🤝｜tech-support

[wicked widget]

civit or github

input the images or select a directory for it to scan and preload values, aim for 2000-4000 steps

2000 steps is what i like to start with

[fickle panther]

I'm using the program but how do i select the dataset?

[wicked widget]

next tab on imager viewer

[fickle panther]

I'm kinda confused, so i have the settings at default with 32 images. Should I change anything or just let it do it's thing?

[wicked widget]

alright sorry needed to bring it up cos i dont even remember how i had it formatted lmao adhd brain
so first will be the number of total images
second will be how many repeats you have it set to, so in onetrainer you would edit this in the concepts tab on the balancing area
third will be the batch size, i typically do 2 or 1 cause vram but it matters in the step calculations
fourth will be target steps, steps are the amount of passthroughs or total cooking time to simplify it, 2000 is generally a good number but you could play with it

now after all that, it will show you under the amount of epochs you should input with all of those numbers known since it does the math for you

example i used here, since my dataset is 113 images, i wanted to know the optimal epochs i would need to do to reach 2000 steps with the images repeating 4 times in the training processs

[fickle panther]

Tried all this, your tips are fantastic!

Model looks much better!

[wicked widget]

glad it helped

🙂

[fickle panther]

you the goat

[wicked widget]

welll i used to be as clueless as the next, CS1o here is the one who guided me in the right direction

[fickle panther]

nooo, most people don't bother helping with certain issues in this server. You did, and you have a program to help with it

you the goat

[wicked widget]

yeah theres a select few that does their best to help, i do what i can 🙂

[wicked widget]

try any of my worfklows

[ornate elk]

It depends on the stuff that a workflow includes.
Zluda emulates cuda but not everything can transform correctly when special stuff is used or needed.
Like face detecting stuff or faceswapper mostly use CPU on AMD as they can't use GPU for that.
Also everything that uses onnx models will use CPU for that.

Normal checkpoints and loras are all compatible

[paper ridge]

Gotcha, so its normal that some models in combination with certain workflows will end up having "current: cpu"

[ornate elk]

Yes but everything that involves the models like generating images or video frames should use the GPU

[paper ridge]

Oh interesting, I was using a basic workflow just to generate images and it still only used cpu with those models

I noticed SD1.5 models tend to say "current: cpu", but SDXL models say "cuda0"

[ornate elk]

Okay, but the images get generated in seconds still?

[paper ridge]

roughly twice as long

like 20 sec instead of 7 sec

[ornate elk]

Hmm

Can you show me the workflow?

I'm not at the PC rn, but later I can do some tests

[paper ridge]

that would be amazing thank you, is it okay if i DM you?

[ornate elk]

Sure

[warped canyon]

Why is it that when I try to inpaint an image i have on the pc, stable diffusion changes the images by like 4 pixels, and i want it to remain the exact size of the original image

[lucid yoke]

Mask blur?

Or mask grow settings

[warped canyon]

ohh, could be that, ill try

so i should set mask blur to 0?

[lucid yoke]

Depending on the ui but yeah

Granted 0 blur/grow on impaint gonna look rough

[granite stirrup]

Heey guys!! I'm getting the following error when loading a upscale model in ComFy UI. Tried multiple models and same error.

"Weights only load failed. This file can still be loaded..."

Have any of you ever faced this error?

[lucid yoke]

looks like a pytorch error, got the full log?

could it be like:

[formal trout]

I'm getting this in Comfy

a SD XL generation

[lucid yoke]

GPU?

[formal trout]

RX 6700

[lucid yoke]

hmm did you have past issues with this or is it a new one

[formal trout]

I used to use the Automatic1111 venv for running Comfy, but now I'm running it from it own thing, no venv it seems

[formal trout]

when I used like that in the past I didn't have that issues

Now I see that message, that generation kind of get a lot of to generated, then it generated, then it generated at a regular ration, with the message

[formal trout]

It seems to work "normaly" after

[foggy dagger]

Was training a Lora with OneTrainer and started doing this in the samples.
Should I stop it an restart it to see if it does the same thing or this could only be the decode and the training still goes correctly?

was running 200 epoch, batch of 2 with 50 images, still goes at 40%, around three more hours to finish

[loud pond]

@ornate elk all my images are coming out weird, do you have any videos on prompt improvement? or is there any way i can self de-bug?

[loud pond]

or anything you would reccomend?

