🤝｜tech-support

[ornate elk]

Hey, no you can also use SDXL models.
But these require a lot more vram.
With directml they can fail sometimes (out of vram)

[real goblet]

Yeah this is defo something

Tysm for all the help man

[ornate elk]

looks great 🙂 np

[visual pilot]

Hello i was thinking of buying a 3090, i'm using currently a 3060 but for larger resolutions and sdxl inpaint it gets me a cuda out of memory, my motherboard is a intel x79 and cpu is a xeon e5-2650 v2 2.60mhz, would be enough for a 3090 or it wont matter?

[ornate elk]

hey, whats in your webui-user.bat
mostly you can improve speed and lessen the vram usage by using the right commandline launch args

[visual pilot]

I'm using forge, according to what i had read, it was not necessary to make changes to its arguments since it did the optimizations by itself, or is it still necessary to do so?

[ornate elk]

Yea also forge is deprecated for normal Users.
If your 3060 has 12gb vram I would switch to Auto1111 as forge will break with the next updates

Forge has 3 launch args that can increase performance that have to be manually added

[visual pilot]

I used to use a1111 but then I found out about forge which was supposedly faster for lower end graphics cards, I tried it and I liked it

I used these arguments in a1111

set COMMANDLINE_ARGS=--theme dark --xformers --medvram-sdxl

Is this ok or am I missing something else?

[ornate elk]

No these are fine

[gaunt goblet]

Hey, just woke up, but so just dont touch forge? I only ask because I was curious if it works for AMD GPU? I got 1111 to work just fine just couldnt get forge to, I only heard great things about it

[ornate elk]

yea its not worth at all now as forge changes its direction in development

[noble belfry]

i think its fair to consider it a dead project in terms of helping people, pity, I like it. Gonna be a bit obnoxious for the next while though when every forge answer is "dont"

[gaunt goblet]

alrighty wel thanks! sirry

[rotund musk]

hey, id like to know if its better if i use the 2 seperate pipelines (StableCascadeDecoderPipeline, StableCascadePriorPipeline) or the StableCascadeCombinedPipeline for using https://huggingface.co/stabilityai/stable-cascade (im going to use it in an Hugging Face Space with ZeroGPU, an A100, i just want to know the real difference)

[broken tulip]

getting this error in comfyui standalone: No module named 'diffusers.models.transformer_temporal'

I installed diffusers

through venv

But still get the error. Anyone can helpß

[unborn bramble]

Alright so I trained an SDXL Lora on the Hollowstrawberry SDXL Lora trainer, but when I use it... it outputs this. I saved 10 epochs for it, but only the 1st epoch gives out results that aren't nightmare fuel... but only if it's at a low weight. (Some images spoilered due to it being nightmare fuel)
Only the first epoch gives out an intelligible image, but I had to set it at a low weight. (The third one is setting the weight to 1)
What did I do wrong? I attached the metadata just in case.

[ember mortar]

im trying to add a lora in the prompt box but it doesn't add it, i click in the box of the lora and nothing happen

[night spoke]

@ornate elkHey, I reinstall the webui user as your guide says. But I encounter another issue, says error code 128

[ornate elk]

OK can you screenshot the full error ?

[night spoke]

@ornate elk

[ornate elk]

oh your from asia, then you need to use a VPN for the installation

[night spoke]

@ornate elkActually I am using VPN

[ornate elk]

ohhh then change the server to a europe or US one

[cold jolt]

hi , i have a problem, What am i supposed to do ??

My previous, stable diffusion got corrupted and im trying to reinstall it completely (from 0)

Normally i fixed It

[ornate elk]

Hey, please follow the install guide from the pinned messages of this channel.
You need git to git clone the repo.
And your antivirus Programm broke the Python installation, so you have to:
Uninstall Python.
disable the Antivirus programm.
Install Python 3.10.11 64bit linked in the guide.
Then enable the antivirus Programm again.
Then delete the venv folder and relaunch the webui-user.bat

Better start with the guide and dont try to fix your broken version. Not using git clone means you won't get any updates.

[real goblet]

Hey there people, so when I try to merge two checkpoints, I end up with a memory allocation error, and it is trying to assign a unrealistic amount of memory. What may be a possible solution to this?

[gaunt goblet]

Hey, so good news, FInalyl got it all working... though I do want to ask and I am curious on... is there a possible way to speed up the process? I don't think it should take a hour 25 mins to load 1 model?

[ornate elk]

Do you try to merge 1.5 or sdxl models?

[ornate elk]

Hey, that's not normal. What does the cmd shows? And what model?

[real goblet]

From what it seems both are 1.5

[gaunt goblet]

I may have figured it out- its a safetensors but the models I got are the old Novel Ai ones

[ornate elk]

they are not good anymore, you can delete them

[ornate elk]

how much RAM do you have?

[real goblet]

32gb

Like to be honest, if I could get a new gpu, I would squeeze it in my server

196 gigs of ram in that thing, would be perfect

[ornate elk]

32gb should be enough to merge models

what are the sizes of the models?

[real goblet]

Ones 7gb, others 5.3gb

[ornate elk]

Hmm that shouldn't be an issue but you could check if your Windows Pagefile settings are correctly (enabled only for C drive and set to system managed)

[real goblet]

Well

[mortal compass]

I'm trying to get custom nodes and workflows working in stable.swarm. I installed comfy manager, and started to install this workflow https://civitai.com/models/119528/sdxl-comfyui-ultimate-workflow-v40. During the install of the custom nodes, I get an error in the manager that something failed to install and to look at my terminal. The terminal had a couple of errors, and I'm only a novice, but the ones that I can interpret are:
Conflicts of some sort [Warning] ComfyUI-0 on port 7821 stderr: [!] ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
Outdated version of opencv-python-headless [Warning] ComfyUI-0 on port 7821 stderr: [!] albucore 0.0.12 requires opencv-python-headless>=4.9.0.80, but you have opencv-python-headless 4.7.0.72 which is incompatible.
Missing files [Warning] ComfyUI-0 on port 7821 stderr: FileNotFoundError: [Errno 2] No such file or directory: 'C:\\Ai_Root\\SD.Swarm\\dlbackend\\comfy\\ComfyUI\\custom_nodes\\comfyui_testing-custom-nodes\\__init__.py'

I'm not sure if those are correct, but still, I do not know how to tackle this without causing more harm. Would appreciate any kind of guidance!

Forgot to say that it seems like [Error] [WebAPI] Error handling API request '/API/GetCurrentStatus': Request input has unknown session id is also causing some issues, like not being able to load workflows

[night spoke]

@ornate elkHey, I changed the server to EU or US, still not work

[ornate elk]

have you deletet the venv too?

you have to d othis before relaunching the webui-user.bat

[night spoke]

@ornate elkI'll delete it now

[thin quarry]

out of curiosity - why it doesn't work on Asian IP's?

[ornate elk]

its not about the IPs, its about what the country or government blocks

Asian ISPs and GOVs often Block Github

[thin quarry]

oh I see

[full dome]

[62/168] Generating 'skimage\graph\_spath.cp312-win_amd64.pyd.p\_spath.c'
[63/168] Generating 'skimage\filters\rank\generic_cy.cp312-win_amd64.pyd.p\generic_cy.c'
ninja: build stopped: subcommand failed.
Activating VS 17.10.3
INFO: automatically activated MSVC compiler environment
INFO: autodetecting backend as ninja
INFO: calculating backend command to run: C:\Users\ravim\AppData\Local\Temp\pip-build-env-i234d2l8\normal\Scripts\ninja.EXE
[end of output]

note: This error originates from a subprocess, and is likely not a problem with pip.
error: metadata-generation-failed

Encountered error while generating package metadata.

pls

help

[night spoke]

@ornate elkStill not work Lol

[broken tulip]

I get this error in comfy: ModuleNotFoundError: No module named 'oss2' . I have oss2 installed both locally and in venv. Any ideas?

[fair oxide]

Hey, I installed ROCm in distrobox and got automatic1111
I followed a 7 Month old guide for Ubuntu.

I managed to launch it with the webui.sh. But my UI looks a bit different and def. lacks options, that the Guide had.

I'm completely new to Stable Diffusion and I'm trying to get this to work on AMD.

[tame yoke]

how do i get rid of generations like this?

the eyes are horrible

[karmic crown]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/13292

your UI looks the same

[fair oxide]

Omg, this might be it.
Thank you very much. ✌️

...

do you think, I can just rename it?

[karmic crown]

it is intentinally pinned here

[karmic crown]

should be enough i guess

[ornate elk]

Use adetailer extension

[night spoke]

@ornate elkWhat can I do NEXT? I really wanna use SD

[ruby yoke]

what issue are you having?

[ornate elk]

You can try not using a VPN and then delete the venv folder again

[ornate elk]

If you want to use Stable-diffusion on windows with AMD, checkout the pinned messages of this channel. There are all my guides.

[fair oxide]

Thanks ✌️
My main pc is on Linux. Currently I'm trying to get it done in a Distrobox, I don't want to install all the dependencies on my Host 😅

But if this doesn't work, I'll boot into Windows and check your Guides 👍

Btw, does a 2070 super perform better than a 6950 xt?
I thin so, right?

[ornate elk]

Nope a 6950xt would perform better

Also double the amount of vram is perfect for Ai stuff

[fair oxide]

okay perfect

[brave frigate]

Is it correct that SD is using most of my RAM but only 14% of my GPU usage?

i thought it mainly uses VRAM instead of anything else?

[ornate elk]

What's your GPU?

[brave frigate]

AMD Radeon RX 7900 XT Hellhound

20 GB

[ornate elk]

okay whats in your webui-user.bat?

GPU usage in % is not VRAM usage

[brave frigate]

@echo off

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

git pull

call webui.bat

[ornate elk]

check the vram usage amount, if its around 5gb or higher its fine

its normal that RAM usage is very high because its loading models into the RAM

[brave frigate]

i gues its fines then huh

[ornate elk]

yes looks normal

[brave frigate]

huh well thank you :)

[ornate elk]

no problem 🙂
also if your upscaling enable the TiledVAE (extension) and for sdxl use the sdxl fp16 vae to reduce the vram amount at the last step

[brave frigate]

im using the pony checkpoints and from my experience VAEs dont really work well with them

they seem to always crash

[ornate elk]

you need the fp16 one, that doesnt crash

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

and if hires fix is enabled, enable tiled vae too

[brave frigate]

ah

[brave frigate]

i tried it and i get a yellow blue picture using t-ponyai

but highres is active lemme try without

nope same outcome without

[fair oxide]

Okay I got it working 🙏
But I probably need better settings or models.

But I'm so glad that it works now 😌

[karmic crown]

so, you just renamed the folder?

[fair oxide]

I tried it, but my distrobox couldn't find the location then.
So I deleted my distrobox, downloaded rocm and the dependencies and connected it to my old home directory. I had all the stuff installed previously and now it works.

But I'll probably do it clean again, once I'm more confident with the process.

But huge thanks for the tip!

[ornate elk]

can you show me an example?

