folder named VENV, and also I noticed in environment.bat that it reads

set SKIP_VENV=1

it seems important because I did what you said and now Auto1111 is working again, but forge still isnt working. How do I reset this one? Do I need to set SKIP_VENF to 0?

also for reference, error message on cmd prompt reads as follows, at the end:
from pydantic.fields import fieldinfo, undefined
importerror: cannot import name 'undefined' from 'pydantic.fields' (path)
Press any key to continue...

Then when I press a key it closes.

[ornate elk]

yes set skip venv to 0

then you get one to delete

you also can launc the environment.bat

[calm cairn]

stderr: path \python.exe: no module named venv
Launch unsuccessful. Exiting.
Press any key to continue...

this was with SKIP_VENV=0

double clicking the .bat file doesn't run anything, or does it so fast I cant tell...most .bat files I can see what they are doing

[ornate elk]

Hmm then maybe your forge is outdated or broken

[calm cairn]

name: automatic
channels:

• pytorch
• defaults
  dependencies:
• python=3.10
• pip=23.0
• cudatoolkit=11.8
• pytorch=2.0
• torchvision=0.15
• numpy=1.23

[ornate elk]

Is there no venv folder in the webui folder?

[calm cairn]

there is not

[ornate elk]

Okay then you can't fix it that easy because it uses the cached python stuff

In appdata

[calm cairn]

in main c: drive or is there an "\appdata" in a subfolder of forge folder?

[ornate elk]

In c

Appdata local Roaming programs Python cache or something

I'm not at the PC rn

[brittle lantern]

Ik tests have not been done but u think the new budget intel card the B580 12gb would be good with stable diffusion?

[ruby yoke]

you want an nvidia GPU - you do not want an intel with integrated gpu

[brittle lantern]

I mean it ain’t bad for £250 anyways

But then again there’s no more other AI tests showing

[unborn mauve]

oka thanks

[terse siren]

I found out what was breaking Forge for me, I tried to back up the external hard drive where Forge was installed on and Backblaze must have been eating up more RAM that I thought it would

once I unchecked the drive off of the Backblaze menu, all the sudden it works now

[supple vessel]

Did anyone know which code can i hide the tabs "textual inversion", "Hypernetworks", "checkpoints" and "Lora" in the UI?

[unborn mauve]

I did a fresh install, but even now it feels glitchy, like when I changed settings then hit apply settings, it infinitely started loading (trying to apply settings)

in console it says loading weights from <model> again which never shows any signs of completion indefinitely

also apparently xformers is not found

[unborn mauve]

cant change models either without some sort of error popping up

[ornate elk]

What's your GPU?

Which forge version did you installed?

[unborn mauve]

3080

[unborn mauve]

A1111

I'm in middle of downloading forge files now

[ornate elk]

Ah okay then you need to make sure to have --xformers --medvram-sdxl in the webui-user.bat at the line commandline_args

Also whitelist the webui in any browser adblocker

[unborn mauve]

do I really need medvram sdxl?

when it first came out, sdxl models worked slowly, but they worked without medvram

[ornate elk]

It can speedup the sdxl gen speed as the 3080 only has 10gb vram

[unborn mauve]

oh

also, do flux loras work with pony models?

[ornate elk]

Nope

[unborn mauve]

figured

[ornate elk]

And flux only works in Forge or Comfyui

[unborn mauve]

i see

also, is flux better or some derivative model of it?

like flux dev vs any model from civitai

[ornate elk]

There are some Custom models like pixelwave flux.
Best for your GPU would be to use Forge and then if you want to try flux get a Q8 gguf model or Q4 with T5 q8 gguf encoder

[unborn mauve]

hmm I'll need to learn a lot in flux

is a1111 completely obsolete now?

[ornate elk]

Not obsolete but slowly in development

They work in the background on stuff for it

But no major release since july

[unborn mauve]

oh

[fallen island]

What's the best GPU to get? I'm considering either 4070 ti super or 4080 super (I can't find a ti super 80 is there not one?)

[ornate elk]

both cards would be really good with their 16gb vram
there is a 4080 super which is only 5% faster than the 4080, and no ti

[fallen island]

Thanks I notice that some AMD cards have higher vram, I guess that would make them better? But I don't want to deal with any issues switching brands.

[unborn mauve]

nvidia gpus have option to use xformers, which boosts efficiency or something

[balmy monolith]

from what I can tell, which someone might like to confirm, but atm, it's about software support when choosing between AMD & NVidia. AMD has much less support, which limits what you can use, mostly older version

[unborn mauve]

is there medvram thing for flux as well?

because it takes so long, just in loading the model

and sometimes it gives AssertionError: You do not have CLIP state dict! error

[ornate elk]

For ai stuff nvidia is better as most ai stuff works with cuda, which is made from nvidia

[ornate elk]

Which model do you use?

You also need the clip l and t5 encoder

[wicked widget]

any idea why the lora i just trained would do nothing

[unborn mauve]

iniverse

[unborn mauve]

i did download them, I've selected 3 things in vae/text-encoder area

[balmy monolith]

typo in the name of the file perhaps

[wicked widget]

on

no

ive trained several loras before, this is the first

[brave jewel]

Is possible to use Stable Diffusion on Cloud freely via API?

[ornate elk]

Thats a pony model

That doesn't need the clip text encoders

Oh nvm they do have a flux variant

[unborn mauve]

so any guesses to improve speed?

[unborn mauve]

yes

you can buy a server, then host a public url using that server and then access that public url

[ornate elk]

Using a smaller model size

Like Q4 gguf

[unborn mauve]

i see

[ornate elk]

And use the T5 Q8 gguf encoder:
https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/tree/main

Needs less RAM so it doesn't fill up so fast

[unborn mauve]

tf is gguf

text encoder?

i'm using these 2 rn

[ornate elk]

Wouldn't use the fp16 encoder. Its for 32gb ram minimum

[unborn mauve]

oof

[ornate elk]

Gguf is a model variant supported by forge and comfyui used by new flux models and sd3.5

[unborn mauve]

clip_I is fine?

oka thanks

[ornate elk]

Np

@unborn mauve here is pixelwave flux
You can try the Q4 or Q6 gguf variants.
Should be faster on 10gb vram
https://huggingface.co/mikeyandfriends/PixelWave_FLUX.1-dev_03/tree/main

[charred rover]

installed a new ComfyUI (portable now) and completely forgot how to point it to my custom folders with checkpoints and loras. Tried writing it in comfy.settings.json and it doesnt work. also, they better be located in the same disk right?

[vapid furnace]

hey can someone help me with finetuning here? or do i have more luck in the general chat? no one seems to respond there

[unborn mauve]

as the main model?

alright

btw cant I just use flux bnb something something?

