🤝｜tech-support

[naive shale]

Is SDXL good at something like this?

[karmic crown]

https://civitai.com/search/models?sortBy=models_v9&query=interior

[naive shale]

yeah I looked around.. not sure which model would actually give me best results for interior design

[karmic crown]

check the examples

see which one suits your need the best

you can ask in #📝｜prompting-help #🏞｜general-with-images

you might get better answer there

but i am almost sure you will need at least a few different models

maybe some loras too

[naive shale]

thanks for all the advice.. I appreciate it

[karmic crown]

btw. "best results" for me and you are most likely not the same thing, whatever we talk about

[naive shale]

yeah.. I know.. just saying that some models are known to be good at certain things

I like Dreamshaper for game-art stuff /fantasy stuff I do

[karmic crown]

i agree, but be prepared that you will need to test bunch of models and loras

i usually download models only considering examples, not what anyone says

i can take a look, but if i do not like example i will not download it

[naive shale]

I'm thinking, I will just draw rough sketch lines in photoshop on the photo of my empty bedroom wall.. and use controlnet.. I think that would be the best choice .. workflow-vise

[karmic crown]

there is no line controlnet model for XL, which would probably be best to use in your case, so you might need to use 1.5 model

[naive shale]

yeah, I'm using 1.5 already..

[karmic crown]

https://www.youtube.com/watch?app=desktop&v=4cRIdxijB-E

i just found this one

[karmic crown]

better ask that in #📝｜prompting-help

[fair oxide]

@naive shale @karmic crown there is line art available for SDXL now, a recent release for controlnet called controlnet union about 2gb file and features many of the common styles, very useful. https://huggingface.co/xinsir/controlnet-union-sdxl-1.0

[karmic crown]

thank you

[fair oxide]

this video could help with setting up comfyui workflow for it, its quite easy to do https://youtu.be/UBFEw1IUX_I

you will need this node from comfyui manager

@naive shale in case you are using a1111 there is t2i adapters for lineart and sketch that works good for sdxl

https://huggingface.co/lllyasviel/sd_control_collection/tree/main

as for recent release of controlnet union there are no support for a1111 atm, only works on comfyui

[karmic crown]

i do not remember this one

[fair oxide]

it came out later with the t2i development

works nicely too

but i removed all those now that i have controlnet union

saves space and better to manage a single file for all styles

this is the only CN model i have for use now controlnet-union-sdxl-1.0.safetensors

and i stopped playing with sd1.5 for several reasons

one being lack of prompt coherence

[karmic crown]

better ask that in #📝｜prompting-help

primary purpose of this channel is to help with installation issues and errors

[quaint idol]

oops

[past tapir]

anyone can tell me if the second "speedbump" is normal in training LoRa? the red graph is the latest version after changing some settings so atleast it got better

oops, guess I cant post a .svg file

[ruby yoke]

that's a really ugly loss curve

[naive shale]

I'm trying to find the Control net model for Recolor_Luminance for SD 1.5 .. I can't find it along where the other models are on hugging face

[naive shale]

I got it nvm, the file name is ioclab_recolor

[fierce hollow]

Guys, I decided to start creating videos in SD anyway - the task is to make realistic videos for example https://civitai.com/images/19251566
Tell me where to start?
I tried the animategif tool, it's completely different (Every frame changes)

[thorn palm]

Hello everyone,

I have been searching for weeks but I can't find how to do it. I would like to automate the creation of images on SD from a Python code (with no webui). I don't know where to start. I can code but I don't know how to create the bridge between my code and Stable Diffusion. Do you have any resources to recommend?

---

[karmic crown]

the main question is if you want to use some backend service over API or just local models

for cloud computing, you can use stable diffusion api or civitai api

for local, it would probably be easiest to use a1111 API. yes you will still have a1111 as a server, but it will be much easier to do it than using the code directly without it

if you do not want to use any api, start here
https://github.com/Stability-AI/stablediffusion

[thorn palm]

@karmic crown @karmic crown thanks,
Local modls hosted on my pc

[karmic crown]

clone https://github.com/Stability-AI/stablediffusion
and check img2img.py and txt2img.py in scripts folder

i do not know what your specific use case is, but i would use a1111 over API for simplicity

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/API

https://gist.github.com/w-e-w/0f37c04c18e14e4ee1482df5c4eb9f53

def call_txt2img_api(**payload):
    response = call_api('sdapi/v1/txt2img', **payload)
    for index, image in enumerate(response.get('images')):
        save_path = os.path.join(out_dir_t2i, f'txt2img-{timestamp()}-{index}.png')
        decode_and_save_base64(image, save_path)

see how easy it is

[fading night]

does SD run on OpenCL?

has anyone tried dopencl or vopencl to "merge" multiple servers into one by running an remote OpenCL instance? I wonder if it would increase the performance or degrate it.

3 years ago I tried both of them (not for SD), one would memory leak so I had to restart it daily, forgot which one... I imagine there may be some problems when the GPU count reaches in hundreds... Like, I had troubles with some opencl libs having a hard limit on the GPU count, e.g., only 64 GPUs 😭

[left wing]

Anyone know how to fix this error?

[quick forge]

are there any good models for generating logos?

[calm raptor]

I currently have this system running as a home server. I'm thinking about getting a used 3090 to put in there for SD so I can remote into it from my terrace. Would that work or would the CPU be too underpowered? https://pcpartpicker.com/list/y4ykpB

[thorn palm]

Thanks a lot i'll check this

[oak thistle]

I have some issue with stable diffusion i can't figure it out how to generate good images. i follow allot of tutorials and everytime i get very weard images. like the sky, clouds etc or so weard. What is the isseus is it because i need to generate more to train it or what is the matter that i get such bad images generated

[long radish]

depth libary not showing

[long radish]

Fix^

[ornate elk]

Thats caused by the LCM lora

If you remove it from the prompt, it should look normal

[quick forge]

ok, so i still can't figure out why a couple of my models/checkpoints just generate a blank grey image

namely, the ones for logo and tshirt design

[ornate elk]

Maybe they are Lora's and not Models

Check their file size

[quick forge]

yeah i think this is the problem. I'm using stuff wrong and have it in the wrong places.....sorting it out now

ok, now the loras aren't showing up under the lora tab in A1111

im a bit lost on how this is supposed to work. so i can loada checkpoint like zavychroma, but then also select a lora?

[ornate elk]

Loras only show up when a compatible model is selected

Most models and loras are based on either 1.5 or sdxl and only work with the same basis

[sharp vessel]

anyone know why my image are blank?

output file found this result failed

https://www.liblib.art/modelinfo/a5817eab96c84225a608df7198b0d92e?from=search
I use this workflow

[dreamy current]

HI, i'm getting a dark img when using R-ESRGAN 4x+ Anime6B model on hires fix any idea? I'm using forge and a GTX 1660. And i get no error on my cmd

[crimson leaf]

How do we train sdxl loras? And how long does it take est?

[oak thistle]

reason i have lcm lora is because if i have it it takes maybe 10 minuts to generate one photo if i don't have it then most of the time it takes above 6 houres and sometimes if fails to generate photos

[cursive acorn]

Reposting issue i had from the other day as I have a simplified version of it now
Inpainting doesn't seem to work. I made an image, inpaint just the eyes, change the setting to only masked, and revised the prompt to solely be about the eyes
As its working, the preview clearly is making new eyes. But right as its done processing, it just gives me the exact same thing as the input which looks terrible.

if you need more info to help me out just ask

dont know what to do, inpainting just not actually outputting as it should be

[solemn moon]

I’ve had this issue a few times where it just disappears.

Check your drivers are updated
Check your extensions are updated
Delete Venv folder
Run .bat again

Maybe pc restart

It’s easier if you copy the text from your terminal and post that too 😎

[ornate elk]

Make sure you have only --xformers --medvram --no-half-vae in your webui-user.bat at the line commandline_args=
That should speed it up

[ornate elk]

Make sure you set the resolution to a sqaure and use Resize to and not Resize by

[oak thistle]

oke thank you for the hulp ill try this out

Clearly the issue was lcm-lora

[cursive acorn]

Ohhhh thats probably it

Ill give that a try

[oak thistle]

Now my next step is using Hires. fix but it seems that there is also weard things happening

[ornate elk]

Did you added the commands I told you?

Hires fix is very heavy on vram usage.
You mostly need to install an extension called tiled vae. To not get out of vram.
Or you can upscale in img2img by using the SD upscale script. That won't run out of vram

[oak thistle]

i forgot to change the fille thanks for reminding me i will also use tiled vae, btw i have a rtx 3090 so i beleev that my pc should not have a isseu with vram but i'm hosting stable diffusion from my laptop to my pc becuase don't trust the programma on my pc so i don't konow if it uses my pc power or laptop

[ornate elk]

In the screenshot from earlier I've saw that its using a 4gb vram GPU.
So its running on your laptop.
It would run miles faster on your 3090.
The Programm is trustworthy as its the most commonly used Programm for stable-diffusion

[oak thistle]

i will think about it if it stays with allot of issues i will do it on my pc , i really apriciat your help

[ornate elk]

No problem 🙂
Also a 3090 only needs --xformers in the webui-user.bat launch args to get the best performance.
Upscaling and using sdxl models isnt a problem for that gpu

[oak thistle]

Great thanks allot! your the best!

this is what i got now lmoa

[vocal burrow]

what does the console log says ?
send the full log here (copy paste log into a .txt file and drop said file in here)

[oak thistle]

Do you mean this:
42% ETA: 04:54
ancient village, (intricate details), hdr, (intricate details, hyperdetailed), cinematic shot, vignette, centered
Negative prompt: Negative prompt: paintings, sketches, (worst quality:2), (low quality:2), (normal quality:2), lowres, normal quality, ((monochrome)), ((grayscale)), skin spots, acnes, skin blemishes, age spot, glans, nsfw, nipples, (((necklace))), (worst quality, low quality:1.2), watermark, username, signature, text, multiple breasts, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, bad feet, single color, ((((ugly)))), (((duplicate))), ((morbid)), ((mutilated)), (((tranny))), (((trans))), (((trannsexual))), (hermaphrodite), extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn face)), (((mutation))), (((deformed))), ((ugly)), blurry, ((bad anatomy)), (((bad proportions))), ((extra limbs)), (((disfigured))), (bad anatomy), gross proportions, (malformed limbs), ((missing arms)), (missing legs), (((extra arms))), (((extra legs))), mutated hands,(fused fingers), (too many fingers), (((long neck))), (bad body perspect:1.1)
Steps: 30, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 7, Seed: 2246970194, Size: 768x768, Model hash: cc6cb27103, Model: v1-5-pruned-emaonly, Denoising strength: 0.7, Hires upscale: 2, Hires upscaler: Latent, Version: 1.9.4

Time taken: 40 min. 29.9 sec

[vocal burrow]

no I mean the console log

the white text in the black window thingy

[oak thistle]

from the cmd

[vocal burrow]

yes

send everything

and I mean everything

from the start

[oak thistle]

stable-deffusion-webui\modules\processing.py:1026: runtimewarning: invalid value encouterd in cast/it] x_sample = x_sample.astype(np.uint8)

[vocal burrow]

no idea what you sent but that s not the console log

(it might be part of it)

[clever wedge]

Pls I need help on editing a picture

[oak thistle]

i

[clever wedge]

Put a space background and wear the monkey a space helmet

[oak thistle]

i'll send the full log but just worry of sending something that may reveal something of my pc detials

[vocal burrow]

apart from your username there shouldnt be much to worry about

[ornate elk]

Set hires steps to 10, set the resolution to 512x512, then set upscale by to 1.5
And set denois to 0.5
Also what's in your webui-user.bat now?