[ornate elk]

yep now checkout civitai.com for some models

[fair oxide]

Thank you!
I checked the channel pins, do you have a guide on how to install models?

Is is just download the model and place it in [?]

[ornate elk]

yep you download a model and put it into models/stable-diffusion
if its a lora it goes into models/lora

[fair oxide]

Hey one last thing, how can I measure my speed?
I saw some ppl discussing it /s. Where can I see it?

And is there a benchmark?

[ornate elk]

you can see it in the cmd or terminal
it/s stands for Iterations per second (like steps per second)

[fair oxide]

Do you think 9 - 10 it/s is normal for a 6950 xt?

[ornate elk]

yes seems good

[ancient plover]

My pc has randomly crashed 3 times today while using SD Zluda. Has anyone else had any issues recently?

[ornate elk]

nope

[copper vine]

{
"errors": [
"An unexpected error occurred while trying to reach an upstream service. Please try again. If the problem persists please contact support and include the id of this response."
],
"id": "81b028d789ba0f412270239fb6a000cd",
"name": "internal_error"
}

[unborn bramble]

I retrained it but it gave out similar nightmare results. I trained another SDXL LoRA on the same day (before) and it works perfectly but this one not so much

[fair oxide]

Hey, idk recently I encounter crashes, I get this error message:

OutOfMemoryError: HIP out of memory. Tried to allocate 2.11 GiB. GPU 0 has a total capacity of 15.98 GiB of which 608.00 MiB is free. Of the allocated memory 7.10 GiB is allocated by PyTorch, and 436.49 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_HIP_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

where can I put this?:
try setting PYTORCH_HIP_ALLOC_CONF=expandable_segments:True to avoid fragmentation.

[ruby yoke]

are you using an AMD card?

[fair oxide]

yes

[ruby yoke]

how much of your CPU are you using for graphics?

[fair oxide]

Idk tbh.
But it seems like my GPU ran out of ram

[ruby yoke]

actually, the error sounds like your configuration isn't set up. go through that document that's linked at the end of that error

[fair oxide]

Alright, I'll check it.
But it worked before... even with higher settings

[ruby yoke]

it says you have quite a bit of memory reserved by pytorch but not allocated.

[fair oxide]

The output also says there is a fix
PYTORCH_HIP_ALLOC_CONF=expandable_segments:True

I just don't know where to put it 😅

[ruby yoke]

i'm not sure where to put it either. are you running comfy? auto11111? something else?

[fair oxide]

auto1111

[ruby yoke]

i ran the error through Claude. it says:

This error message indicates that you're encountering an out-of-memory issue on your GPU when using PyTorch with HIP (likely on an AMD GPU). Let's break down the error and suggest some potential solutions:

Error type: OutOfMemoryError
Attempted allocation: 2.11 GiB
GPU total capacity: 15.98 GiB
Free memory: 608.00 MiB
Memory allocated by PyTorch: 7.10 GiB
Memory reserved but unallocated by PyTorch: 436.49 MiB

The error suggests that your GPU doesn't have enough free memory to allocate the requested 2.11 GiB. Here are some steps you can try to resolve this issue:

Set the environment variable as suggested:
Copyexport PYTORCH_HIP_ALLOC_CONF=expandable_segments:True
This may help avoid memory fragmentation.
Reduce batch size: If you're training a model, try reducing the batch size to decrease memory usage.
Use gradient accumulation: This technique allows you to simulate larger batch sizes while using less memory.
Optimize your model: Remove unnecessary layers or reduce the size of your model if possible.
Use mixed precision training: This can significantly reduce memory usage.
Clear GPU cache: You can try clearing the GPU cache periodically in your code:
pythonCopyimport torch
torch.cuda.empty_cache()

Monitor GPU usage: Use tools like nvidia-smi (for NVIDIA GPUs) or equivalent for AMD GPUs to monitor memory usage in real-time.
Consider using a GPU with more memory if these solutions don't work and you need to process larger batches or models.

i'd start with clearing gpu cache

[fair oxide]

Does anyone know if adetailer supports xyz prompting?

or is adetailer s/r ad 1st in xyz prompt meant to represent the positve prompt in the first adetailer model?

[gaunt goblet]

okay, this is more of me having a curious question, If stable uses my VRAM why is my memory sky rocketing at times? im at 98% useage at times Lmfao

[inner dawn]

Hello. I have this kind of problem. The list of textual inversions is not displayed. And it happens strangely. If I finished the program with one version of checkpoint, the textual inversions are displayed in the list, but if I finish the program with another version, they are not in the list when I start the program. For example, if I end the program with checkpoint dreamshaper 8 and restart the program, the textual inversions are listed, but if I do the same with ponyDiffusionV6XL, the list is empty. If you update the list, nothing happens, but if you change the model to dreamshaper 8 and update the list again, the textual inversions are displayed in the list.

[empty cipher]

I have some questions about LoRA and Model training, and how they relate to styles

For one: if my dataset contains only - as an example - headshots, would the resulting LoRA have difficulty drawing anything else? My understand is basically that when you supply an image to the trainer it will use its vision to identify everything in the image, generating a list of tokens in the process. Comparing this list to the caption, the trainer identifies all the tokens your captions say SHOULD be there, but that it didn't recognize, and adjusts its weights accordingly

In a nutshell, you're teaching the model to learn what "thing X" looks like in your style. For anything not present in the dataset, you aren't conditioning the model to be able to recognize it in the style provided.

To what degree is there bleed over? Suppose I'm training on images of foxes. To what degree would my style transfer over to wolves or dogs or cats?

Second: why would I train a LoRA instead of a model? I read an interesting article on Civit where someone spoke about how he trained a model instead of a LoRA, and then just extracted the differences. This makes a lot more sense to me because, as he states in his article, your style is going to have certain requirements if its going to be versatile. Men, woman, different ages, all different angles, hands in all sorts of configurations, facial expressions, backgrounds, low light, bright light, fire, water, plants, rocks, interior scenes, exterior scenes, common objects, vehicles, buildings, and so on

To me, training a model on these concepts individually rather than all at once in a single massive dataset seems smarter. This brings me back to my subquestion about

Question three: when I train on X, to what degree are Y and Z affected? Suppose I'm training on a dataset filled with pictures of people in various positions. How much are "hands" being affected, considering their a component of most "people" images. To what degree is "animal" being affected - given their both living things and there is some overlap? Does training a LoRA on X implicitly train on adjacent concepts? How would it differ if training a model? If training entirely on landscape, to what degree would the new style model be able to apply to people?

[ruby yoke]

a lora updates a small set of weights, not the entire base model. you can think of a LoRA as an errata sheet for a game. Let's say you are playing monopoly. you play by the rules it comes with. But you have an errata sheet for how to buy houses. When you buy a house, you use the revised rules on the errata sheet for that specifically, but only that. That's what your lora does for your generation and why it has to run with the model it was created for. So if you created a lora for SDXL, and your only trained it on headshots, then SDXL base would draw the rest of the person and the LoRA would take over drawing the head.

[ruby yoke]

It's much easier to train a LoRA as you can do so with a small data set and you don't need a lot of compute power. if you want to train an full model, you'll need a very large data set unless you want that model to only do one thing. Maybe you just want to make pictures of pizzas. you could make a small data set of several hundred images of pizzas from all different angles and train a pizza picture making model. but if you want a model that'll function like Stable Diffusion base does, that's millions, if not a couple billion, images and text caption pairs. And even just a small check point model that is built off the base model requires a decent sized data set and quite a bit of GPU time to train.

[empty cipher]

Could you be a bit more technical regarding how this works? I it sounds as though you're saying "only those weights concerned with what's in the training dataset are stored"

Would that be an accurate way of putting it? My motivation behind asking is that I'm trying to understand how to train flexibility into a model (checkpoint or lora, whichever)

[ruby yoke]

the base model contains an entire set of weights. you load that into memory. then you call the LoRA with the keyword. you then ovewrite the weights in memory with the weights the LoRA trained

[empty cipher]

In general: how would I go about training something capable of replicating a specific style across many different concepts?

[ruby yoke]

you'd want to use a dataset of images in that style. example: let's say you paint watercolors and you want to be able to have the AI generate images that look like you created them. you'd scan a bunch of your watercolors into your computer, lable them, and then train a LoRA with those. Let's assume you like SDXL so you create the LoRA to work with sdxl. and you give it the keyword of MyWatercolor. when you're ready to use it, you put in the keyword at the start of your prompt, then put in whatever you want for a prompt - like: MyWatercolor a painting of a barn in a cornfield.

or MyWatercolor six people having a dance on the beach, dynamic lighting

and the AI will create the image based on the prompt modified by the information in your lora

if you go to youtube, search on how to train a lora, scroll down till you find the video Scenario put out, it's a good tutorial

[terse urchin]

If you get yours going some time let me know. Waiting on github to get back to me

[ornate elk]

Normaly you can add this in the webui-user.bat. But I don't think it will help.

You just run out of vram. Adjust your resolution. Check your txt2img settings. If you use hires fix set the hires steps to 10.

[ornate elk]

That depends on the models loaded in. Sdxl requires a lot of RAM

Thats a normal behaviour.
Embeddings as well as loras are only compatible with 1.5 or sdxl based checkpoints.
They will only show up then.

[gaunt goblet]

yeah, I think I dont think 32GB is gonna cover it for that it feels like sometimes

[copper vine]

Is there anyone having a problem like me regarding this? Even my ticket is not answered. Has the community here experienced this before?

[dense tapir]

Somebody please send me the correct Adetailer hand detector model and tell me where to put it

Current one just leaves blurs

[ebon moat]

I am having some trouble installing SD. Can anyone help me with this error?

these are my system specifications.

[vocal burrow]

your gpu is too old for that, you would have to use cpu mode only which is very very very slow.

[ebon moat]

will you tell me how to do that?

[mystic quail]

How do you fix this error, sorry I am fairly new

[vocal burrow]

find the set COMMANDLINE_ARGS= in webui-user.bat and put --use-cpu all --precision full --no-half --skip-torch-cuda-test after it (same line)

[vocal burrow]

can't tell without the workflow but you're probably not using the Loracaption node correctly / not feeding it the correct value

[vocal burrow]

there s a space between full and --no-half

[ebon moat]

yeah i corrected that and then after the command prompt was done typing, it took me here " http://127.0.0.1:7860/ "

[vocal burrow]

looks good to me

[ebon moat]

okay .. thanks for the help

[ornate elk]

32gb should be enough

I have that amount too

[empty harness]

hi all

[ornate elk]

Hey

[empty harness]

Hello everyone, I'm creating an mobile app (photo editor with AI) based on Stability AI. and it is important for me to understand the following questions regarding AI stability:

• What face data does Stability AI collect?
• How do you use the collected face data? Provide a complete and clear explanation of all planned uses of this data.
• Will the data be shared with any third parties? Where will this information be stored?
• How long will face data be retained?
• Where in your privacy policy do you explain collection, use, disclosure, sharing, and retention of face data? Identify the specific sections in your privacy policy where this information is located.
• Quote the specific text from your privacy policy concerning face data.