[ornate elk]

There is the file custom model path yaml .example which you have to rename and to edit

[ornate elk]

You can use that too yes

Nf4 or bnb but they dont work as good together with loras than gguf

[unborn mauve]

good to know

[charred rover]

also, they better be located in the same disk right?

[ornate elk]

Doesn't matter

Only the loading time is affected by storing models on a hdd

Not the generation time

[wicked widget]

be ready to wait ages if not on ssd tbh

[wicked widget]

to further this, lora seems to work fine in comfy/swarm but not forge/a1111?

[balmy monolith]

is it better to train the SDXL base model or Lightning (Juggernaut-XI-byRunDiffusion-Lightning-Civitai-Version)?

[upper silo]

Hello, how do I install sadTalker in stable diffusion python 3.10.6, I tried to install it, I spent a whole day installing it and nothing, I get thousands of errors, I tried to install it for python 3.8.1 and nothing, can I install it now? I do not know much of programming

[paper dagger]

Did you try the install requirements.txt?

[upper silo]

This is another one of the errors I got, it won't install either, the tutorial videos are extremely ambiguous.

I was talking to chat gpt which is the best I had, but every solution he gave me was presenting me with more errors.

[ornate elk]

Did you trained in in a specific way?

[wicked widget]

in what ways could i do that? i just fired it in onetrainer

[ornate elk]

Hmm idk

[wicked widget]

yeah same xD

could very well be my a1111 being weird too, idk

[vapid furnace]

Does anyone have access to SDXL's inpainting finetuning script??

or if anyone has developed it??? I'd greatly appreciate it

[ornate ruin]

Hi 🙂 does anyone have an idea why sometimes control net doesn't take your images into account? Basically as if you didn't even upload/enable face id.

Using a 1.5 model
-Lora for face id for SD 1.5 is in the positive prompts
-Control net unit 0 is enabled + low vram
-preprocessor is
IP adapter face id plus
-modsl is
Ip Adapter faceid plusv2 sd1.5

Only way to fix this I found yet was to restart stable diffusion

[ornate elk]

whats your gpu? AMD has a problem with face id if you also have Reactor installed

[ornate ruin]

rtx 4060

But I don't have reactor installed anyways

[ornate ruin]

It's weird that it just sometimes works and sometimes not for me

Especially when I clear out the images and upload new ones it often no longer works and I don't have a way of getting it to work anymore other than rebooting

[ornate elk]

make sure the webui is whitelisted in any browser adblocker

[ornate ruin]

it says error face not found in the console, which is weird because all images are close ups of my face in diffrent angles vut always both eyes sharp and visible

And I just whitelisted the url of my SD window but that didn't help either

[ornate elk]

ah okay yea you need more a wider shot then

if its to close up it wont work

[ornate ruin]

Oh okay? I'll try it without that one, one sec!

It does seem to have done the trick.. wow I never realized that's the image I've chosen was what stopped it from working 😭 thank you lmao.

[ornate elk]

no problem ^^

[iron lantern]

why is my forge sd ... starting a image gen .. when i press shift enter to add another line in the prompt box

[lunar perch]

hello, I finally got kohya_ss instaleld on arch linux, and no idea where to start next, I want to train a basic lora, any tips on what to do enxt?

( I had to use linux bc I have an AMD GPU - 7900xtx )

thank you so much

[topaz ocean]

Has anyone run across a resource or document that covers these features found in FORGE UI?

[dusky bone]

does anyone know to speed up flux image generation, i'm on a 6gb vram

[shut geyser]

my queue button menu

dissapeared

wtf

[dusky bone]

Flux [dev] dosn't want to work in comfy

[shut geyser]

did your queue button also dissapear like in my screen?

[dusky bone]

nope

try using the legacy menu

[shut geyser]

a node?

[dusky bone]

nope

go to settings then

[shut geyser]

legacy menus position?

[dusky bone]

Comfy > Menu > Use New Menu > Disabled

here

[shut geyser]

thank you

[dusky bone]

wait

tell me if it works

[shut geyser]

at least the queue button is there

yes it does work

:D

[shut geyser]

do you know where can i see my workflows in this menu?

[dusky bone]

idk 💀

I only know how to load or save a workflow

[shut geyser]

it just keeps going

the prompts

even if i restart the program

[wicked widget]

be sure to run the ae vae

[wicked widget]

alternative hit ctrl enter to gen on comfy

[dusky bone]

I found out the issue, its bc the node that I am using dosn't support the flux model that i'm using

cuz I'm using Flux-NF4

[wicked widget]

ah

[foggy sail]

my stable diffusion cant work with its extensions, the comfyUI cant connect to the server, OpenPose is ignored, AnimateDiff claims it cant find the modules... do i need to reinstall stable diffusion? i already deleted the venv folder one time

[foggy sail]

ok so i think i fixed the openpose problem; i was using SDXL checkpoints and the OpenPose models were SD, so thats why they did not work.

[ornate elk]

whats the error with animated diff?
make sure you also have the correct model files

[foggy sail]

this was the console message:
AnimateDiff - WARNING - No motion module detected, falling back to the original forward. You are most likely using !Adetailer. !Adetailer post-process your outputs sequentially, and there will NOT be motion module in your UNet, so there might be NO temporal consistency within the inpainted face. Use at your own risk. If you really want to pursue inpainting with AnimateDiff inserted into UNet, use Segment Anything to generate masks for each frame and inpaint them with AnimateDiff + ControlNet. Note that my proposal might be good or bad, do your own research to figure out the best way.

[ornate elk]

thats normal

can be ignored

[foggy sail]

oh i see

the checkpoint im trying to use is ponyDifussionV6 dsxl

and the module is animateDiffMotion_sdxlv10

[ornate elk]

should work

[foggy sail]

its ignoring the animateDiff extension and just making the image

[ornate elk]

Can you show the full cmd log?

[river tangle]

how to open the software?
tried downloading from stability.ai public releases, there is no word "download" in their website. Tried some random model, that asks for some ui, downloaded some ui, has no exe file. was recommended matrix, downloaded that, it has two exe files, one simply flashes before closing and python.exe just shows some credits

[ornate elk]

you installed the webui thw rong way (by downloading it as zip)

[ornate elk]

checkout the pinnes messages for the webui install
Start with Automatic1111 or Forge webui

[river tangle]

the 111111111 version has no exe files, just some bat files. which does some flashing, not sure it even opens anything

[foggy sail]

oh damn how did you notice that?
you mean i should download from the extensions tab?

[ornate elk]

Your webui is named -master at the end. That only happens when you download it as zip and not trough git.

[ornate elk]

The webuis are all bats and no exe files

They use the browser as Interface and the cmd as service

You can checkout the install guides in the first link of the pinned messages

[river tangle]

@mystic basin yeah i do have cuda downloaded as an app and it works cause blender can render thru cuda.

