🤝｜tech-support

[steady thistle]

I've been searching but can't figure how to install it

I found the codeformer file but don't know where to put that or to get this: NOTE: To use this node, you need to download the face restoration model and face detection model from the 'Install models' menu.

Oh, I think I got it. Sorry, super new with comfy

[ornate elk]

If you use the reactor node then you dont need to install it

Just select Codeformer in the reactor Node

[steady thistle]

It didn't show up in the reactor node, I was only given gfpgan, but I'm downloading it through the models now, hopefully

[ornate elk]

And you clicked on the little arrow?

[steady thistle]

Yeah, but adding the model put it in the pull down menu. Before that, I just had gfpgan or none.

[ornate elk]

Hmm strange

I'm off now gn

[steady thistle]

Thanks! It does work now. GN

[vocal sundial]

updating forge is just regrets

everything breaks

managed to go back to old forge

[upper basin]

I ran a stable diffusion XL checkpoint on a Macbook Air M3 with 8GB memory. At first it work fine with images (512x768) generating within 45 seconds, but suddenly when I restart it, it requires up to 7 minutes just to load the models and 15 minutes to generate an image with the same size. Any thought on it about why is it suddenly that slow?

[tidal thunder]

Hello all, i need some help, i want to enable or add a pipe.enable_vae_slicing() but i don't know where i need to write that (https://huggingface.co/blog/simple_sdxl_optimizations#:~:text=So if you run SDXL,memory and take 72.2 seconds!&text=This isn't very practical,generating more than 4 images.)

[solemn moon]

this is my bat file

forge UI..

[ornate elk]

What's your GPU?

[solemn moon]

1660ti

i was trying to repath the control net folder to D drive

[ornate elk]

This is how mine looks like:
`@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=C:/AI/stable-diffusion-webui
@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

call webui.bat`

[solemn moon]

@echo off

set PYTHON= "C:\Users\o0\AppData\Local\Programs\Python\Python310\python.exe"
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --precision full --no-half --lowvram --opt-split-attention --always-batch-cond-uncond --api

@REM Uncomment following code to reference an existing A1111 checkout.
@REM set A1111_HOME=Your A1111 checkout dir
@REM
@REM set VENV_DIR=%A1111_HOME%/venv

--ckpt-dir %D:/stable-diffusion-webui-master%/models ^
@REM --hypernetwork-dir %A1111_HOME%/models/hypernetworks ^
@REM --embeddings-dir %A1111_HOME%/embeddings ^
@REM --lora-dir %A1111_HOME%/models/Lora
--controlnet-dir %D:/stable-diffusion-webui-master%/models/Stable-diffusion ^
git pull
call webui.bat

sorry i didnt copy all

Ive deleted my C Drive Auto folder but have the models for control net stored on a backup D drive Auto folder

[ornate elk]

Better move everything together in an Auto1111 webui and then only link that folder

[solemn moon]

k lemme try that

[ornate elk]

You see in my example how it should look like. So you only have to change the A1111_Home path to the Auto1111 webui folder

[solemn moon]

yup think its working now, and i learnt summin with setting a folder then calling for it with %%

[oak jay]

Hi all, I just installed and updated Forge for the first time, I've got it running with SD1.5 and XL models, but can't get it to work with FLUX, I keep getting this error - AssertionError: You do not have CLIP state dict! What am I doing wrong?

[ornate elk]

hey, do you use the nf4 flux dev model?

[solemn moon]

omg that was a rabbit hole, i think im generating but very slow, also no preview, is this normal for forge?

[solemn moon]

you need to instal clip into /sd/text encoders (i think)

dwnld clipI
https://huggingface.co/comfyanonymous/flux_text_encoders/tree/main

and 1 of the t5's

[solemn moon]

and place them here

[open wharf]

trying to install auromatic 111 according tot he instruction i need to go to the file explorer and type CMD, how am i opening the CMD from the file explorer bar?

[floral crow]

anyone else getting issues using Input Audio on https://www.stableaudio.com/generate ?
if i generate with an audio input, the generation is permanently stuck on pending and i can't generate new songs

[ornate elk]

By clicking into the file explorer bar and type cmd

[open wharf]

ok got it

another question: that line right here --xformers --no-half-vae its should go after the set COMMANDLINE_ARGS= right?

also as for the Vram comend, i have arround 64GB of ram.
as for i understand i dont need to edit anything f or it.
after i added the --xformers --no-half-vae am i correct here?

[ornate elk]

Vram is not RAM

Vram is GPU memory

[open wharf]

ohh

wait a sec then

10GB Vram

in that case, i will be needing to add that line: --medvram
were should i place it?

jus to cope paste it or does it need to be added to one of the exsisting lines?

[ornate elk]

Then you need --medvram-sdxl

And xformers and no half vae

[open wharf]

if understood correctly it should be like this: set COMMANDLINE_ARGS=--xformers --no-half-vae

[open wharf]

were do i place that one in the file tough?

[tidal thunder]

Hello all, i have a problem with photomaker (space) in forge, may someone help me please? on github no one answer's me : (
https://github.com/TencentARC/PhotoMaker/issues/186

[ornate elk]

In the same line with --xformers

[open wharf]

allright

COMMANDLINE_ARGS=--xformers --no-half-vae --medvram-sdxl so like this?

[ornate elk]

Yep

[open wharf]

allright

thanks!

[ornate elk]

No problem!

[open wharf]

i installed Python 3.10.11 but that what im getting when luinching webui-user.bat

any idea how to solve it?

[onyx solar]

When you went through the install, at the beginning there was a checkbox to add python to your path. Did you check that? If not, probably the easiest way if you're not tech savvy is to uninstall/reinstall and ensure you choose the box for that. If you did choose it, you may need to reboot.

[open wharf]

this probebly it

[solemn moon]

true sad atory

[fair oxide]

what alternatives to the functionality of pytorch are out there?

Anything C/C++ ideally, I'm looking for the building blocks so I can finally ditch python/pip

[ornate elk]

Run the python installer again, click modify, next, and then check "add python to environment variable's"

[autumn osprey]

Any help with flux training?

[true snow]

hey guys any help with "FileNotFoundError: No checkpoints found. When searching for checkpoints, " when using web11111

[ornate elk]

Hey, you need to download one model (checkpoint) from Civitai.com and move it into the models/stable-diffusion folder

[true snow]

i have a lot of models there already

oh wait

#facepalm

@ornate elkThank you!

[steady thistle]

My AnimateDiff images are not linked together, they're not one image animated but a series of totally different images. I used the ComfyUI workflow and got motion models, but I can't see how mine is different from the sample

[charred bramble]

yo can someone help me, it feels like whatever my prompt is it like 30% of the time tries to create a face? just a second ago it had a blue box around something COMPLETELY unrelated to a human/face and said face0,6

[steady thistle]

This is what I mean, the images aren't related...

[ruby yoke]

which comfy worflow did you use?

[steady thistle]

I've used 3 different ones.

https://civitai.com/articles/2379/guide-comfyui-animatediff-guideworkflows-including-prompt-scheduling-an-inner-reflections-guide

I used #3 from this site, but got the same issue.

I used both mm_sd_v14.ckpt and v3_sd15_mm.ckpt, thinking the motion module was the issue.

[solemn moon]

does forge have a working flux control net?

[solemn moon]

tried looking but not finding it o0

[marsh veldt]

hi guys, whenever i tried to use GPU/directml (using the --use-directml arg) and generate an image, i got the bsod VIDEO_MEMORY_MANAGEMENT_INTERNAL
this happened to both a1111 and sdnext

Ryzen 5 7600x + 7900xt
i've updated the driver to the latest one if that matters

its working well with the CPU although its very slow

anyone can help please?

[ornate elk]

Hey, you should follow my AMD Automatic1111 with ZLUDA Install Guide.
Its on the first Link of the pinned messages

That should work much better than directml

Hey, that means the lora isnt compatible with the selected checkpoint

The lora and the model need the same base type.
Like 1.5 or sdxl

You can use the filter on civitai.com

[marsh veldt]

i tried to use zluda before but for some reason it uses CPU instead of GPU
but i'll follow you instruction and let you know

[ornate elk]

Feel free to ask if any step isnt clear

[ornate elk]

In the positive prompt you write everything they ai should try to generate in the image.
In the negative you write everything you don't want to see

For example you can type
A man wearing a shirt.
Then in the negative you type Yellow
And then the ai will generate the man with a shirt that is not yellow

[marsh veldt]

works like a charm! wondering how to optimize the generate speed? i got like 3-4s/it at the moment

[ornate elk]

Can you show me an example with the cmd

You should get it/s not s/it

You wouldn't get anything yellow than too.

[marsh veldt]

sometimes its it/s

[ornate elk]

Oh did you enabled hires fix ?

[marsh veldt]

nope

[ornate elk]

Hmm

[marsh veldt]

is that slower than it supposed to be?

[ornate elk]

If your using an sdxl model 3 it/s is good

If your using an 1.5 model then not

Also better use 30 steps

Thats enough most of the time

[marsh veldt]

i use sdxl, only got 1.5it/s though

[ornate elk]

And which model did you used for the 3 it/s ?

What resolution did you used ?

Screenshot the txt2img settings

[marsh veldt]

i never get the 3it/s, it was 3s/it in the image

[ornate elk]

Oh true

Can you still show the txt2img settings you used ?
Like resolution sampler etc

And model

[marsh veldt]

wait it got cut off

is that enough

[ornate elk]

Yep

[ornate elk]

To benchmark the max speed you have to try the following:

Resolution: 1024x1024
Sampler: DPM++ 2M
Scheduler: Automatic
CFG: 7
Steps: 30
Prompt: Cat,
Negative prompt: Blurry,

[marsh veldt]

close

[ornate elk]

That looks about right.
Also make sure to not run any Games or Wallpaper engine in the background

What reduces the it/s is:
Higher resolution,
bigger prompt (more tokens)
Using loras, embeddings,
Upscaling

[marsh veldt]

ahh got it, thanks!

i tried increasing batch sizes, and it slows down too

[ornate elk]

Yes only use batch count

Batch size is "generating multiple images at once"

Batch count is "one after one"

[marsh veldt]

ahh got it

thanks a lot!