[oak thistle]

fatal: detected dubious ownership in repository at 'C:/Users/veton/stable-diffusion-webui'
'C:/Users/veton/stable-diffusion-webui' is owned by:
INGEBOUWD/Administrators (S-1-5-32-544)
but the current user is:
LAPTOP-OG181BR8/veton (S-1-5-21-4290696012-3046138489-988074773-1001)
To add an exception for this directory, call:

    git config --global --add safe.directory C:/Users/veton/stable-diffusion-webui

venv "C:\Users\veton\stable-diffusion-webui\venv\Scripts\Python.exe"
fatal: detected dubious ownership in repository at 'C:/Users/veton/stable-diffusion-webui'
'C:/Users/veton/stable-diffusion-webui' is owned by:
INGEBOUWD/Administrators (S-1-5-32-544)
but the current user is:
LAPTOP-OG181BR8/veton (S-1-5-21-4290696012-3046138489-988074773-1001)
To add an exception for this directory, call:

[severe sand]

Anyone see this before?

[oak thistle]

git config --global --add safe.directory C:/Users/veton/stable-diffusion-webui
fatal: detected dubious ownership in repository at 'C:/Users/veton/stable-diffusion-webui'
'C:/Users/veton/stable-diffusion-webui' is owned by:
INGEBOUWD/Administrators (S-1-5-32-544)
but the current user is:
LAPTOP-OG181BR8/veton (S-1-5-21-4290696012-3046138489-988074773-1001)
To add an exception for this directory, call:

    git config --global --add safe.directory C:/Users/veton/stable-diffusion-webui

Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr 5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Version: 1.9.4
Commit hash: <none>
Installing sd-webui-controlnet requirement: changing opencv-python version from 4.10.0.84 to 4.8.0
Launching Web UI with arguments: --xformers --listen
ControlNet preprocessor location: C:\Users\veton\stable-diffusion-webui\extensions\sd-webui-controlnet\annotator\downloads
2024-07-15 10:38:39,626 - ControlNet - INFO - ControlNet v1.1.452
Loading weights [cc6cb27103] from C:\Users\veton\stable-diffusion-webui\models\Stable-diffusion\v1-5-pruned-emaonly.ckpt
2024-07-15 10:38:40,166 - ControlNet - INFO - ControlNet UI callback registered.
Running on local URL: http://0.0.0.0:7860
Creating model from config: C:\Users\veton\stable-diffusion-webui\configs\v1-inference.yaml

To create a public link, set share=True in launch().
Startup time: 64.8s (initial startup: 0.1s, prepare environment: 21.7s, import torch: 14.4s, import gradio: 6.1s, setup paths: 9.1s, initialize shared: 1.5s, other imports: 3.8s, setup gfpgan: 0.1s, load scripts: 3.0s, create ui: 0.6s, gradio launch: 4.4s).
Applying attention optimization: xformers... done.
activating extra network lora: TypeError
Traceback (most recent call last):
File "C:\Users\veton\stable-diffusion-webui\modules\extra_networks.py", line 145, in activate
extra_network.activate(p, [])
File "C:\Users\veton\stable-diffusion-webui\extensions-builtin\Lora\extra_networks_lora.py", line 18, in activate
p.all_prompts = [x + f"lora:{additional}:{shared.opts.extra_networks_default_multiplier}" for x in p.all_prompts]
TypeError: 'NoneType' object is not iterable

Model loaded in 8.0s (load weights from disk: 4.3s, create model: 0.5s, apply weights to model: 2.3s, apply half(): 0.1s, calculate empty prompt: 0.6s).

[oak thistle]

i didn't yet restarted my bat i was just working with the settings after this last generation is finish ill update my bat fille

[ornate elk]

I see you already use --xformers, then you only need to add --medvram
And then try out the settings I mentioned

[oak thistle]

i know that i ahve done that before and removed it because it didn't fix anything

but i can try it again

[ornate elk]

Did you copied the webui from an other PC?

[oak thistle]

yes

[ornate elk]

Ah well, that explains the permission errors

[night sigil]

did anyone know why this symbol not loaded?, I'm using stable next, it's the symbol on the galery filter

[vocal burrow]

what exactly did you copy paste from another pc ?

[regal nebula]

Guys, I'm a newbie, tell me pls, why don't I have these parameters (Controlnet) in my interface? My interface is in the second img

[ornate elk]

Hey, did you installed Controlnet via the extensions tab?

[regal nebula]

I didn't

[ornate elk]

Then thats why it isnt there.
Controlnet is an extension that needs to be installed.
Click on extensions tab, "available" then click on "load from"
Then you get a list. Search for
sd-webui-controlnet and install it.
Then restart

[oak thistle]

i'm not sure if i understand what the question was i thought if i copied the text from another place i just copied the text for my bat file

webui-user.bat text

everthing els i just donloaded from hugging face the filles, for control net i made my own controlnet map to place the filles inside

[vocal burrow]

ok that shouldn t cause any problem then

I m not at home so I cant help much but I would suggest two things :
1/ delete venv folder, when in doubt it can t hurt to start with a clean venv
2/ check in your settings tab for a "extra network" setting or something alike and report its value.

[oak thistle]

oke thanks for your help

[regal nebula]

I installed this, restarted via stability matrix , but the interface did not change, maybe something else is required?

[oak thistle]

https://github.com/Mikubill/sd-webui-controlnet.git use this link

[regal nebula]

I got it: "Extension "sd-webui-lama-cleaner-masked-content" requires "sd-webui-controlnet" which is not installed."

[vocal burrow]

use the search box, type "extra" in it and see what shows up

[ornate elk]

Your missing sd-webui-controlnet extension

The lama cleaner one isnt needed

[regal nebula]

Yes, i instaelled it and it works, thx so much

[ornate elk]

Np

[regal nebula]

Can you please also tell me why I don’t have this sampler? Should I install it too?

[oak thistle]

you have it

[oak thistle]

i can see it here on the photo

[regal nebula]

Should I choose it like this?

[oak thistle]

yes

@ornate elk the Hires steps at 10 fixes the problem

[vestal anchor]

does my workflow save automatically in comfyui? and where can I find it

[ornate elk]

Nope, you have to hit Save

Then it will download a .json file

[vestal anchor]

[vocal burrow]

alternatively it should be embedded by default in the picture your produce.

[jagged glade]

pony doesnt work in 3060 12gb vram?

[regal nebula]

Guys, can u pls tell me, why does the image not match the canny sketch?

[vocal burrow]

should work just fine

[vocal burrow]

no controlnet model selected

[regal nebula]

there is nothing here

i've downloaded model, put in models folder, but there is still none

@vocal burrow should i download something else?

[amber cloud]

did you refresh after downloading the model into the correct directory ?

[regal nebula]

i restarted two times via stability matrix

[amber cloud]

guys anyone knows how to fix this ? RuntimeError: jaxlib version 0.4.30 is newer than and incompatible with jax version 0.4.26. Please update your jax and/or jaxlib packages.

[regal nebula]

put it here C:\Stability Matrix\Packages\stable-diffusion-webui\models\Stable-diffusion

[amber cloud]

i'm sorry. i thought you only needed to refresh the menu cuz i sometimes do that

[vocal burrow]

cf https://stable-diffusion-art.com/controlnet/

yes you need to download controlnet model, each ControlType requires its own model.
And you cant mix and match SD1.5 model with SDXL controlnet models (and vice versa)

[vocal burrow]

need full console log

[amber cloud]

@vocal burrow ok

[vocal burrow]

the full log, including what s before that

[vocal burrow]

try updating your extensions and then delete (or rename to venvBACK or whatever) venv folder

[amber cloud]

this is what happened when i tried to update the extensions

@vocal burrow is there a way to disable clip-interrogator-ext before running Stable=Diffusion ?

[jagged glade]

It's giving me really results

[vocal burrow]

delete its folder in extensions to uninstall it (and delete venv folder to be clean)

[amber cloud]

awesome

[regal nebula]

I already downloaded it, restarted the application many times, but there is still “none”, what else can I try?

[vocal burrow]

because those are not stable-diffusion model, those are controlnet models, you re not putting them in the correct folder
stable-diffusion-webui\extensions\sd-webui-controlnet\models

cf the guide I gave you

[light prism]

Hey there i am wondering if there are similar solutions for hopefully stable diffusion or at least other open source tools that are similar to the midjourney "Blend" command? Any ideas or hints are appreciated!

[solemn moon]

i2p?

[regal nebula]

omg, u re right, thx so much

[light prism]

i am not familiar with this? Inaproppriate Image Prompts (I2P)? how does this help me?

[solemn moon]

control net model

copies style

Ip

dang

[vocal burrow]

multi controlnet style transfer, ip2 adapter, image-mixer (https://huggingface.co/lambdalabs/image-mixer standalone), etc
That blend feature seems too nebulous to give you a proper, one stone to kill them all solution.

[solemn moon]

ip2 😅

[vocal burrow]

close enough :p

[solemn moon]

🫡

[light prism]

ahh okay 😄

[vocal burrow]

cf https://stable-diffusion-art.com/ip-adapter/

[light prism]

ah okay thank you, i stubled across this and found a demo for from this repo and since the demo was no longer working i did not continue this further https://huggingface.co/spaces/lambdalabs/image-mixer-demo

[vocal burrow]

I didn t say it would be an easy solution :p

But yup there are solutions out there. But without knowing what you re trying to achieve we can t help you much more.