[ornate elk]

Means it can't find Python.
My guide says to first install python 3.10.11 64bit

And Git 64bit

[river tangle]

but i got blender running, it runs python just fine.

[mystic basin]

alright so the error isn't cuda related

it could be that you have the wrong python version for automatic1111

[ornate elk]

You still need it... Because its a different version.
And you have to check "add Python to path"

[mystic basin]

don't ask me how to setup different python version for different app because i don't know how

[ornate elk]

He mostly doesn't need to

And if so, he can just add the path to Python into the webui-user.bat

@river tangle if you want an all in one launcher then go for the Forge Webui

[river tangle]

its now doing this. i just need this whole thing to generate... top down layout pngs.

[ornate elk]

Commit hash none doesn't look good

Did you downloaded Auto1111 as zip?

[mystic basin]

i'm pretty sure he/she forgot to add --xformers to his/her webui-user.bat

it should still work but it would be slower

[river tangle]

i did download it as zip, unzipped it

if i press any key to continue, instead of opening the damn software, it gets closed

[mystic basin]

oh

[ornate elk]

What's your GPU?

Also you shouldn't download it as zip

[mystic basin]

you should have created a specific folder, then go in the research bar in the folder area

write cmd

then enter : git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git

in cmd

the setup would have been easier and faster

if you have git of downloaded on your pc

[river tangle]

i aint downloading git, i hate everything about it.

[mystic basin]

i don't think i can help now, i don't know enough to fix the issue

[ornate elk]

Then dont follow the zluda guides. They are for AMD
You need: --xformers in your webui-user.bat

[mystic basin]

oh that's why it wasn't working

i didn't even realize

[river tangle]

its downloading something. 4 gb in size. idk what tho

[ornate elk]

The base sd1.5 model

[river tangle]

i already downloaded some random model tho

.ckpt file

[ornate elk]

Put it into the models/stable-diffusion folder and relaunch the webui-user.bat

Then it skips the base model

But .ckpt is old stuff

Not good

[mystic basin]

oh yeah, it's downloading a base model so if you haven't downloaded any, you can test if the app is working or not

oops, i was too late

[river tangle]

as long as it does whats needed, i dont care if it does so by bashing rocks together and praying to thunder god for fire.

[calm cairn]

came back, didn't fix problem from the other day, wanted to get deeper

Forge: I installed an addon for coloring black and white pictures in both Auto1111 and Forge, and it made both unable to load. However, I was able to delete venv and the addon folder in Auto1111 and it works again, where Forge doesn't have a venv folder. This is where I am now.

I have done some cursory searches in Windows to find relevant files in appdata but have turned up nothing. Anyone know where to go now?

[paper dagger]

so much to learn here

[copper shore]

the stable diffusion crashed after opening for a few secs

[ornate elk]

What's your GPU?

[copper shore]

gtx 1050 ti

i dont have the issue anymore after i closed some apps but now i cant load the illustrious safetensors model

[ornate elk]

Then you should use Forge Webui with that GPU

With that you can use sdxl/pony/illus models

[calm cairn]

hi again, I was here yesterday--looking for a way to make Forge work again (after an addon failed to work)

any ideas? Or should I just reinstall from scratch?

[hollow widget]

hello! I am beginner of SD
I just lean about controlnet, and having some problem. Controlnet isn't working, after trouble shooting, if discovered that some model isn't supporting controlnet, is that true? or it is other error?
if there are some model cant use controlnet, can i fix it?

[ornate elk]

Hey, You need compatible Controlnet models for the normal models

Sdxl needs xl controlnet models and 1.5 needs 1.5 based ones

[ornate elk]

Yep best is to reinstall then

[desert vector]

I tried mergin two Illustrious models in Forge UI and it didn't work

*** Error loading/saving model file
Traceback (most recent call last):
File "C:\Users\Miket\forge\stable-diffusion-webui-forge\modules\ui_checkpoint_merger.py", line 21, in modelmerger
results = extras.run_modelmerger(*args)
File "C:\Users\Miket\forge\stable-diffusion-webui-forge\modules\extras.py", line 135, in run_modelmerger
primary_model_info = sd_models.checkpoints_list[primary_model_name]
KeyError: 'ntrMIXIllustriousXL_xi.safetensors'

---

what did I do wrong here?

[ornate elk]

Make sure the webui is whitelisted in any browser adblocker.
Also set the bottom option to
"Dont"

[desert vector]

it didn't work. same error

[calm cairn]

Alright, I don't think it has an uninstaller package (since it wasnt installed the way most windows programs are), and I'm not really proficient in bash, so like what do I do? Do I just install a new one over top? Or do I have to remove part or all of the current Forge install first?

[dusky bone]

Does anyone knows how to limit forge ui ram allocation cuz each time I get my full memory used, so my desktop starts to lag

[ornate elk]

How much RAM do you have?

[ornate elk]

You can download it and unpack it into a new folder or delete the old one first (but safe your models and outputs)

[ornate elk]

There is no limit option but if you only have 16gb ram you should increase your Windows Pagefile

[sullen atlas]

Hey Bro! Can you help me? I have Amd Ryzen 9 5900HX with Radeon graphics and Nvidia 3070, which type of ComfyUI to install should I choose?

[ornate elk]

Hey, you need the nvidia one

[lilac spire]

Hey everyone, im trying to install ForgeUI however i cannot find any tutorial for AMD GPUs, can someone please send me a proper one for 2024?

[ornate elk]

Hey, here it is:
Forge with Zluda:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

Feel free to ask if any step isnt cleae

[lilac spire]

i tried that one but even tho i can launch forge, the program refuses to generate any images, accusing a CUDA error

[ornate elk]

Then show me your full cmd log and what's your gpu?

[lilac spire]

i dont have the entire log but i took a screenshot, may i DM you the picture?

[ornate elk]

Sure

[primal marsh]

@xzaguer send me a dm for guidance and assistance

[fair oxide]

Hi, im new to all of this. i am trying to run Flux on webforge. I dl'd flux_dev.safe... is that all I need or do I need additional files ? It ran a few times with just this but the result seemed less impressive than I heard about flux

[ornate elk]

You need flux dev fp8 or q8 gguf.
Then clip l and t5 fp8 or q8 text encoders.
And the ae.safetensor VAE file

[fair oxide]

hmm thanks. where do i put clip l and tf fp8?

[calm cairn]

Will do. This time, when I install Forge, is there any way to have a VENV folder locally? Or are all Forge installs always configured to keep that part of its function in appdata?

[fair oxide]

do you know if I dl the flux1CompactCLIPAnd_Flux1DevFp16 file what I should set the vae text encode to? ae.safetensor?