[ornate elk]

No problem, what I also recommend is using the fp16 VAE for sdxl and to install the Tiled VAE extension.
Both can speed up upscaling and lower the vram amount of the VAE step

[marsh veldt]

how do you do that?

[ornate elk]

For the sdxl VAE:
Download the sdxl.vae.safetensor from here:
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main
go into the Settings - User interface - Quicksettings
There add sd_vae
And then hit apply and reload ui.
Then you can select it in the dropdown menu

For the tiled VAE extension go to the extensions tab, click on Available, then on "Load from"
There search for Tiled VAE and install.
Then relaunch the webui-user.bat

[hoary warren]

i can't seem to use loras in comfyui, i tried different lora loaders and they simply don't do anything, when the exact loras work on forgeui. I even downloaded the xlabs nodes for the lora and they give me "list index out of range error". Is there anyone who is using gguf flux dev and is able to use loras in comfy ui ?

[unreal lagoon]

this comes up as the only search result for skeletal animation 😄 , i'm looking to generate body parts correctly for a cartoon 2d game, to animate them in spine, i found plugins called Misto Line and Anyline on huggingface but cannot make it work correctly to understand body parts on same image, have anyone tried something like this? maybe you saw it and could share?

[wet flame]

idk i would love to know it too

i hate making characters but i need them in my game

i got those hand painted look in my environment

those nice cool stylized 2D envrionment

but i can't find nice characters to fit them

i saw something but i forgot now

@unreal lagoon i can't dm you maybe check this
https://civitai.com/models/429580?modelVersionId=478616

I need to train a model or make lora based on my art but my pc can't handle it for some reason

[unreal lagoon]

wow how did i miss it on civitai, was searching those terms and it didn't show up anything 😮

[wet flame]

is flux really good will work on 16 gb vram gpu?

i have not been using AI for a while

lot of press button not getting the result i want

so been doing manually most of it

[unreal lagoon]

the best way is mistoline, check them out, some preprocessor to use for exact match to sketch and can use for pose as well, if combined with openpose and controlnet stuff possibly will give good results, but only humanoid and dog shaped quadruples are available in openpose at the moment, cannot do dinosaurs or dragons etc, and my game will have things like fish walking on 2 hands 😅

[wet flame]

i need side scroller characters

I prompt character walking to the left on a white background in mid journey

I basically need the character in part so i can rig them

[unreal lagoon]

you make a sprite sheet, not for skeletal animation correct? yeah i saw on youtube that sprite sheet consistency is possible

[azure magnet]

Hi, Is there a way for make comfyui upgrade before the start like automatic1111?

[thorn badge]

is there a way in forge img2img to turn anime/cartoon images into realistic ones with flux?

[marsh veldt]

hi, do you know what causes this issue in a1111?

[ornate elk]

hey, would need to see your txt2img settings again

[marsh veldt]

i managed to get it working by restarting the terminal 🫡

[ornate elk]

alright, when it happens again show the values you used

[fierce hollow]

Why don't I have models displayed ?

https://imgur.com/a/yUhzh0H

[solemn moon]

What UI is this?

[fierce hollow]

That's all there is to Forge....

[solemn moon]

Did you change your .bat file by setting the home folder?

[fierce hollow]

Sorry, I'm not a neural network expert like you, which file with .bat resolution are you talking about ? The one I turn on to use the neural network?

[solemn moon]

I’m no expert, you have to download models and put them into the models folder

[fierce hollow]

Check out the link - I did, there are two images in the link.....

[solemn moon]

I missed that 😅
Maybe copy pasta your log from the command prompt so people can better see what’s going on 🫶

[fierce hollow]

How do I do that? You mean this?

https://imgur.com/a/XPgEenK

[solemn moon]

Yea copy all of that and post it here

[fierce hollow]

Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug  1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: f2.0.1v1.10.1-previous-525-gefe6fed4
Commit hash: efe6fed4994e6d29627d5172aecde71e7e496255
Launching Web UI with arguments:
Total VRAM 8192 MB, total RAM 32609 MB
pytorch version: 2.4.0+cu124
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 3070 Ti : native
Hint: your device supports --cuda-malloc for potential speed improvements.
VAE dtype preferences: [torch.bfloat16, torch.float32] -> torch.bfloat16
CUDA Using Stream: False
C:\Ai\Forge\system\python\lib\site-packages\transformers\utils\hub.py:127: FutureWarning: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead.
  warnings.warn(
Using pytorch cross attention
Using pytorch attention for VAE
ControlNet preprocessor location: C:\Ai\Forge\webui\models\ControlNetPreprocessor
2024-09-09 18:37:44,898 - ControlNet - INFO - ControlNet UI callback registered.
Model selected: {'checkpoint_info': {'filename': 'C:\\Ai\\Forge\\webui\\models\\Stable-diffusion\\pilgrim2DSDXL_v60.safetensors', 'hash': '2a3484d4'}, 'additional_modules': [], 'unet_storage_dtype': None}
Using online LoRAs in FP16: False
Running on local URL:  http://127.0.0.1:7860

To create a public link, set `share=True` in `launch()`.
Startup time: 30.6s (prepare environment: 5.2s, import torch: 13.8s, initialize shared: 0.2s, other imports: 0.9s, load scripts: 3.6s, create ui: 4.3s, gradio launch: 2.4s).
Environment vars changed: {'stream': False, 'inference_memory': 1024.0, 'pin_shared_memory': False}
[GPU Setting] You will use 87.50% GPU memory (7167.00 MB) to load weights, and use 12.50% GPU memory (1024.00 MB) to do matrix computation.
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug  1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: f2.0.1v1.10.1-previous-525-gefe6fed4
Commit hash: efe6fed4994e6d29627d5172aecde71e7e496255
Launching Web UI with arguments:
Total VRAM 8192 MB, total RAM 32609 MB
pytorch version: 2.4.0+cu124
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 3070 Ti : native
Hint: your device supports --cuda-malloc for potential speed improvements.
VAE dtype preferences: [torch.bfloat16, torch.float32] -> torch.bfloat16
CUDA Using Stream: False
C:\Ai\Forge\system\python\lib\site-packages\transformers\utils\hub.py:127: FutureWarning: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead.
  warnings.warn(
Using pytorch cross attention
Using pytorch attention for VAE
ControlNet preprocessor location: C:\Ai\Forge\webui\models\ControlNetPreprocessor
2024-09-09 18:40:14,639 - ControlNet - INFO - ControlNet UI callback registered.
Model selected: {'checkpoint_info': {'filename': 'C:\\Ai\\Forge\\webui\\models\\Stable-diffusion\\pilgrim2DSDXL_v60.safetensors', 'hash': '2a3484d4'}, 'additional_modules': [], 'unet_storage_dtype': None}
Using online LoRAs in FP16: False
Running on local URL:  http://127.0.0.1:7860 ```

To create a public link, set `share=True` in `launch()`.
Startup time: 29.7s (prepare environment: 5.2s, import torch: 13.5s, initialize shared: 0.2s, other imports: 0.8s, load scripts: 3.5s, create ui: 4.2s, gradio launch: 2.2s).
Environment vars changed: {'stream': False, 'inference_memory': 1024.0, 'pin_shared_memory': False}
[GPU Setting] You will use 87.50% GPU memory (7167.00 MB) to load weights, and use 12.50% GPU memory (1024.00 MB) to do matrix computation.```

[solemn moon]

@fierce hollow did you restart since you put the models in the folder?

[fierce hollow]

Yes

[solemn moon]

I’m new to forge, and I’m not sure what’s what just yet, hopefully someone can help 🫶

[loud pond]

how do i make a lora?

[lament fog]

yes, I had a 16g card and even the full 16bit dev model worked for me, but there are other options if you want to reduce your vram. there's fp8 models, NF4 models and gguf models now that can run on even less vram. you can research these to see what is the best fit for you, but the main point is yes, you'll have no trouble running flux.

[ruby yoke]

simple answer - you assemble a collection of images, assign captions to each one, then use a lora trainer to create with. longer answer - you should probalby watch some youtube videos on how to create loras as it's a somewhat complicated process when you're first getting started

[loud pond]

Alright, got any suggestions of vids to watch?

[ruby yoke]

olivio has a couple on his channel, along with a lot of other good tutorials https://www.youtube.com/@OlivioSarikas/videos

[loud pond]

thanks ill give it watch

[fierce hollow]

I want to do face replacement in cartoon style using ContolNet, installed a clean version of SD, threw a few models.
What could be the problem, please help
P.S, open the link there are a lot of pictures.
https://imgur.com/a/m2UGpAa

[spare wigeon]

Hello, there's a problem im currently facing. Yesterday and the day before, I could generate a picture rather quickly (about 30s). But today, it's been 10 minutes and it's still saying [99% ETA:5s] and the time keeps going up. And when it gets stuck at 99%, my computer is slowed down like hell 😦

[ornate elk]

Hey, what txt2img settings do you use?

[buoyant sierra]

Recently I've used this service and impressed by AI engine.
https://www.nterview.me/
https://www.youtube.com/watch?v=AfDn_Esqgg8

[spare wigeon]

teps: 40, Sampler: Euler a, Schedule type: Automatic, CFG scale: 7, Seed: 883947904, Size: 768x1152, Denoising strength: 0.4, Clip skip: 2, Hires upscale: 2, Hires upscaler

[ornate elk]

Ah you used hires fix.
Always keep the hires steps at 10 or Max 15 or you will get that freez.
Because hires steps on 0 = the amount of normal steps on top.