[empty harness]

@ornate elk may you help us?

[fair oxide]

does anyone have a guide on how to use prompt matrix or prompt from file or textbox? It doesn't seem to be working for me

[vocal burrow]

cf https://stability.ai/privacy-policy and we re not lawyers so we cannot provide such informations safely and reliably. Moreover you ll probably have to add stuff to the legal pile depending of your country and where your applications will be available.
Also any kind of "does Stabillity AI collect xxxxx?" is too imprecise as Stability AI provide multiple online services on top of giving you the possibility to host your own local AI stuff. Leaving aside the numerous third parties using SD models to provide some services.

[vocal burrow]

cf https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#prompt-matrix

[fair oxide]

and prompt from file/textbox?

also thank you

Because I'm trying to automate entirely different prompts for large jobs

prompt from file seems to be stuck at 40% for some unknown reason, only generating 10 images

[vocal burrow]

check the console log for any error

[fair oxide]

No errors. I guess prompt from textbox saves all generated images at once...? That's kinda weird.

[empty harness]

I know that you are not lawyers 🙂 but that does not mean that technical support should not inform us 🙂 (I am a paying customer and have already paid enough money for your service to get answers to simple questions). Regarding legal matters within my country, I will consult my lawyer, don't worry. Speaking of Stability AI, I meant the cloud platform Stability AI and the product Stable Diffusion XL 1.0. Please answer the questions above

[fair oxide]

yeah, memory usage balloons nearly 10x during script run, which makes me feel the images are being held in memory before being dumped to disk

[grave fulcrum]

I’m thinking about getting a 4070 super 12gb. Up to what size images can you make using it?

[vocal burrow]

There s no limit, theorically speaking even with 8gb model.
But 12gb is better as it will allow you to load the entire pipeline on your gpu at once whereas 8gb models can struggle to load everything with sdxl, sd3 models. And therefore it will have to load part of it, do its work, unload, load another part, keep on working, etc.
With 12gb you ll be mostly limited by the dataset resolution used during training (512x512 for sd1.5 768x768 for sd2, 1024x1024 for sd3,sdxl). Those are the native resolutions, you can usually get descent resolution if you vary the resolution up/down to 50% of the native resolution. After that you ll have to do an upscale pass if you want higher res.

[vocal burrow]

might be the grid generation progress that things slow down but I would expect it to hit at the very end :/ not 40% that s weird. share the complete log and complete webpage setting please.

[fair oxide]

I have grid gen turned off

ima gunna experiment a bit some more and will open a bug later if needed

[vocal burrow]

This is not an "official stability.ai" support channel, this is a community driven one, trying to answer SD related tech problems not legal ones. We're not SAI staff, we're not paid by them (nor you).
Also "stability cloud" is still too vague as stability is partenering with multiple partner for cloud hosting. I would advise to contact them (be it stability.ai and/or amazon,intel,etc depending of cloud service) directly for legal matters.

[grave fulcrum]

If I were using an sdxl model to make 1024x1024 images, about how long it take to generate an image? Would it also be able to make up to, say, 2048x2048 if I used upscaling as well? Apologies if these are dumb questions, I’m not very familiar with most of this

[karmic crown]

this is not the right place for your question. my advice is to write to the official support over the mail, or even better, to raise a proper ticket if possible

[vocal burrow]

np we re all dumb at start. There s this page keeping tracks of multiple user benchmarks https://vladmandic.github.io/sd-extension-system-info/pages/benchmark.html you can easily filter 4070 gpu results from it and get an approximate idea of the speed you should get. Keep in track that speed scale down with the output resolution and depends of the model / prompt / number of lora / etc used.

upscaling process can take a mere second if you use some "easy / non SD method" or double the initial processing time depending of your settings ^^

[empty harness]

thanks bro!

[empty harness]

bro thanks!

[solemn moon]

is it just me or have 1.5 models gotten really fast

[undone coyote]

please anyone help me i'm getting so mad over this simple thing :

AttributeError: 'NoneType' object has no attribute 'lowvram'

everything was working alright like 2 days ago and i touched nothing since but now today i try running my webui and when i just try to generate an image it just outputs me this error

webui version 1.9.4 (last one) i can't downgrade on previous versions because SD3 don't work on the other versions for some reasons and also because i don't understand everything was working fine and now it doesnt without any changes between now and then

but the problem is that nothing works like i tried it and no model works

[ornate elk]

SD3 is only supported when Auto1111 is used on the SD3 branch

But thats experimental so better use stable

[undone coyote]

do you know how to use it on the sd3 branch ?

[undone coyote]

yea but even when i don'tuse sd3 nothing works for some unknown reasons i tried even fresh install

[ornate elk]

Can you show me a full cmd log?

[undone coyote]

ok

@ornate elkhere it is i forgot to ping

[ornate elk]

I checked it.
3 issues here:
You didnt used the correct way to install the Stable diffusion webui.
You need to git clone the repo and not download it as zip.
Then second issue is you placed and used it inside the Downloads folder. That can cause issues. And the third issue is that SD3 won't work until you switch to the SD3 branch which is only doable by using Git.

So please follow the install guide from the pinned messages of this channel.
And then wait for the update until SD3 is supported by Auto1111 in the stable branch. Or if you really want to switch to sd3 experimental branch

[undone coyote]

but idk for what reason but he was working like 2 days ago very correctly eveything was fine and now for no reasons at all it stopped working its so confusing that it can get from working to not working just by changing nothing

i'll try to clone the repo on my desktop

[ornate elk]

Read my guide. Desktop is also not the right place ^^

[undone coyote]

rlly ? but i always used on desktop like for other ancient versions

[ornate elk]

Here is the guide:
#🤝｜tech-support message

It can work on desktop but its not recommended as it can cause permission issues. Some people also use OneDrive and that breaks everything on desktop, downloads etc

@undone coyote My guide covers also the needed performance improvements for the best speed and compatibility

For example xformers was missing in your webui-user.bat

[brave frigate]

my friend @tall siren is struggleing with the generation time can someone help?

He's got 1650 ti

12GB

16GB RAM

and in his user.bat is this:

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --medvram --no-half --disable-nan-check --update-check

git pull

call webui.bat

[tall siren]

With upscale to HD and Sampling steps 40 takes 2h

[ornate elk]

remove --disable-nan-check

[solemn atlas]

Hi, i have a problem since some time, esp. after I installed my new ssd and cloned everything. Generating images makes my PC soooo slow. This never happend before but now it doesn't even work properly when I'm generating. my pc freezes

Just a few weeks ago I could generate 100 images in a short time without my pc freezing but now I need t o

to wait till one image finishes and then my pc works again

[ornate elk]

a 1650ti has only 4gb vram right?
thats to less to run Pony models with a good speed.
better to use 1.5 based models that are 2-4gb in size

If you want to use Upscale make sure to always set the hires steps to 10

[brave frigate]

he said he has 12GB vram idk if thats true tho

[ornate elk]

cant be true

[vocal burrow]

no chance

he's confusing ram and vram

[brave frigate]

then he'd need --lowvram i guess?

[ornate elk]

yea if he really want to use sdxl models. but i wouldnt use them with that card. SDXL and Pony will take ages to generate

[brave frigate]

@tall siren

[jagged turtle]

Can some clarify what the difference is between ControlNets (Diffusers SD1.5) and ControlNets (SD1.5) in the Stability Matrix list of models?

I've been searching around and I can't tell if one is newer or an extension of the other 🤷‍♂️

[thick bay]

GM, I have a question about training a Lora with multiple objects, I trained a Lora with images of bicycles, skates and motorbikes, how do I call each trained object in my final image?
ex: put a person riding roller skates, a person riding a motorbike and a bicycle on the ground

[bronze hornet]

can you help me everytime I click on the pony sdxl model I intalled the stable difusion just ignores that I clicked on it after loading and just goes back to whatever model I was using

[vast crest]

Hi,stable diffusion no longer opens for me, can someone help me. H:\stable-diffusion-webui>git pull
Already up to date.
Couldn't launch python

exit code: 9009

stderr:
Python non trovato; Esegui senza argomenti per installare da Microsoft Store o disabilitare il collegamento da impostazioni > Gestisci app Alias di esecuzione.

Launch unsuccessful. Exiting.

[ornate elk]

Delete the venv folder and relaunch the webui-user.bat

[ornate elk]

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

[warped basin]

Does anyone know why it's not inpainting the transparent area?

[ornate elk]

Set the inpaint area to only masked.
Then set the resolution to 768x768

And your not using xformers. Has nothing to do with Inpainting but your missing out on performance.

[warped basin]

I just tried that and it still just puts white where the transparency is. Except this time it changed everything except the transparency lol

[warped basin]

And thanks, I'll try to find out what that is 👍

[ornate elk]

Thats strange inpaint area mask only should not change the background

Make sure the webui is whitelisted in any browser adblocker

Also what's your GPU?

[warped basin]

I just whitelisted it and still no luck. My GPU is a 4070 super

[ornate elk]

Okay then in the webui-user.bat at the line commandline_args=
Add: --xformers
To get the best performance.

[warped basin]

Like this right?

[ornate elk]

Yep

For the inpaint thing try not changing any settings for test.
Only set inpaint area to mask only

[warped basin]

Cool thanks. Does it just make generations faster or also better?

[ornate elk]

Only faster and uses less vram so you can upscale or Generate at higher resolution

[warped basin]

Nope, same thing happens

Do you mean reset the settings under the Settings tab also? Is there a way to do that

[ornate elk]

You can reset the settings by deleting the config_ui.json

And then relaunch

If that doesn't help then try deleting the venv folder and then relaunch the webui-user.bat again

[warped basin]

I tried doing both, with both inpainting area settings and same result :/

I wanted to upgrade to SDXL anyway, you think that'll help or is it something else causing it

[ornate elk]

Okay that's strange. Can you try using an other model?
Any extension installed ?

[warped basin]

I tried a few other models, and no, only the built-ins

[ornate elk]

Only the built ins work?

[warped basin]

I think so?

[ornate elk]

Oh you were referring to the extensions

Yep thats okay

Its late here, I need to go off sry. We can fix it tomorrow.
If you want to test a clean SD install you can also follow my guide from the pinned messages of this channel.

[warped basin]

Np and ty! I'll probably try that thanks

[undone lance]

Hey! I had to factory reset my PC

I restored my stable diffustion from a backup, and installed python

When I ran the automatic1111 launch, it said something about the user being in-correct and not finding python despite me installing python.

[undone lance]

I think i forgot to check add to path during python instillation

So i figured it out.

It basically gave me the answer in the error. Typed in the command and it ran properly for me. As for the python install problem don't forget to add it to Path during the install process.

[sand laurel]

please tell me what does this option do?

[empty cipher]

Google colab question
I was using an A100, then decided to switch down to an L4
And I noticed the quality of my images dropped like a rock even though I was using the same parameters and the same resolution
What's up with that?

[empty cipher]

Hey, another question

I saw a conversation on another board where someone said if you want to train for a specific character in different clothes (or other, similar situations where variability matters)

You should fine-tune a model rather than train a lora

I think I have a pretty decent understanding of why, in my bones

But could someone explain to me so I can be sure?