[solemn moon]

can only imagine the amount of generations to mix 5 styles to your liking

[vocal burrow]

This "blend" feature seems to be a "black box" without any clear purpose apart from "trade offer : you give me 5 images, you receive 1 image somewhat close to them"

[light prism]

the main idea is producing designs for objects e.g. cars, by multiple persons and then in the end selecting promising (potentially very different images) and merging them into one (or a few) designs that include the most important attributes

so i am not actually interested in just keeping styles but core features of objects

[solemn moon]

suppose the strength of control nets could play a good role in achieving that, never done 5 controls though..

[light prism]

jep i encountered that as well and would definately prefer building on top of open source, so there are less issues sharing my solutions in the end

thanks for all the input already 🙂 really appreciate it

[green merlin]

hellow guys , i try to do no an image to image and i get " Connection errored out" any solutions ?

[solemn moon]

https://tenor.com/view/raised-eyebrows-napoleon-dynamite-kip-gif-5766210

[solemn moon]

post your pc specs and bat file config and command prompt report

[vocal burrow]

you re not closing the console after opening the webpage, are you.

[green merlin]

a person told me i just need to do this " export COMMANDLINE_ARGS="--no-gradio-queue" any idea where to use it ?
bat file config is this
set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --xformers --no-half-vae --medvram-sdxl

call webui.bat

pc specs are 4060rtx , 32gb ram , windows 10 , 3gb cpu

[solemn moon]

wait, this... @green merlin

[green merlin]

i dont close it , i used it always fine but now i get the issue with image to image . i still can generate

[vocal burrow]

then provide the full console log

copy paste everything in a .txt and drop said file in here

[green merlin]

everything in that webui black sreen after i launch the process ?

[solemn moon]

yea

[green merlin]

when i start it , nothing is written , its immediatly error

[vocal burrow]

your stable-diffusion-webui is not updated.
Open a cmd in D:\Stable diffusion\stable-diffusion-webui\ and run git pull then try again.
Also make sure that your 127.0.0.1 webpage is whitelisted in whatever adblocker you re using

[amber cloud]

great, it worked. Thanks a ton!

[solemn moon]

or edit your .bat file to automatically update

by adding git pull

[vocal burrow]

dont copy paste its commandline_args please

Also doing that means it will try to update whenver you launch it. So whenever an update is pushed your extensions are most likely gonna break until they re updated. And that s if the update does not break core functionalities.

Your call.

[solemn moon]

k will delete

[vocal burrow]

don t need to, you just need to explain the pros and cons

you can use --update-check command line args, it will tell you whenever an update is available.

[green merlin]

ok thanks guys , just curious . where can we use set COMMANDLINE_ARGS=--no-gradio-queue ? just in case ?

[vocal burrow]

you shouldn t need to use that

that s mainly for gradio debugging purpose.

[past tapir]

anyone got an idea why my LoRa safetensor file doesnt show up in A1111 after training with Onetrainer but works fine if I train same dataset in Kohya? I got no errors and training completes in onetrainer, but when I move over the finished file nothing shows in SDA1111 :S

[ruby yoke]

you sure you've got the correct model loaded in a1111 first?

[past tapir]

yea, even swapped to a SDXL just to check if some setting made a wront LoRa

https://tenor.com/view/simpsons-homer-simpson-gif-13518564799186478937

wait....nevermind...I wasn't here... (accidently named the file .safetensor and not .safetensors)

[ruby yoke]

i've done that 🙂 glad you got it fixed

[karmic crown]

this time, you are wrong 🙂
that will disable websockets for some of the back/front communication
I had big issues with sockets behind the reverse proxy on linux
so, that one might help in case of timeouts in webui

[vocal burrow]

yeah I know, but normally you don't need it. And if you do need it, it's either because someone f** something up in the code or because you are. Your case definitely seems like an edge one.
Vulgarization is hard :p

But yeah I was imprecise

[karmic crown]

only with those errors

for anything else, do not touch it

[vocal burrow]

Those can be caused by many things. antivirus/adblocker, closing the server, server crashing for some reasons, etc.

[karmic crown]

agree

but the first thing i would try is always --no-gradio-queue
so it is highly possible that i was the one who told them to try that

[vocal burrow]

that would explain it :p

[karmic crown]

that is optimistic stub in the dark 🙂

[vocal burrow]

it was pretty popular at some point too because of some gradio bug

So I wouldn't be surprised if they dug up some 10 months old reddit post talking about this

[karmic crown]

but i would add, if that not help, you are really f... as you said 🙂

[lost onyx]

I've got an issue (using Automatic1111):

I want to generate some variations of an image, and for that I'm putting various loras into a dynamic prompt.
So my prompt looks something like this:
tag1, tag2, tag3, {lora:loraname:0.7 | lora:loraname2:0.7|lora:loraname3:0.7}

The issue is that this results in horribly cursed images.
I think I even identified the culprit:
highres fix pass doesn't use the dynamically generated prompt, but the full prompt, combining all loras for the upscale step.
How do I fix this?

[karmic crown]

i guess easiest would be to create images without upscale, and then to do img2img
unless someone else has some better idea

maybe it is better asking in #📝｜prompting-help

[ripe patrol]

is there such a thing as facefusion but for more explicit content?

[rough plank]

Hi everyone! 🙂 I'm kinda new here, getting familiar with Stable. Could someone tell me which AI API does Stable use? Is it public info or kept behind bars?

[karmic crown]

Stable Diffusion use it's own API and models

but most of us here do not use API at all. we use local installations on our computers

[rough plank]

Hmm okay, thank you!

[past tapir]

Does anyone know where/how I can check onetrainer to log accuracy for tensorboard?

[karmic crown]

I just released a new version of a1111-setup (simple and easy Automatic1111 and Forge install script for macOS)
https://github.com/viking1304/a1111-setup/releases

[rough plank]

Thanks for the github link, I'm diving in!

[karmic crown]

what is your GPU and OS?

[rough plank]

Win 10 x64 Pro, 3080 Ti

[karmic crown]

https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides

you can use CS1o's guide

my script is for Mac only

[rough plank]

Does this include the Stable Audio as well? Or is it the Stable-video-diffusion? To me it seems the latter. (tough I'm interested in both)

[tidal spruce]

This is code I wrote/modified to run IC Light: https://github.com/lllyasviel/IC-Light/blob/main/gradio_demo_bg.py in modal. The text to image one works fine, but the model refuses to use my background when trying to use an existing background. Any idea why? Thanks, and please ping me if u respond

[misty marten]

can anybody help me? All my images generated in the anime or cartoon style model look like this, with realistic eyes and mouth, causing a very strange effect.

In this image, I'm not using any VAE or Lora, I just selected the model and generated it, but even using it, the result is the same

[azure star]

could ... someone help me get stable diffusion working on a windows platform ... with an AMD Gpu ... >.>

[ornate elk]

Hey sure, checkout the pinned messages of this channel.
There you find my AMD install guides.

[ornate elk]

Make sure Face restore is disabled

[fossil dew]

is there a extension that helps with prompts? Like, makes suggested words pop up? What's it called?

[ornate elk]

Yea its called Boorutag autocomplete

[fossil dew]

can I just install it with the extension tab

then just reboot?

[ornate elk]

Yes

[fossil dew]

cooooool ty

I can't find Boorutag autocomplete 🤔

I found this 🤔 but nothing else

nvm!~

ty n_n

[ornate elk]

Np 🙂

[bleak phoenix]

Hello, guys, I am just learning right now on how to use SD. I recently used an app on phone to make an anime art style pictures with ai, it had a very cool style and was very accurate to my original photo, same pose, clothes, colors, face expressions, body shape and so on. But from now it doesn't let me make NSFW. Do you know how can I make an img2img picture to anime very very accurate and close to the original?

[zealous leaf]

How to get pricing formula for stable-diffusion-v1-6 ?

In this page SD1.5 is mentioned, but no 1.6 https://next.platform.stability.ai/pricing
However in this page SD1.6 is mentioned: https://platform.stability.ai/pricing, but pricing is rather vogue. I need to have pricing based on the image size and step count, to have precise reporting.

Help me please, I remember that months ago it was there.

[burnt salmon]

so ive been using stable diffusion with automatic1111 for the last few days and it was working fine until it recently crashed and now whenever i try and use controlnet i get a number of errors,

[E:onnxruntime:Default, provider_bridge_ort.cc:1745 onnxruntime::TryGetProviderInfo_CUDA] D:\a_work\1\s\onnxruntime\core\session\provider_bridge_ort.cc:1426 onnxruntime::ProviderLibrary::Get [ONNXRuntimeError] : 1 : FAIL : LoadLibrary failed with error 126 "" when trying to load "A:\stable-diffusion\automatic1111\stable-diffusion-webui\venv\lib\site-packages\onnxruntime\capi\onnxruntime_providers_cuda.dll"

that is one example but they are all very similar. any idea what is going on here, im assuming i need to reinstall something.

[ripe patrol]

is there such a thing as facefusion but for more explicit content?

[ornate elk]

Are you using an AMD GPU?

[burnt salmon]

no RTX 2080ti

[ornate elk]

Hmm very strange then

[burnt salmon]

i reinstalled python and cuda and i am no longer getting those errors but now im getting this

*************** EP Error ***************
EP Error D:\a_work\1\s\onnxruntime\python\onnxruntime_pybind_state.cc:891 onnxruntime::python::CreateExecutionProviderInstance CUDA_PATH is set but CUDA wasnt able to be loaded. Please install the correct version of CUDA andcuDNN as mentioned in the GPU requirements page, make sure they're in the PATH, and that your GPU is supported.
when using ['CUDAExecutionProvider', 'CPUExecutionProvider']
Falling back to ['CUDAExecutionProvider', 'CPUExecutionProvider'] and retrying.

---

[ornate elk]

Make sure your graphics driver is updated too.
Then try deleting the venv folder and relaunch the webui-user.bat

[burnt salmon]

graphics driver was up to date and deleting venv folder didnt fix it, CUDA is in the PATH, must be this cuDNN thing. no clue how im supposed to install it for the version of CUDA i have.

[ornate elk]

What cuda version do you have installed?

Also when does the error appear ?

Only with Controlnet? Which model?

[burnt salmon]

CUDA 12.1, seems to only appear if i use Controlnet, its a ponyxl model

[ornate elk]

And your controlnet models are for sdxl?

[burnt salmon]

yes

it was working fine no clue what happened

[ornate elk]

Is your webui updated?

[burnt salmon]

yeah looks like it

[ornate elk]

can you show the controlnet settings you tried?

[burnt salmon]

same settings i have been using for a while. was working fine

[ornate elk]

So its only when using IP-Adapter face Id?