[spare wigeon]

Thanks! So this -> Hires upscale: 10?

[fair oxide]

im using forge all of the images are being put in temp gradio how do i stop it from doing that? i just want it to go in my temp folder not in to gradio

[ornate elk]

Hires Steps: 10

Hires Upscale by 2x

[vague surge]

how do I fix this I alreeady did the upcast setting

[ornate elk]

What's your model?

[vocal burrow]

and gpu ?

[thorn pendant]

i tried fix hand with impainting but smh it not work, anyone can help?

[ornate elk]

What's your denois?

[thorn pendant]

denois?

[thorn pendant]

oh 0.9

og img have 0.45

[ornate elk]

Try lower it

[hazy wigeon]

getting this error at the Apply Controlnet (Advanced) node

[thorn pendant]

i/m ussing controlnet too, is that ok sir?

[ornate elk]

for inpainting?

[thorn pendant]

yes sir

[ornate elk]

only if you use the inpaint controlnet model

[hazy wigeon]

here's the log

[thorn pendant]

is that right sir

[ornate elk]

i thought you wanted to use inpaint

you have to disable controlnet to use inpaint when you mask here hands

[thorn pendant]

i i just don't know what to do 😢

[thorn pendant]

omg i didn't know that

[ornate elk]

ypi have to mask her hands, then set inpaint area to "mask only"

then prompt for hands, perfect hands, etc

start with denois 0.5 and adjust if its not good

[thorn pendant]

do i must use adetailer?

[ornate elk]

no

[thorn pendant]

ty sir, it's better now ❤️

[thorn pendant]

so i can use impainting or controlnet right? and iff yes then what's better sir?

[solemn moon]

inpaint is painting in, control net is controlling what to paint, then you get a controlnet inpaint model..

omg so confused

[thorn pendant]

so i can use one of them to make better hand?

[solemn moon]

i use img to img w/ inpaint and control

[thorn pendant]

both at same time?

[alpine wasp]

Hello everyone, I'm having quite some trouble here figuring out what's wrong, so i'm using automatic1111 and webui for it, and when i try to switch to a model i recently downloaded it switches back to the old model and the console gives me this

[ornate elk]

Hey, What's the model and what's your GPU?

[solemn moon]

Yea but I’m talking 1.5

[hazy wigeon]

Hi I'm getting an error at the Apply Contronet (Advanced) node. Here's the logs, I'm also using PonyXL checkpoint with controlnetxlCNXL_bdsqlszLineart

[proud lion]

This is preventing me from generating images in the browser, it doesn't allow me to generate any images

[ornate elk]

Hey, you need to download a checkpoint from civitai.com

Then place it into models/stable-diffusion

[proud lion]

I'll try soon, thx

[little bough]

Hi, does anyone know the route cause of this controlnet error, it won't load anymore for me in a1111, forge or reforge, in the forges it won't stay in enabled in ext settings but it is enabled in a1111 and the error remains. ControlNet preprocessor location: C:\A1111\stable-diffusion-webui\extensions\sd-webui-controlnet\annotator\downloads
*** Error loading script: controlnet.py
Traceback (most recent call last):
File "C:\A1111\stable-diffusion-webui\modules\scripts.py", line 515, in load_scripts
script_module = script_loading.load_module(scriptfile.path)
File "C:\A1111\stable-diffusion-webui\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 "C:\A1111\stable-diffusion-webui\extensions\sd-webui-controlnet\scripts\controlnet.py", line 101, in <module>
global_state.update_cn_models()
File "C:\A1111\stable-diffusion-webui\extensions\sd-webui-controlnet\scripts\global_state.py", line 76, in update_cn_models
found = get_all_models(sort_by, filter_by, path)
File "C:\A1111\stable-diffusion-webui\extensions\sd-webui-controlnet\scripts\global_state.py", line 43, in get_all_models
fileinfos = traverse_all_files(path, [])
File "C:\A1111\stable-diffusion-webui\extensions\sd-webui-controlnet\scripts\global_state.py", line 27, in traverse_all_files
f_list = [
File "C:\A1111\stable-diffusion-webui\extensions\sd-webui-controlnet\scripts\global_state.py", line 28, in <listcomp>
(os.path.join(curr_path, entry.name), entry.stat())
FileNotFoundError: [WinError 3] The system cannot find the path specified: 'C:\A1111\stable-diffusion-webui\models\ControlNet\control_v11e_sd15_ip2p.pth'

Not sure what's going on, it started not working in the forges last month after I updated them to try flux and i hadn't updated a1111 in a good while where it worked before and it isn't working either

[vast pulsar]

bro, pllls tell me wtf denoising does, i've been always using that on 0,55 cuz... just following examples lol

[ornate elk]

Denois means "how much should the output change from the input"

Its important for upscaling to not set it to high

Mostly between 0.35 and 0.5 is good, depending on the upscaler

[tulip violet]

people i become mad... i got currently on ksampler and also ultimate sd upscaler the error " ComfyUI Error Report
Error Details
Node Type: UltimateSDUpscale
Exception Type: AttributeError
Exception Message: 'NoneType' object has no attribute 'model_options'
Stack Trace" and i cant find anything about that error in the hole internet... can anybody help?

[ornate elk]

Best would be to ask on #🧣｜comfy-ui too

[ornate elk]

Have you updated controlnet extension too after updated auto ?

[tulip violet]

maybe not 🙂 i will try thx

but it is a flux workflow without controllnet

https://www.youtube.com/watch?v=j-9m2hOcyiU

[serene basin]

How do people train Loras with multiple people's faces and actually have it generate different unique faces instead of basically one "averaged" face over and over? Do I have to caption each feature of each person's face?

[ruby yoke]

i didn't. i didn't even caption the ones I created, I just let the trainer i was using do that.

[serene basin]

Oh with an auto captioner? I manually caption all of mine but if I wanted to train a "handsome man", and one guy had a big nose, one a small, one a square jaw, one narrow eyes, etc, what I'd be able to generate would be basically one face that looked like an average of all of them. Driving me nuts, really don't want to have to caption every lasat feature.

which i would think would erase the qualities of the face anyway. I've tried just captioning "handsome face" for example, but it makes no difference

I'm assuming you mean something like BLIP. My Loras got much better when I switched to manual caps

[little bough]

yup, updated on all of them same error.. A google search for the error brought me this but it's talking about FastSD from thelastben...

Is it possible this command could save me? the issue is I enter it into command prompt in the venv folder and the command is invalid, I feel like I'm missing something.. any ideas?

Thanks

I've just tried disabling all other extensions other than controlnet and it still doesn't work, I thought maybe there was a clash

[steady thistle]

Ack, I just tried to install ComfyUI-AnimateAnyone-Evolved through the Manager and now ComfyUI fails to launch. I restarted but no good.

[ornate elk]

You can try to reinstall Controlnet

[ornate elk]

Comfyui is so easy to break because it doesn't use a venv

[steady thistle]

Hmm, dang. Perhaps I can backtrack by removing whatever was installed with animateanyone

[ornate elk]

Or you can use my Comfyui start bat, modify it and then youll get a venv

[steady thistle]

Oh, yeah, that would be great. I can find that .bat in the install guide?

[ornate elk]

Here;
https://github.com/CS1o/Stable-Diffusion-Info/blob/main/Resources/Start-ComfyUI.bat

You can define which torch version it should use in the bat.
Default is 2.3.0

[steady thistle]

Should I use that as the default startup?

[ornate elk]

Yep, put that into the comfyui folder

And then it will create a venv

So if anything breaks you can easily delete it and relaunch

Like with auto1111

[steady thistle]

Okay, the root folder where the "run_nvidia_gpu.bat" is I take it

[ornate elk]

Directly in the comfyui folder

I'm on mobile rn, can't check the folder structure

[steady thistle]

Okay, I have the comyui_windows_portable folder and a Comfyui folder below that. I'm starting with the top one, but I'll just fix if that's wrong

[ornate elk]

Then it goes into the second folder

That only is called comfyui

[steady thistle]

Oh, oops. After download I'll drag it down one, or do I need to restart?

[ornate elk]

Drag it down

[steady thistle]

Cool

[ornate elk]

And then launch

[steady thistle]

Yep, will do! Crossing fingers but I was really surprised at this.

[ornate elk]

Delete the old venv if it has created one in the top folder before

[steady thistle]

This is my first venv with ComfyUI.

Random question tho, is Flux worth the 20+ gig download if one likes realism?

[ornate elk]

Go for the nf4 model or the fp8 dev model. These are 12gb

[steady thistle]

Okay, I've seen those. I have a 4070 Ti Super with 16 vram and 32 onboard ram

[ornate elk]

I think flux is good with realism

[ornate elk]

Should work good

[steady thistle]

And I want the best results I can get, I don't mind the big DL if it's worth it

[ruby yoke]

it was for me - i'm using the fp16 model downloaded from bfl's hugging face

[steady thistle]

Good, I'm excited about that

[ornate elk]

I think its not worth using the 23gb model over the nf4 or the 12gb version. You can still use the clip t5 fp16 text encoder if you want

[steady thistle]

Is there any install guide for it? I thought it was pretty straightforward but wasn't sure

[ornate elk]

Here is a good info:
https://comfyanonymous.github.io/ComfyUI_examples/flux/

[steady thistle]

Thanks!

Gah, that didn't work. I moved venv down a level and ran the startup bat again and it just went poof too fast to read.

Activating virtual environment...
Running main.py with --auto-launch argument...
[START] Security scan
[DONE] Security scan

ComfyUI-Manager: installing dependencies done.