[ornate elk]

can you show an example?

[loud pond]

oh alright lemme make a new one ( I usually delete all my horrible images)

[loud pond]

i generated a character by i got this 💀

[ornate elk]

can you show me the settings you used?

[brave frigate]

Can someone help me with this? I keep getting noise images eventho everything should be fine

[loud pond]

send images and settings

[ornate elk]

can you try an other model?

ponyv6 is the worst pony model (not very compatible)

[loud pond]

what model would you suggest? and would pony lora's work on any other model?

[ornate elk]

yea any other pony based model

like this:
https://civitai.com/models/439889/prefect-pony-xl

[ornate elk]

hey, can you show an example + your settings used?

[loud pond]

oh got it, is autism sdxl ok?

[ornate elk]

yep would work too

[brave frigate]

i have the amd setup if it helps, i got a AMD Radeon RX 7900 XT 20GB, AMD Ryzen 7 7800X3D and 32GB RAM

[loud pond]

can you try it with a lower denoising strength for yout upscaler?

[brave frigate]

looks like it doesnt do anything to make it better, still that bad

i tried .4, .2, .1, and even .7 and 1 all just noise

[ornate elk]

whats your GPU?
The model is a so called v-pred model. its known to work in Forge / reforge and comfyui but idk about sdnext

any none v-pred model will work

[brave frigate]

AMD Radeon RX 7900 XT 20GB

I'll try the epsilon_pred version of the noobai see if that helps

[loud pond]

it gave me a feet pic (it has full body in the prompt, i didnt ask for this 💀)

^^^^ yeah its not generating characters

[ornate elk]

remove everything from the prompt and try cute cat

if that works, the fault is your prompt / loras etc

[brave frigate]

looks like a densoising issue to me or you need to add some negative prompts

[loud pond]

should i remove the neagtive too?

[loud pond]

what would you suggest?

[ornate elk]

nope seems fine

but the wildcard

and the loras

test without them

[brave frigate]

whats your current ngprompt?

[loud pond]

blurry, deformed, malformed, low resolution, bad anatomy, fused body parts, bad proportions, extra limbs, bad hands, bad feet, watermark, text, signature, cropped, flat, smudged, pixelated, distorted face, wrong perspective, multiple torsos, merged bodies, low detail

[brave frigate]

hmm looks good to me as well, you've got merged bodies so it should work fine

I'd add "morphed body parts" but i believe this might not be a big change

[loud pond]

yeah only thing is my pc tends to take a while to generate an image

[loud pond]

got it!

[brave frigate]

if it takes that long you might need to lower the res and use an upscaler

[loud pond]

how does that work?

could you please explain ive always wanted to do that method but its never worked before for me

[brave frigate]

eg i use 540x960 as width and hight and have the 4xUltrasharp upscaler

so you have an option called "Hires. fix"

thats your upscaler

it already has a couple in there

[loud pond]

is that from civitai?

[brave frigate]

When you use upscaler make sure to lower the denoising to 0.4-0.5 otherwise it'll morph the body or completely change the output after 50%

no i use Stable diffusion

[loud pond]

Generated this, i think its my prompting then

[ornate elk]

yep, maybe to much loras with a lora weight of 1

[loud pond]

alright, ill give it a try thank you

[loud pond]

what would you suggest a good weight?

[ornate elk]

depends on the loras

[brave frigate]

0.8 is always a good choice imo

0.5-0.8

[ornate elk]

yea 05.-0.8 is good

[loud pond]

alright thank you!

[brave frigate]

looks like epsilonPred works thanks!

[loud pond]

what style are you using btw? it looks so good

[ornate elk]

np, also a question, do you use only sdxl based models or also flux and sd3.5 on sdnext

[brave frigate]

https://civitai.com/models/1009396/redtvpes-illustriousnoobai-styles?modelVersionId=1131623

[brave frigate]

I use mostly, illustrious, sdxl and pony

those are my checkpoint i have installed xD

[ornate elk]

ah okay and your on windows or linux? zluda i guess?