[cursive acorn]

hi, 2 questions

1. I am still on sd1.5 with a1111 (waiting till next year to upgrade my PC), can I upgrade SD at all or do I need a better card? I only have 8 gigs of VRAM (RX 590)
2. img2img Inpainting always fails. It looks like its working in the preview, even up to 55%, and then at the very end it just reverts whatever change I was seeing and the final output is always the original image. Any clue why? again, sd 1.5 a1111
   thanks

[ornate elk]

Into text-encoders or clip

[ornate elk]

Normaly it should create one

[ornate elk]

Hey do you use the directml or the zluda amd webui?

[cursive acorn]

directml

[ornate elk]

Okay directml has a problem with inpainting and its slow too.
If you switch to the zluda version you can use Inpaint and also sdxl models too

[cursive acorn]

!!! woah

ok let me google zluda

thank you!

ive been stuck not using SDXL models this whole time 😭

[ornate elk]

You can follow my AMD Zluda guide for RX 570/580/590 cards here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides#rocm-hip-sdk-57-with-zluda-setup

Feel free to ask me if any step isnt clear.

[cursive acorn]

Unzip as Folder and move it onto the C: Drive and rename the Folder to ZLUDA.
C drive mandatory? And just straight up C:\zluda ?

[ornate elk]

You can do it on any drive just make sure you later add the correct path to it

[cursive acorn]

ok thanks

[ornate elk]

As the zluda files are only a few MB in size I recommend C

[cursive acorn]

HIP_PATH is just the rocm 5.7 bin folder right

i just win+r'd it and it took me there

[ornate elk]

Yep

[cursive acorn]

sweet

[ornate elk]

you can add the %HIP_PATH%bin or the path directly

[cursive acorn]

also rocm 6.1.2 exists but i did 5.7.1 anyway im guessing some compatability reason

[ornate elk]

Yep rocm 6.1 dropped support for the RX 5xx gpus

[cursive acorn]

ahh

bastards 😭

im going im going!!! i need another year!!

[ornate elk]

Next year you have the choice of new GPUs

[ornate elk]

@cursive acorn I'm off, can help tomorrow if you run into issues.

[cursive acorn]

looks to be working at the moment

thank you!

[ornate elk]

Ah okay, also for sdxl you can try add --medvram-sdxl to the webui-user.bat

Also if you have only 16gb of RAM you need to increase the Windows Pagefile

That fixes model loading errors

[cursive acorn]

oh ok, if I run into that I'll give that a try

[gritty loom]

@ornate elk Hi, I'm trying to install forge-webui with Zluda. I followed your guide but when I went to gen my test image I got this error:

RuntimeError: CUDA error: operation not supported CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1. Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

Is this a common error or did I do something wrong?

[queen marten]

i forgot to cancel my subscription the free trial and i was charged how do i contact to get a refund

[misty nymph]

okay, so I followed the tutorial for ZLUDA for the first time, I think I did everything correctly, I got a nit confused on the part of changin the path though

because, is it supposed to be like C:\ZLUDA%HIP_PATH%bin or what??

I also then tried running the webuser batch, and gave this error. I have python installed, but I dont know if I need to download a specific version of it or what

I followed other video tutorials, and this is as far as I got

[gusty harbor]

my install is not working guys

what should i do 🥲

stuck at commit hash

[mystic basin]

I'm not really good at this but could it be that python isn't a variable?

In the same place where you added the Zluda variable in path

For me, I see python as a variable with the path of where python is located

In the windows area where you added the variable for zluda (C:\ZLUDA%HIP_PATH%bin)

If this isn't you error then i'm not sure what could cause the error

[ornate elk]

Hey, can you share the full cmd log?
Then I can see where the issue is

[ornate elk]

Out of memory error.
Maybe try a 768x768 resolution

[ornate elk]

What's your GPU and do you use a VPN?

[gusty harbor]

RAM-->20GB
VRAM-->16GB
GPU-->NVIDIA RTX A4000
CPU-->INTEL CORE BROADWELL

its working now sir

it just says failed to run model so i guess i just need to add a model

right?

[ornate elk]

Hey nope the variable like that is not correct.
Checkout these images to see how they should look like:
https://github.com/CS1o/Stable-Diffusion-Info/issues/7#issuecomment-2408541054
Then for the python error make sure you have Python 3.10.11 64bit installed and checked "add Python to path"
Then delete the venv folder and relaunch the webui-user.bat.

[ornate elk]

Yes

Put any model that is 2gb or larger into the models/stable-diffusion folder

[gusty harbor]

okay thanks sir for responding have a nice day

[ornate elk]

Np you too

[mystic basin]

can xformers work with ZLUDA? i know it doesn't work with directml but can it with ZLUDA?

i don't think so but i'm curious

[gritty loom]

Here's the full log.

[ornate elk]

Xformers can't work with ZLUDA and directml

[ornate elk]

What's the model you try to use?
How large is the file?

[mystic basin]

i knew about directml but since ZLUDA performance is close to CUDA performance, i though i could probably be able to

thanks for the info

[gritty loom]

Any model I've tried isn't working. I've tried Mistoon anime v10 (around 2 gigs), a variant of SD 1.5 (around 7.5 gigs), and others in between. Older models from about June last year when I originally started to get SD working on my machine.

[ornate elk]

Okay, did you installed any extension by yourself or did you used a lora etc?

Which settings did you used? Resolution etc.
Please try only with the 2gb model first

[gritty loom]

No Loras, no extensions, default settings. The guide said the first gen would take an extended period to make so I just wrote "Mario" and hit generate.

[ornate elk]

Okay thats correct.
Does your CPU has an integrated igpu?
Also is your driver updated?
Also open up a cmd and run
Pip cache purge
Then update Python to 3.10.11 64bit
Then delete the venv folder. And relaunch the webui-user.bat

[gritty loom]

No integrated graphics on my CPU.
I updated my drivers to latest before installing.

My Python should be up to date. 3.10.11 released before I even built this PC. But I'll see if that's an issue.

I'll do the other stuff and report back.

[gritty loom]

Thanks for the help, I got it working now.

Turns out I had Python 3.10.6.

[ornate elk]

Yep Ive seen it in the cmd log.
Good that it works now

[mystic basin]

i'm not sure if it goes in this chat but i have a lot of time problems with hands and other thing so i downloaded an extension for automatic1111 named ADetailer. i found out what it does and how it work but there's a lot of setting i don't understand and right now the extension doesn't seem to work quite well. could anyone help me? (if this doesn't goes in this chat, i'm sorry)

[sullen atlas]

Hey Bro! I have this problem...

[ornate elk]

Hey adetailer just needs to be enabled. Then it will fix faces

[mystic basin]

so i just turn it on and keep the settings as they are? also ADetailer didn't seem to work because i had ADetailer detector off in all the 4 tabs so the prompt given weren't that useful

ty for the help

[ornate elk]

Yea default settings are alright.