Do you also have the lora needed for face id?

[burnt salmon]

maybe not. what lora?

[ornate elk]

Loras for face id:
You can find them here: (.safetensor files)
https://huggingface.co/h94/IP-Adapter-FaceID/tree/main

[burnt salmon]

yeah just remembered i had that but i wasnt using it lol. is that supposed to be in the prompt? either way im still getting that error

seems to only happen with ip adapter

[ripe patrol]

Is there any way to remove the photo blur from explicit content face swaps on FaceFusion? If not are there any types of alternatives that offer this?

[carmine ibex]

I'm getting this error in comfyUI with the facedetailer node

[chilly knot]

i download t2i adpter models put them in cn model but i not see preprocessors

[lost onyx]

repeating my question from yesterday, with more details this time:

I want to generate some variations of an image, and for that I'm putting various loras into a dynamic prompt.
So my prompt looks something like this:
tag1, tag2, tag3, {lora:loraname:0.7 | lora:loraname2:0.7|lora:loraname3:0.7}

The issue is that this results in horribly cursed images.
The problem seems to be the highres fix pass, which doesn't use the dynamically generated prompt, but the full prompt, resulting in a combination of all loras for the upscale step.
How do I fix this?

Pictures related, shows how the highres fix step fucks everything up. Also I'm pretty sure that this didn't use to be this way in the past.

[solemn moon]

add a new prompt to hi res?

no ref for sd3? no vae?

[lost onyx]

I don't want to add a prompt to hi res, it's supposed to use the original one, after dynamic prompting does it's thing.
So "{tag1| tag2| tag3|} {lora:loraname1:0.7 | lora:loraname2:0.7|lora:loraname3:0.7}" might get transformed into "tag1 lora:loraname2:0.7" by dynamic prompts, so high-res should also use "tag1 lora:loraname2:0.7" for its own upscale prompt in that case. If I enter a value there, it'll apply the same upscale prompt to every generation, which isn't what I want.
VAE is set to automatic, and refine isn't needed.

[solemn moon]

i was asking for me lol.

Hi res should just include "non object" tags (for lack of a better wording)

if im right o0

[lost onyx]

like "highres", "good quality", etc.?

[solemn moon]

yea exactly

or it will try to fill in objects into objects

[lost onyx]

hmm, it's a slight improvement, but it causes the opposite problem, it now removes some smaller objects into generic background :/

[solemn moon]

maybe play with the denosing strength o0

[solemn moon]

no bokeh

[meager carbon]

@ornate elk hey, remember how we could not get my system to work with zluda,,, i found the issue.

[ornate elk]

Hey, ah did it worked by adding this to the webui-user.bat?

[meager carbon]

no

[ornate elk]

Normaly disabling the integrated gpu should also fix it

[meager carbon]

i disabled it systemwide

thing is

i like my integrated graphics

i use it for apps that run in the background

[ornate elk]

Have you added this with the word
set in the front?

set HIP...

[meager carbon]

not sure what you mean

[ornate elk]

You have to add this line to the webui-user.bat in an empty row:
set HIP_VISIBLE_DEVICES=1
It won't work without the set

[chilly knot]

help i dont see all preprocessors

[ornate elk]

Sdxl or 1.5?

[chilly knot]

1.5

on civitai base on 1.4 sd

it can work on 1.5?

[ornate elk]

Yes it works with 1.5 but not with sdxl models

[chilly knot]

where my preprocessors how i install it?

[ornate elk]

It will only show the ones available for the model.
T2i is so old that I guess that some of the preprocessor didnt worked and got removed

[chilly knot]

t2i open pose better?

[meager carbon]

well im running into another issue for some reason...

[chilly knot]

you think i can use default

[meager carbon]

yet... i have that lib

[chilly knot]

i can replace funtions t2i adapter?

[ornate elk]

Have you run git pull before?

[meager carbon]

i did

[ornate elk]

Normal openpose is better or dw-pose

[meager carbon]

boop

[ornate elk]

I see the webui got updated 8 hours ago and added support for Rocm 6.1

Maybe something broke

[meager carbon]

devs not testing their shit smh

[chilly knot]

thx for answer

i will use what i have now

[ornate elk]

You can install the module manually into the venv

[meager carbon]

im going to try to revert to a previous commit

i really hate that ppl dont know how to write commit messages

get your shit together

[ornate elk]

xD yea sometimes he breaks the webui but fixes it in a day

[meager carbon]

the fuck is this

[ornate elk]

The revert commit should work fine

[meager carbon]

i can try

works!

[ornate elk]

Perfect!

[twilit hemlock]

Has there been a change with the Stable Image Ultra endpoint? Generations starting today with the same seed and prompt are creating slightly different results

[night pagoda]

I just installed Stable Diffusion WebUI, I've gotten past a lot of issues, but now whenever I try to change models it just processes infinitely, and whenever I try to generate images the progress bar for generating the image just stays at zero, as well as the prompt just not calculating how many tokens it's using. I don't have any launch arguments and I haven't made any changes to the code, this is my first time trying to use Stable Diffusion

[twilit hemlock]

I am noticing the results on the ultra endpoint appear to be losing some detail causing less realism, as well as the higher saturation/contrast you would associate with higher cfg. Possibly a slight cfg increase or a change in the way the negative prompt works

[night pagoda]

Also when I click Generate it doesn't give me any errors or anything, it just sits there

Oop, welp, apparently something changed, now it's giving me

NansException: A tensor with all 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.

It somehow fixed itself

For image generation at least, it still just processes forever when I pick a model

Now it's just doing nothing again

I haven't even changed anything

Is anybody at all willing to help me with this? I really wanna get into image generation, and I'm getting incredibly frustrated because nothing I try works

This is just making me feel like an idiot

[vocal burrow]

What s your gpu ?

[night pagoda]

GTX 1660 SUPER

I've gotten it to generate one image

So I know it's possible on my machine

[vocal burrow]

What client / UI are you using ? Automatic1111's stable-diffusion-webui ?

[night pagoda]

Yes

[vocal burrow]

What command line arguments are you using ?

[night pagoda]

Because it told me to in an error, --no-half

Before then, I had nothing

[vocal burrow]

Add --medvram --xformers

[night pagoda]

It's installing xformers

[vocal burrow]

Should work faster that way

[night pagoda]

It's not even working at all, it's just stuck at zero and doesn't send anything to the console

[vocal burrow]

If there are still errors please screenshot your webpage and share the console log

[night pagoda]

Alright, gonna try generating an image again, default checkpoint

[vocal burrow]

(for the log it s better if you copy paste into a .txt file and drop said file here)

[night pagoda]

Just like before it isn't calculating the tokens in my prompt

That's not a good sign

Pressed Generate

Nothing is happening

Still stuck at zero, nothing sent to console

[vocal burrow]

Let me see it then

Screenshot + log

[night pagoda]

Right after I take a screenshot it starts working

I don't understand this

[vocal burrow]

Let me see nonetheless

[night pagoda]

Do I just have to wait a million years for it to even give me an indication that it's doing anything

[vocal burrow]

No it should be pretty obvious when it s computing stuff

[night pagoda]

It just does absolutely nothing for minutes at a time

And then eventually decides to start doing something

[stable dust]

question, how do I know if a downloaded model isnt corrupt?

[night pagoda]

I'm going to try switching models now

[stable dust]

Cyberdream said y was 6.99gb, then it went down to 3.99gb, this is after I dowbloaded the model in ys entirety

and when I tap open file it just says "save to" instead of "this archive is corupt" so i guess im good? idk

says the file is downloaded completely in google chrome

so idk whats going on

Autismmix had the same exact thing. at first it shoed up as the right file size (after dowbloading) refreshed on 7zip and they all went to 3.99gb

[vocal burrow]

Looks fine to me, down to the "artistic" se1.5 vanilla output. Took 18s to output something. It s a bit long but at the same time gtx1660 is not really powerful and does not support "half" type.
Screenshot + log again in case something goes wrong

[night pagoda]

It's taking much longer to load a model than it's supposed to, and this is all it's sent into the console

I have a feeling it's doing absolutely nothing again

[vocal burrow]

You can check its hash, civitai and huggingface shows the hash for each of its file.

[night pagoda]

Also that's 18 seconds after the multi-minute long wait after clicking generate

[stable dust]

can I do that on mobile?

[vocal burrow]

Loading from a hard drive? How much ram do you have? Not doing anything else at the same time?

[night pagoda]

Over 5 minutes now, I have 16GB of RAM and I don't have anything else intensive open

[vocal burrow]

Probably, how to tho will depend of your phone. There s probably an app for that in both apple and Android.

[night pagoda]

And yes, I'm loading from an HDD, that's where I store all of my programs, should I move it over to an SSD and that'll magically fix all of my problems?

I'm legitimately getting very frustrated at this

[vocal burrow]

Definitely yes, at least move your models to an SSD.

[night pagoda]

Python is taking 5.6 million memory and it's just sitting here doing nothing

[vocal burrow]

It s reading the whole model and copy pasting it to your gpu's vram

It s not doing nothing

But reading 7gb (at least) is gonna take a while for sure on hdd

[night pagoda]

It would be nice if it gave me some kind of progress update in the console

[stable dust]

is there an easy way for me to be certain these files arent corupt

I dont wannna redownload rhe same models only for them to download corrupt

[vocal burrow]

You screenshot the progress thingy

[night pagoda]

It's just counting up seconds, that's it

That can just be done on a render loop by comparing the start time to the current time

[vocal burrow]

Estimating precisely file transfer is not as trivial as it seems. And definitely not a priority for A1111 as virtually everyone is storing their models on an SSD already.

Anyways, try moving your sd models to an SSD
And use --ckpt-dir to indicate the new address
Cf this page for the list of command line arguments btw
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings

[night pagoda]

I know it's not trivial, obviously, but just knowing what it's doing at all in the console is better than printing absolutely nothing

[vocal burrow]

Apart from trying to use it or comparing hashes. No

[stable dust]

oof guess ill find out in an hour

[night pagoda]

What parts of the model folder do I move, the models folder or the models/Stable-diffusion folder?

[vocal burrow]

2nd

[stable dust]

it would be the md5 hash right?

[vocal burrow]

Yeah it s usually md5

[vocal burrow]

You can cycle through different hashes on civitai

Just click the arrow next to it

And for HF it s state pretty clearly that it is sha256

[night pagoda]

So set COMMANDLINE_ARGS= --no-half --medvram --xformers --ckpt-dir D:\SDModels would work?

