🤝｜tech-support

[empty cipher]

This might seem a bit dense, but if the goal is for the encoder to learn an accurate representation of an image in a lower dimensional space, wouldn't skip connections essentially be "cheating," and giving the network an easy way out instead of actually learning a representation?

[ruby yoke]

nope. i did a tutorial on FreeU - which allows you to adjust the skip connections. https://youtu.be/1FMIZNR25jA?si=VZooq5jjwVqQtpsQ this will give you a good idea of why you might want to do that

and that should also tell you what is going on with the neural network even when it trains

[modest blade]

anybody knows a fix?

fresh install of sd and also insatlled all the requirements and torch and all but getting this error

[ornate elk]

Hey, whats your GPU?

[modest blade]

amd integrated

[ornate elk]

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

[modest blade]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--use-directml --disable-model-loading-ram-optimization --opt-sub-quad-attention --lowvram --disable-nan-check

call webui.bat

[ornate elk]

Remove --disable-nan-check
And add --upcast-sampling
And then delete the venv folder and relaunch

[modest blade]

ok will try now

[modest blade]

@ornate elk no issues now, thanks g 🫂

[clever cargo]

I'm not sure what's happened :l

queueing anything will quickly result in the "reconnecting" notification and then I get noooooothing .... forever...

Anyone available to help?

[ornate elk]

What's your GPU?

[clever cargo]

I'm not home at the moment :v

I think RTx 3060 2060?

I can pull it up when I get back.

I've been using it all just fine up until a couple days ago. Made thousands of images but just out of nowhere I get nothin. And I couldn't find error logs x_x

[clever cargo]

my mistake 😅

[ornate elk]

Okay so it worked before, normaly

Then make sure you have whitelisted the webui in any browser adblocker

Also try to use an 1.5 based model to see if that works

[clever cargo]

I dont use an adblocker that i know of. I shouldn't how to worry about that, right?

[ornate elk]

Okay yep, then try an 1.5 model

[clever cargo]

my download speed is garbage so I'll update you when I get to try it out. Thanks for helping so far ^_^

Thanks man 😭 That's all it was, i guess?

it's at least generating images.. which is a step in the right direction :v

[white thorn]

Hello, trying to use flux with forgeUi,I have an RTX4080 16Gb. what models should I be able to run and what is too much? just wanna to be sure I match the standart performance for my card.like is it normal that the dev fp16 model is struggling?

[ornate elk]

Okay, if you have only 16gb of RAM you can increase your Windows Pagefile.
That helps to load sdxl and pony models

[ornate elk]

Yes thats normal because its 22gb

You should take a look at the fp8 model (gguf version) and the t5_fp8 gguf Text Encoder

[white thorn]

oh there is a gguf one ! i only got the safetensor , I also have to learn all about the new text encoder stuff and all the new options i dont know in forgeUi

[iron lantern]

are there any apps that help with promting what to write in sd ??

[ruby yoke]

since flux and SD3 both use the same encoders, use this https://huggingface.co/spaces/Nymbo/FLUX-Prompt-Generator

[olive lintel]

Is xinsir controlnet union inpainting supposed to work in A1111? I always get NaN when trying to inpaint using it and any of the inpaint* preprocessors (but eg. canny preprocessor doesn't result in NaNs). GPU is Tesla V100 (on vast.ai).

"modules.devices.NansException: A tensor with NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check."
(Upcast cross attention layer to float32 is set in A1111 options)

[karmic crown]

https://github.com/DominikDoom/a1111-sd-webui-tagcomplete.git

[calm cipher]

I decided to give ForgeUI a try after being told its better than automatic, but so far its been pretty bad. Every image it makes has really off lighting, as if it wasn't using a VAE, but I can clearly see I have the correct VAE selected. Any ideas? (or if ForgeUI not all its cracked up to be?)

[ruby yoke]

if you want to make your life eaiser, just install SwarmUI and then run comfyUI inside it

[ornate elk]

What's the model? And if its sdxl dont use bnf fp4

[karmic crown]

and what you have here

do not use too high cfg

that can result in "overcooked" images

use corect vae for choosen model...

[calm cipher]

SDXL 1.0, and yeah I was just trying random things to try and get something to work. I'm looking into SwarmUI now.

[olive lintel]

Doesn't lightning model require a different sampler / scheduler also?

[karmic crown]

depends on the model

blue pencil lightning

i like to use this vae for xl anime

its always best to check recommendations by the authors of model and lora (if used)

examples by the author are always a good start

[olive lintel]

I wonder how BluePencilXL LCM would react if you suddenly changed the scheduler from SGM Uniform to Karras...

[ruby yoke]

do it and find out ;)

[olive lintel]

Pretty sure Zylo just did that (with poor results)

[karmic crown]

i used DPM++ SDE Karras, and euler a with automatic, karas and sgm, and some others 🙂

[ruby yoke]

the latest version of comfy has a new scheduler

[karmic crown]

which one? beta?

or some new one?

[ruby yoke]

just run the update.py file in /comfyUI/update and you'll get the latest one.

[karmic crown]

what is the name

i do not want to install comfy just to see which one you are talking about

i guess its Karras Exponential

[ruby yoke]

it's linear_quadratic

[distant narwhal]

@ornate elk Sorry for the ping, but can you help with this?

[ornate elk]

Which settings did you used?

[distant narwhal]

i found it always did it no matter what the settings, just with any of the DPM samplers

oh and upscaling

[ornate elk]

Yea thats normal because comfyui with ZLUDA has much worse vram handling when it comes to upscaling than auto1111 or forge with ZLUDA.

DPM 2M shouldn't be a problem for normal tasks

[cold kernel]

steps: 55%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▊ | 1917/3497 [54:19<44:46, 1.70s/it, avr_loss=0.122]

getting these speeds on a 4090 with batch size 1, 13 epochs, and 269 images with 1 repeat, prodigy optimizer, ive been told its not supposed to be this slow, how can i fix it?

[balmy shore]

whats like the easy way to apply "ai filter" to a video? is it the mov2mov extension?

[balmy shore]

ive tried a few of the video extensions and gotten like sortof okay results but i wanna figure out how to take like a video of myself dancing or a dumb gif and basically apply a filter on it, im not sure how to go about it
and never tried actually applying it to an initial source

also wondering whats the easy/best way to edit or reinterpret already created images
guessing its something with controlnet but i s uck at understanding how it works

[ornate elk]

You can also use img2img to get different variations of the image

Play around with the denois level

[balmy shore]

ok i need to just play around with it sum more maybe

and learn the masking ive never tried the thing where u can draw on the image

[ornate elk]

the draw thing isnt that good, but inpaint can help for face fix etc

[mighty mist]

Hi guys, Ollama and Open WebUI are already up and running as LXC containers on my Proxmox server i am serving it on https://ai.takeone.bh. I'd like to add image generation capabilities using Stable Diffusion and Automatic 1111.
Can anyone help me with the following:
1. Installing Stable Diffusion and Automatic 1111 as LXC containers
2. Configuring them to autostart when the host machine boots up
3. Integrating these containers with my existing Ollama and Open WebUI setup, running on separate LXC instances
Let me know if you have any guidance on this.

[half island]

Hoi, what causes these issues? Downloaded a upscale model trained on alpha with images, but it doesn't work when upscaling sprites

[ruby yoke]

is the upscale model trained for the same base model that whatever checkpoint you're using is trained for?

[half island]

Not using a Checkpoint. Just direct image to 'upscale with model", then output image. But found out that stable diffusion isn't able to do alpha through upscaling

[keen raptor]

guyz i have a question

if i click ‘low vram’ will it affect parameters like image generation results, or only take a little longer to generate

[ornate elk]

It will only take longer

[keen raptor]

o

thanks

[ornate elk]

What's your GPU?

[keen raptor]

4060

[ornate elk]

With 8gb ?

[keen raptor]

yeaa

;-;

[ornate elk]

Do you use auto1111?

[keen raptor]

yes

[robust heron]

Hello, I have a problem that on sushiswap I changed the link to usdt and all the money just went to a third-party wallet, what should I do in such a situation?

[keen raptor]

i think the auto1111 is the easiest to use

[ornate elk]

Then you don't need lowvram

You need --xformers --medvram-sdxl --no-half-vae

In the webui-user.bat

Then you won't get any vram problems

[keen raptor]

i have it to —xformers only

[keen raptor]

if the vram is full i can just restart Ui right.

[ornate elk]

Yea

[keen raptor]

o

[ornate elk]

But just add --medvram-sdxl --no-half-vae

That makes you run sdxl and pony models faster

[keen raptor]

after the - - xformers?

[ornate elk]

Separated with a space

[keen raptor]

o

ogay

THANK YOU DUDE WITH THE LUFFY PFP

[ornate elk]

No problem 🙂

[robust heron]

?

[ornate elk]

This isnt the right place for any crypto stuff

[half island]

@robust heron Here we use our gpu power for funny image generation, not for fake currency that somehow carries economic value

[noble belfry]

@ornate elk ^

(soz if theres a mod i should be pinging instead)

[gilded sand]

Hello I put a message in general chat but the question is about development what is the best location to talk about that

[ruby yoke]

what sort of development?

[gilded sand]

@ruby yoke the one you read i think in general chat about game engines pairing with video gen

[ruby yoke]

oh. that's probably the best place for it. it's a general sort of topic

[gilded sand]

Do the coders/devs review the posts in there I'm curious about implementation

[ruby yoke]

what coders/devs?

[gilded sand]

It says DEVS in members list are they part of Stability or programmers?