[brave frigate]

windows

zluda bc of AMD

if i had NVIDIA i could use the normal setup

[ornate elk]

ah nice im on a 7900XTX with Zluda too

[brave frigate]

yea

[ornate elk]

but i mainly use Auto1111 or Forge

[brave frigate]

well mine is Auto1111 as well

just in combo with zluda i think its part of the installation

[ornate elk]

yours is SDnext

its a fork of auto1111

not the original

[brave frigate]

huh where can i see that?

cuz i just went after the tutorial in this chat here that you sent in ages ago

the [AMD]

wait bruh

[ornate elk]

oh wait you just changed the color

i was confused

check the bottom of the webui to be sure

[ornate elk]

sry, yea thats auto1111

all good

[brave frigate]

but i just realized smth

[ornate elk]

your on an old torch version

[brave frigate]

i installed the one for NVIDIA Gpus and not for AMD GPUs xD

[ornate elk]

2.3.1 is the best one

[brave frigate]

yea i did it like 9 months ago might need to update it

[ornate elk]

no its fine for zluda but 2.2.2 is not recommended

[brave frigate]

can i just update the torch version? or do i need to reinstall the whole thing..

[ornate elk]

you can update your webui by adding git pullinto and empty row in the webui-user.bat
then delete the venv folder and relaunch

it will download the latest compatible version then (2.3.1cu118)

[brave frigate]

I'll do that in a sec

need to remember to do that once in a while

[ornate elk]

with git pull it will auto update everytime you launch

also if your on rocm 5.7 and and an old zluda version i would recommend using the newer stuff

[brave frigate]

i have the git pull command in, just never deleted the venv folder

doin that rn been loading for a bit

[brave frigate]

so whats the difference between 2.2 and the new one?

[ornate elk]

the new one is more stable

[brave frigate]

i see, well its all done updating so I'll just try one i did before and see if something is wrong xD

Ah

I have errors

[ornate elk]

these two errors are normal and can be ignored

[brave frigate]

yea but for some reason it won't start to generate an image

[ornate elk]

yea because zluda is compiling

it will take 15-30 minutes

happens when a torch version is changed or the gpu driver is updated or you change rocm and zluda files

[brave frigate]

hm alright, I'll let it do its thing for a while, see if smth happens

[ornate elk]

yep let it run until you get an image

[brave frigate]

cuz there's like no indication on how far it is, just a repeated "Compilation is in progress. Please wait..."

do i need to clear the cache as well after the updates are fully done?

[ornate elk]

nope

[ornate elk]

you can try

[onyx comet]

How did you get the image? I want some pics for nails

[loud pond]

it was an error generation (meaning it produced a totally random image which was different from the prompt)

this is the prompt for it tho

[onyx comet]

How did you get the prompt? I haven't been here for a while. Things change a lot

[loud pond]

oh i think the main thing what caused the generation was this lora - https://civitai.com/models/412632/views-and-composition

[loud pond]

using it you should be able to change what is the main thing generated in your prompt, and how close you want your characters

[noble idol]

@ornate elk so i edited to ui config and nothing happened

edited the ui config*

what do i do

[ornate elk]

You can try edit the config.json maybe that works

[noble idol]

alr, lets see

just a sec

it worked!

sweet thanks

[wicked widget]

how much vram again?

if below 16 no keep it

[wicked widget]

you could do fp16

i mean i do it with my 12gb even if adds abit of strain cos it understands the clip better than fp8

[reef heron]

does anybody know if there is a WSL ROCm RDNA2 solution or work around yet, please?

[wicked widget]

@ornate elk

look above. seems a question youd know

[meager quiver]

For flux, there's the pro, pro1.1 and ultra who are all training datas, so if I download them what should I do with them?

[cyan mountain]

i downloaded hunyuan ai in comfui as a gguf. how do i add like audio synced to the video within comfyui ?

[ornate elk]

Nope, not that I know one. But you can use Zluda if you want a good speed on Windows with AMD

[gilded yew]

Hey guys i am facing issue with automatic1111 i am using gtx 1060 6gb with r5 3600
And when i starting useng stable duffion like a week ago my pc can generate 512x512 image in 20 second easy but now when i try to generate that same again it take more they 1min and results also is some werd graphics glitches

I search on internet maybe some one have the same issue but i couldn't find any

I have try using medvram lowvram and some the like xformers but still no improve ment

Any advise would bee relly appreciated and sorry for my bad English

It all started when i start messing with comfyui

[ornate elk]

Hey I wouldn't recommend Comfyui for that GPU.
You will get the best performance on Forge Webui with that GPU.

[gilded yew]

But why would it slow down

[ornate elk]

Can you show the workflow you used?