Inside that folder I have the .safetensors files themselves

[vocal burrow]

Something like that yes. If not try putting double quotes around the path

[night pagoda]

I'll put quotes to be careful

Now that I've moved the models over and tried to change models, my entire computer has frozen

[vocal burrow]

Probably because it s not bottlenecked by hard drive anymore, so it can work much more.

Loading sdxl models with only 16gb of ram can be tricky depending of well maintained your windows is, how many stuff is running along SD

Try increasing your system pagefile size (only for c:)

launch sysdm.cpl from the Start menu search or run box (Win+R) and navigate to Advanced –> Performances -> Settings –> Advanced –> pagefile/virtual ram -> Modify

And set its minsize to 8gb and max size to 32

Cf for more details
https://www.thewindowsclub.com/increase-page-file-size-virtual-memory-windows

What is pagefile ? Tldr when your ram is filled it will overflow into the pagefile.

But your hard drive or SSD is much much much much slower than your ram. So you will feel it when it s being used. But at least it won't crash.

Anyways it s 2:30 am. Good luck and good night

If nothing works. Fall back to sd1.5 or sd2.1 based models for now.

[night pagoda]

I currently cannot access my computer

I'm just giving up, if it kills my entire PC just to switch models there's no way in hell I'm going to be able to generate anything

I can run LLMs on my PC just fine, but I guess image generation crosses the line

[stable dust]

OMFG. just went tomy pc and yep. Files are fucking corrupt

ughhh

what the fuck google chrome for ansroid

lying about dowbloads beig conplete alog with their finished download size

and then my ohone lies about it before I refresh it on 7-zip

fucking hell I cannot wait to move somewhwre where I can have internet

fucking furious rn godfucking damnit

was looking forward to trying out the different models I had downloaded

[summer smelt]

i just got a new gpu (4070 super) and im getting avg of 4 it/s. my current user config is --xformers --medvram --no-half --autolaunch
this was my config for 1660

[drowsy creek]

how much RAM do you have?

[old parrot]

Is this pre-built worth it with 16gb VRAM? Will only be doing AI and Solidworks/Blender, my current 1660 Super finally died

[ruby yoke]

that's very close to the one i have and it works very well

[old parrot]

Appreciate that! I think I may wing it

[ruby yoke]

I got mine from NewEgg though, i know nothing about the seller of that one

[old parrot]

Seems decent enough, not excited to uograde the ram though I'm lazy

[ruby yoke]

i only have 16 - so far, no issues running anything

even blender screams

[old parrot]

Interesting! Everyone in the PC build discords are acting like I'm sinning by buying it lmao

Can't be that bad

[thorny iron]

16 VRAM is good. I run 12 and it works but wouldn’t go any lower.

[ruby yoke]

you have to remember that if you're in a hardware build group, the only thing they really care about is bigger and better. or just bigger.

[snow lance]

just grab more system ram, its fairly cheap

[drowsy creek]

any good stable diffusion gimp extensions that work on linux?

every one I tried does not seem to work

[sleek zinc]

Hi Guys! Is there an API for Stable Assistant?

[vocal burrow]

try to delete your venv if you changed hardware.

[vocal burrow]

16gb of ram is a bit short for loading unloading sdxl model without hiccups but it should works. Is it worth the money, don't know depends of your country/market.

[ruby yoke]

just don't run other stuff at the same time

[vocal burrow]

yup

[jovial inlet]

Hi , Can someone help me with making async calls to the api: ```
import aiohttp
import aiofiles
from aiohttp import ClientSession
import asyncio
import os
from dotenv import load_dotenv
load_dotenv()

API_KEY = os.environ.get('STABLITY_AI_API_KEY')
import requests

import aiohttp
import asyncio
import os

API_URL = "https://api.stability.ai/v2beta/stable-image/generate/ultra"
OUTPUT_FORMAT = "webp"
HEADERS = {
"authorization: Bearer": API_KEY,
"accept": "image/*",
# "content-type": "multipart/form-data",
}

prompts = [
"Lighthouse on a cliff overlooking the ocean",
# Add more prompts as needed
]

async def fetch_image(session: ClientSession, prompt, output_file):
formdata = aiohttp.FormData()

data = {
    "prompt": prompt,
    "output_format": OUTPUT_FORMAT,
}
for field in data:
    formdata.add_field(field, data[field])
formdata.is_multipart= True
print(formdata.__dir__())


async with session.post(API_URL, headers=HEADERS, data=formdata) as response:
    if response.status == 200:
        with open(output_file, 'wb') as f:
            f.write(await response.read())
        print(f"Image saved: {output_file}")
    else:
        print(f"Failed to fetch image for prompt: {prompt}")
        print(f"Status: {response.status}")
        print(f"Response: {await response.text()}")

async def main():
async with aiohttp.ClientSession() as session:
tasks = []
for i, prompt in enumerate(prompts):
output_file = f"./image_{i}.{OUTPUT_FORMAT}"
tasks.append(fetch_image(session, prompt, output_file))
await asyncio.gather(*tasks)

if name == "main":
asyncio.run(main())

[open otter]

can somebody tell me if this works cuz i'ts not working for me

[ornate elk]

Nobody wants to run a random .bat file

Maybe explain what it should do and paste the content as cleartext

[open otter]

it's meant to shutdown a program i think if my code is working properly however my batch file dosen't work please help me

[ornate elk]

This #🤝｜tech-support channel is only stable diffusion related.
What's the "Programm" that should be shutdown?.
I still wont download any random .bat file.
Also you can ask chat GPT.

[open otter]

understandable i'l ask chat gpt

also are random batch files dangerous?

[ornate elk]

Yea if you just download and run them they could do anything

[timber drift]

Hi,
I installed Rembg with the extension tab, as shown on the official github, and it didn't work. There is nothing in the Extra tab. Can someone help me?
I have attached the error log.

I saw on internet it might have something to do with english and utf-8. But how to change that on SB?

[timber drift]

I succeed to install it, when changing all the locales and now I have this error.

[simple tartan]

Not super specific to SD, but does anyone know which font or package OneTrainer was intended for because it looks a little off on my Arch system

[ornate elk]

Do you have tensor rt extension installed too?

Because there is an error with it

[elfin gate]

Hey all, I am trying to train a lora and I keep hitting an issue. When I train the lora, I take sample images during the training process and I can clearly see the influence of the training on the sample. Fast forward to after training and I go to use the lora and no matter what I try I can't seem to find any sign of the training. I can't seem to figure out what the issue is. If I download and use other loras, I don't have any trouble using them and they work fine... I am at a bit of loss.

[vocal burrow]

you're not mix matching sd1.5 loras and sdxl models ?

[neat prairie]

Guys, can anyone tell me about these errors? (I train LoRA in Kohya_ss)

[elfin gate]

No, using the same model to train and generate. Both 1.5. I have tried a few models to make sure it was not a model specific issue and it seems to be the same across the board.

[elfin gate]

Looks like it can't find the images to train on, check that the path you provided is correct.

[elfin gate]

What's inside the "sharkimage" folder?

[neat prairie]

a photograph in .jpg format and a description for it in the form of .txt are called the same

[elfin gate]

That's your issue, inside the "sharkimage" folder it's looking for a folder with a number_promptKeywork format. It looks like that is your "100_LoraSharkChanged" folder. Not sure what other files you have in your other folders but you need the right folder structure. Check this out: https://vancurious.ca/generative-AI-Kohya

Not sure about the rest of the tutorial but update your folders to match the folder structure section.

[elfin gate]

Thanks for your reply. Let me know if you think of anything else. Banging my head on my desk on this one.

[vocal burrow]

does it show up in the listing, in the output log does it show up in the list of used lora ?

[elfin gate]

When generating an image, I only see this output in the console log:
Total progress: 100% 80/80 [00:06<00:00, 13.84it/s]

Same output when I use other loras that work vs my trained one that appears to have no effect. I am using A1111 for generating.

[vocal burrow]

it should list the loras detected and used in your prompt under the preview zone

if it's not there it should give you a first hint about what's going on, it's not detected as a (compatible) lora.

[deep heron]

any hint to why when i use sd 1.5 checkpoints its giving kinda grey images?

[vocal burrow]

you're not using any VAE (apart from the default one)

most models recommend using such or such VAE on top of it.

[deep heron]

oh

[vocal burrow]

VAE quick guide :
1/ What is a VAE.
It's a part of the stable-diffusion pipeline that encode/decode information from tokens to latent space and from latent spaces to pixels. Aka it transforms math statistics in pictures
2/ Where do I put my VAE ?

• VAE with .vae.pt, .vae.ckpt, .vae.safetensors extensions go into the models\Stable-diffusion folder
• VAE with .pt, .ckpt, .safetensors go into models\VAE
  3/ How do I use my VAE ? Three possibilities :
• Either you name it similar to another one of your model (eg : Anything-V3.0.safetensors + Anything-V3.0.vae.pt), by doing that it should automatically load the VAE when you load the associated model.
• You manually load your VAE by going to Settings -> Stable-Diffusion -> sd_vae and selecting your VAE
• You add an easily accessible VAE dropdown at the top of your page to quickly switch back VAE by adding sd_vae to your Settings -> User Interface -> Quicksettings list

[elfin gate]

Ah, checking there I do see my lora listed after "Lora hashes:". I was checking in the console for a message.

JFXPL, girl, masterpiece, best quality, in white shirts, upper body, looking at viewer, simple background lora:JFXPLLora:0.3 lora:lift:1
Negative prompt: low quality, worst quality, bad anatomy,bad composition, poor, low effort
Steps: 80, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 26, Seed: 3038799731, Size: 512x512, Model hash: ec41bd2a82, Model: photon_v1, Lora hashes: "JFXPLLora: 1f5f1d934f0c, lift: 5ce36857badc", Version: v1.9.4

[vocal burrow]

try only using your lora then

I see there's two of them here

[elfin gate]

I try mine, no effect. In fact, if I keep the same seed the image is the same with and without the lora. If I use the other lora with the same seed it produces a new image that matches the intent of the lora.

[vocal burrow]

even if you put the strength of your lora to crazy high level such as 2 or 3 ?

[elfin gate]

Tried all sorts of numbers from .2 to 8, identical image each time

It's like something between the sample process and when the lora is actually saved is a problem... makes no sense to me

[vocal burrow]

hmmm I don't know then.

maybe try asking in #🔧｜finetune ?