[little bough]

is sd3 working in a1111 yet? i updated my install just now and noticed an SD3 update but tried loading SD3 model and got the same 'lowvram' error' i did before?

[solemn moon]

came here for this.. frothing..

[little bough]

do you also get this error when trying to generate with the sd3 model? (i got this error before the latest update)

[solemn moon]

naa i got the update from git-pulling so was wondering, havent tried yet, tryna setup pony xl to give it a go

attempting to try sd3 in auto only gives me the sadz, so waiting for somewone to say "hey it works"

[vocal burrow]

@little bough only available in the sd3 branch for now

I d advise you to wait for that guy to come and say "hey it works"

[solemn moon]

yesss, holding thumbs, cant wait

[karmic crown]

SD3 is only available in sd3 branch of a1111, and it is not working for everyone. there are currently several serious bugs and limitations

[marsh atlas]

In the API from stability AI the results of using SD3 are clearly cropped, any one seeing the same? here it is an example

[trail pumice]

guys why i am getting this error over and over again

do you have any clue how to fix this?

ERROR: Could not install packages due to an OSError: [Errno 2] No such file or directory: 'C:\ComfyUI_windows_portable\insightface-0.7.3-cp311-cp311-win_amd64.whl'

[ornate elk]

You need to download the insightface whl file and put it into the Comfyui folder

[trail pumice]

thats it?

[ornate elk]

Maybe you also need to install it via the command you used

Here is the link to the insightface file

https://github.com/Gourieff/Assets/raw/main/Insightface/insightface-0.7.3-cp310-cp310-win_amd64.whl

[trail pumice]

thank you🙏

[ornate elk]

Np 🙂

[trail pumice]

nah bro didnt work

[ornate elk]

You need to change the command

Copy the file name of that file

Its not 311 its 310

[trail pumice]

yeah it worked

but it gived some warninngs

i guess for now i will ignore theese

[feral otter]

Hey @ornate elk what type of cmd log do you need? I'm finding that with A1111 on the dev branch I'm getting about 10it/s but with fooocus I can get 1.20it/s so my computer is capable of more but A1111 just doesn't play nice.

I was using forge previously, but it's not really supported anymore so I was trying to switch back to A1111 and it's just super slow in comparison.

[deft stone]

Hi can anyone tell me how i could put in my HF token key with diffusers?

[ornate elk]

10 it/s is more than 1.20/ its

so auto1111 should be faster for you.
But you mostly missing the Performance launch args for auto1111.
Edit the webui-user.bat
at the line COMMANDLINE_ARGS=
you need to add: --xformers --medvram-sdxl --no-half-vae
then save and relaunch the webui-user.bat

[feral otter]

Thanks, I added those but I'm still getting way worse results on a1111

I just ran a test with the exact same image prompt/model/settings.
A1111 took 1:34 and 1:18
Forge took 0:27 and 0:24

[ornate elk]

Can you show the settings you used?

[feral otter]

you want the full prompts, sampler etc?

'score_9, score_8_up, score_7_up, score_6_up, wise bearded man, wizard robe, contemplative, peace pipe, candid, philosophical
Negative prompt: score_6, score_5, score_4
Steps: 30, CFG scale: 2.5, Sampler: Euler a, Seed: 263937623, Size: 832x1216, Created Date: 2024-06-25T0319:43.9263828Z, Clip skip: 2'

I input that on both webui interfaces with the exact some model (pony v6) and the time to generate is as mentioned above

[fair oxide]

Hi sry to bother but fighting with this install problem on mac m3

tried to runtime all and this is result

cant install xformers even I dont have cuda but wheels cant be build... any idea how to solve this

[ornate elk]

can you show me your cmd log from auto1111 ?

[ornate elk]

xformers is for nvidia only.
if you want an easy stable diffusion webui setup, follow this guide for MAC:
https://github.com/viking1304/a1111-setup

[fair oxide]

ok thank you will give it a try

[feral otter]

I think this is what you are looking for, I just ran it again and it seems to get to 97% in about 20 seconds (similar to Forge) but then it takes another 60 seconds to finalize it.

[ornate elk]

that model load block error is strange. can you try to delete the 1.5-ema-pruned model and then relaunch the wbeui-user.bat?
Then you can go into the settings and set the VAE type to TAESD
That could be the setting forge uses for the speed at the end.

[feral otter]

For VAE type i have options for encode and decode? Should I switch both of them from Full to TAESD?

[ornate elk]

there should be an option to only set TAESD for Previews

maybe yes you need to toggle both, i cant check it rn

[feral otter]

I changed both of them to TAESD, however when I paste that script in adds and override to make them "Full" again. I deleted the overrides and it generated in 26 seconds!

[ornate elk]

ah perfect, yea than thats what forge uses probably

a user a few days ago had the same issue and he fixed it with that too

[feral otter]

Thanks for your help, hopefully the prompts I get from Civit don't continue to force override to full...but this certainly seemed to help a lot!

[ornate elk]

yea you'll never know what settings or webui they may used or even gpu ^^

[weak whale]

can anyone help me? For some reason all the loras i put in the folder without 2 are not showing up does anybody know why?

[crisp vortex]

I have been experimenting with terrariums for a while. It was all going great until something happened. Now, all I can get out of it is these gray squares. Any ideas?

[karmic crown]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/16092

[ornate elk]

Yes they will only show up when you use a compatible model.
1.5 loras only work with 1.5 models.
And sdxl Lora's only with sdxl models.

[dreamy current]

Hi, I1m getting this message on my cmd: extra {'cond_stage_model.clip_l.text_projection', 'cond_stage_model.clip_l.logit_scale'}
left over keys: dict_keys(['alphas_cumprod', 'alphas_cumprod_prev', 'betas', 'embedding_manager.embedder.transformer.text_model.embeddings.position_embedding.weight'...
What does it mean?

[ornate elk]

Hey would need the full cmd log

Also what webui are you using?

[dreamy current]

forge

[dreamy current]

do you want it in dm or should i post it here?

[karmic crown]

https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/801

"Forge will then be turned into an experimental repo to mainly test features that are costly to integrate."

"These updates are likely to break almost all extensions, and we recommend all users in production environments to change back to upstream webui for daily use."

[dreamy current]

i know, but i didn't uploaded it

[karmic crown]

basically - go back to a1111

[ornate elk]

And what did you tried to do?

[dreamy current]

Nothing, it's a message that jumped when i started it, i think that it happens when i try to load the ghostmix model

[karmic crown]

do git log and post

[dreamy current]

How do i do that?

[karmic crown]

open cmd/terminal and go to folder where forge is installed

type what i asked and post the result

[dreamy current]

ok

Wait a moment, i'm trying to use the cd command

commit f06ba8e60b7a6c0a18979b2b7f2bc4aed68715d0 (HEAD -> main, origin/main, origin/HEAD)
Author: lllyasviel [e-mail]
Date: Thu Feb 8 10:13:59 2024 -0800

Significantly reduce thread abuse for faster model moving

This will move all major gradio calls into the main thread rather than random gradio threads.
This ensures that all torch.module.to() are performed in main thread to completely possible avoid GPU fragments.
In my test now model moving is 0.7 ~ 1.2 seconds faster, which means all 6GB/8GB VRAM users will get 0.7 ~ 1.2 seconds faster per image on SDXL.

commit 291ec743b603fdcd9c58e60dc5ed3d866c53bc4c
Author: lllyasviel [e-mail]
Date: Thu Feb 8 02:00:54 2024 -0800

use better context manager to fix potential problems

commit 760f727eb985cff1df2f164bab0a776f0e5518bd
Author: lllyasviel [e-mail]
Date: Thu Feb 8 01:51:18 2024 -0800

use better context manager to fix potential problems

commit 4c9db26541bdca195cff89977dfbf601f8193ac6
Author: lllyasviel [e-mail]
Date: Wed Feb 7 21:42:56 2024 -0800

support controlnet_model_function_wrapper for t2i-adapter

There you go

[karmic crown]

looks good

it is not updated

################################################################
Launching launch.py...
################################################################
Python 3.10.14 (main, Mar 19 2024, 21:46:16) [Clang 15.0.0 (clang-1500.3.9.4)]

what this part says when you run it?

3.10?

[dreamy current]

3.10.6

Or should i run launch.py

[karmic crown]

you should run webui-user.bat

or webui-user.sh if you are on linux or mac

[dreamy current]

What does launch.py does? Because i started running it

[dreamy current]

Then it says 3.10.6

[karmic crown]

do you have an error when you run that?

or not?

[dreamy current]

No, just that message

It's actually downloading things

[karmic crown]

can you copy/paste the whole output and post the file?

than wait if it downloads

[dreamy current]

From the webui bat?

Or form launch

[karmic crown]

do not run launch

and wait to finish installing

if it install something

[dreamy current]

webui wasn't installing anything, but when i started launch because i thought you were talking about that file it did

should i keep letting it download things?

[karmic crown]

launch.py should not be run directly, webui.bat and webui-user.bat will do that for you

[dreamy current]

they never do

[karmic crown]

webui.bat is the basic version, without optimizations

but for forget it should be enough in most cases

[dreamy current]

I use the user-bat

Okay, i have the webui opened and with the same message

Im copying it

[karmic crown]

left over keys: dict_keys(['alphas_cumprod', 'alphas_cumprod_prev',
i have never seen this

[dreamy current]

I can use the model perfectly but that message is jumping everytime i use this model, and sometimes others

[karmic crown]

i found that error only for comfy
https://github.com/comfyanonymous/ComfyUI/issues/1142

[dreamy current]

Well, i don't think it's and error because i can use everything just fine but i don't really know

[karmic crown]

hmm

https://github.com/lllyasviel/stable-diffusion-webui-forge/issues/240

can you try any other model? do you have the same output or not?

[dreamy current]

extra {'cond_stage_model.clip_l.text_projection', 'cond_stage_model.clip_l.logit_scale'}
left over keys: dict_keys(['model_ema.decay', 'model_ema.num_updates'])

With another model

With yet another model: extra {'cond_stage_model.clip_l.text_projection', 'cond_stage_model.clip_l.logit_scale'}

[karmic crown]

https://github.com/comfyanonymous/ComfyUI/issues/1100#issuecomment-1664824148

here it is

[dreamy current]

so basically it's nothing no?

[karmic crown]

just ignore it

[dreamy current]

okay, thanks!

[haughty bluff]

what does it mean when it says ERROR Decode VAE: Given groups=1, weight of size [128, 3, 3, 3], expected input[1, 4, 128, 128] to have 3 channels, but got 4 channels instead. ?

[bitter kraken]

What is the hardward requirements for stable diffusion 3 🙂

[ruby yoke]

16gig vram

[bitter kraken]

Will mbp m3 works?

[shy charm]

apparently i need to apply this to my forge install to use DoRA LoCONs but i don't understand how to
https://github.com/lllyasviel/stable-diffusion-webui-forge/pull/608/files

[modern current]

can anyone help??