You can define the face a bit if you give adetailer a prompt like
Yellow eyes etc

[mystic basin]

alright, i'll keep the default setting because i'm not sure what each settings does

[mystic basin]

can it make hands have 5 fingers?

instead of having less than 5 fingers or more than 5 fingers

[ornate elk]

Nope its not that good as hand fix but you can try with the hand model

For good hands best is to use sdxl/pony models

[sullen atlas]

Hey all guys! Can anyone to help a mamont? 🙂

[mystic basin]

alright, i think i'll switch to another model

ty for the info

[mystic basin]

i've never used ConfyUi

i'm not sure why the import fail

did you install the yourself? with automatic1111 you can download extension directly from the ui and once you restart the ui, you have the extension of downloaded and (working most of the time)

you might be able to do the same with ConfyUI

[sullen atlas]

Yes bro! Thanks! Will try after work!

[mystic basin]

np

[ornate elk]

What does the cmd shows?

[sullen atlas]

Bro, I installed everything from the Mickmumpitz video. So as not to bother you I will try to do everything as you have written in this instruction and if nothing works then I will come back to you.

[brittle lantern]

I have a question for someone is it possible to run Flux on a MacBook Pro with 16gb of ram

And get “decent” generation speed from it

[swift inlet]

Hi,
I am using Stability Matrix for my SD WebUIs. Currently I am using reForge, but ControlNet doesn't seem to be working. It is up-to-date.

But I am getting following error:

*** Error loading script: controlnet.py
    Traceback (most recent call last):
      File "E:\Data\Packages\reforge\modules\scripts.py", line 533, in load_scripts
        script_module = script_loading.load_module(scriptfile.path)
      File "E:\Data\Packages\reforge\modules\script_loading.py", line 13, in load_module
        module_spec.loader.exec_module(module)
      File "<frozen importlib._bootstrap_external>", line 883, in exec_module
      File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
      File "E:\Data\Packages\reforge\extensions-builtin\sd_forge_controlnet\scripts\controlnet.py", line 24, in <module>
        from lib_controlnet.controlnet_ui.controlnet_ui_group import ControlNetUiGroup
      File "E:\Data\Packages\reforge\extensions-builtin\sd_forge_controlnet\lib_controlnet\controlnet_ui\controlnet_ui_group.py", line 16, in <module>
        from lib_controlnet.controlnet_ui.openpose_editor import OpenposeEditor
      File "E:\Data\Packages\reforge\extensions-builtin\sd_forge_controlnet\lib_controlnet\controlnet_ui\openpose_editor.py", line 6, in <module>
        from annotator.openpose import decode_json_as_poses, draw_poses
    ModuleNotFoundError: No module named 'annotator'

How can I fix that?

[swift inlet]

Update: I installed "annotator" through the built-in python package installer. But now I am getting this:

*** Error loading script: controlnet.py
    Traceback (most recent call last):
      File "E:\Data\Packages\reforge\modules\scripts.py", line 533, in load_scripts
        script_module = script_loading.load_module(scriptfile.path)
      File "E:\Data\Packages\reforge\modules\script_loading.py", line 13, in load_module
        module_spec.loader.exec_module(module)
      File "<frozen importlib._bootstrap_external>", line 883, in exec_module
      File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
      File "E:\Data\Packages\reforge\extensions-builtin\sd_forge_controlnet\scripts\controlnet.py", line 24, in <module>
        from lib_controlnet.controlnet_ui.controlnet_ui_group import ControlNetUiGroup
      File "E:\Data\Packages\reforge\extensions-builtin\sd_forge_controlnet\lib_controlnet\controlnet_ui\controlnet_ui_group.py", line 16, in <module>
        from lib_controlnet.controlnet_ui.openpose_editor import OpenposeEditor
      File "E:\Data\Packages\reforge\extensions-builtin\sd_forge_controlnet\lib_controlnet\controlnet_ui\openpose_editor.py", line 6, in <module>
        from annotator.openpose import decode_json_as_poses, draw_poses
    ModuleNotFoundError: No module named 'annotator.openpose'

[ornate elk]

Is your reforge version updated?

[swift inlet]

Yes

[wicked widget]

did youn restart the ui after updating

(sometimes i forget)

[swift inlet]

Yeah, it updates before it starts the webserver

I also restarted it multiple times

[ornate elk]

Hmm then maybe try a reinstall or install the normal Forge version

[swift inlet]

Forge is not being development anymore, no?

That's why "reForge" is there

nvm, forge is still working. But most people recommend reForge, when you not using Flux

I will reinstall reForge now

[wicked widget]

yeah theres no ui that "one size fits all" and multiple uis each has their pros and cons, thats why i recommend people to download and use Stability Matrix to be able to easily switch uis and keep things linked between them to conserve on space and time moving files between folders

[ornate elk]

Forge is in active development. It just shut down some of the legacy features of old forge.
Reforge adds them back.
So if you dont need the old stuff, normal forge is recommended

[swift inlet]

A reinstall worked! Thanks!

[night orchid]

I am trying Flux for the first time, so probably I do something wrong, because SD dont put out any image or error message

any points in the right direction?

[ornate elk]

Which webui do you use?

[night orchid]

A1111 Forge

[ornate elk]

Okay and what's your GPU?

And which flux model did you tried?

[night orchid]

It is a Nvidia RTX 3060 12 Gb

Flux model is: flux1-dev-bnb-nf4-v2.safetensors

[ornate elk]

Okay that should work

Can you share the cmd log?

[night orchid]

its embarrsing but where do I find that log ?

🙂

ah the CMD window ?

[fair oxide]

do you know anything about stabe forge ui? i have some flux model and trying to understand what VAE/Text Encoder to select since it has the option to select several. i do not know if i just select ae.safetensors or also clip and and fxx file

[ornate elk]

You need clip l and t5 and ae

[fair oxide]

Thanks. its all confusing haha.

[ornate elk]

Np!

[night orchid]

Flux are a very different model, a little more komplicated than other models

@ornate elk that is my log

or what sd does

As far as I can see: the says that I don't have a CLIP state dict!

isn't that a some kind of VEA or Textencoder ?

[ornate elk]

You need the t5 clip text encoder.
Download the Q8 t5 Encoder here:
https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/tree/main

Put that into models/text-encoder

[night orchid]

I did put does in

[ornate elk]

Can you show a screenshot how it looks ?