** ComfyUI startup time: 2024-09-11 16:03:44.519863
** Platform: Windows
** Python version: 3.10.11 (tags/v3.10.11:7d4cc5a, Apr 5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
** Python executable: C:\Users\alton\AppData\Local\Programs\Python\Python310\python.exe
** ComfyUI Path: D:\Ai\ComfyUI_windows_portable\ComfyUI
** Log path: D:\Ai\ComfyUI_windows_portable\ComfyUI\comfyui.log

Prestartup times for custom nodes:
0.5 seconds: D:\Ai\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Manager

Traceback (most recent call last):
File "D:\Ai\ComfyUI_windows_portable\ComfyUI\main.py", line 88, in <module>
import comfy.utils
File "D:\Ai\ComfyUI_windows_portable\ComfyUI\comfy\utils.py", line 20, in <module>
import torch
ModuleNotFoundError: No module named 'torch'
Deactivating virtual environment...
'deactivate' is not recognized as an internal or external command,
operable program or batch file.
Press any key to continue . . .

I added 'pause' and this is what I got

[ornate elk]

Delete the venv as I said before

Then relaunch

[steady thistle]

Oh sorry, thought you meant from the upper folder

[ornate elk]

Yea

You moved it?

That won't work

[steady thistle]

Ah, okay.

Okay, deleting old and redoing from correct folder

[ornate elk]

It searches for the requirements text file that is only in the comfyui folder

[steady thistle]

Makes sense. I was just hoping to save the bandwidth. heh

Slow connection

[little bough]

tried, still doesn't work, same error lol. very bizarre, found a few people having the issue on the controlnet github so will have to wait for an official response, only some seem to have this issue, do you think it could be a torch/python issue?

[ornate elk]

What's your torch and python version?

[little bough]

how can i check what version of torch? python is 3.10.9

[ornate elk]

It should show at the bottom of the webui in browser

[little bough]

ah nice for a1111 its

nothing out of the ordinary, right?

[ornate elk]

thats an old torch version

[little bough]

oh

[ornate elk]

Open up a CMD.
Type
Pip cache purge
Then delete the venv folder
Then relaunch the webui-user.bat

[little bough]

all a1111, forge, reforge assuming i am installing them all seperately as i do, dont interact with each other at all, right? (aside from sharing the same overall version of python that i am using? ) so if i updated forge for example it shouldn't have messed with my a1111 and reforge, right?

[steady thistle]

It started up again, so that did work. Previously working nodes are now red as if not installed. I'll restart my PC and see if that takes care of it and otherwise I'll try to uninstall and reinstall the nodes. After that, we'll just go from there.

[ornate elk]

Alright 👍

Yea it could be that the nodes need a reinstall

If they place stuff into the venv

[ornate elk]

Yes. They use their own venv

So only python is shared, but that doesn't mess up

[little bough]

ok thats what i thought, just checking.. i am installing again but notice when its downloading, isn't this the same version of torch it is downloading?

[ornate elk]

Yep that's the old version lol

That shouldn't happen

[little bough]

lol damn

maybe ill try a fresh install of a1111 then?

not sure if that will make any difference though

[ornate elk]

Could help yes. But then run the pip cache purge command again before doing it

That ensures nothing old stays in the python cache

[little bough]

ok, when wiping the python cache is that the python cache globally or just for that install that i run it in

[ornate elk]

Globaly

If you want you can also update your python to 3.10.11

Thats the latest 3.10 python version

[little bough]

ye i was thinking maybe its time to update.. wasn't there a reason to not update to .11 before?

[ornate elk]

Nope

[little bough]

i think last year at least, something didn't work with it and it wasn't recommend to update

hmm ok

[ornate elk]

I'm using 3.10.11 since its release

3.11 was not recommended

[little bough]

yeah thats it

so can i safely just install 3.10.11 and it it will override 3.10.09 or are there any caveats? i dont want to break anything more lol

[ornate elk]

Yep its just an overwrite

Like a small update

[little bough]

ok thanks

[ornate elk]

No problem

[steady thistle]

Okay, I tried to redo some nodes and I think it included AnimateAnything, and it's crashing again immediately, however, the bat screen vanishes too fast to read and I don't know where to put the 'pause' command

[ornate elk]

Then you can delete the node and the venv folder and it should work again

Without downloading because its cached

[steady thistle]

Okay, going to delete the node, but I need to delete venv as well? Does that mean I go back to the other bat?

[ornate elk]

You can try run the "run nvidia bat" but I dont know if it will work again.

[steady thistle]

Nope, didn't work. Deleting venv and then it'll rebuild venv again, right?

[ornate elk]

Yes

[steady thistle]

Got it

[ornate elk]

Thats the trick if something breaks

[steady thistle]

Ah, yeah, rebuild the venv without the culprit

Okay, that got me back into ComfyUI! Walking away from AnimateAnything and see if AnimateDiff with some controlnet will work for me.

[ornate elk]

Alright 👍 maybe animate anything just needs some fixes xD

[mighty mortar]

hey can anyone help me using stable diffusion for the first time? im on a macbook pro 14" 2022 model with 16gb of ram. i have python installed and my application is image to video. when i go to hugging face, i cant find a model to download. do i need to download another model and then install this as an add on? if someone knows of the best resoucrces and videos to help me with broader subjects, please let me know! thanks!

[ruby yoke]

where on huggingface are you going? what model are you wanting to use?

[rain vessel]

Hi, I've subscribed to the standard plan, but still unable to use Stable Assistant:

when subscirbing, I've used my current email (@yahoo) but, I might have used a @gmail account in the past to log into Stable Ai.

• when going into the account details under stable assistant I see the yahoo (good) account
• when I click and use the Feedback form on the website, it shows me the gmail (bad) account when submitting my message

I have the feeling, dispite showing yahoo as the account for the standard subscription, Stable Ai still having my old gmail under the hood somewhere

Could anyone please help me? Thanks!

[little bough]

hi, so i created a new install for a1111 using stability matrix and the same controlnet error occurs, same old version of torch too but i forget to wipe the cache again.. Stability Matrix has a way to see and modify python packages, so I can update torch from the list I think, what version of torch should I go for? Thanks

[ornate elk]

Hey, torch 2.3.1 is good to go for

[little bough]

ok... i selected 2.3.1 but for some reason it installed 2.4.1, i then downgraded to 2.3.1 and loaded it up now i get

any ideas? lol nightmare

[ornate elk]

Hmm strange

Make sure your GPU driver is updated

Then maybe delete the venv folder and relaunch

[little bough]

ye just updated that earlier today and i just deleted the venv and tried reinstalling, it immediately tries to install torch 2.1.2 again lmao

I guess it is the A1111 dev i should ask

thanks for your help anyway, much appreciated

[ornate elk]

Thats really strange. What GPU do you use?

[obtuse drift]

Is 65,9s/it fast enough for a AMD 6700xt with 12gb combined with a i5-14600? Could i improve the speed with some tweaks?

[little bough]

3070ti

[ornate elk]

That should be supported by any new torch version.
What you could try is to install the webui not with stability but with my guide from the pinned messages

But you need to do a
Pip cache purge
In a CMD before

[obtuse drift]

that are my my settings ...

[ornate elk]

Hey, your not using your GPU power with this setup.
For AMD you need a specific version of the webui, compatible with AMD.
Checkout my AMD Guides for the setup with the best performance/usage on windows:
AMD Automatic1111 with ZLUDA:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides

You should then get 2-3 (or 3-4) it/s and not s/it

For any questions feel free to ask me 🙂

[obtuse drift]

wow and thx. I will check 🙂

[tulip field]

Hiring someone that can reduce the AI detection % of some AI photos by using several techniques

[obtuse drift]

i need further help.

something got wrong

Is this right? i am unsure ...

[ornate elk]

Nope thats wrong

Delete that entry

You have to click on the Path entry and then Edit

Also not on HIP_Path.
Only "Path" is the right one

[obtuse drift]

this ok?

[ornate elk]

No

And there make 2 new entrys

[obtuse drift]

perhaps i accidently deleted some windows standart settings?

[ornate elk]

You may did

You should put these 4 variables at the top again

Only the 4 top ones starting with %

And then the zluda folder and the hip path

[obtuse drift]

thanks for your help, windows restart as usual ... and i needed to reinstall phyton and git

[ornate elk]

Alright good

You now need to download a checkpoint, because the webui doesn't ship the 1.5 model anymore

I need to update my guide on that, but you can just download for example Dreamshaper v8 and put it into the models/stable-diffusion folder.
https://civitai.com/models/4384/dreamshaper

Or any other model you like. It just needs to be at least 2gb in size

[obtuse drift]

ok, thx i downloaded

than it say i have two tensor devices ... How to select AMD GPU and not INTEL onboardGPU?

or could i use both together?

i try to put wrapper_CUDA__index_select into arguments

How to put in this argument for selecting my AMD gpu?
set COMMANDLINE_ARGS= --use-zluda --update-check --skip-ort --wrapper_CUDA__index_select

[ornate elk]

You don't need that argument

Try deleting the venv folder and relaunch the webui-user.bat

If you still get an error, copy the full cmd log in here

[obtuse drift]

kk

another question about where to put Optimised_ROCmLibs_gfx1031.7z, in 5.7 or 6.1 Folder?

[ornate elk]

In the 5.7 folder

[obtuse drift]

ok

[ornate elk]

If you installed hip SDK 5.7

[obtuse drift]

i installed both

[ornate elk]

Oh

You only have to install one of them

[obtuse drift]

okok i see %)

[ornate elk]

So you can uninstall 6.1

But I will update my guide in a few days for older GPUs to work with 6.1 too

[obtuse drift]

thanks for your effor 💝

[ornate elk]

No problem

[obtuse drift]

i think i need to redo everything from beginning. .. %)

[ornate elk]

Oh okay

[obtuse drift]

it dont want to work ...

[ornate elk]

What error do you get ?

You can show me the whole cmd log

[obtuse drift]

the second pic is the first page log

[ornate elk]

Looks normal

Has the webui opened in browser?