[karmic crown]

it should work. i am using m3 pro

run this command
system_profiler SPHardwareDataType | head -n 10 && echo "" && system_profiler SPDisplaysDataType && sw_vers

double-check if your serial is displayed in the output (remove it if it is, but it should not be) and post the result

you can use my script to install stable-diffusion-webui
https://github.com/viking1304/a1111-setup

[noble knot]

When starting stable i get the output
"[ERROR]: unable to create share (unable to create share: [POST /share][401] shareUnauthorized)
fatal: No names found, cannot describe anything"

Anyone got any idea?

(Using zrok)

[little bough]

Hi, not sure if anyone can advise with this but I have a USB audio interface with my computer and often times when starting A1111 or changing models in A1111 or even upscaling, it is somehow 'tripping up' my audio interface and cutting off the audio (maybe the drivers) and I have to switch off and on my audio interface.. i have put up with it but I'm thinking if there is maybe a solution to this then I can put an end to it.. does anyone know some reason why A1111(python ? ) would be clashing with USB audio? my other USB devices are finem Thanks

[noble knot]

Figured it out, i just had too many shares connected to 1 enviroment

[bitter kraken]

Is there a specific specs for m3?

[karmic crown]

any ARM Mac should work, the question is "how fast/slow" and what can be done

please show me this info, using the commands i sent

      Model Name: MacBook Pro
      Model Identifier: Mac15,6
      Model Number: MRX73ZE/A
      Chip: Apple M3 Pro
      Total Number of Cores: 12 (6 performance and 6 efficiency)
      Memory: 18 GB

      Chipset Model: Apple M3 Pro
      Type: GPU
      Bus: Built-In
      Total Number of Cores: 18
      Vendor: Apple (0x106b)
      Metal Support: Metal 3

ProductName:        macOS
ProductVersion:        14.5

[light prism]

Hey i am new here and was wondering if anyone could hint me in a direction for merging multiple Images automatically. I got multple images of variations of one object and want to get one result with the most important attributes. Inpainting does not work for me since this needs to be automatic when there are too many inputs and i can specify every attribut. I Saw that you can input a array of images in the StableDiffusionImg2ImgPipeline but it seams that it just uses the first one for the first output and then the second for the next and so on...

[light prism]

Or is this just not possible?

[modern current]

someone pls reply

[ornate elk]

Dont use Forge, its outdated.
Switch to Auto1111 by using my Install Guide from the pinned messages of this channel.

If you really want to use forge you need to do this to fix it:
put albumentations==1.4.3 and pydantic==1.10.15 in the requirements_versions.txt file
Then relaunch or delete the venv

[modern current]

got it

[light prism]

Or is this not the right place to ask it? If yes any tipps where i find an answer?

[modern current]

git pull not working

[ornate elk]

Can you show the full cmd log?

[vagrant ore]

how to add SD to my server?

[fair oxide]

is there nsfw channel for nsfw pics?

[thin quarry]

no

[fair oxide]

ok

[shy charm]

apparently i need to apply this to my forge install to use DoRA LoCONs, but i don't understand how to apply it. i assume i use git but literally all i know how to do with git is clone and pull
https://github.com/lllyasviel/stable-diffusion-webui-forge/pull/608/files

i don't know how to apply a pull request

[solemn atlas]

what does that mean?

[ruby yoke]

means you apparently need to go download the insightface.app model

[solemn atlas]

https://github.com/deepinsight/insightface I downloaded this but it still says no module

[ruby yoke]

try using the comfy manager to install from github url

[solemn atlas]

okay I'll try

[solemn atlas]

but I have automatic1111

[ruby yoke]

you can run both, as long as you have the space to install both, but if the error is from auto1111 then installing it with comfy won't fix that.

[ornate elk]

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

[pale salmon]

sup, every time i prompt something, i can this filter

[ornate elk]

Hey, thats caused by a Wrong or missing vae

Sdxl and 1.5 need different vaes

[pale salmon]

i use autismmixsdxl

it recommend this

https://github.com/hako-mikan/sd-webui-prevent-artifact

but idk how to install

[ornate elk]

You don't need this

You only need a VAE file

This one:
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main

Put that into models/VAE folder

Then in the webui go to
Settings/ UserInterface/ Quicksettings
There add sd_vae
then hit apply and reload ui
Then select the sdxl VAE in txt2img at the top.
Then you can generate without artefacts

[pale salmon]

which one ?

[ornate elk]

The top one

[pale salmon]

okay ty i will try

fixed ty!

[ornate elk]

No problem!

[light prism]

Hey i am new here and was wondering if anyone could hint me in a direction for merging multiple Images automatically. I got multple images of variations of one object and want to get one result with the most important attributes. Inpainting does not work for me since this needs to be automatic when there are too many inputs and i can specify every attribut. I Saw that you can input a array of images in the StableDiffusionImg2ImgPipeline but it seams that it just uses the first one for the first output and then the second for the next and so on...

[ruby yoke]

how's the AI going to know what the most important attributes are?

[light prism]

That is the callenging part of course, but i was hoping that there is a way to maintain the parts that occur in most images for example. e.g. i got 10 images of a person with long black hair and only two with short hair that the output will be similar to the part that is more often present. Like having multiple locations in the latent space and the output needs to be close as close as possible to all input images

Generally i am open to suggestions of all kinds. My idea is just to have multiple people generating images and then finding a way to come up with a solution to combine these to a single output image as the "ultimate design"

[ruby yoke]

you'd have to have a way of telling the AI what you considered the most important aspects. otherwise it would have to guess what you thought was most appropriate, and there's no real possibility it would guess correctly. so first you're going to have to write something that'll let you tag each part of an image. And at tht point, you might as well just cut those parts out in photoshop or something.

[light prism]

hm okay thank you for the response. if found this "image mixer" thing here but my results were just very abstract and i couldn't even recognize the images in the output :/ https://huggingface.co/lambdalabs/image-mixer or https://www.youtube.com/watch?v=3q_9DS26qcw&ab_channel=NerdyRodent

[timber holly]

Are only "pony" loras useable with the pony XL base model? Just downloaded it & trying to understand why the generations are so messed up

[ornate elk]

@light prism You would need to use the multiple tools and coding your own script maybe.
It could be possible by using Segment Anything, Inpaint Anything and Controlnet.
Also checkout Grounding-Dino.
But there isnt any Script or tool that can do that automatically.

[ornate elk]

Normal sdxl Lora's can work with pony too. But not all.

[solemn atlas]

thank you so much, now it works. I wanted to generate something with 3 characters but get this

[ornate elk]

NP, With which Extension?

[timber holly]

Ah, all my loras are made for 1.5 I believe. I'm new to XL, it only takes XL loras I assume?

[solemn atlas]

regional prompter

[ornate elk]

Pony is based on SDXL so yep, only pony or sdxl Lora's will work

[ornate elk]

With the mask paint module?

Or the normal separator

[timber holly]

Are VAEs the same or are those model agnostic?

[solemn atlas]

the normal one, the thing is, it worked for one time and now i get this . I use xl checkpoint, vae and lora for all 3 characters

[ornate elk]

VAEs also have different model compatibility.
Pony and Sdxl needs sdxl VAE and won't work with an 1.5 vae

[ornate elk]

Hmm and what did you changed after the first image?

[solemn atlas]

I had two characters and wanted 3 so i put another one with a new lora and I get this error

[ornate elk]

Is the lora compatible?

Try without it again

[solemn atlas]

it doesn't work. With or without Lora

these are my settings

[ornate elk]

And what's the image resolution?

[solemn atlas]

it's 768:512

[ornate elk]

Then you need to set that in Regional prompter too

In your screenshot its on 512x512

[solemn atlas]

I tried that too but then i get this

[ornate elk]

Hmm strange.
Do you have 3 BREAKs in your prompt?

[solemn atlas]

yes,

[ornate elk]

Can you try set the resolution to 1024x768 ?

For Regional prompter and the image resolution

[light prism]

Thank you for the Response!!! i will take a look at it and maybe find a way to realize this and then share it 🙂

[solemn atlas]

I tried, but it still doesn't work

[ornate elk]

Okay, can test around tomorrow when I'm at the PC

[solemn atlas]

Sure, thank you

[safe hull]

need some help with a random nan error that keeps popping up randomly. keep getting an error similar to this:

for reference, using A1111, hires fix 4x ultrasharp 1.5 to 1536x1536
doing ADetailer with face_yolov8n.pt
dynamic prompts and only using 13GB VRAM of 16

no startup arguments aside from listen occasionally

tried disable nan check, and upscale to fp32 too

it's weird though, since ill rerun the same combo with the same seed and it'll run fine. just seems to pop up randomly

updated nvidia drivers etc, and good python etc

[safe hull]

after further testing it may be something with adetailer. turned it off and so far 100 generations without the error.

when using dynamic prompts, do you guys use the same prompt with wildcards for adetailer? i left it as is and just used the same one

nvm still giving error randomly

[upbeat pewter]

How do i fox this issue here? I’m on Windows, can someone please guide me.

[ornate elk]

What's your GPU?

[ornate elk]

What's your GPU?

[upbeat pewter]

can i dm u the pic of it?

[ornate elk]

Sure

[ornate elk]

Okay. you need to update your python version to 3.10.11 64bit.
Then you need to install the latest Nvidia GPU Driver.
Then in the webui-user.bat at the Commandline_args=
You only should have:
--xformers --no-half-vae
Then save.
Then delete the venv folder.
Then relaunch the webui-user.bat

[upbeat pewter]

okay ima try thanks

btw im new to this

how i install the nvida gpu driver?

???

[karmic crown]

https://www.nvidia.com/download/index.aspx

GPU = graphics processing unit

[rancid nymph]

how do you make img2img more accurate to your original image

[ornate elk]

By lowering the denois

Or by using Controlnet canny

[safe hull]

my gpu is a 4080 non super

[ornate elk]

Then you should add
--xformers --no-half-vae
In your webui-user.bat at the Commandline_args=

[safe hull]

thank you! ive tried no-half-vae in the past and it seriously slowed down performance. do you think just doing xformers would help avoid the error popping up?

i know you suggest deleting the venv a lot, i can also try that after backing up some stuff

[ornate elk]

You can also try first without --no-half-vae and only add --xformers

It will boost the performance a lot and allows higher upscaling

[safe hull]

thank you! ill give that a go tonight for the next batch and see if it runs without errors

[ornate elk]

Np, let me know.
Also make sure your webui is updated and your extensions too

[little bough]

Hi, i was just installing the new anydepth extension for a1111 and noticed this error, anything to be concerned about? says it has been installed in a1111 Thanks

[cosmic furnace]

How much harmful for a PC can be reboots due to lack of power for GPU ? 🤔

[manic flume]

For wildcards, would you get the webui wildcards extension, the dynamic prompts extension, or both?

I guess dynamic prompts, but is webui wildcards necessary?

Another thing im wondering is do wildcards work for Pony if they aren't specifically Pony wildcards

[manic flume]

I assume so, it's just text files

[ornate elk]

You only need the wildcards extension

If you want to use dynamic prompts too then both

[manic flume]

Okay, thanks

[ornate elk]

Yes

These are just words in txt files

[amber pollen]

how cam i generate pictures using this server?

what section and what command i have to use?