[ruby yoke]

they're stability developers. they may or may not read the channels

but wouldn't posting the question on a video game company discord be more likely to get video game companies interested in the idea?

[gilded sand]

I think it might if they were also gearing up with AI but I don't think that is too common. Since SD is AI and making video gen directly I wanted their thoughts

[royal nova]

Hi! i am trying to use stable diffusion in a docker container but it gives me a lot of problems... the container is using the gpu because when using nvidia-smi the graphic comes out, this is my dockerfile in case it works:

FROM nvidia/cuda:11.7.1-cudnn8-devel-ubuntu20.04

ENV DEBIAN_FRONTEND=noninteractive
ENV PYTHONUNBUFFERED=1
ENV PATH=/opt/conda/bin:$PATH

RUN apt-get update &&
apt-get install -y --no-install-recommends
wget git libglib2.0-0 libsm6 libxrender1 libxext6 &&
apt-get clean && rm -rf /var/lib/apt/lists/*

RUN wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh &&
bash miniconda.sh -b -p /opt/conda &&
rm miniconda.sh

WORKDIR /workspace

RUN git clone https://github.com/CompVis/stable-diffusion.git
WORKDIR /workspace/stable-diffusion

COPY environment.yaml /workspace/stable-diffusion/environment.yaml

RUN conda env create -f environment.yaml

RUN echo "source activate ldm" > ~/.bashrc
ENV PATH /opt/conda/envs/ldm/bin:$PATH

COPY dreamlike-diffusion-1.0.ckpt /workspace/stable-diffusion/models/ldm/stable-diffusion-v1/model.ckpt

CMD ["python3", "scripts/txt2img.py", "--prompt", "a fantasy landscape", "--plms", "--n_samples", "1"]

[ruby yoke]

not sure why you are using docker, but consider dumping docker, install SwarmUI, let it handle all the technical stuff, and run comfyUI inside. generate with stable that way, it'll make your life a whole lot eaiser

[royal nova]

I forgot to comment an important detail, I am creating a final degree project, in which from a wordpress I will be able to generate images, and I thought about docker because in the future I would like to upload it to aws to make a scalable architecture.

at the moment I am testing on my windows with docker desktop

[ruby yoke]

wordpress?

um... why?

why not react?

[royal nova]

because I am not a web developer jaja

[ruby yoke]

you're gonna be from the sounds of it ;)

backend and front end

[royal nova]

mmmmmm

can i use automatic1111 but instead of the web you provide modify it to create the api that communicates the container with the wordpress?

[ruby yoke]

probably, though i'd personally advise you use comfyUI not auto1111

[royal nova]

con comfyui puedo hace mi propia api?

[open fox]

Si

[royal nova]

grande

pues le daré caña

[open fox]

You need to active the develop mode in the options to get the api informations

[primal wigeon]

Is flux avaliable for forge?

[primal wigeon]

Noice

[royal nova]

hi is comfyUI compatible with .safesensors models?

[balmy shore]

i havent tried them but it seems capable

i was trying the model in the huggingface ui last night but its like 20gb so i doubt i can run it locally

[ruby yoke]

comfyUI runs safetensors, yes

[brittle lantern]

is it possible to run flux 1 dev on 6600 xt using optimized models or no?

[balmy shore]

when deleting a git clone install of a ui, do i just delete the folder via windows explorer or are there other files it downloaded somewhere

[safe inlet]

ya clone just copies the files you can delete it. That would not remove and requirements you may of installed if you did

[balmy shore]

ok

do u know if theres a way to clean up the stuff it may have downloaded

i have a pip folder in appdata thats like 12gb but i dont think i can just delete it

[safe inlet]

can uninstall packages, pip list would show them or pip freeze > whatevernameyouwant.txt would make that txt file with a list. there is a cache you can clear also, pip cache purge

[balmy shore]

ah thanks, basically just want to clear junk files that seems like the way

wondering if i install a different UI, will it create additional junk folder files for similar dependencies or will it use the already installed stuff

(if the dependencies are identical)

i will install the other UI then try cache purge

does it matter which folder im in when i run it

[safe inlet]

no

have venv activated if you have that

[balmy shore]

ok, it did say it removed 9 files but C:\Users\katie\AppData\Local\pip\cache still has 12gb

5 .body files in the /http-v2/ dir

gpt says this so it seems like its mostly safe to delete the folder manually probably

thats actually not too bad some programs litter files all over the computer

cant seem to find any other bloated folders

thanks

[ornate elk]

Maybe a very small gguf variant of flux could work

[ornate elk]

There are smaller flux variants

[balmy shore]

ah ok

[ornate ruin]

So there's Hires.Fix and Img2Img upscaling. Which one do you use when?

In my scenario I am creating 20-30 images of my self via IP Adapter, or general human portrait style images of which I later want to pick the 2-3 best looking ones

For my use case hires fix is probably better no? It takes way much longer for the generations to finish but il already get a 'finished' picture to choose.

Maybe? I don't know, they are a bit confusing to me

Thanks for any help

[abstract sonnet]

I've been trying to generating ultra-realistic human photos using Flux dev and a dataset of about 40 images. While the results are decent, the skin still looks noticeably AI-generated. Any tips for improving this?

[ornate elk]

Make sure you set hires steps to 10-15 if you use hires fix. If its set to 0 it will take much longer

[ornate ruin]

I usually always do 15, yea

[fair oxide]

Flux on Forge UI returns black generations on any sampling method aside from 'euler simple.' Any tips on how to fix this?

[ornate elk]

try forge realism

[hushed niche]

Hey guys, any time i try to combine two loras (face and concept) i get widely deformed results, not related to any of the loras or the prompt, i tried to do so with forge, as well as comfyui, lora stackers, anyone has any idea why, what im doing wrong and how to fix it?

[brittle lantern]

which gguf model would u say to go for or does NF4 works on amd now

[ornate elk]

nf4 still doesnt work, whats your gpu vram amount?
you can try the Q4:
https://huggingface.co/city96/FLUX.1-dev-gguf/tree/main

[fair oxide]

Do you mean [forge] flux realistic? That still returns a black screen

[ornate elk]

thats strange

[fair oxide]

No issues at all with euler simple. Thanks for the suggestion anyways!

[brittle lantern]

8gb i have 6600 xt

[calm cipher]

In comfyUI is there a way to store a generated image for use in later attempts, other than manually saving it and creating a load image node? (specifically asking this for control net open pose generated poses)

[past rose]

can anyone help with controlnet issues on Forge? I can only get canny to work, the others throw error messages. thanks

Any help with what is causing this error when I try to use controlnet?

Traceback (most recent call last):
File "D:\Users\XXXX\Documents\webui_forge_cu121_torch231\system\python\lib\site-packages\gradio\queueing.py", line 536, in process_events
response = await route_utils.call_process_api(
File "D:\Users\XXXX\Documents\webui_forge_cu121_torch231\system\python\lib\site-packages\gradio\route_utils.py", line 285, in call_process_api
output = await app.get_blocks().process_api(
File "D:\Users\XXXX\Documents\webui_forge_cu121_torch231\system\python\lib\site-packages\gradio\blocks.py", line 1923, in process_api
result = await self.call_function(
File "D:\Users\XXXX\Documents\webui_forge_cu121_torch231\system\python\lib\site-packages\gradio\blocks.py", line 1508, in call_function
prediction = await anyio.to_thread.run_sync( # type: ignore
File "D:\Users\XXXX\Documents\webui_forge_cu121_torch231\system\python\lib\site-packages\anyio\to_thread.py", line 33, in run_sync
return await get_asynclib().run_sync_in_worker_thread(

File "D:\Users\XXXX\Documents\webui_forge_cu121_torch231\system\python\lib\site-packages\anyio_backends_asyncio.py", line 877, in run_sync_in_worker_thread
return await future
File "D:\Users\XXXX\Documents\webui_forge_cu121_torch231\system\python\lib\site-packages\anyio_backends_asyncio.py", line 807, in run
result = context.run(func, *args)
File "D:\Users\XXXX\Documents\webui_forge_cu121_torch231\system\python\lib\site-packages\gradio\utils.py", line 818, in wrapper
response = f(*args, **kwargs)
File "D:\Users\XXXX\Documents\webui_forge_cu121_torch231\webui\extensions-builtin\sd_forge_controlnet\lib_controlnet\controlnet_ui\controlnet_ui_group.py", line 775, in run_annotator
result = preprocessor(
File "D:\Users\XXXX\Documents\webui_forge_cu121_torch231\webui\extensions-builtin\forge_legacy_preprocessors\scripts\legacy_preprocessors.py", line 103, in call
result, is_image = self.call_function(img=input_image, res=resolution, thr_a=slider_1, thr_b=slider_2, **kwargs)
File "D:\Users\XXXX\Documents\webui_forge_cu121_torch231\webui\extensions-builtin\forge_legacy_preprocessors\legacy_preprocessors\preprocessor.py", line 238, in midas
result, _ = model_midas(img, a)

etc...

[idle cairn]

What are the last lines of that log?

[oblique osprey]

Can someone please help me? I've been using webui reForge for a while, but now for some reason changing checkpoints has become a huge pain

It takes around 100s~300s to change a single checkpoint, and sometimes it just straight up crashes my PC

With a 100% CPU, RAM and Disk usage

I've tried a lot of solutions, such as disabling memmapping for loading safetensors, but nothing seems to help

And the weird thing is, even though I'm using it on an HDD instead of an SSD, both Forge and reForge used to be fast before

I have no idea how to proceed and it's making SD unbearable because I have to risk crashing my pc every time I run webui

[ornate elk]

Do you have an SSD?