[obtuse drift]

yes

but than nothing happens ...

when i start generate ....

[ornate elk]

Thats ok

Read the last line of the guide

First image will take 15-40 minutes

Only for the first time

After that its fast

[obtuse drift]

ok, i was irritated because there are no messages in the gui or cmd that say it is working ....

ok thx

[ornate elk]

Yea it will load and then it will show something in the next 15-40 minutes

If the generate button is grayed out and the last line in cmd is the dogetexx, its working.

[obtuse drift]

ok thx, i wait than 🙂

[ornate elk]

No problem, let me know if it worked 🙂

[obtuse drift]

its working now 🥳

[ornate elk]

Perfect 🙂

[steady robin]

hi i am struggling to generate images on dreamstudio

it's unworkably slow, any idea what's going on

[south vigil]

Hey everyone! Is there a way to automate a list of character names through a specific part of a prompt in ComfyUI? To clarify, I have 20 characters, and I want to generate one pose for each. Since I'll be frequently changing the poses, I'm wondering if there's a way to automate inserting the character names into the text prompt node, rather than manually typing them each time. Any tips or workflows to streamline this process?

[obsidian carbon]

Hiya

I trained a PONY lora and used some settings and when I try generating any epochs or weights, i am getting ONLY black images - any one ever have this?

[brittle lantern]

forge with AMD i get this error i need fix

[ornate elk]

you need to use the non gguf model

[brittle lantern]

How? u downloaded the one of ur guide so i don't know how many diff versions there are

[ornate elk]

fp8 flux dev https://huggingface.co/Comfy-Org/flux1-dev/blob/main/flux1-dev-fp8.safetensors

[ornate elk]

my guide doesnt list any flux model

oh wait

[brittle lantern]

So I have to download flux to get into the UI?

[ornate elk]

you get that error when launhcing ?

[brittle lantern]

Yes

[ornate elk]

ohhh

okay let me do a quick test on mine

[brittle lantern]

Ok

[ornate elk]

yep gettting the same error after updating

[brittle lantern]

Ah ok

I just kinda installed it just now

And got it so ill assume it's installed recent update

[ornate elk]

i fixed it by going into forges folder. then clicking into the explorer bar
type cmd and hit enter
then run the following two commands one by one:
venv\Scripts\activate.bat
pip install gguf

[brittle lantern]

Done I now open it?

I also did the linking my stable diffusion folder for the models and nothing shows up now

In forge UI

[ornate elk]

can you show me your webui-user.bat ?

[ornate elk]

yea thats not correct, you have to remove the @worthy smelt

[brittle lantern]

Ok

[ornate elk]

like here

only remove the ones i did, not all

[brittle lantern]

It still doesn't show in models in forge

[ornate elk]

can you show it again? and did you relaunched?

[brittle lantern]

i did relaunched

[ornate elk]

you have to change the \ to / at your main path

[obtuse drift]

While rendering under windows my mouse curser hangs every ~12 seconds. Is there a tweak against?

[ornate elk]

make sure you dont run any games or heavy programms like wallpaper engine in the background

[brittle lantern]

like "D:/" instead of "D:\ "?

[fallen saddle]

Hi there, does anyone have a good effective tutorial/guide link on how to improve performance on webui forge? Sorry if this is not the right place to ask, I'm new at this 😅

[ornate elk]

@obtuse drift if you still get lags you need to show me your txt2img settings

[ornate elk]

What's your GPU ?

And hey

[fallen saddle]

NVIDIA GeForce GTX 1050 ti

[ornate elk]

Ohhh

You used batch size

Dont do that with sdxl

Only use batch count

[obtuse drift]

kk

[ornate elk]

Batch size means = generating images at the same time (needing a lot of vram)
Batch count means = generate one by one (no problem)

[obtuse drift]

ah, good to know, thx 👍

[ornate elk]

Np!

[ornate elk]

Try enabling fp8 mode when using sdxl models.
and in the webui-user.bat add this at the commandline args: --cuda-malloc --cuda-stream --pin-shared-memory
that could help

[unreal anchor]

did this but still got FileNotFoundError: No checkpoints found. When searching for checkpoints, looked at: - file "------------------------------------------------------------- - directory--------------------------------------------------diffusionCan't run without a checkpoint. Find and place a .ckpt or .safetensors file into any of those locations.

[ornate elk]

can you show your webui-user.bat?

also try putting one model into models/stable-diffusion

[unreal anchor]

@echo off

set PYTHON=C:\Users\Atanok\AppData\Local\Programs\Python\Python310\python.exe
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --no-half-vae --api
--models-dir "F:\Stable Diffusion Output Backup\models"

call webui.bat```

[ornate elk]

It has to be in the same line with xformers to work

Not in a seperate row

[ruby yoke]

no hard line breaks in your bat file between command line arguments

[unreal anchor]

Ohh

[burnt coyote]

Hi folks, question: is there a "best practices" way to completely wipe everything to do with Stable Diffusion? A1111, ComfyUI, all the python, everything but the models? I've been doing this a long time relatively speaking, and my installs are increasingly messy and redundant, consuming a huge amount of drive space, and getting less and less stable and more prone to errors. I'm looking to start fresh.

[ruby yoke]

you could do what purz does - install to a virtual environment, then just nuke that when you want a fresh install

[burnt coyote]

This is definitely what I want to do on the reinstall. Right now though I'm just trying to clean this up without nuking the drive itself, which is not an option.

[ruby yoke]

what I did was move my models to a different drive, uninstall python/cuda/etc - and just delete the folder that comfy is in

you don't have to worry about windows registry or anything

[burnt coyote]

Do python, cuda, etc uninstall cleanly? Or does it turn into a scavenger hunt across the hard drive for rogue files?

I've just discovered that I have no fewer than eight copies of diffusion_pytorch_model.bin, for example

[ruby yoke]

thats intresting. comfy uses an embedded version of python. not sure why you've got that many copies of it - maybe you've got other things that use their own embedded version

[burnt coyote]

Yeah, it turns out that doing this for this long, without starting with the knowledge of how to keep everything neat and tidy, leads to a lot of redundancy

Until quite recently I was under the impression that everything would share dependencies, realize that, for example, torch was installed and just... use it

instead of grabbing its own copy

But that is quite evidently not the case!

[ornate elk]

Open up a CMD and type
Pip cache purge
To get 5-10gb back.
That deletes the temp packages for stable diffusion installations.

[ruby yoke]

now if you knew how to switch the google drive cache file to a different drive other than C and specify how large IT should be ...

[young cipher]

hey not sure if this is the right place but any ideas why this is happening? when i generate and image it goes from this during it, then the colours go weird?

[ruby yoke]

please explain what yo umean by the colors go wierd

[young cipher]

look at how the colours change

[ruby yoke]

sure. the first one is a sketch, the last one is the refined verison.

[young cipher]

yeah im asking how to stop the colour change

[ruby yoke]

what nodes are in your workflow?

[ornate elk]

Thats an VAE issue

You need to use a custom vae:
For example this:
https://civitai.com/models/23906/kl-f8-anime2-vae

[opaque viper]

Question. I'm trying to use the fp8 flux-dev for the first time on Forge. And I got this error: linear(): argument 'weight' (position 2) must be Tensor, not NoneType

Any idea what's happening? I've previously used gguf no problem

[ornate elk]

Where did you used gguf?

[opaque viper]

On Forge

[ornate elk]

And now which model gets the error?

[raw vessel]

So looking for some help. I had SD on my PC for a long time. All the way up to 2.1 (or was it 2.2?) I used AUTO111 to run it. Well my drive died and its been a few months and I finally reinstalled everything. I got the SD3.0 checkpoint added...etc.

Problem is the only sampler that seems to work is the Heun (Heurn) one for generating anything. All the other samplers, no matter the size, CFG, steps...etc just give me super blurry colored blotches, sometimes ROUGHLY in the shape of what I asked.

Any idea what I did wrong? I even followed the directions here: https://github.com/AUTOMATIC1111/stable-diffusion-webui

I'm at a loss. I know LONG ago when SD first went public, like 0.5 version maybe? I had black squares at that tiem until someone helped me.

[opaque viper]

Sorry I'm so sporadic in respnding. The one giving me the error is the fp8 model

[karmic crown]

find the guide for your GPU here
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides

[ornate elk]

Better use SDXL instead of SD3.
SD3 is not that good supported by auto1111

[last tide]

Why did it fail to load ZLUDA?

And the model failed to load for some reason?

Clicked the button next to the Stable Diffusion checkpoint but sends a error aswell "Error completing request"

[karmic crown]

do you have rocm 6.1 installed and does that file exists?

[last tide]

i downloaded and installed the AMD HIP SDK 6.1.2

if thats what it means

[karmic crown]

open program files, then amd, than rcom.... and check if file is there

[last tide]

Ah

[karmic crown]

go to bin

[last tide]

checking now

[karmic crown]

and see if you have that file, looks good so far

[last tide]

i am missing the other bin folder

[karmic crown]

wait

i havent seen that is shows 2 of those

this is not right

it must be one

[last tide]

correct

[karmic crown]

your installation is fine, something else is the problem

[last tide]

just the git clone

xD

[karmic crown]

https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides#amd-automatic1111-with-zluda

did you followed this guide?

[last tide]

Yes thats the one i followed

"[AMD] Automatic1111 with ZLUDA"

i have a 7900 gre if that might be a problem

[karmic crown]

Add the C:\ZLUDA folder and %HIP_PATH%bin to your PATH in the System Variables.

[last tide]

I did

[karmic crown]

pleace check if you have it 2 times

[karmic crown]

this is your problem

use %HIP_PATH%bin

[last tide]

hmm, weird that it shows like that.