[ruby yoke]

You can use the #🗣｜artisan-support-feedback artisan channels. Read the information at that link first

[solemn atlas]

Maybe you know what is wrong with my SD?

[river bough]

Does anyone know how to fix problem in the img2img section?

I've tried everything, and the connection error keeps appearing, or it simply doesn't display or generate any image. I don't know what else to try to solve the problem. Any help would be appreciated

[fossil tendon]

Hello guys, I would like to know what are the requirements in terms of hardware to use Stable Diffusion? I want to install either A1111 or Forge on my Synology NAS (Intel Celeron J4125 with SSD and 18Gb RAM) but not sure it will actually works

[karmic crown]

you need windows/linux/macos
i think Synology have some other OS

[fossil tendon]

I run docker containers on my synology, it's linux based

I saw some docker version of stable diffusion, so that's what makes me think it could be possible - not sure about performances though

[karmic crown]

even if you manage to run it on that it will be extremly slow

i tried to run it on VDS with 10 CPU, 60GB of RAM and SSD and it worked slow as hell

even extra slow 1660 super worked much faster

[fossil tendon]

yeah i guess without GPU it's not worth it :/ wasn't sure so i figured i'd ask

[karmic crown]

i would suggest local install on your computer

what do you have?

[fossil tendon]

Mac M1 Pro (Apple Silicon)

[karmic crown]

ok

i am mac user 🙂

[karmic crown]

run this command
system_profiler SPHardwareDataType | head -n 10 && echo "" && system_profiler SPDisplaysDataType && sw_vers

double-check if your serial is displayed in the output (remove it if it is, but it should not be) and post the result

you can use my script to install stable-diffusion-webui
https://github.com/viking1304/a1111-setup

[sage scroll]

I am using The replacer extension and animatediff is not working. Here is the result and here are my settings (its an example) Can somebody help me?

[fossil tendon]

Hardware:

    Hardware Overview:

      Model Name: MacBook Pro
      Model Identifier: MacBookPro18,1
      Model Number: Z14V001EMD/A
      Chip: Apple M1 Pro
      Total Number of Cores: 10 (8 performance and 2 efficiency)
      Memory: 32 GB

Graphics/Displays:

    Apple M1 Pro:

      Chipset Model: Apple M1 Pro
      Type: GPU
      Bus: Built-In
      Total Number of Cores: 16
      Vendor: Apple (0x106b)
      Metal Support: Metal 3
      Displays:
        Color LCD:
          Display Type: Built-in Liquid Retina XDR Display
          Resolution: 3456 x 2234 Retina
          Main Display: Yes
          Mirror: Off
          Online: Yes
          Automatically Adjust Brightness: Yes
          Connection Type: Internal

ProductName:        macOS
ProductVersion:        14.5
BuildVersion:        23F79

[karmic crown]

nice, you can use that without problem

it will not be as fast as you would like, but it will be usable

you have 32GB of unified RAM, so mac can use 24GB for graphics without problem

you can use my script to install forge or a1111, but i recommend using a1111

[fossil tendon]

Cool, thanks for the info. And nice you made a script. I will definitely take a look at it. I'm just starting with generative AI, was playing with Leonardo.AI but then realized with Stable Diffusion you can have all the models you want, it's a bit more technical but I work in IT so I don't mind

[karmic crown]

download my script
run it like this

bash a1111-setup.sh -f all

and wait

depending of your internet connection speed it will take a shorter or longer

btw, you will be able to run sd 1.5 on that mac without any issues, but sdxl might be a little slow

my m1 had 16GB, so i am not sure how much faster yours will work, since you have more ram

[fossil tendon]

yeah maybe worth trying and fallback on 1.5 if it's really not usable

[karmic crown]

m3 pro with 18GB a lot works faster, than my old m1

you can use lcm and lightning

you can get images with less steps, which will take less time

btw. i love this thing
https://civitai.com/models/171159/sdxlrender
and i used it a lot with 1.5

[fossil tendon]

thanks for the tip, will definitely take a look at it! so far i really like epiCRealism but I only been playing with this for one day and I'm a complete beginner

[marsh smelt]

Hey everyone I am having error with this line
Tag Autocomplete: Could not locate model-keyword extension, Lora trigger word completion will be limited to those added through the extra networks menu.

It's messing with my Stable Diffusions ability to read Loras, I think.... not entirely sure

I just know I can't gen my previous prompts

[fossil tendon]

@karmic crown between a1111 and forge you said you would recommend a111 better?

[karmic crown]

yes, because forge is outdated and not recommended to end users

some extensions might not work

[fossil tendon]

oh alright, I see. thought it was the opposite and forge was a newer version of a1111 haha

[ornate elk]

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

[karmic crown]

a1111 works faster on my m3 pro than forge with default params, i had to do a lot of tweaking until i finally made forge to work about 3% faster

[marsh smelt]

Where can I go to get someone to help me with my issue?

[karmic crown]

you will just get a headache without any real benefit. forge makes miracles on some nvidia systems

[river bough]

I did that but the problem persists, now when I click on generate, it just keeps loading until no image appears.

[ornate elk]

what does the cmd shows?

[fossil tendon]

alright, thanks for all the info!

[sage scroll]

Animate diff is working fine when i use it normal (with the lora) but replacer doenst work with Loras (i think)

[river bough]

Only the number : 1006

[ornate elk]

whats your gpu and whats inside your webui-user.bat?

[river bough]

Rtx 4070 and inside my webui is :

[ornate elk]

okay looks good.
can you relaunch the webui-user.bat and screenshot or copy the full cmd log ?

[river bough]

Yes here is it : venv "C:\Users\Rt2\Documents\Escritorio\Aplicaciones\Stable Diffusion\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.9.4
Commit hash: feee37d75f1b168768014e4634dcb156ee649c05
Launching Web UI with arguments: --xformers
Loading weights [8d0ce6c016] from C:\Users\Rt2\Documents\Escritorio\Aplicaciones\Stable Diffusion\stable-diffusion-webui\models\Stable-diffusion\sd_xl_refiner_1.0_0.9vae.safetensors
Creating model from config: C:\Users\Rt2\Documents\Escritorio\Aplicaciones\Stable Diffusion\stable-diffusion-webui\repositories\generative-models\configs\inference\sd_xl_refiner.yaml
Running on local URL: http://127.0.0.1:7860/

To create a public link, set share=True in launch().
Startup time: 9.9s (prepare environment: 2.2s, import torch: 3.4s, import gradio: 0.9s, setup paths: 1.0s, initialize shared: 0.2s, other imports: 0.6s, load scripts: 0.8s, create ui: 0.3s, gradio launch: 0.5s).
Applying attention optimization: xformers... done.
Model loaded in 4.4s (load weights from disk: 0.4s, create model: 0.3s, apply weights to model: 3.4s, calculate empty prompt: 0.1s).

[karmic crown]

this looks fine

[ornate elk]

your using the refiner model of sdxl. dont use that

[karmic crown]

i havent seen that

[ornate elk]

its not for generating images

[karmic crown]

good catch

i was just checking if there are any errors

[ornate elk]

yea it looked normal at first

[river bough]

even if I choose some other model like SDXL Base or AniverseXL no image is shown but I will check i again anyway

[ornate elk]

how much RAM do you have?

[river bough]

Now the connection error appears

[river bough]

36 gbs

Ddr4

[karmic crown]

can you try other browser?

[river bough]

I try using google chrome but it still does not generate any image or gives connection error

[karmic crown]

you are getting this timeout on all browsers?

[marsh smelt]

Hey does anyone know why Automatic1111 would suddenly start generating the same prompt differently?

[karmic crown]

does generation ever works?

[sage scroll]

I would try restarting the ui

[river bough]

yes or if it does not appear, it simply does not show anything.

[marsh smelt]

what does ur cmd prompt say?

[karmic crown]

do you use mcaffee or any other 3rd party antivirus?

[river bough]

Yes generation of images with text works fine

[marsh smelt]

automatic1111 is just the gui, the cmd prompt will tell you why

[karmic crown]

hmm, in that case it is not an AV or blocker issue

[ornate elk]

specific changes made by updates of the webui can change the seed

[river bough]

Idk what is cmd prompt but here is it : Applying attention optimization: xformers... done.
Model loaded in 12.1s (create model: 0.5s, apply weights to model: 9.7s, apply half(): 0.2s, move model to device: 0.2s, load textual inversion embeddings: 1.2s, calculate empty prompt: 0.2s).
Reusing loaded model aniverseXL_v10.safetensors [a62aa94c13] to load sd_xl_base_1.0_0.9vae.safetensors [e6bb9ea85b]
Loading weights [e6bb9ea85b] from C:\Users\Rt2\Documents\Escritorio\Aplicaciones\Stable Diffusion\stable-diffusion-webui\models\Stable-diffusion\sd_xl_base_1.0_0.9vae.safetensors
Applying attention optimization: xformers... done.
Weights loaded in 11.6s (send model to cpu: 2.5s, load weights from disk: 0.3s, apply weights to model: 6.7s, move model to device: 2.0s).
1006

[marsh smelt]

Thanks for the response, it happend just today, yesterday was working fine, how do I go back to the previous update?

[river bough]

And i also disable my antivirus but nothing works

[marsh smelt]

thats the entire line when you trigger the error?

[karmic crown]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/9136

[marsh smelt]

thats the entire line when you trigger the error?

Replicate the error and then send a pic of ur cmd prompt

[marsh smelt]

And I genuinelty mean thanks for the response, ive been typing for the last 10 minutes and felt invisible lol

[ornate elk]

we dont know if its from the update.
can you share the orignals 2 images, so i can comapre the meta data

[marsh smelt]

uhm... they are pony and very very nsfw lol

are you okay with that?

[ornate elk]

okay then load them into the PNG info tab. and copy the data into a text file and compare both lines

[uncut relic]

please put spoiler if you do so

[marsh smelt]

exactly the same

[ornate elk]

nsfw is not allowed here

[marsh smelt]

I have both the png info tabs next to eachother and they are 100% identical

[river bough]

This is all my cmd prompt

I will try now disabling my AV and trying in other browser

[karmic crown]

you can get connection issues without an error in windows/linux/mac terminal, but the console in web browser is probably completely red 🙂

[marsh smelt]

Any reason that you are using such an old python with a new automatic1111?

[marsh smelt]

true but its a better place to start than starting from no where

[karmic crown]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/9136

this is his problem

websocket returns 1006

[river bough]

because to install automatic it asked me for that python so I installed it.

[river bough]

Oh thank you i will check it

[karmic crown]

i posted that twice 🙂

since you didnt react when i posted that few minutes ago 🙂

[ornate elk]

@river bough delete the config.json, config_ui.json and relaunch the webui-user.bat

[river bough]

sorry, I didn't know that was the problem XD

[ornate elk]

have you changed any setting in the webui?

[marsh smelt]

so any sort of theory why the two images with identical prompts are different?