[elfin gate]

Worth shot, thanks for at least validating I am not insane lol

[ornate elk]

What's your GPU?
And yes forge is outdated and not supported for normal users anymore.
Read this and best is to switch to Auto1111:
https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/801

[hollow quartz]

Oh ok, thanks for the info, I have a nvidea gtx 1080

[ornate elk]

Ah okay, that shouldn't have any problems with Auto1111.
But Animatediff is really resource hungry and will take a long time

[hollow quartz]

Yeah, I know, planned on having it run in the background 🙂

I'll try using A1111, and will see if the error keeps being there👍

[ornate elk]

In my old PC I have a GTX1080 too, used it a long time for SD.
But its not that fast for training or video stuff

But it works

[hollow quartz]

Yeah, I might need to upgrade soon 😅

[ornate elk]

The card is still good for FullHD gaming ^^ played cyberpunk on it.
But for ai stuff the newer ones are better xD

@hollow quartz In Auto1111 you need
--xformers --medvram-sdxl --no-half-vae
as launch args in your webui-user.bat
To get the best performance.

Next Auto1111 update should also helps in performance

[hollow quartz]

Wait, you mean in webui-user.bat, correct?

I currently have --xformers --autolaunch --medvram --theme dark, should I just add --medvram-sdxl --no-half-vae ?

[ornate elk]

Change --medvram to --medvram-sdxl

Because then its faster for 1.5 models and only uses medvram for sdxl where its needed

[hollow quartz]

Ok, I'll do that, thanks a lot for the advice

[ornate elk]

No problem

[fair oxide]

can you point me towards a working version of forge, keep trying to install it (with good wifi) and it keeps giving me yellow text, ik it has smth to do with the venv and pytorch but its not working and im losing it 😭

[ornate elk]

What's your GPU?

Forge is outdated and not recommended

[fair oxide]

dawwwggg

people keep telling me to use forge because its easier on bad computers 😭

[ornate elk]

Its okay for GPUs with less than 6gb vram

@fair oxide if you still want to use forge. Follow my install guide from the first link of the pinned messages of this channel

But still what's your GPU?
You may be fine with using auto1111

[fair oxide]

my gpu only has like 6 gigs of vram

im using the guide rn

lets see if this works 😭

[ornate elk]

Is it a gtx1660 ? Or a 2060 ?

[fair oxide]

2060

[ornate elk]

Then Auto1111 will work fine

[fair oxide]

yeah but i want to run pony/xl models

[ornate elk]

That can work too

[fair oxide]

on a1111 its taking like 40 minutes per image

i was told forge is like waayyy faster

[ornate elk]

Did you edit the webui-user.bat?

For performance improvements like xformers ?

[fair oxide]

yep

i have xformers, lowvram

the whole shebang

[ornate elk]

Only these two?

40 minutes isnt right

There is a wrong setting used then

[fair oxide]

i have these three

but i get 40 minute image times, sometimes even hour long times

[ornate elk]

You can try change --lowvram to --medvram

[fair oxide]

hmm alright

[ornate elk]

Then show me your txt2img settings when it takes that long again

[fair oxide]

aight

thatll have to wait until im home though

im at mcdonalds again for their wifi

its weird because forge was downloading properly last night right

except when it hit the whole "you have shitty internet" thing

so i came here to install it and now it hits me with yellow text walls

this shiittt

[ornate elk]

Does it still opens in browser?

[fair oxide]

no

it just keeps repeating that yellow text over and over aand over and over

[ornate elk]

There seems like an error in your WiFi. Like an blocker.
SSL cert error
Best try over ethernet

[fair oxide]

well fuck

😭

[elfin gate]

That "secure-login.attwifi.com" makes me think you are behind a captive portal on public wifi, are you? lol

[craggy oxide]

Guys, is it enough an 4070 super for sdxl?

[ornate elk]

Hey sure it will work

[craggy oxide]

Thanks

[stable dust]

guys

what does y mean when stable diffuuion forge
gives you a type error

[ornate elk]

Np, 12gb vram is good for sdxl.
8gb vram will work too.
But wouldn't recommend 8gb cards in 2024

[stable dust]

It says "TypeError "nonetype" object is not iterable

oof doesstable diffusion forge outright not support pony?

cause any other model works

also wtf does forge need internet?

it outright refuses to render without connection

also idk if any model even works

[ornate elk]

It should work with pony but forge is deprecated for normal users

So better use Auto1111

[stable dust]

it says it failed to load face and body model wtf

welp, oof Ivant download ajtomatic 111

used my friends hotspot and ys about to run out

I cant download it with my data cause it would take forever and it would disconnect

[ornate elk]

Then do it when you have a stable connection or in McDonalds WiFi ^^

[stable dust]

oof, id have to move my entire pc near mcdonalds

which isnt doable

also, where would I put custom adetailer faces?

cause I have a face file

nvm oof

[ornate elk]

Inside models/adetailer

[stable dust]

my pony model outright refuses to work in stable diffusion

ughh wtf

fuck me man

went through all this trouble downloading forge only for it not to work wth pony models

[ornate elk]

Can't help much with forge since its change.
Auto1111 is your best bet

Pony will work there even with a lower end gpu

[stable dust]

oof

I cant dowbload auto 111

I could try rn with my slow ass phone data but it'll prob take hours if not all day

[ornate elk]

Can you show the cmd log when getting that error?

Also does an 1.5 model work?

[stable dust]

1.5 works no issue

huh now it works

any way I could force it to download the missig files gw for adetailer?

[ornate elk]

It should do that automatically

[glass sable]

I'm a rube and just now getting into SDXL. Where do I pick up my CNET models. Can't find them.

[ornate elk]

Here:
https://huggingface.co/lllyasviel/sd_control_collection/tree/main

And here;
https://civitai.com/models/136070?modelVersionId=267493

[glass sable]

As always, CS1o, youdaman.

[ornate elk]

Np 🙂

[fair oxide]

LMAO

mcdonalds wifi hates installing pythn stuff, i had to move to walmart wifi 😔

[fair oxide]

now its just sitting here not doing anything during the install

idk why it copied the prompt i was using cuz i copied the image but 💀

there we go

https://tenor.com/view/greg-sb-gif-sponge-bob-menacingly-sb-menacingly-sb-patrick-menacingly-fc-blairhall-primary-gif-gif-25308554

@ornate elk

3 HOURS

[fair oxide]

1 hour

💀

i switched it to medvram

[ruby yoke]

don't remember how powerful you said your machine was?

[fair oxide]

It's got like 6 gigs vram

It's more lower end

I think its 2060

[ruby yoke]

i might leave it run overnight then. or a couple days.

[fair oxide]

💀

[ruby yoke]

in the past i've had to leave a machine running for a week in order to finish a render

[fair oxide]

💀 x2

Naaaahhhhhhh

[ruby yoke]

there's a reason that high end nvidia GPUS with at least 16 Gig of ram is recommended

[fair oxide]

My laptop bluescreens twice a day there's no "letting it take its time"

[ruby yoke]

maybe it's time to take that machine to a computer tech and find out what is going on?

[fair oxide]

It's 4 years old and likely needs cleaned/new paste

But

Idk how to fix all that

And I don't have many to pay people to fix it either

I live in poverty bro that's why I'm using Walmart wifi and an offline ai image generator LOL

[ruby yoke]

then you don't try to fix that.

[ruby yoke]

you might want to use mage.space then. you dont' get all the bells and whistles with a free account, but you get a good generator and you get unlimited images

[fair oxide]

I will try this

[ruby yoke]

let me know if you get stuck or have questions. feel free to DM me

[fair oxide]

I don't even need all the bells and whistles, just unlimited gens, maybe upscaler access, some loras

[fair oxide]

Thanks 😭

[neat prairie]

who is training LoRA on a GTX 1080 video card on Kohya_ss, send me the config, I can’t configure it normally, it always says that there is not enough video memory, although I have 8 GB of video memory and 8 GB of RAM

[ruby yoke]

think you're gonna have to upgrade that memory

[simple tartan]

anyone have kohya working for Lora training on AMD?

[steep cape]

Hey everyone, very new SD user here. I have done a lot of research, watched a lot of videos and guides and feel like I have gotten running pretty quick. I just have a common question: how fast are you guys able to render a new txt2img prompt at a resolution like 768x512 on a GPU like RTX 2060 Super 8GB?

For me at that resolution, using Euler a Karras, 35 sampling steps, CFG scale 7 and nothing else - it takes about 4:30 minutes. Is this normal or could I speed it up somehow?

[ruby yoke]

that's a really slow setup you have. i'd expect it to take a while

[steep cape]

Yeah I know the 8GB VRAM is very low, I have been looking at 12GB cards or even the 24GB 3090 but for now I'm stuck with the 2060. There is no other way to speed up the process, right?

[ruby yoke]

not really.

[steep cape]

Ok, thanks for quick reply! I read somewhere that increasing either batch count or size could improve speed, but I haven't been able to achieve this

[ruby yoke]

@ornate elk might have some options, but as far as I know, no.

[steep cape]

@ruby yoke Do you have have GPU upgrade recommendation? Also, what card do you have and how fast can you render with the settings I mentioned in my first post?

[ruby yoke]

https://www.newegg.com/abs-aqa14700kf4060ti16g-stratos-aqua/p/N82E16883360436 this is the machine i have

[ornate elk]

Thats not normal. Only if your using sdxl or hires fix

Please show your txt2img settings

[simple tartan]

@ornate elk since you seem to be the expert, at least for all my questions, any idea why kohya wont train my loras, 7900XTX, plenty of ram, but keeps bombing with some parameter trying to run cuda, well no shit, its AMD, not nvidia>

ValueError: torch.cuda.is_available() should be True but is False. xformers' memory efficient attention is only available for GPU

[ornate elk]

Hey, Kohya won't work with AMD GPUs.
You have to use OneTrainer (it supports Zluda)

[simple tartan]

well that would explain it 😄

i'll try OT again then

[simple tartan]

hmm

RuntimeError: Expected one of cpu, cuda, ipu, xpu, mkldnn, opengl, opencl, ideep, hip, ve, fpga, ort, xla, lazy, vulkan, mps, meta, hpu, mtia, privateuseone device type at start of device string: zluda

reinstalled zluda and trying again, suprised we are doing "rocm"

[ornate elk]

Zluda needs HIP SDK (ROCm)

[simple tartan]

i'm going to end up having to pay civitai to train

i have rocm install and zluda

comfy works fine, ollama works fine

OT is the only thing i cant get working

i'm even using the nightly of rocm with no issues

[ornate elk]

Windows or Linux?

[simple tartan]

arch linux

[ornate elk]

Ahhh okay

[simple tartan]

https://tenor.com/view/archpepe-gif-20110294

[ornate elk]

Then idk how to fix it.
I only use Zluda on windows

[simple tartan]

ah, damn

[ornate elk]

OneTrainer should work with rocm natively I guess. But you can ask in their discord

[simple tartan]

yea, let me try that

[ornate elk]

Let me know what you find out 👍

[simple tartan]

will do

[neat prairie]

I don’t understand why the training takes so long I have 8GB of video memory and 8GB of RAM

[simple tartan]

wtf, 8gb of ram? was your pc built 20 years ago?