Is the hdd internal or external?

[oblique osprey]

Yes, but also no
I don't have enough space to use it for Stable Diffusion

[oblique osprey]

Internal

It was working fine before so

[ornate elk]

Make sure you have at least 15gb free space on C drive.
Then you have to check if the windows pagefile swapped over to your hdd. Then set it back to use ssd

[dark socket]

Can someone here who uses efficiency nodes in comfyui help me with how to do image to image?

[oblique osprey]

There's around 50 gb free space and the pagefile is using the ssd, I've allocated 26gb to it as well

[ornate elk]

Okay good. Make sure the pagefile is only enabled for the SSD and not for the HDD too

[oblique osprey]

Just did so

[ornate elk]

And then the best you could do is to move forge to the SSD and leave only the models on the HDD. Then you can link the models in the webui-user.bat.

[oblique osprey]

That does not make any sense

It would imply the problem is with the HDD

But it was working before

[ornate elk]

Did you uses the same models too? Because sdxl and pony load longer

And flux etc

[idle cairn]

No, it implies that Forge would be using the fast drive, and models would be stored on the slow one... But I don't understand either. If the models are on the HDD, how would it be faster? and if it was faster before in this configuration, why would it change now by changing where any of the files are?

[oblique osprey]

Yes, I didn't change models

[idle cairn]

To be honest, in your current configuration, the only thing I would think of is fragmentation...

SSD's don't have this issue, but if you're still using a HDD...

[oblique osprey]

To answer the first one, the thing responsible for the loading speed is the SD itself, not the model
So

[oblique osprey]

Too bad SSDs are expensive af in my country

[idle cairn]

But to move the model into the VRAM, if the models stay on the HDD, it would take the same time.

It does it with every model? Or just some of them?

[oblique osprey]

Every checkpoint, specifically

I'm not entirely sure on this, but I think it slows down dramatically when loading weights

[idle cairn]

I couldn't tell you about that. For example, I have everything on NVME drives and I have a 4090. It took 63 secondes to load a new Flux checkpoint

You were talking about 100s loads?

Ah... 100-300... Sorry

[oblique osprey]

Yeah

[idle cairn]

Have you tried checking your drive's fragmentation status?

[ornate elk]

What's in your webui-user.bat?

[oblique osprey]

I'll take a look at in a minute

[oblique osprey]

Nothing, I've tried it in clean installs

[ornate elk]

and whats your gpu and the model? and how much ram?

[oblique osprey]

It's a 3050, 8GB VRAM and 16GB RAM
This happens whenever I want to swap checkpoints

I'll try and do another clean install again and a defragmentation

[dark socket]

what's a good custom node for resizing images in comfyui?

[ornate elk]

ok make sure you have --xformers --medvram-sdxl --no-half-vae in the webui-user.bat
and make sure the pagefile is set to 16000min and 24000max or higher

also reboot

[hallow latch]

How do I add "clip-skip" to Auto 1111?

[ornate elk]

Make sure the webui is whitelisted in any browser adblocker.
Also try move forge onto C path, it can have issues when its in a system created folder like documents or desktop, downloads etc

[ornate elk]

Settings, User interface, Quicksettings there add clip_stop_at_last_layer

Then apply and reload ui

[hallow latch]

Then I can change it everytime I do a prompt?

There is a lot of settings there I see now, anything more I should add ? 🙂

[ornate elk]

SD_vae

[sleek escarp]

hi, im trying to reinstall an old version of forge-ui but am getting the following error: ImportError: cannot import name 'Undefined' from 'pydantic.fields' (E:\stable-diffusion-webui-forge\venv\lib\site-packages\pydantic\fields.py)

[ornate elk]

Hey, did you downloaded the forge version from the forge github as zip?

[sleek escarp]

no, git cloned

[ornate elk]

Ah okay because here is the old version linked directly:
https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/849

[sleek escarp]

okay, ill try that. not really sure why the current version is giving me overcooked images and noise every other generation as well as taking much longer, seems like the kind of thing you would eliminate as you update

[ornate elk]

Oh thats strange, the new version works good for the most users

[ruby yoke]

cfg too high

[sleek escarp]

lol no, i've been using this stuff since 1.5, definitely not that simple

[ornate elk]

Does it happen randomly?

With the same settings

[sleek escarp]

seems to. I'm using a variant of ponyv6 which i remember doing that sort of thing when it first came out

[ornate elk]

Ponyv6 can give very bad outputs sometimes. It might be the first but there much better ones now xD

[sleek escarp]

which im using. in my old forge 17 setup it worked fine, no full noise results ever

[ornate elk]

Make sure you don't use to much () or :1.4 in.the prompt

That causes pony models to go full noise

Without any word strength it should work always

And yea its random sometimes my prompt with (((word))) worked and then adding another word causes it to go full noise until I removed the ()

[oblique osprey]

Tried it in a clean install

This time at least it didn't crash lol

But still takes a LOT of time and used up all my memory and disk

Although this time I didn't disable memmapping

So maybe there's that

[oblique osprey]

What the hell is going on

disabling memmapping made it worse somehow

I'll try regular forge instead of reforge

I think it was working

[ruby yoke]

maybe it's time to switch to comfyUI

[oblique osprey]

Possibly
But man I hate the node interface lol

[ruby yoke]

you'll get used to it

[calm cipher]

So what am I missing about comfyUI here? I have a random integer being entered into my K sampler (running multiple in parallel and want them all using the same random seed) But this doesn't work! It produces the same image over and over as if it were given only the same seed. I am watching the random number change, but the image does not. What am I missing? ( I feel this is the issue, because if I disconnect it and let the Ksampler make its own seed it starts working again.)

[oblique osprey]

I uh

I must be tripping

Guess I'm using regular forge now

[ruby yoke]

maybe you should let the ksampler make its own seed then

[calm cipher]

Already said why I would rather not have to do that.

[safe inlet]

i think that way it will change seed after generation is completed meaning the whole workflow

[calm cipher]

what do you mean the whole workflow? I have watched it run to completion multiple times if that is what you mean; and it only returned the exact same image each time. I have also turned off the other parallel generations and the single one produces the same image only.

[tranquil vigil]

?

[ruby yoke]

doesn't seem to be working for you. you sure this is what it's supposed to do?

[safe inlet]

i dont know that node but wonder maybe you have to set min and max, its at 0,0 maybe generating the same seed /shrug

[safe inlet]

random zero generator

[calm cipher]

was looking for a node that did that, would display the output, thanks! Guess that solves that mystery.

[hallow latch]

Same settings as before and was fine but now when I try to render

"NansException: A tensor with NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check."

??

Can someone ban this guy? ^

[ornate elk]

Which webui and which GPU?

[hallow latch]

Auto 1111 and 3060 - 12 GB

[ornate elk]

Okay hmm when do you get this error?

Any extension used?

[hallow latch]

I did 10 images before and it was in a que with others so all was aborted. Tried to run again but same as before, I use ADetailer and Tiled VAE with 1536 x 96 and Clip Skip 2 and SD Vaae Anything.PT

I restarted the console and it working now with the same stuff :/

[visual pilot]

The history in my state manager disappear, when i make a new image it is saved for the moment but when i restart the webui it disappears from the history, what could it be? i'm using forge

[brittle lantern]

if i download Q4 FLUX GGUF (dev) would i need to download the normal flux text encoders and VAE or is there a diff text encoder or VAE needed for GGUF models

[ornate elk]

You can use the same as for normal flux but for the t5 I suggest using the t5 q8 gguf Text Encoder

Reduces RAM usage by 10gb

[brittle lantern]

where would i grab that

same guys hugging page profile?

[ornate elk]

https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/tree/main

[brittle lantern]

and for my 6600xt should i use comfy or forge?

[ornate elk]

Forge

[brittle lantern]

ill assume reduces vram and ram usage or is it easier to use on AMD?

[ornate elk]

There also enable the tiles VAE option at the "never oom" at the bottom

[brittle lantern]

yes i remember that

ur tut is pinned correct?

[ornate elk]

Yes in the first link

[brittle lantern]

what should i do?

[fair oxide]

how i can start doing animations in stable diffusion ?

[iron lantern]

how or what would i prompt for 2 girls in the same image .... for a pic

[twin ermine]

It's been like this for more than an hour I think, without continuing from 63%...

help pls

[ornate elk]

thats an internet connection problem

open up a cmd and run
pip cache purge
then delete the venv folder and relaunch the webui-user.bat

[brittle lantern]

ah... idk where that came from then

[ornate elk]

hey dont use easy diffusion (its outdated), better use Automatic1111 webui

youll find a install guide in the pinned messages of this channel

[twin ermine]

It's too late for that, I'm not going to start all over again xd

[ornate elk]

okay then maybe tomorrow xD

[twin ermine]

But do you know why it gets stuck?

If necessary I will follow your guide...

[ornate elk]

in your screenshot its downloading with 123kb/s and needs 600mb to finish

so its just really really slow.

but i dont want you to start with a bad webui experience when there are much better options out there.
easy diffusion got their last big update in august 2023

[twin ermine]

That's because I'm far from the router and I use wifi, besides having only adsl... I'll put it on ethernet and it will go faster. But if it doesn't work after that, then I'll follow your guide.

thx

[ornate elk]

okaynp , it should still work, but its not that feature complete as Automatic1111

[mental hazel]

hey guys, im trying to get into comfyui and am currently using mickmumpitz' workflows to try and generate a character sheet with multiple emotions, as of right now im running into an error

"expected scalar type Half but found BFloat16"

some solutions say to update xformers, but i am not sure where to go to complete this. can anyone help? if anyone knows of other solutions im open to those as well. thanks in advance :)

[fair oxide]

do you have any video that can help me make videos on stable diffusion ?