I should clairfy they are only very very slightly different

But I feel like the resulting image is slightly less understanding of the prompt

[karmic crown]

even the , makes a difference

[marsh smelt]

they are the exact same

infact i found the problem by pressing send to txt2img

all my old images aren't identical gens

[river bough]

connection error again

[ornate elk]

if you do that check for override options at the bottom

[marsh smelt]

yeah there was no overides

[ornate elk]

slightly like what? a changed hairstrain?

[karmic crown]

all i can tell you is to try the things they mentioned there, one by one 🤷🏻‍♂️

[marsh smelt]

like a girl having cat ears when she shouldnt

[uncut relic]

that's a lot of information in a bit of time

[marsh smelt]

or there being 2 werewolfs instead of 1

things u would expect from a different seed

[ornate elk]

okay then delete the venv folder and relaunch the webui-user.bat

[marsh smelt]

but they have the exact same info

i can send u what im seeing but like I said its nsfw

[ornate elk]

send them to me in dm

[marsh smelt]

yeah of course

[river bough]

I delete that two times and not working

[ornate elk]

true ^^

[ornate elk]

can you try if it works in firefox?

some people mentioned it on github

[river bough]

I will try that

also i will try first adding --no-gradio-queue

[ornate elk]

if nothing helps please reinstall SD and not in the Documents folder.
Follow my install guide from the pinned messages and also upgrade python to 3.10.11

[river bough]

FINALLY

--no-gradio-queue Works

Thank you so much all!

[sage scroll]

Is there the possibility, that someone can help me? Ive searched online and didnt find any solutions...

[proper swift]

I got the new version of SD.Next that supports sd3, but it wants to connect to hugging face to download a config file. so i put the hugging face token into the options and it still won't work and is saying i'm not authorized to download.

anyone have an idea how to make this work

or a work around like where to put the downloaded config

[boreal vector]

im trying out regional prompter but sometimes it merges two characters into 1 with two halfs instead of two characters, does anyone have a solution?

[ornate elk]

@solemn atlas Here is how you can get 3 chars with 3 different loras:

[ornate elk]

you need to adjust the resolution in regional prompter to match the resolution from above

[boreal vector]

if im doing a 512 x 512 gen so it should also say 512 x 512 in the regional prompter?

[ornate elk]

yes

[boreal vector]

but they are the same

[ornate elk]

but 512x512 is not a good resolution for 2 or more chars

you need a wider width

like 768 or 1024

[boreal vector]

ok ill check with them

[ornate elk]

then always click on this after chaning the resolution in regional propmter

[boreal vector]

what about with high res fix

also double it?

[ornate elk]

depends on your gpu

i use upscale by 2 with 10 hires steps, on 0.5 denois

[boreal vector]

i did 4 test gens with 768 and 3 of them have two characters and 1 is merged

can i share the grid

[ornate elk]

yea

[boreal vector]

its swimsuits so idk if it counts as nsfw

[ornate elk]

whats the prompt?

and the regional prompt settings

[boreal vector]

lora:naipf:1, score_9, score_8_up, score_7_up, source_anime, beach, 2girls, straight-on,

BREAK yae miko,
BREAK raiden shogun,

[ornate elk]

try remove the lora

or reduce its strenght to 0.3

[boreal vector]

its an art style lora

[ornate elk]

yea but 1.0 is to strong

@boreal vector also add 1girl, after the BREAKs

for each girl

[boreal vector]

still similiar results

[ornate elk]

have you tried without the lora?

[boreal vector]

that was with reduced strength

now trying without it

[boreal vector]

ok the lora was at fault, can i invoke it after the break command seperately?

[ornate elk]

you have to test, but i guess it wont work with common prompt but with base prompt setting of regional prompter

[boreal vector]

so i have to check both of these?

[ornate elk]

then you would need a additoonal break

[boreal vector]

where?

[ornate elk]

in the prompt

at the top

[boreal vector]

this is with style lora invoked after break

[boreal vector]

Here? score_9, score_8_up, score_7_up, source_anime, beach, 2girls, straight-on,

BREAK 1girl, yae miko, lora:naipf:0.8,
BREAK 1girl, raiden shogun, lora:naipf:0.8

[ornate elk]

@boreal vector

[boreal vector]

yeah im following this guide. I was just confused when a single merged character kept popping up instead of two

[fair oxide]

can anyone help me with automatically creating folders for each disparate prompt/generation config? For example, I generation 512 images with prompt "white dog", I want them in a folder automatically called "white dog"

[boreal vector]

im having better luck with the other setting

[safe hull]

it worked! thanks for your help, was able to do a full batch of 1000 with no issues

[ornate elk]

Perfect 🙂

[cosmic mulch]

after i make one image, it keeps saying "connection timed out" and i can't even view the image

version: v1.9.3-amd-26-g50d3cf78  •  python: 3.10.6  •  torch: 2.3.1+cpu  •  xformers: 0.0.23.post1  •  gradio: 3.41.2  •  checkpoint: ac17f32d24

also whenever i load the model it takes like 150 seconds ish and in that time frame the pc lags a loottt

can someone help?

[ornate elk]

Hey it seems that you installed the wrong webui version and either using directml or CPU mode.
What's your GPU?

I've made several AMD install Guides. If your GPU is supported by ZLUDA try out an the install guide in the pinned messages of this channel.

[cosmic mulch]

direct ml. on a 6700xt

[cosmic mulch]

alright thanks

[ornate elk]

Ah that GPU is supported by ZLUDA, you can find the guide in the pinned messages on the first link. Automatic1111 with ZLUDA

[sage scroll]

Does anyone know what i do wrong there?

[ornate elk]

Motion module 3 needs a motion lora

[cosmic mulch]

i did all of the steps but i get this error

i have the two folders

made sure to edit the webui-user batch file

[ornate elk]

Can you show your path settings?

[cosmic mulch]

i edited the path below for system variables

should i edit the path for user variables?

i was a bit confused with this part. from what i understood, i did the following

1. copy the library folder and name it old_library.
2. paste the files from the optimized_rocmlibs_gfx1031 into the library folder, not the old_library folder
3. restart my pc

is that right? or did i miss something?

this is my webui-user file

i the SDXL 1.0 model in the models folder for stable diffusion

this is the software i got from the amd SDK link, and i have zluda on the 3.8 version from the github page

i have not done anything regarding this step

I did pip install requests but i still get the same erorr? do i need to restart

[final fractal]

I'm trying to get the comfyui extension to work in automatic1111 (python 3.11.9.)
However the version that gets installed seems to be running into a issue

To create a public link, set `share=True` in `launch()`.
[sd-webui-comfyui] Started callback listeners for process webui
[sd-webui-comfyui] Starting subprocess for comfyui...
Startup time: 17.9s (prepare environment: 4.7s, import torch: 6.8s, import gradio: 1.2s, setup paths: 0.5s, initialize shared: 0.2s, other imports: 0.4s, load scripts: 1.9s, create ui: 0.9s, gradio launch: 0.8s launch: 0.8s, app_started_callback: 0.4s).
[ComfyUI] [sd-webui-comfyui] Setting up IPC...
[ComfyUI] [sd-webui-comfyui] Using inter-process communication strategy: Shared memory
[ComfyUI] [sd-webui-comfyui] Started callback listeners for process comfyui
[ComfyUI] [sd-webui-comfyui] Patching ComfyUI...
Traceback (most recent call last):
  File "C:\Users\***\sd\stable-diffusion-webui\extensions\sd-webui-comfyui\lib_comfyui\comfyui\pre_main.py", line 98, in <module>
    main()
  File "C:\Users\***\sd\stable-diffusion-webui\extensions\sd-webui-comfyui\lib_comfyui\ipc\__init__.py", line 41, in wrapper
    return function(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\***\sd\stable-diffusion-webui\extensions\sd-webui-comfyui\lib_comfyui\comfyui\pre_main.py", line 43, in main
    patch_comfyui()
  File "C:\Users\***\sd\stable-diffusion-webui\extensions\sd-webui-comfyui\lib_comfyui\ipc\__init__.py", line 41, in wrapper
    return function(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\***\sd\stable-diffusion-webui\extensions\sd-webui-comfyui\lib_comfyui\comfyui\pre_main.py", line 84, in patch_comfyui
    paths.share_webui_folder_paths()
  File "C:\Users\***\sd\stable-diffusion-webui\extensions\sd-webui-comfyui\lib_comfyui\ipc\__init__.py", line 41, in wrapper
    return function(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\***\sd\stable-diffusion-webui\extensions\sd-webui-comfyui\lib_comfyui\webui\paths.py", line 9, in share_webui_folder_paths
    from folder_paths import add_model_folder_path
ModuleNotFoundError: No module named 'folder_paths'
Applying attention optimization: Doggettx... done.
Model loaded in 5.1s (load weights from disk: 0.7s, create model: 0.6s, apply weights to model: 3.5s, apply dtype to VAE: 0.1s, calculate empty prompt: 0.1s).
[sd-webui-comfyui] Updating comfyui at C:\Users\***\sd\stable-diffusion-webui\extensions\sd-webui-comfyui...
[sd-webui-comfyui] Already up to date.```

For trial I did try to install the portable version of comfyui seperately from automatic1111, which does seem to run just fine.
But targeting that version in the webui's install path does not seem to help in getting the extension to actually work in the end either.

[sage scroll]

Ty, it kinda fixed it. Now its not just random noise. However, every frame has some noise somewhere. Do you know any fix? There are no errors in the log

[ornate elk]

That driver version is outdated

You need the latest

AMD Adrenalin software 24.6.1

Zluda works only with anything newer than 24.0.0

[ornate elk]

I didn't used replacer sry

[cosmic mulch]

okay

[cosmic mulch]

so what do I do now? I search sdk and get the latest version?

[ornate elk]

Nope

Thats not the sdk

Thats your normal AMD GPU driver

[cosmic mulch]

oh crap

https://www.amd.com/en/support/download/drivers.html

i can get it from here?

[ornate elk]

Yep

[cosmic mulch]

so i need pro edition as well as adrenalin edition for amd huh

[ornate elk]

No

In my guide it says "don't check pro"

Pro driver is not needed

[cosmic mulch]

during the installation the only mention of pro was "install pro drivers" and i checked it to "don't install pro drivers"

i must have done something else wrong

[ornate elk]

Nope

You did everything correctly

Your normal GPU driver is just outdated

[cosmic mulch]

this should be it right?

[cosmic mulch]

oh i see... well it makes sense because i just reset my pc. thank you so much! im gonna install and then see if it works

[ornate elk]

Also restart the PC after Installing it

[cosmic mulch]

if i still get a different error ill send it here soon

[ornate elk]

Alright

[cosmic mulch]

same error bro :(

i restarted after the update

[ornate elk]

Delete the venv folder and the .zluda folder and relaunch the webui-user.bat

[cosmic mulch]

.zluda folder?

[ornate elk]

Its in the stable-diffusion-webui-amdgpu folder

[cosmic mulch]

there is only a .github folder

[ornate elk]

Oh ok

[ornate elk]

Ok. Have you replaced the rocm library files?