[gilded yew]

For comfyui its default text to image

Model i use is
Ponyrealism
Henmixreal

currently i am using text to image and then upscale and its takes 15min to complete whole process

[lucid yoke]

Upscale by 2.66? Cant read it well

[gilded yew]

this one is image to image than upscale

[ornate elk]

I see your using an pony (sdxl) model.
These are slower than 1.5 based models you used in auto1111

[gilded yew]

by bad i was 2 orginelly

last system crash and i just quit

[gilded yew]

automatic 1111 just say out of vram and stop

[ornate elk]

Would still recommend forge webui for you. As its faster than Comfyui with your GPU

[ornate elk]

Mostly because you didnt set hires steps to 10 and didn't added the tiled VAE extension

Both needed for upscaling sdxl with 6gb vram

[gilded yew]

but model Henmixreal i first start on also gives same error now

i have 8gb ram + 12+12 pegging ram from ssd is that causing the issue?

[ornate elk]

you alsoneed --medvram --xformers --no-half-vae in the webui-user.bat and nothing more

[ornate elk]

yes for sdxl you need 16gb ram and/or increase your windows pagefile to 24gb

increasing pagefile should work for you

[gilded yew]

i am using --medvram --xformers in automatic1111

[gilded yew]

let me give that a try

[gilded yew]

is it for comyui ?

[ornate elk]

then you need to increase your windows pagefile

[ornate elk]

no thats for auto1111

[gilded yew]

is that loading time normal ?

[brazen bramble]

Hi this might be a dumb question but I wanted to try out illustrious XL and was wondering do I need to downalod another vae or is the one I use for Pony going to be alright ?

[gilded yew]

6min to generate this image i don't think that normal

[lucid yoke]

i use the base SDXL vae and its working fine 👍

[ornate elk]

Is your webui updated?

[gilded yew]

how i check that

[brazen bramble]

Good to know thank you

[ornate elk]

send me the full cmd log

then i check it

[gilded yew]

like copy past whole thing here

venv "F:\ai11\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.10.1
Commit hash: 82a973c04367123ae98bd9abdf80d9eda9b910e2
Launching Web UI with arguments: --medvram --xformers --no-half-vae
CHv1.8.13: Get Custom Model Folder
ControlNet preprocessor location: F:\ai11\stable-diffusion-webui\extensions\sd-webui-controlnet\annotator\downloads
2025-01-18 21:40:21,377 - ControlNet - INFO - ControlNet v1.1.455
Loading weights [4c1e096b94] from F:\ai11\stable-diffusion-webui\models\Stable-diffusion\bigLust_v16.safetensors
CHv1.8.13: Set Proxy:
2025-01-18 21:40:22,472 - ControlNet - INFO - ControlNet UI callback registered.
Running on local URL: ** i delete that**

To create a public link, set share=True in launch().
Startup time: 79.4s (prepare environment: 34.0s, import torch: 17.3s, import gradio: 7.0s, setup paths: 2.4s, initialize shared: 0.9s, other imports: 2.9s, list SD models: 0.6s, load scripts: 6.3s, initialize extra networks: 0.2s, create ui: 1.9s, gradio launch: 6.1s, add APIs: 0.2s).
Creating model from config: F:\ai11\stable-diffusion-webui\repositories\generative-models\configs\inference\sd_xl_base.yaml
F:\ai11\stable-diffusion-webui\venv\lib\site-packages\huggingface_hub\file_download.py:795: FutureWarning: resume_download is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use force_download=True.
warnings.warn(
Applying attention optimization: xformers... done.
Model loaded in 214.7s (load weights from disk: 8.7s, create model: 1.9s, apply weights to model: 193.0s, apply half(): 0.5s, apply dtype to VAE: 4.2s, hijack: 0.3s, load textual inversion embeddings: 1.2s, calculate empty prompt: 4.6s).
100%|██████████████████████████████████████████████████████████████████████████████████| 11/11 [02:36<00:00, 14.24s/it]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 11/11 [02:42<00:00, 14.76s/it]
Reusing loaded model bigLust_v16.safetensors [4c1e096b94] to load v1-5-pruned-emaonly.safetensors [6ce0161689]2.44s/it]

[ornate elk]

looks normal and its the latest version
also the IP is a local IP so only accesable from your home network. its the same ip for everyone.
What size does the model (checkpoint) you used have?