[karmic crown]

I do not dare to try that on M3 Pro 18GB

I would not recommend trying on anything less than 16GB RAM + 16GB VRAM with at least 3xxx GPU. for decent speed, at least 4xxxx with 24GB is needed, but I might be wrong

can someone with more experience confirm this, or tell me that I have no clue? 🙃

[glacial flax]

Hi all, can you please tell me how to fix these errors?
I recently installed Stable Diffusion but I can't figure out what to do, I don't know much about code.

[karmic crown]

you have error here on top

i am not sure if that extension works with new versions of a1111

you might need to remove that extension

do you have some errors before that?

roop is replaced by reactor (if i remember correctly)

[glacial flax]

Everything is fine before this error, it's the first one on the list.))
Can you tell me how to remove this extension? Don't think I'm stupid, it's hard for me to figure out, but I want to deal with it

[karmic crown]

go to folder where stable-diffusion-webui is, find extensions folder inside and remove the extension

c:\stable-diffusion-portable-main\extensions\sd-webui-roop

[glacial flax]

I'll give it a try!

Ok, that error is fixed, but there are more errors, can you help with the rest?

[karmic crown]

i can't help much with onix installation
it looks like some dll which is part of onix cannot be loaded corectly
you might need to reinstall onix related things

you can try one of CS1o's guides for windows if you want

https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides

looks like zluda works best with AMD, but the initial image generation can take up to 40 minutes (yes i know it sounds wrong), but after then they say it works pretty fast

[glacial flax]

Thank you very much, I'll try to figure it out!)

[karmic crown]

if you have any problem, you can always ask here

[ornate elk]

What's your GPU?

[glacial flax]

nvidia geforce gtx 1650

[oak thistle]

why are my images so weard using same prompts like others but i get only such things

[vocal burrow]

1/ you're using the ANCIENT vanilla SD1.5 model
2/ denoiser strength is relatively high for upscaling
3/ the prompt is not the only thing dictating the quality of an output

[ornate elk]

Okay, best would be to reinstall Stable-diffusion by using my Nvidia Automatic1111 install guide.

[glacial flax]

thanks

[ornate elk]

NP, the guide is the one Viking linked, or in the pinned messages of this channel.
After a clean installation you can reinstall the extensions too and it should work

[glacial flax]

Okay, I got it, thanks for your help.

[oak thistle]

can you find the prompt of a image using stable diffusion

[vocal burrow]

prompts and settings are stored in exif data. You can find everything by either dragging the picture in the positive prompt field, in the png info tab, or any other (online) tool able to read exif data

[oak thistle]

it says parameterse none

[vocal burrow]

then someone either stripped the exif data or the software used was set to not save anyyhing in exif

[oak thistle]

thank you for your information

[simple tartan]

boom, and i'm training in linux on my 7900xtx!! @ornate elk
create a dedicated conda/venv and pip the requirements-rocm.txt and the requirements-global.txt
I had an order issue with the rocm and and to manually install the onnxruntime-gpu==1.18.0 but after that, training is running now, will take a while though 😄 163 images with 25 repeats...

[ornate elk]

Nice! THX for the information.
Do you train a model or lora?

And Sdxl or 1.5 ?

[simple tartan]

i'm training my character lora, sdxl

[karmic crown]

the list of temp fixes in my script is getting larger and larger
I hope we will get 1.10 soon

[shell crown]

Anyone have any tips on getting camera movement with animatediff hotshotXL? Things like zoom in, pan left/right etc? Would i need a lora for this?

[stable dust]

any way I can make stablediffusion webui forge use less ram? i read its more memory efficient than automatic111 and yet my ram usage stays at around 90/99% and crashes a lot while geberating

[steep cape]

I am not using Hires fix but I am using various different models, not the default SD 1.5 or anything like it. So using a model like SDXL would make the render time I'm getting more normal?

[steep cape]

This looks super nice! My PC is a bit on the older side now, but today I purchased an RTX 3090 24GB 2nd hand, so super excited to see the boost I will get from it

[ruby yoke]

the one i have works extremely well, i'm really happy with it. i expect you'll be just as happy with your new machine

[steep cape]

Great to hear, really excited! 🙂 I just have another question if you don't mind. When you make a txt2img render, do you get the best result by starting at a low resolution like 512x768 and then upscaling afterwards, or should you render straight into a high resolution like 1080p?

[ruby yoke]

i'll use the low resolution when creating. i usually don't need more than that, anyway, for what i'm doing. but if i need an image larger, i'll upscale it with something OTHER than the AI I'm generating with - usually with magnific but also at times with the image upscaler on capcut's magic tools, or topaz

[steep cape]

So when you are done rendering in a lower res, the image is pretty much how you want it? All my renders so far at low res always look very rough and "template" like, and only after doing an upscale afterwards, all the details seem to be added. I normally use around 35-40 steps, this is too low maybe?

[ruby yoke]

for what i'm doing, usually, yes. i rarely have anything i want to change or fix. if i do, i use other tools than the AI i created it with - i tend to use 32 steps

[steep cape]

Alright, I'll keep tinkering with it. Thanks for your help!

[heavy perch]

Hi, any solution for this?

[ruby yoke]

what is it you're trying to do that got you that error?

[heavy perch]

open Fooocus

[somber oriole]

hello, i'm quite new to stable diffusion I was wondering what version of stable diffusion what would be best for Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz , gtx 1070 8gb and 16gb of ram

as i've tried running comfy ui with stable diffusion xl, and it just keeps on saying connecting when i queue a prompt also with connection errors, so i'm wondering which version of stable diffusion would be best

[ornate elk]

yes using sdxl model takes a lot more resources, but still 4:30 is to long. Whats inside your webui-user.bat?

your mostly missing the performance adjustments

[ornate elk]

hey, i would use Auto1111 with this card

it needs --xformers --medvram-sdxl --no-half-vae as launch args in the webui-user.bat
to get the best performance and compatibility

[stable dust]

Anyone have a solution?

[ornate elk]

how much RAM do you have?

[stable dust]

16gb

im not sure but it feels like comfy uses less? but my system does hug a bit while rendering so maybe it also uses alot? ive just noticed forge crashes more idk, ugh

I really like forge tho cause its easy to use an adetailer

[ornate elk]

like i said before your best at using Auto1111

but using sdxl requires 16gb RAM or more so you need to increase your windows pagefile

[stable dust]

I would but Ivant download it, maybe I can figure out a way

maybe I can use my other phone and use the cox wifi's limted 60 min time period and aee jf it works with that phpne

both of my other phones logged out after an hour oof

also, I read that forge was more ram efficient than auto111

[ornate elk]

RAM isnt a big problem when you can increase the windows pagefile

that mostly saves you

[stable dust]

ah ok, what should I set it to you think?

I have forge on my 970 evo, got 60gb left free on it

and got like 20gb left on my boot drive

[ornate elk]

16000 min and 24000 max for C drive (if its an ssd) and disabled for any other drive

[stable dust]

ah ok so 16gb to 24gb

wouldnt that make eveything much slower?

[ornate elk]

no, for some people it even made it faster

[stable dust]

maybe not too noticable tho if I havw it set on my 970 evo (same where I have forge)

huh, odd

im guessing it wouldnt make that much of a difference whether ts on my 850 evo or 970

[ornate elk]

best is to have only one pagefile enabled. then pick the fastest ssd on the system

[stable dust]

should I have any launch arguments?

like --xformers or whatnot?

[ornate elk]

that makes a big difference as the 850 is an old sata ssd with 500mb max read/write and an 970 evo is an nvme ssd with 2700 read/write

so better set pagefile to the 970evo

[stable dust]

idk, I just never noiced big loading sifferences when t came to games lol

maybe a second or two faster

[ornate elk]

yea but for paging it will help

[stable dust]

would that allow me to render at higher resolutions too?

or is that entirely vram dependent?

[ornate elk]

xformers is a must

but your using forge, it should already use it or similar stuff

[stable dust]

oof

oh wait nvm

I think it does use xformers

[ornate elk]

yea

[stable dust]

cause it was installing xformers when I was downloading it

[ornate elk]

thats vram dependent

[stable dust]

big oof

I used to have a 980 ti but it died from rdr2

[ornate elk]

but the key is always use upscaling to get highre resolution

[stable dust]

yea I use hires.fix

with datx4 for the most part

havent touched sd 1.5 in a few days lol

i noticed when I enabled Cuda Stream it would just crash lol

Does forge automaically save all generated images?

cause I did a few good gens but forgot to hit save

would be nice if it was like Inference for Comfy where it just saves all of them with their metadata

apparently forge also uses less vram

oof

i guess I'll check myself if I get auto downloaded

but then Ivotta redownload adetailer

[ornate elk]

Normal yes

[stable dust]

does the metadata save too?

i had afew past gens that were really good but forgot what prompts I had used

[ornate elk]

Yes meta data is in there too

[stable dust]

ah cool

how would I get that?

like is it under details in the file or?

[ornate elk]

You can drop the image into the PNG-Info tab

[stable dust]

ah cool, neato

what is refiner for?

[ornate elk]

To get them.
Or any image viewer with Exif data tool

[ornate elk]

Refiner was for the base sdxl model

Isnt used anymore since the community models are good without it

[stable dust]

ah ok cool

downloaded TunixStyle pdxl rn

Deepdream pdxl does really really good with fur

tried indigoxl and it looked 2d, Echo alpha looked almost like plastic

[ornate elk]

How do you download new models with your internet situation?

[somber oriole]

thank you i'll look into this

[somber oriole]

hey i've messaged you if that's okay

[heavy perch]

@ornate elk hi again 😅

[ornate elk]

Hey

[heavy perch]

Can you help again

[ornate elk]

sure, whats the issue?

[heavy perch]

@ornate elk

can we go dms btw, easier

[ornate elk]

when do you get that error?

[steep cape]

I have these in my webui-user.bat:
--allow-code --theme dark --xformers --skip-torch-cuda-test --no-half-vae

Anything I should change or add?

[ornate elk]

yea remove --allow-code --skip-torch-cuda-test
and instead add: --medvram-sdxl so that you have only:
--theme dark --xformers --medvram-sdxl --no-half-vae

[steep cape]

Damn, that helped a lot, it went from minutes to seconds 😮

[ornate elk]

perfect 🙂

[steep cape]

Thank you! I have a much better GPU on the way, so after I install that, I just remove the --medvram-sdxl line and leave it as is?