[ornate elk]

nope sry

[fair oxide]

aww :/

[brittle lantern]

@ornate elk im also doing comfy ui and ive gottern a cmd error launching the bat file but it closes so quick idk what it says what do i do

[ornate elk]

its self closing?
then open up a cmd and run
where python
and then
python --version

and show a screenshot

[ornate elk]

okay and you installed hip sdk 6.1?

[brittle lantern]

yh

[ornate elk]

and forge works?

[brittle lantern]

yh

[ornate elk]

okay then delete the venv from comfyui again

[brittle lantern]

done

i open bat again?

[brittle lantern]

it did it again

[ornate elk]

okay then you need to get rid of the microsoft store python version

it causes this issue

go into system and then under APPS uninstall the python version fro mthe microsoft store

after that delete the venv folder again.

if it still closes, then you need to reinstall python 3.10.11 64bit

[royal nova]

Good people, is it normal that it takes me 25 minutes to generate an image with stable diffusion with a GTX 1650 Ti?

[ornate elk]

Hey, nope, what's in your webui-user.bat

[royal nova]

im using docker and api

[sonic pulsar]

getting this issue while trying to start lora training using https://github.com/derrian-distro/LoRA_Easy_Training_Scripts

[brittle lantern]

forge is giving me this with SDXL

and it took 10 min for a gen

whats the fix?

[ornate elk]

what settings did you used?

screenshot the txt2img page

[proper zenith]

pretty new to all this just curious as to how fast a 4090 should be able to make a 3 second video at 24fps? its taking me 6-7 mins just seems so long

[serene kelp]

hey guys, can someone point me in the right direction how to set up diffusers with multiple controlnets with start_step, end_step and strength settings like I can do in comfyui or automatic1111? I can provide the code I already tried. I have not found any implementation like that anywhere :S

I'd basically like to configure my controlnets with a list of dicts like this:

    {
        "model_path": "monster-labs/control_v1p_sd15_qrcode_monster",
        "start_step": 10,
        "end_step": 22,
        "strength": 0.25,
    },
    {
        "model_path": "latentcat/control_v1p_sd15_brightness",
        "start_step": 1,
        "end_step": 50,
        "strength": 0.2,
    },
]```

[snow swallow]

I follow the tutorial pinned for installing Confyui in my AMD ryzen 4650g (no gpu). https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides#amd-comfyui-with-directml

Apart from an error "?set_tracing@SavedTensorDefaultHooks@at@ @SA_N_N@Z" it loads comletely, starts comfy, I am able to generate one imagem, but when I try a second one the terminal just closes becore I can see what error caused it. Can someone guide me trough solving this issue?

[ornate elk]

I guess its an out of vram error

[winter salmon]

How do I place my prompt

[proper zenith]

I'm getting this "GET was unable to find an engine to execute this computation" when cogvideox tries to decode. anybody have this issue before?

[snow swallow]

is there any way to pretent this crash?

[native ridge]

i have this error what can be ?

Installing requirements
Legacy Preprocessor init warning: Unable to install insightface automatically. Please try run pip install insightface manually.
Launching Web UI with arguments:
Total VRAM 8192 MB, total RAM 24508 MB
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 2060 SUPER : native
Hint: your device supports --pin-shared-memory for potential speed improvements.
Hint: your device supports --cuda-malloc for potential speed improvements.
Hint: your device supports --cuda-stream for potential speed improvements.
VAE dtype: torch.float32
CUDA Stream Activated: False
D:\stable-diffusion-webui-reForge-main\venv\lib\site-packages\transformers\utils\hub.py:127: FutureWarning: Using TRANSFORMERS_CACHE is deprecated and will be removed in v5 of Transformers. Use HF_HOME instead.
warnings.warn(
D:\stable-diffusion-webui-reForge-main\venv\lib\site-packages\timm\models\layers_init.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers
warnings.warn(f"Importing from {name_} is deprecated, please import via timm.layers", FutureWarning)
Traceback (most recent call last):
File "D:\stable-diffusion-webui-reForge-main\launch.py", line 51, in <module>
main()
File "D:\stable-diffusion-webui-reForge-main\launch.py", line 47, in main
start()
File "D:\stable-diffusion-webui-reForge-main\modules\launch_utils.py", line 542, in start
import webui
File "D:\stable-diffusion-webui-reForge-main\webui.py", line 19, in <module>
initialize.imports()
File "D:\stable-diffusion-webui-reForge-main\modules\initialize.py", line 49, in imports
from modules import shared_init
File "D:\stable-diffusion-webui-reForge-main\modules\shared_init.py", line 5, in <module>
from modules import shared

File "D:\stable-diffusion-webui-reForge-main\modules\shared.py", line 6, in <module>
from modules import shared_cmd_options, shared_gradio_themes, options, shared_items, sd_models_types
File "D:\stable-diffusion-webui-reForge-main\modules\shared_items.py", line 4, in <module>
from modules import script_callbacks, scripts, ui_components
File "D:\stable-diffusion-webui-reForge-main\modules\script_callbacks.py", line 11, in <module>
from modules import errors, timer, extensions, shared, util
File "D:\stable-diffusion-webui-reForge-main\modules\extensions.py", line 9, in <module>
from modules import shared, errors, cache, scripts
File "D:\stable-diffusion-webui-reForge-main\modules\scripts.py", line 10, in <module>
from modules import shared, paths, script_callbacks, extensions, script_loading, scripts_postprocessing, errors, timer, util
File "D:\stable-diffusion-webui-reForge-main\modules\util.py", line 220, in <module>
file_name: str | None = None,
TypeError: unsupported operand type(s) for |: 'type' and 'NoneType'

that the full error

[calm cipher]

dumb question out of the way first, did you try to run pip install insightface manually?

[native ridge]

I don't know what that is, I'm ultra new to everything related to AI.

[calm cipher]

try opening a command line in that folder and run pip install insightface

I am new too, but thats what the first line in that error asks, so maybe that will work? Otherwise the real error looks like TypeError: unsupported operand type(s) for |: 'type' and 'NoneType' and I have no idea what that means.

I need some help. On swarmUI I try to use the midas preprocessor and get ComfyUI execution error: [Errno 2] No such file or directory: 'C:\\Users\\___\\Documents\\GitHub\\SwarmUI\\SwarmUI\\dlbackend\\comfy\\ComfyUI\\custom_nodes\\comfyui_controlnet_aux\\ckpts\\lllyasviel\\Annotators\\.cache\\huggingface\\download\\dpt_hybrid-midas-501f0c75.pt.501f0c75b3bca7daec6b3682c5054c09b366765aef6fa3a09d03a5cb4b230853.incomplete'

but I have no idea what that file is or how to get it. I tried placing the midas PT file there, but that did not work. Anyone know how to fix this?

[safe inlet]

looks like failed auto download

[spice stratus]

I dont know if this is the right channel but has anybody used roop for putting you face into generated images? I am able to get one face pretty well but if i try to do me and my friend, then it just uses my face twice. I even got a clearer photo but that didn't help. Any ideas or other extensions yall use for adding your face to stable diffusion created images?

[ornate elk]

Hey, roop is outdated. Better use the Reactor extension

[untold flame]

Hey people, I am having a very weird issue with comfy, and I have no idea how to fix it

effectively, all nodes that have to do with upscaling are just gone, and will not load or install. I have no idea when or why it happened, but every single one is gone. Upscale to size, load upscale model, esrgan upscale, ultimate upscale, none of them will load at all

I tried updating, git pulling, diving through the files and I just can't find it

I am trying to use krita AI image generator backend, and it refuses to connect until all of those nodes are restored

I am gonna try to make a new venv and see if that works, but I am not sure

[silk agate]

last time i use ai last year theres only 1 which automatic111. now there are 4 additional webui, which 1 is beginner friendly?

[untold flame]

beginner friendly? I have no idea. I use very high level AI and I use comfy and other interfaces

been doing this for 2.5 years now 😅

[ornate elk]

Very strange issue. If nothing helps, reinstall comfyui. Thats also good if your torch is on the older 2.1.2
New one uses 2.3.1

[untold flame]

I was able to get it working with a venv. Not sure what happened

I am on 2.5.1 torch 😅

[ornate elk]

Ohh

[silk agate]

i want to try each 1..do i just delete the folder if i dont use it later?

[ornate elk]

Yep

[silk agate]

thanks

[cold kernel]

how do i add a tagger model into kohya?

[weak eagle]

Hey everyone, I'm just playing around with generation again after taking about 1.5 years out.

I've set automatic1111 webui back up and plugged in some test prompts but all of the new models/checkpoints I've picked up just take an absolute age (20-30 minutes) per image, and usually just output a complete garbled mess.

Confused, I grabbed some old checkpoints I had backed up and with the same prompts they're generating images exactly as before in about 30-40 seconds per image.

What's funadmentally changed?

[fringe flicker]

Quick question I’m useing forge and was wondering where I put embedding files like easy negative

[ornate elk]

Hey what's your GPU?

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

[ornate elk]

In the embeddings folder in the webui folder

[ornate ruin]

When a checkpoint says VAE baked in, does that mean I have to select 'none' SD VAE? Or can I leave it at automatic?