[last tide]

I think this is what i did

will try

[karmic crown]

if it does not work, manualy edit to be correct, with one bin

[last tide]

yes thats it

imma just remove the bin at the end of the value

[karmic crown]

just meke it have one bin at the end 🙂

[last tide]

Looking good

[karmic crown]

i would just remove venv
and let a1111 create it again

to be sure you have correct files there

[last tide]

ah i see

will do

[karmic crown]

it is posible that it alrady instaleld wrong torch (cpu only) and similar things

do you have any models already or this is your first install?

[last tide]

First install, just saw i had to download some models

[karmic crown]

ok

the default model that was downloaded was removed from huggingface

so, you will need to download some model manually

for example from civitai

https://civitai.com/

for example https://civitai.com/models/4201

[last tide]

Anything worth knowing as of names or tags?

[karmic crown]

just be sure to foloow the instrauction of the author, which samplers to use

and check what they used in examples

[last tide]

Alright

[karmic crown]

click on "i"

[last tide]

How long does it usually take for the webui to start?

[karmic crown]

show me what you have

[karmic crown]

it probably downloading required python modules

hmm

have you stoped it, remove venv and rerun it?

[last tide]

yes i did

[karmic crown]

since i do not see any installation after run of venv

[last tide]

ill do it again, might fix it

[karmic crown]

close cmd completely

[last tide]

yep

[karmic crown]

CS1o any other thoughts?

[ornate elk]

Thats not correct

Your in user variables

[last tide]

Will do it in system variables

[ornate elk]

You need to add them in the Path setting

[last tide]

Oh i see

[ornate elk]

There create the 2 new entrys

[last tide]

This would be correct

[karmic crown]

and remove from the other place (i guess)

[last tide]

Already did :)

[karmic crown]

@ornate elk I hope you have seen this https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/16460

[last tide]

Its installing again so ill just wait untill something happens

[ornate elk]

Yep

Thats good xD

[karmic crown]

so you are aware about the default model problem 🙂

no defult model - no bad images? 🙂

[ornate elk]

True xD

[karmic crown]

hahahaha

[ornate elk]

No more questions why the first image looks so bad

[karmic crown]

thats why forge does not use that one 🙂

[oak jay]

I finally managed to get Flux working using Forge, but, it's painfully slow, anything I can do to speed it up? I have a RTX 4080 GPU so shouldn't have an issue, and I don't with 1.5 and XL models.

[ornate elk]

You need to use the nf4 flux model

Thats the fastest you can get

[ornate elk]

Let me know if it works

[last tide]

Will do!

[oak jay]

Thanks, I'll try that, but, is there anything I can do in the settings to speed up the other models?

[last tide]

Looks like its working, gonna upload a model now and restart the ai

Can i find a place with the differences between each model? (codeformer, karlo, Lora, VAE, etc)

[ornate elk]

Try enabling fp8 mode when using sdxl models.
and in the webui-user.bat add this at the commandline args: --cuda-malloc --cuda-stream --pin-shared-memory
that could help

[oak jay]

SDXL models are fine, I can create pictures really quickly, it's just on Flux models it's taking forever

[ornate elk]

Which checkpoint do you use?

How big is the size of it

[karmic crown]

ask ChatGPT, i am not joking
4.o mini is good enough for things like this

but read this when you have time
https://stable-diffusion-art.com/lora/

[oak jay]

I've used flux_dev, fluxunchained-dev-q8-0, fuxCapacityNSFWFluxBaseModel, the bf16 and fp8 versions and jibmixflux_v10, all are taking about 20 mins for one image!

[last tide]

Also it looks like there are 99/100 anime models xDD

[karmic crown]

you can check https://stable-diffusion-art.com/ for some explanations, but as i said for a quick reference, just ask my friend chatgpt 🙂

[last tide]

Will do, reading it rn, looks interesting

[ornate elk]

thats definitly not normal

[oak jay]

Yeah!

[oak jay]

And, I'm getting this error at times too!!

[last tide]

Just curious, what are the regular times with a 4080?

[ornate elk]

can you show me the settings when using the flux fp8 16gb model?

[oak jay]

Is this what you mean?

[ornate elk]

yes

[ornate elk]

when using any flux model above 12gb you dont need to use clip or vae and text encoder that will slow down

and then choose a resolution of 1024x1024 or smaller or a variant like 860x1200

[oak jay]

Ok, I'll try and let you know

[oak jay]

So, if I remove the clip, vae and text encoder, I get this error, 'AssertionError: You do not have CLIP state dict!'

[ornate elk]

hmmm have you used flux fp8 dev (16gb) ?

[oak jay]

I used this one, 'fuxCapacityNSFWFluxBaseModel_v03Fp8Un.' No matter what model I use, if I do so without clip, vae and text encoder, I get a version of that error

[ornate elk]

hmm okay strange

i will test my flux model again

[oak jay]

Ok, thanks

[last tide]

Uhm, is this normal? tried different prompts but all look so pixelated as a 2011 meme template

[ornate elk]

yes thats normal for SD3, you should use any other model

or use SD3 with a specific sampler. but its not worth

[last tide]

Any recommended model for random stuff (simple logos like chatgpt or something)

Gonna give juggernaut XL a shot

[ornate elk]

this for example is an allrounder:
https://civitai.com/models/84040?modelVersionId=395107

but yea juggernaut xl is recommended for realism stuff

if you use sdxl, you need to use a resolution of 1024x1024

[last tide]

Understood

[karmic crown]

or 768x1024

1024x768

just do not use strange values like 941x987

[oak jay]

Did you come up with any suggestions?

[ornate elk]

hmm you can try set your gpu weight limit to 14500mb

because right now its on 12gb

[acoustic sky]

--use-zluda --debug --autolaunch

[oak jay]

Seems the max I can do is 12281. I assume I just use the slider in the setting at the top?

[ornate elk]

the first image can take 15-40 minutes

if its not working, you should follow my Zluda guides for Auto1111

[acoustic sky]

its been 20 minutes

i will wait for 20 mins more if it doesnt work then i will

[oak jay]

Ok, so I set it to the max and I get these warnings '[Low VRAM Warning] You just set Forge to use 100% GPU memory (12281.00 MB) to load model weights.
[Low VRAM Warning] This means you will have 0% GPU memory (0.00 MB) to do matrix computation. Computations may fallback to CPU or go Out of Memory.
[Low VRAM Warning] In many cases, image generation will be 10x slower.
[Low VRAM Warning] To solve the problem, you can set the 'GPU Weights' (on the top of page) to a lower value.
[Low VRAM Warning] If you cannot find 'GPU Weights', you can click the 'all' option in the 'UI' area on the left-top corner of the webpage.
[Low VRAM Warning] Make sure that you know what you are testing.'

[last tide]

Just watch a serie or a movie, thats what i did xD

[ornate elk]

hmm but you said you have a 4080 (16gb) why does forge thinks it only has 13gb ?

[oak jay]

I don't know, this is what is says in 'settings' - Name NVIDIA GeForce RTX 4080 Laptop GPU

[ornate elk]

ohhh laptop gpu

[oak jay]

Oh, is that an issue?

[ornate elk]

it has only 12gb vram, not an issue but i didnt know

can you try the nf4 flux model?

[oak jay]

Oh, I didn't know that! That's annoying! I'll give it a try, just waiting for the last generation to finish

[oak jay]

Oh, the nf4 download failed, I've started again, but for some reason, downloads of Flux models are being really slow

[ornate elk]

you neeed the v2:
https://huggingface.co/lllyasviel/flux1-dev-bnb-nf4/tree/main

[oak jay]

Ok, thanks. It's downloading now

[oak jay]

Oh, should I reduce the GPU weight limit or leave it at the max?

[ornate elk]

dont leave it at max

make sure you have like 500mb at least free

[little bough]

has anyone had this error in forge trying to use flux(edit i think it's all models in general, same issue when swithching to sdxl)? been getting it for the past few days after an update a few days ago, i've keep it updated but it has not resolved itself, not sure what is going on, sampler is euler and throws the error no matter what sample i use. I t happens after clicking generate and ofc nothing generates, just the error

[oak jay]

Hey, now I'm just getting black images. Haven't tried using the model you suggested as it's still downloading!

[ornate elk]

with vae and clip or without them?

[oak jay]

With

[ornate elk]

then remove them and try again

[oak jay]

That worked, thank you

[ornate elk]

No problem

[acoustic sky]

is there any way to change the ".cache" folder of python? i dont have any spaces in my ssd

[oak jay]

I've discovered that if I change the 'Diffusion in Low Bits' to 'Automatic (fp16 LoRA)' it speeds things up a bit.

[ornate elk]

With ugly registry hacks maybe. That could break something. So better leave it there. But after the installation of SD open up a CMD and run
Pip cache purge
To get 5-10 GB back

[acoustic sky]

i have 14gb storage and its still not enough (120 gb ssd)

[karmic crown]

i would move models to other drive (if you have it)

[oak jay]

@ornate elk finally downloaded the nf4-v2 model, and no longer need to use any VAE/Text Encoder, but, it's still taking 3-4 mins to generate an image, any ideas on how to improve that? I can do in image in about 30 seconds using 1.5 and XL models

[acoustic sky]

i dont and it doesnt work anyway

[karmic crown]

if you do not have other drive, forget my comment

[ornate elk]

how much RAM do you have?

if you only have 16gb you can speed up by increasing the windows pagefile

[oak jay]

I have 32gb

@ornate elk so I'm pretty much stuck with the current speed?

[ornate elk]

How long does it take with nf4 ?

Without loras

1024x1024

[oak jay]

So, with the parameters you said and sampling steps set at 20 it took 4-5 mins

[ornate elk]

Okay

Hmm

It seems to long for me

But idk

[oak jay]

Well, that's what I was thinking, especially as I get about 30-40 seconds for 1.5 and XL. I thought Flux was supposed to be quick too!