[ornate elk]

yea, you then can also remove --no-half-vae
so only --xformers and --theme dark (or just change windows to dark theme)
But you then also need to delete the venv folder before launching it with the new Card and better install the latest GPU driver on top.

[delicate citrus]

any stable version of python works for SD?

[simple tartan]

i think everyone is still building in 3.10 so best to use conda or venv and do the same

I might have to tune my OT a bit more, woke up expecting the job to be finished...not even close:

[delicate citrus]

thanks!

[delicate citrus]

looks cool, what is OT?

[simple tartan]

OneTrainer

[delicate citrus]

hmm cool

[simple tartan]

i'm going to try to integrate it into my AI Garage once I finish the current updates

[delicate citrus]

AI Garage, this is new to me, what is this?

[simple tartan]

tool I wrote, one sec, i'll link my post

#💬｜general-chat message

[oblique vapor]

This may be a super dumb question, but is there a correct way to download models from Civtai? Or are there other components or something that I might need if I see a model I like?

[simple tartan]

if you're using A1111, use the extension, if using comfy or something else, just downlaod it directly

[delicate citrus]

interesting front-end, I started follow your repo

remember me st and kobold

[oblique vapor]

I think I figured out my issue, it was something really dumb. Still giong to run some tests, but seeing MUCH better results

[viral knot]

why are you posting all this in this channel?

[marsh tusk]

My seemingly stable Fooocus program stopped working today, it just wont open, can anyone help please, thank you

[ornate elk]

Check the latest update of fooocus and do the Update steps there;
https://github.com/lllyasviel/Fooocus/releases/tag/v2.5.0

[marsh tusk]

Thank you CS1o, it says 'Already up to date'

[ornate elk]

Then try the steps from 2.

[west rune]

Im having massive trouble training an embedding in with a1111
I was actually gonna ask why it keeps giving me errors when i commit to the training but today it wont even let me create a new embedding 🙄

Im at a loss with a1111. I feel like im doing exactly as the tutorial videos im watching tells me to do. Yet i keep encountering errors fucking everywhere. They seem to appear almost at random with the only things ive done being restarting my computer or simply going away for an hour.

The best i've ever had a1111 working was when i installed it from scratch through git. But god that was hell with the python versions and old repository version. And im not good enough at python to understand what im doing when im setting up an anvironment with ex. anaconda.

How much do you guys battle with just getting it to work?

Tried installing locally from git. Getting the same error.
I have a gtx3070 with 8gb vram and a Ryzen 7 5800X and 32gb ram

[delicate citrus]

can I allocate ram do use or must be only vram? I'm looking for a pc to buy

[vocal burrow]

1/ that log is incomplete (copy paste everything into a .txt and drop said file in here)
2/ You should nt have to setup anything manually. And I d also recommend throwing anaconda in the garbage bin and sticking to regular python (3.10.6 to 3.10.11)
3/ embedding are not really worth the hassle, I d recommend jumping straight to Lora with Kohya_ss or OneTrainer.
4/ Tutorial videos are often outdated but youtubers keep them up anyways (hello Olivio) because views, money and whatnot. Prefer official instructions on github whenever possible or at the very least recent text tutorials as they re easier to keep updated
5/ your first screenshot seems to indicate that you re using forge and not a1111. Given that forge was more of a proof of concept that hasn t been updated in 5 months or so. I can t recommend it.

[west rune]

Im gonna purge forge immediately. I went to try it because a1111 gave me the same error messages.
When you say the log is incomplete, do you need everything from i press launch to the very last byte?
Is there a difference between using stability matrix and downloading the sd.webui.zip from the a1111 git? Its all just a premade virtual environment right?
Is it better to run it off python with the dependencies installed instead of through a virtual environment?

[vocal burrow]

• log is incomplete, yes I want to see everything.
• stability matrix is just an additional UI sitting on top of whatever you re launching. Unless you re a power user constantly switching back and forth between different UI (comfyui, a1111, sd.next, fooocus, etc). I don t see any use for it. It s just gonna add another layer of potential problems.
• sd.webui.zip, don t use the .zip, install it properly by using git clone (cd guides in pinned comment or read the a1111 page on github)
• better to run it from a venv as that way it will keep all its mess within one venv folder and won t propagate to your entire system

[latent wave]

Logs [Count:Whole number]

[-channel channel:Channel - Optional channel to log instead]

Invalid arguments provided: "is incomplete, yes I want to see everything.\n- stability matrix is just an additional UI sitting on top of whatever you re launching. Unless you re a power user constantly switching back and forth between different UI (comfyui, a1111, sd.next, fooocus, etc). I don t see any use for it. It s just gonna add another layer of potential problems.\n- sd.webui.zip, don t use the .zip, install it properly by using `git clone` (cd guides in pinned comment or read the a1111 page on github)\n- better to run it from a venv as that way it will keep all its mess within one venv folder and won t propagate to your entire system" is not a whole number

[west rune]

Thanks for the tips! I'll try that when i get back home again. However im using comfyui for all generation so for the training im gonna give onetrainer a go. Thanks again :)

[vocal burrow]

there s no corelation between the trainer and your sd client.

you can use any lora, embeding, etc regardless of where they come from / which trainer was used.

[west rune]

Oh yeah i know. It's just i dont use a1111 except for training textural inversion. As i dont use it for anything else and i have so many problems with it i might just scrap it and use onetrainer instead 😛

Can you recommend anyone/anywhere who's good at explaining what im actually looking at with the clients and what the terms actually mean. I find the majority of tutorials/youtubers just parrot the same things without actually explaining what the things means. It seems to me like the clients are made for people who arent pro programmers but that it still feels like you need a phd in computer science to effectively use them.

[vocal burrow]

For Lora this one seems good enough https://vancurious.ca/generative-AI-Kohya

[ornate elk]

Vram is the important part for ai. Best is to have 12gb or more. 16gb ram is also required

[delicate citrus]

ok thanks

[dense tapir]

It's already in my txt file, but I don't think it works (?)

Also I hope you have a good day

[karmic crown]

it needs to be exact path where your models are
if your models are in D:\AI\SD\models, you must use

--ckpt-dir "D:\AI\SD\models"

if it is D:\stable-diffusion-models than you need

--ckpt-dir "D:\stable-diffusion-models"

...

[ancient geyser]

I use two PC's for AI generation - one with a 4070 and one with a 3080 TI. Which one should be my main machine? It looks like the 3080 TI might be slightly more powerful but with the 4070 being a newer card, maybe its more optimized for the task?

[indigo pike]

Can you mix 2 images with Stable Diffusion, similar how Artbreeders Mixer does it?

[ornate elk]

compare the speeds with the same image. they are both that "new" that they support everything needed for ai stuff

[west rune]

Appreciate it

[waxen zenith]

I'm trying to give my Hyper-V hosted VM access to my GPU and I'm unable to.
I tried following tutorials such as https://www.youtube.com/watch?v=XLLcc29EZ_8, but in my VM's Device Manager, the card isn't showing up (well, there is a "Display Adapter" with the yellow question mark.)
Anyone has ideas about how I could proceed?

[steep cape]

Not sure if this is the place to ask, but I have a question regarding Inpaint and fixing things in my images. I am making a comic and in one of the frames the color on a character is wrong. How would I go about changing this? I don't want the motive itself to change, ONLY the color. I have tried with mask/no mask, messed with the different settings of mask mode, scale and denoising strength but it often ends up changing the motive completely or make new random glitches. Is the Inpaint feature the right thing to use here?

[deft violet]

@oblique meadow automatic111

I am planing to use

[oblique meadow]

are you sure, comfy ui or swarm ui is a lot faster

[deft violet]

on prompting ?

does my image come faster

[oblique meadow]

yeah

[deft violet]

why

it is just ui platform

I thought the model was the important one

[safe inlet]

backend also that executes the stuff on the UI

[deft violet]

so you recommend comfy ui

[oblique meadow]

a few things with automatic111, backend is slower as kagi said, frontend is also slower bc of gradio.
comfy ui and swarm ui are probably the best options

[ornate elk]

but the speed difference is like 1-10 seconds

[deft violet]

do I just install sdxl from civitai

or do I need to consider on clip stuf

[ornate elk]

whats your gpu?

[deft violet]

rtx 2060

[ornate elk]

then i would recommend using auto1111

[deft violet]

why

[deft violet]

.

[ornate elk]

because it has more settings to get the most out of the 6gb vram.

sdxl should also work

but sdxl models are not that fast on 6gb gpus

[deft violet]

you said sdxl models

arent there single one

[ornate elk]

There is one BASE Model thats called SDXL 1.0
There is also one Base Model thats called 1.5-ema-pruned.
Both Models are made by Stability AI and got published to the community for Free.

The Community uses these Models and train their own Versions on it.
These Community Models are then also either Based on SDXL or 1.5
On Civitai.com you can find hundreds of models made by the community.
These Models are usually better than the inital Base models from stability.

Automatic1111 and Comfyui etc are so called "Stable Diffusion Webuis"
in these tools you can load and use multiple Models and switch between them

[deft violet]

what is the difference between using sdxl itself or

sth based on it

[ornate elk]

the community models are better

the base models are just a base. people then trained on that base in certain directions like, anime, or realism etc

the output images have a much better quality

[deft violet]

o

which one do you recommend for animes

or realism

[ornate elk]

For Realism try Juggernaut XL
And for Anime its AAM XL (Anime Mix)

[deft violet]

sure thx

[ornate elk]

NP, and in the pinned messages of this channel you find Installation guides for Auto1111 and Comfyui

[terse walrus]

For automatic1111 I keep getting this dumb error all the time which stops me from using it and always going back to comfy. I followed a long tutorial for the whole installation thing and the webui was working great, but once I tried switching models to sd3 it just broke and did this thing over and over again like how it always did. I'm so frustrated cause it feels like I'm the only one with this issue as I've tried searching online and either no one has a fix or no videos pop up

[unborn bramble]

Alright guys so I want to train a 49 character pack. How many images is good for a character? And does the amount of images for the folder the character matters (like if one has 43 and another one has 256)
Also does the amount differ in SD1.5 (classic) and SDXL

[ornate elk]

That error is normal when using an unsupported model like SD3.
The next Auto1111 update will support it.

[terse walrus]

Well sd3 kinda bad anyways

As soon as you ask it for a blue pillow it just blows up

[ornate elk]

Feel free to show me your settings when you get that error again

[unborn bramble]

How does one train their own AI model? (Not a Lora, but a full out image model)
How many images are recommended?