[ruby yoke]

it means the VAE is included in the checkpoint, so you don't use another VAE

[ornate ruin]

I also can't find anything about the prompt style that Pony Realism uses, heard it's pretty good for realism fantasy

So it says

Prompt Style:
Positive: score_9, score_8_up, .. etcetera
Negative: score_4, score_5, score_6

If someone has a link about it or an easy explanation 😅

[ornate ruin]

So it's saver to put SD VAE to none in those cases, alright thanks

[ruby yoke]

did you look on the civitAI page for it?

[ornate ruin]

Yes there it says those prompts are used for this checkpoint

[ruby yoke]

pretty sure the exact same prompt style for base pony is used for pony realism.

[ornate ruin]

I just learned that pony is the base model much like SD1.5 or XL.. okay that just makes it so much more confusing

[ornate ruin]

So Pony in it's self just uses a different prompt style? I use Automatic1111

[ruby yoke]

yes. and there's a link on the main pony page on civit that takes you to an article the dev wrote that explains why those tags

this is really not the right channel for pony, however, or probably the correct discord

[ornate ruin]

Ahh okay well thank you. I didn't even realize pony was the base model and wondered how my results were so much worse than expected. Well now I can look up base pony for further information haha

[rugged summit]

hey guys im using forgeui with pony, dpm++2 karras 30 steps and highsres 15step, generating at 896x1152, it takes around 20 mins per picture with a gtx1650 4gb, is this the best time achievable? any suggestions to make it a bit faster?

[ornate elk]

Hey try to use a lower resolution and 10 hires steps

Also enable the tiled VAE at the bottom

[rugged summit]

hires is latent forgot to say

i'll try to lower the resolution and enabling tiled vae, but the pictures are gonna be lower quality right? is this a trade between quality and speed or should i upscale more etc

(from the command box, my generation is about 10s/it while generating, and around 40s/it while upscaling

[ornate elk]

With a 4gb vram card you can't get better speeds. Only forge can even let you use pony with that amount

But make also sure if you have only 16gb of memory, to increase the Windows Pagefile size

That can help getting a bit more speed too

[rugged summit]

how do i do this?

[ornate elk]

https://www.tomshardware.com/news/how-to-manage-virtual-memory-pagefile-windows-10,36929.html

Enable it only for the C drive and disable it for any other drive.
Set it to customized: 16000 min and 24000max.
Then restart the PC.

[rugged summit]

thank you

[iron lantern]

having a problem of prompting two girls one with glasses and one with none why do both keep coming up with glasses any help is gratefull thanks ❤️

in the same image

[ruby yoke]

what model are you using?

[primal wigeon]

Can loras be used for other base models they weren't made on

[ruby yoke]

no

LoRAs are a small set of instructions that update the specific weights of the model they were trained for. no other model is going to have the same weights

[clever cargo]

...what....what am i doing wrong..?

[ruby yoke]

what are you trying to get them to do? and what is up with this negative prompt?

[spice stratus]

Yup, that works amazing!

[iron lantern]

pony xl

[ruby yoke]

well that would be why. Pony, all versions, doesn't have the comprehension necessary to do that, you'll hve to create them withotu glasses and then inpaint the glasses onto one of them

[clever cargo]

all i'm trying to do is get a decent looking image... every iteration is low quality and messy. I've tried adjusting steps, cfg, sampler, scheduler, loras/no loras

[ruby yoke]

what model are you using? and what is up with this negative prompt?

and this is all you have for a positive prompt:

[clever cargo]

I was trying out negative textual embedding. I don't quite understand it but the creator of it said to use it like that so that's what I tried.

I've had great looking stuff without ANY negative prompts whatsoever

Now I'm trying to back up as much as I can and just work with the bare minimum before I start adding things.

[ruby yoke]

dont' do that. best practice, never use negative prompts at all, but especially if you're not really sure what they do and why you use them. they have a very specific use. craft a good, positive prompt, with good details, don't ramble, and don't ask the AI to draw stuff it cant' - like what someone's thinking.

[clever cargo]

alright, I'll try again several times and I'll let you know if I'm still struggling -_-

[primal wigeon]

I'm not really happy with the armor results i'm getting so far. Trying to replicate this exact armor from Skyrim.

Should I use img2img for the top one and try to make some kind of LoRA out of it?..

[rugged latch]

i used to be an arrow like u, til i took an adventurer to the knee

[snow breach]

Hey everyone, what are we using for training lora and generating SD 3.5 these days? Seems like Kohya, A1111, and Forge are not yet caught up?

[ornate elk]

Hey, OneTrainer supports 3.5 lora training.
And comfyui for generating.
But forge should support 3.5 soon

[snow breach]

Thanks, I'll check out OneTrainer. Glad Forge is going to work soon, I still don't like ComfyUI 🙂

[iron lantern]

how do i do that

[twin ermine]

Ok, I decided to uninstall easy diffusion and move on to installing automatic1111 via a recent guide (webui release v1.0.0-pre... I should have done it from the beginning xd). I get these errors, can anyone help me pls?

I think I uninstalled easy diffusion correctly, so I don't understand those errors u.u

pytorch was still installing for 2.1 gb of 2.5

[ornate elk]

Hey, you should have used the guide from the pinned messages

But for this error I need to see the whole cmd log.

[twin ermine]

okay 2 secs pls

I'm going to click "run" again

xd it's reinstalling everything for me

😭

I followed that guide because it seemed to be the most recent and easiest way to install it, but if you tell me to follow the pinned messages, I believe you xd, you are the expert @ornate elk

I would just need to know if I have to search for pytorch or certain elements and uninstall them first (maybe it gives me an error because the program overwrites them or something, I don't understand this)

[ornate elk]

Now it looks better. Only your Download speed could be an issue here.

[twin ermine]

yes, sometimes it downloads in half an hour and other times it takes me 2... for the same thing, it depends on the day xd

[ornate elk]

You don't need to search for leftover stuff.
Stable diffusion only uses and download what is needed.
If you want to start completete clean you can open up a cmd and run
Pip cache purge
And then folllow the guide from the pinned messages.

[twin ermine]

okay, I will do it

thx! 😄

[ornate elk]

No problem ^^

[alpine kelp]

hello?

[ornate elk]

Hey

[alpine kelp]

where i start to learn this

[iron lantern]

whats going on here ???

[alpine kelp]

Hi, so where 2 start?

[vocal burrow]

Can tell, log is very incomplete from this screenshot. Most likely you re loading an incompatible model with whatever workflow you re using. Need to see full log.

[vocal burrow]

• https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki
• https://stable-diffusion-art.com/
• https://www.youtube.com/@Aitrepreneur/videos
• etc

[alpine kelp]

So this is free right?

[iron lantern]

Put it in a word document?

[vocal burrow]

in a .txt, and drop said file in here

[fair oxide]

why automatic1111 is generating pictures so slow ?

[ruby yoke]

What gpu do you have?

[fair oxide]

1060 gtx

its wierd bc on other stable diffusion with same setup (steps,resolution) it took like 10 minutes to generate picture

on a1111 it takes me 30 minutes

to make one

[vocal burrow]

what command line arguments are you using ?

[fair oxide]

what

i think none

[vocal burrow]

that would be why.

[fair oxide]

how to add them ?

[vocal burrow]

open webui-user.bat with a notepad. find the set COMMANDLINE_ARGS= line and add --medvram --no-half --xformers to it

then relaunch the same .bat

[fair oxide]

so i need to tun on webui-user.bat?

[vocal burrow]

turn on webui-user.bat ? what do you mean ?

you run it.

Just like you did before.

[fair oxide]

i am using webui.bat to turn on

[vocal burrow]

yeah that s not the correct way to do it. you d want to run webui-user.bat (which in turn will call webui.bat, cf last line of your screenshot)

[quartz musk]

I just started using this AI agai nafter a few weeks and every ckpt I use has this load error, cant seem to find a resolution online. Was working fine previously?

GPT saying the comfyui may be out of date, im not sure where or how to check if it needs updating

[fair oxide]

now its saying OutOfMemoryError: CUDA out of memory. Tried to allocate 40.00 MiB. GPU 0 has a total capacty of 3.00 GiB of which 0 bytes is free. Of the allocated memory 9.86 GiB is allocated by PyTorch, and 180.28 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

[vocal burrow]

screenshot your webpage so I can see what settings you re using to generate a pic.

[fair oxide]

dm

[vocal burrow]

enable fp8 for sdxl models in the settings tab

[fair oxide]

i am using pony

[vocal burrow]

it is an sdxl model

[fair oxide]

which

[vocal burrow]

pony model is based on sdxl model. It s "just" trained even more with some extra pictures.

so enabling "fp8 mode for sdxl models" will work with pony models too

they share the same architecture.

[fair oxide]

oh kk

[ornate elk]

you have a 3gb gpu?

[fair oxide]

yep

[ornate elk]

then you should definitly use forge

and not a1111

[fair oxide]

oh

well

whats the diffrence between a1111 and forge ?

[ornate elk]

forge has more optimisations for vram and ram management allowing a better performance for low end to midrange cards 2-6gb vram

[proper zenith]

Anybody know what this means? "Given groups=1, weight of size [3072, 16, 2, 2], expected input[36, 32, 60, 90] to have 16 channels, but got 32 channels instead" im using cogvideox and it happens when i try to do anything but 49 frames for a img2vid

[ruby yoke]

https://discuss.pytorch.org/t/runtimeerror-given-groups-1-weight-64-3-3-3-so-expected-input-16-64-256-256-to-have-3-channels-but-got-64-channels-instead/12765/2

torch isn't happy with you

[vocal burrow]

you're feeding something unexpected to something else. Can't tell more without any context.