[cosmic mulch]

yes

i clicked replace all while keeping the old libraries

i copied the library and renamed it old_library, and on the original library folder, i kept all of the files and just dragged and dropped the others from the zip and clicked replace all

[ornate elk]

Okay good

Then try delete the venv folder and relaunch the webui-user.bat

[dense tapir]

Hey CS1o

[ornate elk]

Hey

[dense tapir]

Hope you're having a good day

[cosmic mulch]

fr, this man helps literally everyone out

[dense tapir]

I know bro is the Stable Diffusion Jesus

[ornate elk]

^^ thanks

[cosmic mulch]

noooo :(

is it because my pip is 23.0.1?

[ornate elk]

Nope

[cosmic mulch]

do i do everything again from scratch

as you saw from my previous screenshots my python version is indeed 3.10.11

and the path variables are set

[ornate elk]

Yep it all looks correct

@cosmic mulch What you can try is the following:
Go into the stable-diffusion-webui-amdgpu folder
Click in the adress bar
And type cmd and hit enter.
In the cmd type
Pip install requests
Then delete the venv folder again

[cosmic mulch]

alright

[ornate elk]

Hmm ok

Can you relaunch the webui-user.bat and show me the full cmd log?

@cosmic mulch do you have a third party antivirus Programm installed?

These can break python Installs

@cosmic mulch it seems like the dev fixed something 4 hours ago.
Thats related to your issue

Go into the stable-diffusion-webui-amdgpu folder
Click in the adress bar
And type cmd and hit enter.
In the cmd type
Git pull

Or maybe that "fix" is causing this issue.
Also check your python versions installed

[lost crescent]

I'm using torch 2.0.0 in ComfyUI - but when I run ComfyUI - it stops half way thru startup - and installs torch 2.3.1, after which ComfyUI opens with a blank screen!!!

It's counter-intuitive; ComfyUI tells you it needs torch 2.0.0 - and once you acquire it - it immediately overwrites it with 2.3.1!!!

[ornate elk]

You should use comfyui with 2.3.1

Make sure comfyui is updated with git pull

Then whitelist the comfyui page in any browser adblocker

[cosmic mulch]

I'm sorry, i went outside

still am outside

when I get home I'll send you the screenshots

[cosmic mulch]

yes, i have done it and it says already pu to date.

[cosmic mulch]

no third party software, firewall is turned on, windows defender is turned on

[cosmic mulch]

my PC keeps freezing while running the webui-user file

it's still frozen

[storm sail]

is this the correct chat to ask questions if im having trouble running inference in stable diffusion webui?

[cosmic mulch]

can't move my mouse

[storm sail]

btw, hello all and hope everyone is well 🙂

[ornate elk]

yes

hey

[ornate elk]

how much RAM do you have?

make sure all other programms are closed (games and wallpaper engine)

[cosmic mulch]

16 gigs

it's still frozen.. should I force stop?

and turn my PC off with the power button

[ornate elk]

what does the cmd shows?

[cosmic mulch]

collecting torch

[storm sail]

@ornate elk ty, im getting an RuntimeError: Windows not yet supported for torch.compile when i try to use dreamshaper model and lcm sampling method

[ornate elk]

is it completely frozen or just stuttery? laggy?

[cosmic mulch]

I can sometimes move my mouse a bit

all actions ar3 delayed by like a minute

[ornate elk]

then let it load

installing torch uses a lot of resources

if its not frozen its good

[ornate elk]

lcm as sampler or the lcm lora?

[storm sail]

i have stable diffusion with openvino acceleration and running on intel cpu cuz im amd, i looked up this issue and found someone say Bypass this on Windows by adding --compile=False to the end of your run statement. Ex:

python train.py config/finetune_conversation.py --compile=False works, whereas
python train.py config/finetune_conversation.py will raise the error

only idk what that means

[ornate elk]

AMD can run stable diffusion better than any intel cpu

whats your gpu?

[storm sail]

my amd is secondary and idk how to make stable diffusion use it

my amd is my gpu

[ornate elk]

yea but whats the gpu model and vram amount?

[storm sail]

only it defaults to my intel cpu/gpu integrated

[ornate elk]

if you tell me that i can make it work ^^

[storm sail]

k, totally appreciated, second

amd radeon r7 m440

[cosmic mulch]

my PC completely froze and it's still at the collecting torch=2.3.0 💀

can't even move my mouse now

[storm sail]

i have 16gb ram but directx thingy shows 4gb vram for the amd

[cosmic mulch]

should I force quit and retry with 0 background apps running?

cuz I can't navigate to close my bg apps since the mouse is stuck

[ornate elk]

yes then force close and then we need to check your windows pagefile to increase your virtual RAM

[ornate elk]

ah alright.
To get it working with the AMD gpu you have to follow this guide from me:
Automatic1111 with DirectML:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides#amd-automatic1111-with-directml

[storm sail]

idk much about computers, despite how much i use the thing, lol, mainly i only play videogames 😛

sec

[cosmic mulch]

okay how do I do thay

my card has 12 gigs of vram so I thought it wouldn't lag this much but nvm

[ornate elk]

its about RAM not VRAM

https://www.thewindowsclub.com/increase-page-file-size-virtual-memory-windows

[cosmic mulch]

ohh

[ornate elk]

the installation needs a lot of RAM

[storm sail]

i think i already tried that repo but i still got error about needing to add --skip-torch-cuda-test, i tried many repos

[ornate elk]

it should work, but let me know if not

[storm sail]

hmm, i havent seen them exact instructions, i will try this

i will be back then after i give that a shot, man, u rule, thanks 🙂

[ornate elk]

you have to make sure the Pagefile is only enabled for the C: drive and not for any other drive.
Then set the Min to 16000 and Max to 24000
then save and reboot the PC

[ornate elk]

alright, no problem 🙂

[fair oxide]

Im pretty sure i can ask here but im trying to install stable diffusion but keep getting this, im not sure what to do

[ornate elk]

hey, can you show the full cmd log?

[fair oxide]

sure, let me grab some screenshots

give me 2

[cosmic mulch]

if i click set

then it gives me this error

and if i click OK, it doesn't save and goes back to "no paging file"

[ornate elk]

oh i see you downloaded it as zip. thats not the right way to install stable diffusion webui.
Follow my install guide (also check the preparations):
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides#nvidia-automatic1111-webui-stable-diffusion-webui

[fair oxide]

oooh

thanks!

[cosmic mulch]

what should i do

[ornate elk]

then check system managed

is C or D an ssd?

[cosmic mulch]

ssd

c is the ssd

[ornate elk]

okay then disable it for D, that slows down your PC if the pagefile is on hdd

[cosmic mulch]

256 gigs of storage, but only 58 free

[cosmic mulch]

how?

[ornate elk]

select the D drive and click "no pagefile"

then select C drive and click system managed

[cosmic mulch]

if i click "ok" after doing that it just goes back to the old settings.. wth

nvm i did it a couple of times and it worked

weird

[cosmic mulch]

it worked. the error is somehow gone now. i won't question it

let me restart my pc, close all my bg apps, and then run the webui-user.bat

[ornate elk]

so its enabled for C only now?

and system managed

[storm sail]

@ornate elk u mind if i add u as friend or whatnot?

to keep track of whos helping me here 😛

i added u 😛

oh, more question, should i delete or whatever the repo i already have and all associated (hugging face cache, etc) before i try that https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides#amd-automatic1111-with-directml

[cosmic mulch]

bro i am actually losing my mind

yeah it's enabled only for c now

and it says 16k to 24k

but i still get the same exact error

the only version of python i have is python 3.10.11

nothing else is running in the background

[storm sail]

will things get screwed up if i have more than one stable diffusion on my pc at same time or???

[cosmic mulch]

nothing that can block the python code

[ornate elk]

hey im at home now and tested a clean installation. and guess what, i get the same error.
Its caused by the devs update a few hours ago and he needs to fix it.
Im already opening an github issue right now.

[ornate elk]

nope

[cosmic mulch]

ohhhh. okay man

sorry for causing you so much trouble then. can you please give me a ping when it's fixed?

[storm sail]

ok, cool, gonna go try that then, ty 🙂

[cosmic mulch]

or is there any way i can know when it's fixed and how i can make it work after it's fixed

[ornate elk]

no problem at all. your now ready and only need to git pull again when the fix is pushed

i will inform you

[cosmic mulch]

ohh okay

[cosmic mulch]

thank you so much, have a nice day man

[ornate elk]

thx you too 🙂

[storm sail]

i still need to do this part first tho, right? (probably answer is yes and is stupid question)

[lost crescent]

I cannot for the life of me get ComfyUI to start!!! Something about the Huggingface Hub?!

[ornate elk]

if you dont have git and python then yes

[storm sail]

and looks like i need uninstall and remove from path etc, delete pip, etc of python cuz i have 3.10.6 and will need 3.10.11

Important: If you tried to install SD before and failed, do this: open up a cmd and type pip cache purge to remove the nvidia stuff. <- i mean, not preperations step (python and git) but clear pip cache thing

lol, that will be part of removing python 3.10.6 i think anyway, lol

or can i just have multiple installs python? something about editing path tho?

think when i get python 3.10.11 all will be well if i keep 3.10.6 only i need to remove 3.10.6 from path and add 3.10.11

[ornate elk]

no

you can just install python 3.10.11

its an upgrade to 3.10.6

no need to uninstall

[storm sail]

no need to do any extra work at all?

3.10.11 will overwrite all?

[ornate elk]

the pip cache purge is recommended

[storm sail]

but still do the purge

hoping im understanding everything correct, ok

ok, ok, i try this then, so much ty for ur patience 🙂

[ornate elk]

no problem, let me know if it worked or if you have any question

[storm sail]

cool, ur the best 🙂

[ornate elk]

@cosmic mulch i fixed the issue (workaround)

[storm sail]

pip cache purge can be done by click start -> cmd -> i ran as admin then pip cache purge, it said 315 files removed. right? not having to open cmd from somewhere else?

thinking i did right

[cosmic mulch]

oh, how?

[ornate elk]

go into the stable-diffusion-webui-amdgpu folder
click in the explorer adress bar (not searchbar)
type cmd and hit enter
copy and paste the following commands one by one:
venv\Scripts\activate.bat
pip install requests
Then relaunch the webui-user.bat

[storm sail]

and the upgrade of python from 3.10.6 to 3.10.11 auto adds that new python to path (it didn't give me any options), then i check group policy to be sure i am enabled long paths (that last is probably optional maybe unneccessary, idk) 😛

[ornate elk]

thats fine

[storm sail]

thinking hmm, i wonder wtf is zluda, lol

[ornate elk]

Zluda is a programm that let AMD gpus run cuda (nvidia) stuff

[cosmic mulch]

is it bad that there is no module xformers? or is that supposed to happen?

[cosmic mulch]

also this worked tysm

altho it's stuck on launching web ui

[ornate elk]

Xformers is an nvidia only tool

[cosmic mulch]

ohh i see

[storm sail]

ahh, zluda sounds neato