Actually, weirdly, without changing anything, I ran it again and it says it took 45.6 seconds

[ornate elk]

No flux is really slow

In comparison to sdxl and 1.5

[oak jay]

I knew that, but I didn't think it was THAT much slower

Well, thank you anyway, you seem to have improved it somewhat, I'm now getting 1 min 14 sec when adding LoRA back in

[ornate elk]

NP, thats a good time

[raw vessel]

So I installed SD again according to advice on here. When I went to generate something it says "
RuntimeError: "LayerNormKernelImpl" not implemented for 'Half'"

[hot spindle]

Depends on expectations, Schnell is pretty fast, but I personally dont fancy the results

It resolves ok in 4 steps

[ornate elk]

Hey, what's your GPU?

[raw vessel]

2070 Super. I know I can still run my old SDXL for some reason.

[ornate elk]

Make sure you have --xformers --medvram-sdxl --no-half-vae in the webui-user.bat

Ans then make sure you don't load a lora as model by accident. Check the file size

[raw vessel]

I see. I have this in my bat:
@echo off

COMMANDLINE_ARGS=--xformers --no-half-vae --medvram-sdxl --precision full

if exist webui.settings.bat (
call webui.settings.bat
)

if not defined PYTHON (set PYTHON=python)
if defined GIT (set "GIT_PYTHON_GIT_EXECUTABLE=%GIT%")
if not defined VENV_DIR (set "VENV_DIR=%~dp0%venv")

Oh, how do you know if its a lora ot not? I think it was something like 6.75gigs.

Ah wait it says "sd3_medium_incl_clips.safetensors"

[ornate elk]

Remove --precision full

[ornate elk]

Everything above 2gb is a model

[raw vessel]

Ah ok cool. I'll try it again

[steady thistle]

@ornate elk You helped me before with the venv because something crashed and wouldn't run anything. I'm using your startup bat. I tried loading an upscaler, RealESRGAN_x4, and it crashes again. I removed that file, but it still crashes. Can I try using that file (I need an upscaler) and then go through another venv install?

[raw vessel]

On a side note, is there some reason you can't launch webui.bat as a admin? Not sure if that even amtters or not

[ornate elk]

What does the cmd shows ? When it crashes

[steady thistle]

It goes off too quickly to say. I will add a pause

[ornate elk]

You shouldn't launch the webui.bat at all. Only launch the webui-user.bat
To avoid permission errors its not recommended to Start it as admin

[steady thistle]

I can't add a pause

I ran the run_nvidia bat and got this

but that was using the old bat

[ornate elk]

and when you use the upscaler it crashes ?

[steady thistle]

I can't start up comfyui at all. It crashed after installing the upscaler with manager and restarting from then, and hasn't restarted since

[ornate elk]

A node or the upscaler model?

[steady thistle]

ComfyUi simply won't start at all. I can't even get in to just uninstall the model.

I didn't get far enough to even add it to a node

[ornate elk]

Where did you put it in?

The PTH file

[steady thistle]

Well, the Manager loaded it. I'll try to remember the actual folder it got put in

It went into the 'upscale_models' folder under 'models'

[last tide]

So i am having a new problem, few hours ago it worked fine but now i am getting a tensors error:

[steady thistle]

I kept the upscaler in the upscale_models folder and deleted venv and am redoing it now. Crossing fingers, it'll work. If not, I'll try something else.

[ornate elk]

when do you get that error

what are your txt2img settings

[last tide]

When generating a txt2img

Ill head to sleep now, might be fixed tomorrow just by restarting the pc

[last tide]

Might be related, idk

[ornate elk]

looks normal

doe you have an igpu ?

[last tide]

Works after restarting now

[last tide]

nope

5800x3d

[steady thistle]

I deleted the update, it did not work. I tried to leave realesrgan in the upscale_models folder and I hit "restart" on Manager and ComfyUI crashed. Trying to restart the .bat just resulted in a shutdown too quick to see, and the 'pause' in the bat didn't work.

[ornate elk]

this cpu has an igpu

you may need to deactivate it

in the device manager

AMD Radeon TM Graphics

[ornate elk]

idk why it happens for you :/ maybe to much custom nodes?

[last tide]

amd says that it doesnt

[steady thistle]

I have no idea. I didn't realize there might be an upper limit, but I'm only installing nodes that various workflows require. There's a new Nvidia driver that I'm going to get, but I'm really bewildered.

[ornate elk]

oh okay i thought it has one

make sure your gpu driver is updated then

[steady thistle]

Okay, 'numpy' is messing up I think. I deleted the upscaler and reinstalled the venv again. Got into ComfyUI and loaded the default workflow. Removed the save node and put a preview node, got a 'numpy' error and then ComfyUI died and won't restart.

This is really insane. I may reinstall comfyui, saving all the big downloads like models and loras and controlnets, and then see if it was something weird. Do you have anything better I should do? I am so confused at this. I thought using the Manager was reasonably safe.

[last tide]

Is there a vram limit? My pc crashed 2 times today because of generating pictures.

[ornate elk]

There is no limit. What's your GPU?

[last tide]

7900 gre

[ornate elk]

Make sure you dont run Wallpaper engine in the background

[last tide]

i dont have that

[ornate elk]

Did you used upscaling?

[last tide]

nope

[ornate elk]

Okay

Is your PSU good enough for the full load of the gpu?

[last tide]

Now that i think about it. I was going to put my other psu in it...

lmao

[steady thistle]

@ornate elk Please review this, I got this by using "run_cpu.bat" and it did pause so I can see the error. It's so tough getting this going.

[karmic crown]

PyTorch 2.4.0+cu121 with CUDA 1201 (you have 2.4.0+cpu) <- this is the problem

why do you use --cpu?

you need different torch for cpu

[calm cairn]

Hello all, last time I was here I got a recommendation for a Pony XL model that was pretty nice. Right now I'm looking for an XL model that is good at inpainting (cartoon/anime)

My experience is that XL models are pretty bad at inpainting, so if there's a good one, please let me know, thanks in advance 🙂

[steady thistle]

Is there something wrong with my Desktop?

[karmic crown]

i am not sure why you want to use CPU instead of GPU

--cpu flag in ComfyUI specifies that the application should use the CPU instead of a GPU

xFormers and Torch are not in sync, one want to use cuda, the other wants to use cpu

[steady thistle]

I was using the CS1o bat, but that crashed so fast I couldn't see the error. I went to the run_nvidia_gpu.bat but that also crashed too fast to find the error, then only the run_cpu.bat lasted long enough to find at least one error.

[karmic crown]

do you use venv or not?

if you do activate venv and do

pip uninstall torch
pip uninstall torchvision
pip uninstall xformers

otherwise just run the commands

and install them again

[steady thistle]

I do. I used the venv after this happened a few days back. Then today I deleted venv and reinstalled it with CS1o's bat with the upscaler I wanted in it. That crashed immediately, then I removed the upscaler and reloaded the venv from scratch yet again. Then I ran ComfyUI long enough to go to the default workflow. Ran that once and it gave an error about 'numpy' and crashed again and won't restart.

But I'll try that. I run those in the root folder, I take it?

[karmic crown]

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

pip install xformers

as long as you activate venv corectly it should be fine

[steady thistle]

Alright.. I'll run those 5 commands in the root folder of ComfyUI.

[karmic crown]

venv\Scripts\activate

first activate venv

[steady thistle]

I haven't done this. Should I try that first?

[karmic crown]

yes

otherwise it will install things globaly later

and it will not remove conflicting versions

[steady thistle]

Gotcha. I'm in that environment.

I got to xformers and it was not installed

Continuing

[karmic crown]

show me what you have on screen now

[steady thistle]

Right now it's collecting torch, I have slow internet, but here...

Microsoft Windows [Version 10.0.22631.4169]
(c) Microsoft Corporation. All rights reserved.

D:\Ai\ComfyUI_windows_portable\ComfyUI>venv\Scripts\activate

(venv) D:\Ai\ComfyUI_windows_portable\ComfyUI>pip uninstall torch
Found existing installation: torch 2.4.1
Uninstalling torch-2.4.1:
Would remove:
d:\ai\comfyui_windows_portable\comfyui\venv\lib\site-packages\functorch*
d:\ai\comfyui_windows_portable\comfyui\venv\lib\site-packages\torch-2.4.1.dist-info*
d:\ai\comfyui_windows_portable\comfyui\venv\lib\site-packages\torch*
d:\ai\comfyui_windows_portable\comfyui\venv\lib\site-packages\torchgen*
d:\ai\comfyui_windows_portable\comfyui\venv\scripts\convert-caffe2-to-onnx.exe
d:\ai\comfyui_windows_portable\comfyui\venv\scripts\convert-onnx-to-caffe2.exe
d:\ai\comfyui_windows_portable\comfyui\venv\scripts\torchrun.exe
Proceed (Y/n)? y
Successfully uninstalled torch-2.4.1

(venv) D:\Ai\ComfyUI_windows_portable\ComfyUI>pip uninstall torchvision
Found existing installation: torchvision 0.18.0+cu118
Uninstalling torchvision-0.18.0+cu118:
Would remove:
d:\ai\comfyui_windows_portable\comfyui\venv\lib\site-packages\torchvision-0.18.0+cu118.dist-info*
d:\ai\comfyui_windows_portable\comfyui\venv\lib\site-packages\torchvision*
Proceed (Y/n)? y
Successfully uninstalled torchvision-0.18.0+cu118

(venv) D:\Ai\ComfyUI_windows_portable\ComfyUI>pip uninstall xformers
WARNING: Skipping xformers as it is not installed.