[vast hull]

do u have swarmui experience? @ornate elk

[ruby yoke]

#🐝｜swarm-ui for swarm questions

[ornate elk]

not much as its prety much comfyui in the backend

[quartz musk]

OH LEGEND THANKS

caps

[vocal burrow]

Also 3440, 1440 resolution is asking too much without any kind of upscaling

sdxl and sd3 are trained for 1024x1024

[ruby yoke]

except don't update comfy itself with manager. use the update script in /comfyUI/update for that

[vocal burrow]

you can deviate a little from its trained resolution but if you want more you'll need some kind of upscaling.

[novel temple]

hey lads, i need help with my workflow
im trying to upscale and sharpen an image but it ends up looking oversharpened and overexposed
here is the workflow i use/made

[ruby yoke]

turn your cfg down to 4

[novel temple]

of which node?

the first Ksampler or 2nd?

[warped canyon]

im trying to add dataset-tag-editor extension to forge, but its not loading

[fair oxide]

How do I fix these weird lines?

The image looks really blurry

[ruby yoke]

both

[vast hull]

can someone hlep me>?????

[ruby yoke]

what happened?

[vast hull]

mcmonkeyt never respojnded to dm or a ticket i made on his server @ruby yoke

[ruby yoke]

he hasn't been online all day. he just about lives on discord, however. he will reply to you. he might have had to work extra today or something

[void dome]

where is the bot

[onyx solar]

#artisan-faq

[fair oxide]

So turning up denoiseing fixed the blurriness

But now I getting duplicate body parts

Like belly buttons

Any idea how to fix?

[safe inlet]

https://tenor.com/view/futurama-yes-gif-take-my-money-shut-up-and-take-my-money-gif-8581127

[visual bane]

is there a way to pass in different width and height for different image in batch in diffusers pipeline?

[languid dew]

Hi, I am using ComfyUI with M1 Max 32 GB. Unfortunately not a very fast environment for Stable Diffusion with shared memory compared to newest NVidias...anyone out there having comparable experiences wth newer Mx processors and higher Memory? Is it worth an upgrade? Thanks Gerald

[rugged latch]

g wront that like an email

lul

[fair oxide]

This looks like a scam

Why you have mod/admin in your about me and not your roles 😭

[fair oxide]

Can someone help me with this?
If anyone replies I will reply in the morning
It's 3am for me now and I'm going to bed

[ornate elk]

Change the upscaler to Resrgan4x+ Anime6b and set the denois to 0.5

[novel temple]

yea that helped, thank you

[celest delta]

So i installed stable diffusion web UI like 2 years ago, and barely used it since then... i want to get back to it, but since it's been 2 years, there's a ton of new stuff... I've been told it's better to reinstall everything fresh with the pinned message

do i need to uninstall what's already on my PC or something?

[ornate elk]

You can install it fresh with the guides from the pinned messages and then move the models over

[celest delta]

yeah but what about what's already installed on my PC (stable diffusion and webui, not the models), will the fresh install replace stuff? since some files are on my PC

[ornate elk]

You can delete it

After moving the models

Nothing more to do

[celest delta]

thanks, will do

[ornate elk]

If you want to start completely clean open up a cmd and type
Pip cache purge

[celest delta]

is that worth doing?

[ornate elk]

That will remove old packages

Yea that gives you back 5-10gb space

[celest delta]

thanks

i'll have more questions later on since there's a loooot of new stuff i have no idea about :p

[ornate elk]

Sure xD

[woeful falcon]

I know this can be a reddit post but I wanted to ask it here in discord. I'm a devops guy and I know Python. However, until yesterday, I did know nothing about stable diffusion or AI in general. What I'm interested in is virtual clothing experience. I've looked at different models and installed ComfyUI so far but I don't know how to proceed from here. I know there is IDM-VTON and OOTDiffusion but I don't know how to use them. I also know that there are some LoRAs involved to fine tune the experience and upscaling methods but I still haven't used. Are there any guide that you can provide. What should I do to start?

I'm working with MacBook Air but I can spin up a GPU instance on Google Cloud and install ComfyUI there. It's not a problem with me. The only problem I'm having is to understand how those technologies fit together.

[clever latch]

Hello,
I am using the "Inpaint" API but I don't know how to generate the "mask" that should be include in the parameter. Any ideas?

[fair oxide]

Yes

[fair oxide]

which stable diffusion i need to get if i got bad gpu ?

[vocal burrow]

anything 1.5 derivated should be the easiest to run.

[fair oxide]

can you send link to this ?

[vocal burrow]

you can find plenty of sd1.5 models using the filtering system on civitai

[fair oxide]

i don't have problem with model

i have problem with stable diffusion

[vocal burrow]

"stable diffusion" refers to the model.

If you re looking for the more lightweight client to use stable diffusion models, comfyui or forge are probably your best bet.

[fair oxide]

i mean stable diffusion like automatic1111

[vocal burrow]

Maybe sdnext too

[fair oxide]

which is the best for making nsfw ?

[vocal burrow]

it does not matter.

[fair oxide]

kk which is fast and makes good pictures ?

[vocal burrow]

same question, same answer; comfyui, forge or sdnext. Pick any of them depending of which workflow click the best with you.

[fair oxide]

oke

[fair oxide]

forge is lllyasviel ?

[vocal burrow]

yes

[ornate elk]

Forge webui

Best for 3gb vram

[fair oxide]

ye but it makes kinda shit photos

1024x1024 50 steps

[ornate elk]

Thats not the fault of the UI

You have to use a good model and prompts

[fair oxide]

idk

i was using other ui and it was making way better pictures

[ruby yoke]

which other ui? what were the settings with it?

[fair oxide]

i don't know which it was

[ruby yoke]

where were you using it?

[fair oxide]

what do you mean

[ruby yoke]

the prevous UI you were using that you were getting better results with. what UI was it? what was it called? were you using it on a website or on your own machine?

[fair oxide]

own machine i don't remember name i only remember how it was looking like

[ruby yoke]

that's unfortunate. were you using the same model and prompts?

[fair oxide]

yes

same resolution and same steps

[ruby yoke]

okay then what i would do is start playing around with the settings. different UIs process differently.

[fair oxide]

and my nsfw prompts are showing rarly

[fair oxide]

is it gonna make any diffrence when i am going to put UI on ssd or hdd disk ?

[fair oxide]

know any other i tried every is bad :/

[ruby yoke]

the problem is going to be the settings you are using. Becasue everyone else uses all of those and gets good images. so it's not the UI, it's how you are using it

[fair oxide]

what should i change in settings ?

[worn vale]

Hi, I'm trying to set up stable diffusion but it keeps coming up with this error

even though I have installed python

[ornate elk]

hi, make sure you checked "add python to path" when installing

[worn vale]

I did

[ornate elk]

you can also run the installer again and click, Modify, Next, and then check "add python to environment variables"

also delete the venv folder from the webui if you have one

[worn vale]

alreayd ticked

and theres no venv folder

[ornate elk]

you installed the wrong python version

[worn vale]

what version do I need

[ornate elk]

you need 3.10.11 64bit

[worn vale]

okay

uninstalling 3.13 now

This one right?

Weird... should I upgrade or custom install?

[ornate elk]

upgrade

maybe you had an older version installed too

[worn vale]

must've.

right lets give it a go

still coming uip with the same thing

[ornate elk]

which webui do you want to install?

[worn vale]

I didn't know there was different versiobns

[ornate elk]

also open up a cmd and type
where python
and
python --version

[worn vale]

It's in AppData/Local. but when I did the version command the same message thats in the webui thing shows up

[ornate elk]

seems like you have a windows store python installed too

try uninstalling that

[worn vale]

bruh

[ornate elk]

and do you want to install Automatic1111 or which webui?

[worn vale]

I think the 1111 one is the github thing i downloaded

yeah

[ornate elk]

did you git cloned it?

[worn vale]

no downloaded to zip

ands un zipped it

[ornate elk]

ah thats not how it works

Here is a short install Guide:
#🤝｜tech-support message

or here: https://github.com/CS1o/Stable-Diffusion-Info

[worn vale]

thanks

[ornate elk]

if you still get the error after that guide. we can fix it manually

[worn vale]

Like this with no space?

[ornate elk]

yea but you need a second - at xformers

--xformers

[worn vale]

I have a 1660 so I added the medvram bit and removed the vae

[ornate elk]

perfect

[worn vale]

gawdamn

[ornate elk]

okay then open up a cmd and type again
where python

thne copy the path it shows that points into appdata local python310

then edit the webui-user.bat again

at the line PYTHON=
you add the path with " "

[worn vale]

Okay, gimme 5 mins I’ll brb

[ornate elk]

should look like this:

[worn vale]

its doing something

[stable birch]

please help me, Im just clean install and add some extention and it happen when Im reload

[ornate elk]

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

[stable birch]

oh wow it' work!!
why did you know? so expert

so don't add Sadtalker and TensoRT right

thanks you

[ornate elk]

sadtalker is outdated and tensorRT can make compatibility issues

and venv folder needs to be deleted when deleting extensions

so it gets recreated clean

[stable birch]

I have used Stability Maxtrix before
now try to install directly

I think it would be better.

[ornate elk]

yea its better

[stable birch]

When using Stability Matrix, even after deleting and reinstalling, there is still a problem with the basic program being incompatible.