In txt2img

[night orchid]

it is working this time, after restart

@ornate elk Tousand thank you

Flux are insane good

[ornate elk]

Np and yep its pretty good

[night orchid]

with the current setup, I am not sure if I can do text

geez its very memory hungry

I could not find that little o 😉

[swift inlet]

I am getting an error when using ControlNet OpenPose:

    Traceback (most recent call last):
      File "E:\Data\Packages\Stable Diffusion WebUI Forge\modules\scripts.py", line 892, in process_before_every_sampling
        script.process_before_every_sampling(p, *script_args, **kwargs)
      File "E:\Data\Packages\Stable Diffusion WebUI Forge\venv\lib\site-packages\torch\utils\_contextlib.py", line 115, in decorate_context
        return func(*args, **kwargs)
      File "E:\Data\Packages\Stable Diffusion WebUI Forge\extensions-builtin\sd_forge_controlnet\scripts\controlnet.py", line 561, in process_before_every_sampling
        self.process_unit_before_every_sampling(p, unit, self.current_params[i], *args, **kwargs)
      File "E:\Data\Packages\Stable Diffusion WebUI Forge\venv\lib\site-packages\torch\utils\_contextlib.py", line 115, in decorate_context
        return func(*args, **kwargs)
      File "E:\Data\Packages\Stable Diffusion WebUI Forge\extensions-builtin\sd_forge_controlnet\scripts\controlnet.py", line 507, in process_unit_before_every_sampling
        params.model.process_before_every_sampling(p, cond, mask, *args, **kwargs)
      File "E:\Data\Packages\Stable Diffusion WebUI Forge\extensions-builtin\sd_forge_ipadapter\scripts\forge_ipadapter.py", line 147, in process_before_every_sampling
        unet = opIPAdapterApply(
    TypeError: lib_ipadapter.IPAdapterPlus.IPAdapterApply.apply_ipadapter() argument after ** must be a mapping, not Tensor

I have a fresh installed WebUI Forge

[night orchid]

you are lacking something

have you also tried update ?

[ornate elk]

Make sure to use the t5 Q8 and not fp8 encoder to save a lot of RAM

[swift inlet]

I just have a fresh installation WebUI. It is up-to-date

[ornate elk]

Controlnet In forge is and always had issue

[ornate elk]

Which openpose model did you tried?

[swift inlet]

I am getting the mask correctly with the preprocessor

[ornate elk]

The mask?

[ornate elk]

Ahh

Ans which model do you use?

[swift inlet]

Which checkpoint do you mean?

A SD1.5 model: epicPhotogasm_ultimateFidelity

[swift inlet]

Does it work better with reForge?

[ornate elk]

wait, ip adapter is not for pose

do you use multiple controlnets at once?

[swift inlet]

No, only one

[ornate elk]

you need to select openpose for pose

not ip adapter

[swift inlet]

I am selecting the radio button "openpose" and the preprocessor is "openpose_full"

Do I have the wrong models?

[ornate elk]

yep if you cant find a openpose model

here are all the models for 1.5:
https://civitai.com/models/38784/controlnet-11-models

[half island]

Hoi, kinda hit a snag somewhere. Installed intel arc a380 to convert most of my movies in the background until my server can use it, so i installed it as my second gpu, but now comfy won't even launch properly or at all.

It launches, then it just stops, no errors

Even attempting set CUDA_VISIBLE_DEVICES=0 && python main.py

First gpu is 3090, second is the arc card. Worked just fine yesterday

Is it because i got ffmpeg running in it's own powershell window?

[swift inlet]

My SD takes so long every time I start the SD server. It does always "Installing requirements" for 2 minutes or so, till it finally boots up. But it doesn't say what it installs. Why is that so

[half island]

Most likely under the hood, it checks everything, sees it's already installed, then moves on

Open up webui-user.bat, and add --skip-install to commandline args if you're using automatic1111

[ornate elk]

Which webui ? And what's in your webui-user.bat?

Maybe there is a wrong launch arg

[swift inlet]

Ah yes, it was because of one extension. It failed to download a bigger requirement. I just removed the extension

Next problem:
I have Forge and reForge installed:
In Forge everything seems to be working fine.
But in reForge I have problems in img2img. Especially when having a picture with an odd resolution, like 1216x1984
It gives me an error on very low denoising (to be exact 0,01 - 0,04). starting with 0,05 it generates me a picture. On the other side, on Forge it is working normally, even with lower denoise. Same picture, same resolution, same model.
The working denoise strength depends on the resolution. Even if I downscale, it throws me errors with higher denoise strenghts. Or vice-versa

[wicked widget]

File "F:\AI\Packages\OneTrainer\modules\ui\TrainUI.py", line 561, in __training_thread_function
trainer.train()
File "F:\AI\Packages\OneTrainer\modules\trainer\GenericTrainer.py", line 702, in train
lr_scheduler.step() # done before zero_grad, because some lr schedulers need gradients
File "F:\AI\Packages\OneTrainer\venv\lib\site-packages\torch\optim\lr_scheduler.py", line 238, in step
values = self.get_lr()
File "F:\AI\Packages\OneTrainer\venv\lib\site-packages\torch\optim\lr_scheduler.py", line 384, in get_lr
return [
File "F:\AI\Packages\OneTrainer\venv\lib\site-packages\torch\optim\lr_scheduler.py", line 385, in <listcomp>
base_lr * lmbda(self.last_epoch)
File "F:\AI\Packages\OneTrainer\modules\util\lr_scheduler_util.py", line 10, in warmup
return lr_lambda(current_step - warmup_steps)
File "F:\AI\Packages\OneTrainer\modules\util\lr_scheduler_util.py", line 35, in lr_lambda
progress = float(current_step) / float(scheduler_steps)

when trying to save my lora

[candid flicker]

hello, can somebodoy please advise how to install this docker image on windows?

https://hub.docker.com/r/rocm/pytorch

i installed docker desktop on windwos

how do I download a docker image inside of this desktop tho?

[ruby yoke]

@floral cradle has some experience with docker

[candid flicker]

since I have an AMD GPU, I need to get the rocm pytorch

thankfully, AMD provides their own prebuilt pytorch docker image

so I shoudl be albe to pull that and then piggyback ontop of it hopefully

i mean it would be so insanely amazing if I could run all this from windows

I have an amd gpu unfortunately, but a very powerful one, so I am fine running comfyui and stabeldiffusion forge w zluda on windows

the one thing im missing is kohya, apparently that only works on linux if u have amd gpu

so I partitioned my drive and did a whole dual boot setup w arch linux, got that far, but then buildign a custom docker image based on teh kohya repo hit some barriers

ok now I see it running in docker.desktop, but it just exits immediately w no logs, what gives?

[ornate elk]

Hey, you can train Loras with OneTrainer + Zluda on Windows.
I made a Guide for that:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Lora-Trainer-Setup-Guides#amd-onetrainer-with-zluda

[wicked widget]

https://civitai.com/articles/3105/essential-to-advanced-guide-to-training-a-lora

i peeked the lora guide suggestion and found it gave bad advice, this one is more reliant

[ornate elk]

But yours is for Kohya which doesn't work with zluda

[wicked widget]

ah

woop

[hallow latch]

Hi, is it possible to see where the program is working on.. I only see that it's upscaling but I have closed down the browser and went to bed.