(venv) D:\Ai\ComfyUI_windows_portable\ComfyUI>pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
Looking in indexes: https://download.pytorch.org/whl/cu121
Collecting torch
Downloading https://download.pytorch.org/whl/cu121/torch-2.4.1%2Bcu121-cp310-cp310-win_amd64.whl (2444.9 MB)
━━━━╸━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0.3/2.4 GB 5.5 MB/s eta 0:06:27

Completed. Going to pip install xformers next

Got an error at the end there.

[karmic crown]

i hate those OSError: [Errno 2]

do you have some 3rd party antivirus installed? mcafee for example?

[steady thistle]

I deleted the bundled mcafee that came installed

[karmic crown]

try again

same error?

[steady thistle]

Should I try to regedit and put that change for long path support?

[karmic crown]

it would not hurt

try again, if it fails do it

[steady thistle]

I'll give that a go

Okay

[steady thistle]

I got this. Should I do a reboot and then try xformers again?

[karmic crown]

No module named 'torch' !?

what the...

[steady thistle]

I have no idea, this is a new PC and I haven't done anything crazy on it.

or try... conda install pytorch-cpu torchvision-cpu -c pytorch ?

[karmic crown]

you are using conda + venv?!

use one of the other

[steady thistle]

I got it off a website, sorry, I'm no good at this

that's the wrong one for me

Forget I said that.

[karmic crown]

pip list

what you get?

do you see torch?

[steady thistle]

torch 2.4.1+cu121

It was there, yes

[karmic crown]

so it is installed properly

what about xformers?

probably no

[steady thistle]

No, not there.

[karmic crown]

pip cache purge

to clear pip cache

[steady thistle]

Ok

[karmic crown]

and then try

pip install xformers again

[steady thistle]

1895 files removed

[karmic crown]

pip install xformers

again was my comment 🙂

[steady thistle]

Same issue. Should I restart the PC?

[karmic crown]

try

but do not forget to activate venv again after that

before installing xformers

[steady thistle]

okay, in the venv environment

Same thing

[karmic crown]

try this

pip install -U xformers --index-url https://download.pytorch.org/whl/cu121

[steady thistle]

It's trying. a 2.4 gb download started

[karmic crown]

ok, that should work (hopefully)

[steady thistle]

That would be fantastic

Installed but incompatability warnings?

[karmic crown]

Successfully installed torch-2.4.0+cu121 xformers-0.0.27.post2

close cmd and try to rerun (normal version - not cpu)

[steady thistle]

Slowly going, no errors yet

ComfyUI is up and running. At the basic workflow and going to test the image

Got an image. Going to restart via manager

Yay! Restarted successfully. This is fantastic

So the issue was xformers not installing properly?

[karmic crown]

xformers and torch were not in sync

[sinful marlin]

How do I use Stable Diffusion with multiple GPU's?

Currently Stable Diffusion is only running with one

[karmic crown]

you can use only one, but you can choose which one

[sinful marlin]

Welp, looks like I just wasted $200

[ornate elk]

You can run stable diffusion on each card at the same time. But not together
You need two installations of a stable diffusion webui for that

[karmic crown]

CUDA_VISIBLE_DEVICES=0
CUDA_VISIBLE_DEVICES=1

0 - first one
1 - second one

[sinful marlin]

So my idea was to use both video cards together to have a combined 16gb instance

[karmic crown]

sorry, but no

[calm cairn]

this looks like my error, maybe I should ask about this
how do I tell if Stable Diffusion is using my onboard graphics at the same time as the GPU? Asking because I get CUDA errors, I'd love to specify only to use the card

[steady thistle]

@karmic crown Thank you again for all your help, I really appreciate it.

[karmic crown]

#🤝｜tech-support message

i just wrote that 🙂

[calm cairn]

what file do I add that to? the batch?

[karmic crown]

let me double check how it should be used

looks like it might need 2 parts

open webui-user.bat

and add --device-id 0 (for the first one)

or --device-id 1
for the second one

but there is a note:

"select the default CUDA device to use (export CUDA_VISIBLE_DEVICES=0,1 etc might be needed before)."

so try like that first

if it not work we will add CUDA_VISIBLE_DEVICES

[calm cairn]

so like this?
@echo off

call environment.bat

--device-id 0

cd %~dp0webui
call webui-user.bat

[karmic crown]

if that does not work
you might need to add
set CUDA_VISIBLE_DEVICES=0
or
set CUDA_VISIBLE_DEVICES=1
before that line with params

[calm cairn]

alright lemme check rn, ill message soon

also, how do I know if I just told it to use only the GPU, or if I just told it to use only the CPU or the onboard graphics?

OK first test failed, will add set CUDA_VISIBLE_DEVICES=0

--device-id is not recognized as a command

[karmic crown]

show me your webui-user.bat

[calm cairn]

@echo off

call environment.bat

--device-id 0
set CUDA_VISIBLE_DEVICES=0

cd %~dp0webui
call webui-user.bat

[karmic crown]

that is not good

what fille is that?

that is not

webui-user.bat

[calm cairn]

its forge not 1111

[steady thistle]

I'm trying to use a Prompt Generator node and it has a model input, but is there some way of finding what folder it's trying to find it? I have a model for it but no idea where to put it.

[karmic crown]

that file calles webu-user.bat

[calm cairn]

oh ok, i got it then

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--theme=dark --ckpt-dir "E:/Life/Projects/ART/Stable_Diffusion/Models/" --lora-dir "E:/Life/Projects/ART/Stable_Diffusion/LoRA/"

@REM Uncomment following code to reference an existing A1111 checkout.
@REM set A1111_HOME=Your A1111 checkout dir
@REM
@REM set VENV_DIR=%A1111_HOME%/venv
@REM set COMMANDLINE_ARGS=%COMMANDLINE_ARGS%^
@REM --ckpt-dir %A1111_HOME%/models/Stable-diffusion^
@REM --hypernetwork-dir %A1111_HOME%/models/hypernetworks^
@REM --embeddings-dir %A1111_HOME%/models/embeddings^
@REM --lora-dir %A1111_HOME%/models/Lora

call webui.bat

[steady thistle]

Thank you

[karmic crown]

that one

remove

--device-id 0
set CUDA_VISIBLE_DEVICES=0

for the previous file

set COMMANDLINE_ARGS=--theme=dark --ckpt-dir "E:/Life/Projects/ART/Stable_Diffusion/Models/" --lora-dir "E:/Life/Projects/ART/Stable_Diffusion/LoRA/" --device-id 0

add it here

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set CUDA_VISIBLE_DEVICES=0
set COMMANDLINE_ARGS=--theme=dark --ckpt-dir "E:/Life/Projects/ART/Stable_Diffusion/Models/" --lora-dir "E:/Life/Projects/ART/Stable_Diffusion/LoRA/" --device-id 0

@REM Uncomment following code to reference an existing A1111 checkout.
@REM set A1111_HOME=Your A1111 checkout dir
@REM
@REM set VENV_DIR=%A1111_HOME%/venv
@REM set COMMANDLINE_ARGS=%COMMANDLINE_ARGS%^
@REM  --ckpt-dir %A1111_HOME%/models/Stable-diffusion^
@REM  --hypernetwork-dir %A1111_HOME%/models/hypernetworks^
@REM  --embeddings-dir %A1111_HOME%/models/embeddings^
@REM  --lora-dir %A1111_HOME%/models/Lora

call webui.bat

[karmic crown]

sorry, i have no answer to that

[calm cairn]

testing now

[karmic crown]

if that is the wrong card try with 1

[calm cairn]

tested, got the same error, let me check to make sure I have it in there exactly like you have it

ah i was missing a line

[karmic crown]

set CUDA_VISIBLE_DEVICES=1
set COMMANDLINE_ARGS=--theme=dark --ckpt-dir "E:/Life/Projects/ART/Stable_Diffusion/Models/" --lora-dir "E:/Life/Projects/ART/Stable_Diffusion/LoRA/" --device-id 1

and be sure that all this is ONE LINE

set COMMANDLINE_ARGS=--theme=dark --ckpt-dir "E:/Life/Projects/ART/Stable_Diffusion/Models/" --lora-dir "E:/Life/Projects/ART/Stable_Diffusion/LoRA/" --device-id 1

you can save this as cuda-test.py in your forge folder

activate venv and run it
python cuda-test.py

it should show you which card is which

[calm cairn]

i got the same error, let me check the thing you just posted

[karmic crown]

i am on Mac so i am getting "CUDA is not available." 🙂

otherwise i would show you the result

i just asked chatgpt to give me example output

Number of CUDA devices: 2
Device 0: NVIDIA GeForce RTX 3090
Device 1: NVIDIA GeForce GTX 1080 Ti

so it should look like this

[calm cairn]

i appreciate the direction but I really don't know anything about python, how do I create the virtual environment and run this? "git bash here"? cmd?

[karmic crown]

do you know how to use cmd?

open cmd

and go to folder with forge

better open cmd at that folder

venv\Scripts\activate
should activate venv

it should look like this #🤝｜tech-support message

than just try python filename.py

[calm cairn]

the system cannot find the path specified, let me see if im messing something up

[karmic crown]

f... 4am

i need to sleep

[calm cairn]

ah sorry to keep you up, I will just put it on hold until someone else can help, get some rest

[karmic crown]

but even without that

you can try both combinations

with 0 and 0

and 1 and 1

one should work

[calm cairn]

ok thanks, I'll see if that's enough to eradicate this error, it's been a pain for a couple weeks

[karmic crown]

you can try to disable the one that you do not need

[calm cairn]

wow, "runtime error"
torch is not able to use gpu

[karmic crown]

than you will not need to mess with this

[calm cairn]

that was on device 1

[karmic crown]

i am sorry but i need to go

[calm cairn]

i think maybe we could keep giong all night though, it's OK for you to go sleep

[karmic crown]

good night

[lost dagger]

i looking for the best choice iam a photographer and i trying to give my photography a extra touch with ai but i dont know what should use comfyUI or Diffusion?

im new on this so before i install i would like to know