When I use git hub, when I clean install everything works fine.

[ornate elk]

yea stability matrix adds just a point of failure. its easy but not needed

@stable birch also make sure you have the correct performance args in the webui-user.bat

[stable birch]

Im newbie just comeback after 11 month ago

[ornate elk]

ah okay, whats your gpu?

[stable birch]

Im follow you on github and follow installation

[ornate elk]

ah nice okay

[stable birch]

RTX4090

[ornate elk]

okay then your ready

you dont need tensorRT with that card

[stable birch]

how to find my spec

or just GPU is fine right?

Before I came to ask I tried it myself for hours with the latest problem lol.

reinstall once and add Sadtalker again 🤣

[worn vale]

https://tenor.com/view/huh-gif-4961456348134506326

[ornate elk]

Sadtalker was nice. But it wasn't updated anymore :/

[ornate elk]

Can you show the full cmd log?

[worn vale]

yeah

I literally have zero clue what any of that means

[ornate elk]

Thats not the full cmd log

[worn vale]

oh yeagh sorry

its verrrry long

[ornate elk]

Then close and relaunch the webui-user.bat

And then copy the log

[worn vale]

it doesnt really matter if it shows my like users name thing right?

[ornate elk]

Yep

[worn vale]

yep it doesnt matter?

[ornate elk]

Yea if its not your real name

[worn vale]

It's part of my name

[ornate elk]

Yea shouldn't be an issue but its up to you

In the log isnt anything what can't be shown

The webui IP is also only a local IP. Its the same for everyone

[worn vale]

Sorry that's a lot of images but I wasn't sure how else to do it

[ornate elk]

Did you relaunched it?

[worn vale]

no gonna do it now

[ornate elk]

What you can try is to open up a cmd and type
Pip cache purge
Then delete the venv folder and relaunch the webui-user.bat

[worn vale]

This is where its at rn

[ornate elk]

Ah okay

Then just let it load

Looks good now

[worn vale]

Now it's doin a dwld

[ornate elk]

Yea thats the default model

[worn vale]

sseet

its done

I just need to add a model and re run it right?

[ornate elk]

You can download models from Civitai.com and put them into models/stable-diffusion folder.
Then click on the blue refresh button on the top left

Models files are at least 1.9gb or larger

Everything smaller is something else like a lora or embedding

[worn vale]

Where would I put a VAE if a model has one?

same folder?

[ornate elk]

Models/vae

[worn vale]

thanks

[ornate elk]

Then in the webui go into settings, User interface, Quicksettings, there add sd_vae and hit apply and reload ui

Then you get a VAE dropdown Menu in txt2img

Im off now

[worn vale]

real quick

tgheres no quicksettings

oh wait

Got it 😎

@ornate elk Thank you so much for the help

[ornate elk]

Dont forget to hit apply

[worn vale]

i did dw

[ornate elk]

No problem 🙂

Have fun

Gn

[worn vale]

Gn

[fair oxide]

Thanks

[stable birch]

Adetailer don't change anything on image but don't have any error too

[onyx solar]

You probably need to test cranking up the denoising strength to a higher value.

[stable birch]

thanks

[rigid notch]

Hello everyone, I'm looking for a controlnet model for InstantID with SD Forge and SDXL checkpoints. Maybe I don't look where I should, because I can't find a proper repository that contains one. Can you help me find one ?

[delicate citrus]

Hi, I Will buy a new PC, what GPU do you recomend me? I wanna a good GPU

[ornate elk]

Hey, What's your budget?

For Ai stuff the general rule is the more VRAM the GPU has, the better.
Also nvidia is better for ai tasks currently

[delicate citrus]

I'm looking for some with 16vram

[uncut jewel]

Hey everyone, i'm having this error every months, litterally, each time i manage to solve it but always forget to pin how i did solve it and now i seems to struggle af to solve that same recurrent error... Anyone can help ? I tried to reinstall torch py -m pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

I guess the issue come from Cuda then

[delicate citrus]

@cs1 something between 340 and 510 USD (2 a 3 mil BRL)

[ornate elk]

In that price range you won't find any nvidia GPUs

With 16gb vram

AMD has some GPUs in that Price range

[delicate citrus]

can be 12vram, not problem

is not a problem I take one AMD, so? For my firs GPU I need vram and low price

[ornate elk]

Its not as easy to run ai stuff on AMD but its possible

[delicate citrus]

hmm not AMD so XD

can be a NVIDIA

[ornate elk]

A nvidia 3060 12gb maybe?

Or much better a 4060 ti with 16gb

[delicate citrus]

👍 perfectly

[ornate elk]

For euro prices it's around 450-520€

Dont know how much it costs for you

But make sure it's the 16gb version and not the 8gb one

[delicate citrus]

its a good price for me 🙂

[ornate elk]

Nice

[bold field]

I have a problem, I'm launching the AUTOMATIC1111 locally on my pc; in CMD it says "Running on local URL: http://0.0.0.0:7860"
And then (I'm using the autolaunch param in webui-user.bat) when it opens in browser, it shows it's on local host.

And yea, i cant access it from another device in the same network.

[ornate elk]

Hey, to use it from other Devices you have to add --listen

[bold field]

I did ):

[ornate elk]

Then its your PCs local IP+Port from the webui

Open up a cmd and type
Ipconfig

[bold field]

Also quick question is missing imports in webui.py a problem

[ornate elk]

What do you mean?

[bold field]

Theres the webui.py file, and u can see all the "couldnt be resolved" warns

[bold field]

Ok

[ornate elk]

Yea because its a script

It will show that in the cmd if it happens

But you always launch the webui-user.bat not the other files

[bold field]

Ik, ik

1. what the cmd says
2. what actually happens

It also works if i use 127.0.0.1 instead of localhost

[ornate elk]

Remove --opt-sdp-attention as it doesn't work when you already use xformers

[ornate elk]

Which browser is that?

[bold field]

Chrome

[bold field]

Ah

[ornate elk]

Strange and when you try to use the PC IP on an other device?

With the port

[bold field]

Which one in the ipconfig is it

Ok its working now!

used the ipv4

Another question, how to set sampling method via the payload?

[ornate elk]

Ipv4

[ornate elk]

My guess would be sampler and scheduler

[bold field]

uhhh wdym

whats scheduler for

[ornate elk]

Sampler would be DPM++ 2M
And scheduler is Karras for example

I can't explain what a scheduler does in that way

[bold field]

ill try it

[ornate elk]

I mean sampler is sampler

DPM 2m would be a value

Scheduler would be scheduler

[bold field]

So what i did is correct or not?

[ornate elk]

I guess here is everything you need to know:
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/API

Maybe the API payload display helps

[bold field]

then i guess its
"sampler_name": "string"
and
"sampler_index": "Euler",

🤔

not sure how to use the index

[woeful falcon]

Is it enough to use SDXL and default ComfyUI configuration to generate images? I tried a few prompts but it's not really good.

I'm trying to understand what the simplest form of workflow would be so that I can experiment later on

[ornate elk]

I can send you later my sdxl base workflow

But the default should be enough

Just make sure you use 1024x1024 for sdxl

As resolution

[woeful falcon]

Oh okay, let me try

I will probably install Comfy in Google Cloud. I'm just experimenting with my MacBook Air M2 locally.

Can't offord another PC just for this, I will rent

Later I need to learn how to use IPAdapter, upscaler for img2img. I'm aiming for virtual try-on experience for clothes. There are YouTube videos but it's too far complicated for me. I'm beginning from scratch 🙂

woman wearing a scarf, desert, hot, sand

Yay, I crated my first image with a basic workflow! Thank you for pointers @ornate elk

Maybe I should include an upscaler for the image now. I'd like to get 2k at least. Which path should I choose?

[ornate elk]

A good path would be to use the custom node called ultimate upscaler

[woeful falcon]

https://github.com/ssitu/ComfyUI_UltimateSDUpscale this one?

[clever cargo]

Using stable diffusion/comfyui requires a SPECIFIC version of python, is that correct?

[ornate elk]

Yep should be

[woeful falcon]

I'm running it on 3.12. I think you should have at least 3.10

[ornate elk]

Yea for stable-diffusion-webui best is 3.10.11 64bit.
Comfyui ships its own python (3.10.6)

[woeful falcon]

Thank you, I will check them out!

[ornate elk]

3.12 can cause issues and is not recommended

[clever cargo]

aaaand, you CAN'T have multiple versions of python, correct?

[ornate elk]

You can have multiple

[woeful falcon]

Oh good to know. I will install it with 3.10 then.

[clever cargo]

:v

[ornate elk]

But not in the same sub version

[clever cargo]

._.

[woeful falcon]

You can have multiple versions of Python. You need to install it into virtual environments called venvs

[clever cargo]

. _ .

[woeful falcon]

And create a virtual environment with specific version of Python that you install ComfyUI

[clever cargo]

. _ .

https://tenor.com/view/kuzco-kronk-yzma-right-wink-gif-22358009

okay, cool; thanks! I'll work on figuring that out xD

[ornate elk]

What do you want to do? ^^

[woeful falcon]

Yeah. Just look for Python virtual environment. Basically you create a virtual environment with the Python you have, e.g python3.10 with python3.10 -m venv venv command (venv is the directory of it). Then you source it with source venv/bin/activate then pip install everything

It sounds complicated at first but it isn't really

[ornate elk]

On windows its not needed for comfyui or Auto1111

Both create their own venv

And will use the installed python from path

In Auto1111 you can also define the path to other Python versions.
For comfyui you have to do it rhe above way

[viral mason]

Problem with Latent Couple in Stable Diffusion API: Overexposure, Artifacts, and Contrast Increase when Using AND Operators

Hi! I'm facing an issue while using Stable Diffusion and the Latent Couple extension through the API. When I add AND operators in the prompt to define different segments, the generated images start to show strong overexposure, artifacts, and increased contrast. The more ANDs I add, the worse these effects become.