So waking up and not sure where it is 😛 Is a bit annoying sometimes hehe 😄

Can I somehow see where it is?

[floral cradle]

i've just woken up, have you installed the dependencies? I'm not familiar with the windows version but it's usually that

[floral cradle]

is there a terminal you can type docker commands in? I normally have to do a docker pull name of thing

then I can run it/make a compose file

I'm not running comfy in docker though, it was easier to run it in a virtual container without docker

[supple vessel]

Did anyone know how to hide the tabs of Textual Inversion,
Hypernetworks, Checkpoints. Lora in txt2img and img2img tabs?

[swift inlet]

Getting an error when startig WebUI Forge. I have installed an extension, but when the error appeared, I removed the extension (by deleting the folder). But error still persists:

[vocal burrow]

delete E:\Data\Packages\Stable Diffusion WebUI Forge\venv\ and let it rebuild itself without the faulty extension's modifications.

@swift inlet

[swift inlet]

Thanks @vocal burrow !

[uncut jewel]

Anyone did install successfully Trellis thing for image to 3D on Comfy ?

[fossil crag]

Hey There,

Which provider is the easiest to host a stable diffusion ai?

[uncut heron]

can anyone help me with training a Lora ?

[grim fog]

yes! I can help you

[grim fog]

How about runpod or vast.ai? the runpod.io is easy to use

[fossil crag]

I've tried with runpod serverless but it seems I can't make it work for some reason, always running into errors

[fossil crag]

do you have experience how to run an sd3.5 with "custom" safetensors on runpod ?

[grim fog]

what is the main problem to use? "custom" safetensors? can you explain about that more detail?

And I think only to host the SD model you want, you don't need to use the runpod serverless and it is okay to use the runpod GPU instance

it is too easy]

[stray thicket]

Does anyone have experience setting up Stable Diffusion to run locally on an Apple Silicon mac?

Does it work? If so, is it relatively simple to achieve?

The machine is capable harware wise (M4 Max) to elimiate that variable!

[ornate elk]

Hey, we have an all in one installer script for Mac in the pinned messages of this channel

[stray thicket]

Ah, amazing. I'll give that a try when I have time

Do you know which version of SD it's based on?

[ornate elk]

you can decide between Auto1111 webui or Forge webui

[stray thicket]

But how does that map on to Stable Diffusion 3.5/3/XL etc.?

Or is it 'any of the above', but the difference is just the UI

[ornate elk]

Auto1111 can use SD 1.5, SDXL and SD 3.0
Forge Webui can use SD 1.5, SDXL, FLUX and in the Future SD 3.5

[stray thicket]

I'll have a play, but if you know before I go digging, does this also play well in a basic code interface? I just want to direct it to a list of prompts for it to generate. Web UI is suggesting to me that it's going to be copy pasting in prompts.

[ornate elk]

you cant do that with the webuis. But you maybe with API provided by Auto1111

[vapid thistle]

really stupid question probably, but I have stable diffusion correctly installed, it's working. But now I wanna try these celebrity models. I have for example found one that works with the pony realism model. I got both models installed, do I need to merge them myself or how does it work? And if so which of the 2 should be the primary model

[vapid thistle]

Nvm imma try out flux, seems a bit more beginner friendly!

[wicked widget]

thats because how hevaily trained the model is, if you want to get conssitence on sdxl or 1.5 for example youll need to start using loras

[drowsy creek]

any good a1111 alternative for linux on CPU/AMD?

since A1111 is abandoned and barely any extensions being made for it anymore

but forge doesn't seem to work on linux

I keep hearing everyone talk about how "forge is basically the successor to A1111 and all the new extensions are being made for it"

it's really annoying if the only big webui that supports running SD on a CPU without nvidia is dead

[pale pagoda]

how can i insert object from image 1 to image 2

[wicked widget]

check the pins, CS1o has a link on how to install for amd

[drowsy creek]

does it work even with a CPU (I have 16 cores and 32 GB ram but my card is not new enough for rocm) like with a1111?

ik it is slow and images take a minute each, but it's better then nothing

[wicked widget]

more than a mionute lol

looking at the guide rn gimme a sec

@ornate elk

need you cos im lost lolol i tried

[drowsy creek]

looking at the guide list in the pins

does the Zluda one for forge allow for CPU generation like on A1111?
again, I have a nice graphics card with 8 GB vram but it is a 2019 model, and I don't wanna have to upgrade JUST for Rocm support

[queen marten]

how long are ticket responses usually

[wooden kelp]

Where are my previous prompts stored? I'm using auto1111 and I can click the blue arrow to get the most recently used prompt but where do I find the older ones? I know you can get them by reading the png metadata but I recently cleaned my output folder and got rid of the older pictures. So any help is appreciated.

[ornate elk]

What's your GPU?

[ornate elk]

The blue arrow only stores the last used prompt

[drowsy creek]

amd radeon rx 5600 xt

it's not listed as supported by rocm

[graceful forum]

I'm having a weird issue with ForgeUI, I've installed plenty of other LORAs without issue, but I just installed one and when I try to add it to a prompt nothing happens

if I manually type in the <lora:etc etc> prompt, nothing happens either, even at the suggested weights

anyone ever had that issue? It's the correct model and everything

[grim fog]

Please use the lora name, if the lora mame is "fuzz", <fuzz: etc etc>

[graceful forum]

can I just link its civitai page? The lora itself is worksafe, but some of its images (as usual) aren't

it's called S1 dramatic lighitng

[grim fog]

I don't interesting about not safe work

[graceful forum]

https://civitai.com/models/661736/s1-dramatic-lighting

I just want to use it because it can do dramatic lighting and I'm trying to do some cool Cyberpunk art so having stark shadows is perfect

but yeah, no luck getting it to work for some reason

[wooden cairn]

I just copy-pasted stability.ai API example code and it doesn't work, returns 400 🤦‍♂️