Has anyone encountered similar issues with artifacts when sending prompts with ANDs through the API? Could it be that the API interprets AND differently compared to the web interface?

Is there a way to properly define separate prompts for each segment without these effects, maybe with a different formatting approach?

I would really appreciate any advice or alternative approaches to working with segmentation through the API!

Note: I'm sending the image in base64 format.

[clever cargo]

I had comfy ui working well for awhile now but recently downloaded a different version of python i think, may have uninstalled the version I had when I started using comfyui..

and now, for video editing purposes, I need to try installing pythonv3.11 for some other subtitle project...

Basically, I'm way in over my head and probably changed things I shouldn't have and am most likely gonna need to reinstall the right versions.. :v

You mentioning that comfyui doesn't work well with 3.12 made SO MUCH sense as to why I have had so much trouble lately

[iron lantern]

what can i prompt to add animals manily dogs to a image ??? nothing seems to work

[bold field]

Hmm this causes an error for some reason:
the payload: {'prompt': 'hi', 'steps': 69, 'cfg_scale': 7, 'width': 512, 'height': 512}

Is there a problem with it 🤔

its just a request error

Hm wait

Ok the generation CMD shows its generated but still request error

ah wait maybe its just timeout

Ok nvm 😭

was just http timeout

[woeful falcon]

I've installed this custom node, I've added the connections but there is this upscale_model is missing. What should I use there?

[ruby yoke]

the word 'dog' is usually good enough

[woeful falcon]

Okay I've found 4x_NMKD-Superscale-SP_178000_G

[woeful falcon]

Should I fill the form for SD 1.5 to download it from huggingface?

I can't seem to access it for free

[iron lantern]

still not working ... i wanted a dog with a girl ... in the same image ??

a little dog looks shocked (puts his paws on a face) , (shocked emotion putting hands on a face) , an opened mouth BREAK studio portrait, high quality studio light setup, shocked, screaming shocked ?? with a girl would work or ?

[ruby yoke]

For what you are doing, or wanting to do, you are going to have to stop using pony and use one of the models that have the actual comprehension necessary for these sorts of complicated concepts. your choices are: flux or sd3.5 large. none of the pony models, none of the previous versions of stable diffusion, understand well enough to give you the image you want without you having to inpaint quite a bit

[iron lantern]

it has worked before ... so i admire the words you give out but i know it works as i have done it before ... so ...

[safe inlet]

i dont use pony but doesnt it want prompting like 1girl, 1dog

[iron lantern]

@safe inlet been trying that lol

guess its time to google and a long weekend lol

[celest delta]

just started using AUTO1111 webUI, any reasons why my embeddings aren't showing up in textual inversion?

[ornate elk]

They won't show up if they are not compatible with the model

[celest delta]

it's supposed to be though

[ornate elk]

And its in the embeddings folder?

[celest delta]

yeah

i have the pony model, and the negative embedding is made for pony

[ornate elk]

Hmm should work then

Have you restarted?

[celest delta]

i put them all in there before starting the webui

https://civitai.com/models/332646?modelVersionId=720175
https://civitai.com/models/257749?modelVersionId=290640

are they somehow not compatible? Did i not understand something?

[ornate elk]

Hm it should work

But its the base pony model

Wouldn't use that its bad

Maybe they work with newer pony models

[celest delta]

like which one? this looked like the most recent

[ornate elk]

This here is recommended for the embeddings you downloaded:
https://civitai.com/models/623819?modelVersionId=697402

Same creator

[celest delta]

i'll try thanks, i picked many models to try anyway, but i don't understand why this one doesn't show up since they should be compatible

when downloading an embedding, what should i check to see if it's the correct model, like the section "base model" on the right?

[ornate elk]

If you have other pony models test with them

Would be still strange that the original pony wouldn't be compatible

[ornate elk]

Yep base model needs to be the same

[celest delta]

tried this model, still nothing... what am i doing wrooooong

[ornate elk]

Can you screenshot the embedding tab?

Also make sure the webui is whitelisted in any browser adblocker.
That sometimes causes issues with displaying loras and embeddings

[celest delta]

how do i do that?

like just using the URL of the webui?

the one with numbers?

[ornate elk]

You can add that to a whitelist

Then reload the site or restart the browser

Also don't use opera VPN or other vpn extensions

[celest delta]

ah that was it!

thanks a lot

[ornate elk]

No problem 🙂

[celest delta]

i still have a few loras that don't show up though (most of them are there)

[ornate elk]

Maybe they are not compatible then

[celest delta]

it seems it's the same number of lora regardless of which model is loaded, oh well i'll figure it out

[ornate elk]

Alright I'm off for now gn

[celest delta]

thanks again for the help!

[ornate elk]

Your welcome

[still mesa]

hey- i have a specific use case i need some help with. i have an image of a unique looking portable blender, and i want to place it in a product photoshoot setting. if i just use inpainting to paint everything around the blender, the model tends to add things like a blender handle to the blender when i just want to retain the original blender (the portable blender does not have a handle). does anyone have any ideas how i can guarantee this or at least increase the odds of the model not attaching stuff to the product?

[ruby yoke]

generate your background and then composite your image over the background in photoshop

[still mesa]

the problem then is that the image just looks superimposed, no? there won't be shadows or reflections, etc.

[ruby yoke]

not if i do the compositing.

[still mesa]

hmm i wonder if there's a way to automate that process? thanks for the input btw

[ruby yoke]

i'm fairly sure all you need is a good action. there are lots of photoshop people out there that are exceptionally good at this, and have been for years. start by getting hold of Unmesh on the piximperfect youtube channel

[still mesa]

i wasn't clear which is my fault. but i'm building something that enables people to do this easily within the web app without having to understand how photoshop works. ideally, they upload their asset, we remove the background, they position it in an image frame, then we just generate everything around the asset (this would probably also include the blending you're talking about) so that it looks like a legit product photoshoot. i'm just trying to figure out exactly what that last step in the process should look like without introducing artifacts or anything

[ruby yoke]

okay, this is what I do in photoshop when compositing with gen fill. i select the object and contract the selection 5 or so pixels so that the edge of the object is outside the selection. then i invert the selection. now my mask is the edge of the object all around it, and when i run the AI, it melds the edge of the object into the background it's generating. you'll probalby want to do something like that

[still mesa]

that makes a ton of sense. so you just have the model inpaint around the edge of the object you've inserted.

[ruby yoke]

yes, exactly

just a little of the edge, a few pixels

remember, when you look at a real object, you see light that's bending around it, and that's basicaly what you want to do with your gen

[still mesa]

that's very insightful and smart. thank you so much for the idea!!

[ruby yoke]

you're more than welcome :)

[half plank]

Does someone is having problem starting A1111 on Paperspace?

[royal nova]

Hi does anyone have experience deploying Stable diffusion in docker? I am developing a final degree project and I need some help and I don't mean using the typical webui like automatic1111.

[livid lark]

Extreme Barf Contest in my pod! Orbital Discharge Measurement! Consume the finest French dairy to stand a chance. See you in orbit!

[woeful falcon]

This was supposed to be realistic but I'm getting the following result from my comfy workflow. Prompt: woman in the bridge looking over sea, night setting, autumn, raining. I'm using realistic vision checkpoint and 4x_NMKD upscaler only on top of default comfy workflow. Where am I doing wrong?

Resolution is set to 512x512 and I'm upscaling by the factor of 4, doing 20 passes with euler.

[woeful falcon]

I think my prompt was missing, I needed to add realistic to it and I kinda get somewhat realistic images

[ornate elk]

Which model did you used?

[woeful falcon]

Realistic Vision

I think it's based on SD1.5

[ornate elk]

Ah okay

[woeful falcon]

I think I have a problem with upscaling. 512x512 images are okay with my prompt but when upscaled it doesn't give nice output

[ornate elk]

Can you show the upscale Settings?

[woeful falcon]

I'm using the default, just changed upscale factor

[ornate elk]

When you click on Mode_type, what are the options?

[woeful falcon]

Linear, Chess, None

[ornate elk]

Ah okay

I don't see the upscaler model here

[woeful falcon]

4x_NMKD-Superscale-SP_178000_G.pth

Reddit said this was the best 😄

Not sure though

[ornate elk]

Afk, driving

[stark yacht]

https://pastebin.com/mAbjViFg model selection error

[ornate elk]

Which model?

Also update the webui by adding git pull to the webui-user.bat in an empty line

[stark yacht]

tponyai

[ornate elk]

Okay, and what's your GPU?

[stark yacht]

6900xt

[ornate elk]

The zluda hip path is not correct

It has to be two seperate entrys in the environment variable's

Check these images:
https://github.com/CS1o/Stable-Diffusion-Info/issues/7#issuecomment-2408541054

You also need to update the webui

[stark yacht]

It's the same

https://pastebin.com/MPqiC6pt

[ornate elk]

Try restarting the PC. Because you corrected the hip path. And then delete the venv folder and relaunch

Then show the log again if it still doesn't work

[celest delta]

so i installed swarmu UI, changed the path for my moddels that were in auto1111 like the guide said, and they still don't show up