[graceful forum]

doh, so I guess the file is broken on their end

that explains that

that's a bummer

[wicked widget]

when using onetrainer, sometimes i get File "F:\AI\Packages\OneTrainer\modules\ui\TrainUI.py", line 561, in __training_thread_function trainer.train() File "F:\AI\Packages\OneTrainer\modules\trainer\GenericTrainer.py", line 702, in train lr_scheduler.step() # done before zero_grad, because some lr schedulers need gradients File "F:\AI\Packages\OneTrainer\venv\lib\site-packages\torch\optim\lr_scheduler.py", line 238, in step values = self.get_lr() File "F:\AI\Packages\OneTrainer\venv\lib\site-packages\torch\optim\lr_scheduler.py", line 384, in get_lr return [ File "F:\AI\Packages\OneTrainer\venv\lib\site-packages\torch\optim\lr_scheduler.py", line 385, in <listcomp> base_lr * lmbda(self.last_epoch) File "F:\AI\Packages\OneTrainer\modules\util\lr_scheduler_util.py", line 10, in warmup return lr_lambda(current_step - warmup_steps) File "F:\AI\Packages\OneTrainer\modules\util\lr_scheduler_util.py", line 35, in lr_lambda progress = float(current_step) / float(scheduler_steps) ZeroDivisionError: float division by zero
but the file still outputs, does still error corrupt the lora in anyway and would be why sometimes the loras i train dont work on forge but do on comfy? its always when being saved on the last epoch this happens

posted on onetrainer discord too crossposting in hopes someone here has an idea too

[ornate elk]

Its supported by rocm 5.7 on Linux and by 6.1 zluda on windows

[drowsy creek]

wait what??? I thought the official site showed it as unsupported??

where does it say it's supported?

dang, if it's actually somehow supported being able to generate images without having to wait a minute a pop would be nice

[ornate elk]

Not officially but with community build rocm library's for that card

[drowsy creek]

oh nice

link???

[ornate elk]

If your on Windows you can try my Zluda guides here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[drowsy creek]

is it in the AUR as a package that I download?

[drowsy creek]

I am on arch

[ornate elk]

Then i dont have a guide

[drowsy creek]

rip, back to CPU genning I guess if it doesn't have linux support yet

[wicked widget]

@sick elk

[sick elk]

yes ?

[fierce glen]

hi can someone explain to me how loras work specifically the zoom slider one?

[wooden kelp]

The other prompts are not stored at all?

[ornate elk]

Nope

[shy charm]

is there a way to change the port a1111 opens on?

[ornate elk]

Yep by adding --port xxxx
To the webui-user.bat commandline_args

[shy charm]

i'm assuming it works the same on forks like forge and reforge

[shy charm]

dual boot windows

[ornate elk]

It should if they didnt removed the args

[wooden kelp]

I see. Thanks for the info. Is it the same situation with the images? They're only stored in the output folder and not anywhere else?

[ornate elk]

Exactly

[wooden kelp]

Got it. Thanks.

[quartz grail]

stableDiffusionV15Bf16Fp16NoEma_v15NoEma.safetensors [e3fa3670bb]

i am using this model

with automatic 1111

and i am only getting grey images

[ornate elk]

Try an other model. This one is over 2 years old

For example try Dreamshaper v8

[quartz grail]

Yeah i tried stable diffusion 1.5

And it worked

Thx

[quartz grail]

what will be the best model for generating photorealistic headshots

from images

[wicked widget]

not 1.5, if hardware allows, sd 3.5/flux dev, if not XL Models

[quartz grail]

yeah 1.5 is generating

really funny headshots

[wicked widget]

because its a super early model, like 2-3 yrs now when sd was hjust starting

if you want current, sd 3.5

[quartz grail]

yeah i would try that

my laptop should be able to handle that

[wicked widget]

allot of people use PonyXL for anime and Pony Realism for phothoreal

[quartz grail]

i just want to generate a photo realistic images for pfp like for linkdln or something

i can make a great side income in my uni from that

i wont be needing the large model right ?

like i am planning to use 3.5 medium

[uncut heron]

can anyone help me with Lora training im trying to do - i did all needed ( set file path - C:\Users\Milos\Desktop\LORA treniranje\Desanka Maksimovic\desanka maksimovic\train_data_dire_1 ; all pictures are .jpg and text documents are .txt ) and im using this scrypt via CMD - cd C:\Users\Milos\sd-scripts
.\venv\Scripts\activate ;;;;; python train_network.py --pretrained_model_name_or_path "C:/Users/Milos/Downloads/backup za model" --train_data_dir "C:/Users/Milos/Desktop/LORA treniranje/Desanka Maksimovic/desanka maksimovic/train_data_dire_1" --output_dir "C:/Users/Milos/Desktop/LORA treniranje/Desanka Maksimovic/output" --resolution 512,512 --max_train_steps 1000 --max_train_epochs 1 --debug_dataset ;;;; ---- yet all im getting is this in CMD -

what do i need to do to fix?

[shy charm]

where do i put these to get reforge to show them in the upscaler dropdown?

on forge i had them in my ESRGAN folder but they don't show up on reforge if i put them there

[paper dagger]

SD3.5 Large Turbo was pretty fun via API

[sick elk]

whaty is my error (amd user)
Command: "git" -C "C:\Users\bruch\stable-diffusion-webui-directml\repositories\stable-diffusion-webui-assets" fetch --refetch --no-auto-gc
Error code: 1

[ornate elk]

What's your GPU and when does this error appear?

[sick elk]

AMD RX6650XT
and i try install diffusion

[ornate elk]

Then you should follow my AMD guides. The ones with ZLUDA are faster and better than directml.
You can find the install guides In the pinned messages of this channel or here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

For your card the best would be Auto1111 with ZLUDA or Forge Webui with ZLUDA

Feel free to ask me if any step isnt clear

[sick elk]

i try thanks

[sick elk]

(Make a new Folder on your Drive (not on Desktop, Downloads, Documents, Programms, Onedrive) and name it SD-Zluda.
Dont use the same if you had installed the DirectML webui before.)

Where is the best placemnt ?

[ornate elk]

on an SSD, preferable C: drive

[sick elk]

C/user etc it's okay ?

[ornate elk]

Yep that works

[sick elk]

when i past that's (git clone https://github.com/lshqqytiger/stable-diffusion-webui-amdgpu && cd stable-diffusion-webui-amdgpu && git submodule init && git submodule update) he said "git" is not recognized

@ornate elk

[ornate elk]

thats because you need to have Git 64bit installed and Python 3.10.11 64bit

[sick elk]

i have

i don't uninstall

[ornate elk]

try install Git again

that should work

if not restart the PC

[obsidian gull]

i am getting this error when i launch webui-user.bat, am i missing something?

[ornate elk]

wrong python version.
Uninstall python 3.12 and install 3.10.11 64bit instead
then delete the venv folder and relaunch the webui-user.bat

[obsidian gull]

oh okay

[obsidian gull]

this one?

[sick elk]

absolutly

(3.1 Download the corresponding .zip File for your GPUs GFX Version from here:
https://github.com/likelovewant/ROCmLibs-for-gfx1103-AMD780M-APU/releases/tag/v0.6.1.2)

how can i dowload ? this is what i need fromm url (rocm gfx1032 for hip sdk 6.1.2.7z
rocm.gfx1032.for.hip.sdk.6.1.2.optimized.Fremont.Dango.Version.7z)

@ornate elk