[gilded yew]

biglust is 6.7gbs but try to load Henmixreal 3gb but it just keep loading

[ornate elk]

1.5 based models are 2-4gb and work faster on your gpu than SDXL/Pony/Illust models (these are 6gb)

[ornate elk]

then maybe the model is broken

redownload it or try an other

[gilded yew]

i am on it

[gilded yew]

i will be damm thats it was model whole time

soo for me stick to 1.5 base models?

what other models can i use?

[ornate elk]

Or use forge. As its faster with sdxl than auto1111

[gilded yew]

i will try that to night thanks man.

[keen falcon]

I have a W11 computer with an AMD GPU and have tried following tutorials on how to get Stable Diffusion running with Automatic1111 but those I've found are old and fail quickly during the process. I should add that I'm not computer savvy, I'm borderline computer illiterate when it comes to this command prompt/powershell type of installations and configurations. Does anyone know of an easy to follow step by step tutorial on how to get Stable Diffusion running with Automatic1111 on a W11 AMD GUP? Because the amount of times I've tried to get it working is in the double digits, and I've even resorted to try and get help from ChatGPT to do it and it failed.

[gilded yew]

you deserve a kiss on your hand 😘

@ornate elk

[cyan mountain]

i downloaded hunyuan ai in comfui as a gguf. how do i add like audio synced to the video within comfyui ?

[ornate elk]

Hey try follow my AMD Zluda guides.
https://github.com/CS1o/Stable-Diffusion-Info
If you have any questions feel free to ask me

[loud pond]

@ornate elk whats better more hires steps and a lower sample steps, or less hires steps and more sample steps?

[neon orchid]

@ornate elk i have a problem

o well, my friend have a problem

he try to run "Webui-user.bat"

and show this message

[ornate elk]

Hey, hee need to reinstall python 3.10.11 64bit and check "add python to path"

If he dont get the option then he had to relaunch the installer and click modify, next, check "add to system environment variable"

[neon orchid]

How strange, I have been guiding him and he did all the steps to the letter

and the part "Add phyton to path"

[ornate elk]

Then do the second step I mentioned

[hallow latch]

I having problem with dynamic prompts.. is there any other extenstion that is better or working?

[ornate elk]

Wildcards is nice

[hallow latch]

But it does not show up a text when using it right? like .. it's fills in automatic

[ornate elk]

In the metadata you see the tags

[hallow latch]

Do I have to go in there and look it up every time?

[ornate elk]

Yep if you want to know which word got used

[hallow latch]

I know with dynamic prompts when I typed in __X .. it showed up what options there was :/ I miss that

[lucid yoke]

You mean autocomplete from a wildcard?

Like if you say "character" it adds other tags based on that?

[hallow latch]

No like .. When I use __XX .. It gives me options but I checked my Tag Autocomplete and dunno .. I did nothing but restart UI and it works.

It had a option for this

but it was alreadya ctive

[neon orchid]

We managed to fix it, thanks 🙂

Even though I have a friend who has a GTX 1070 / 4GB VRAM and wants to know if he can install Stable Diffusion

Is it possible?

[ornate elk]

Sure but he should install Forge Webui

[neon orchid]

ohhh kk

Zluda no, right?

[ornate elk]

Nope zluda is for AMD

[neon orchid]

ohhh kk

ty CS1o 😄

[hot thorn]

Hi. Can someone help me instal local stable diffusion?
I formatted my PC and my old files dont work, I did some "work" on it after the format, using chat gpt for help, but I think it's even more messed up now, and I'd like to make a clean install.
Is there anyone who can walk me through and help me fix thing as we go?

[ornate elk]

Hey you can follow the install guide on the first link of the pinned messages here.

[hot thorn]

oh i alreadyd di that, but it doesnt work for me

[ornate elk]

Where do you stuck?

[hot thorn]

at this point im even having chat gpt walk me through

when its all downloaded, and the IP opens up in browser, it just severs the connection, and i cant generate anything

[ornate elk]

What does the cmd shows then?

[hot thorn]

ive run few different versions, one from before format, and some new ones, all of them were slightly different and I just deleted them all, so I cant say what they showed

[ornate elk]

What's your GPU?

[hot thorn]

gtx 1070

[ornate elk]

Okay then you should try out setting Forge Webui up

[hot thorn]

i want to use reforge specifically

[ornate elk]

Should work too

Its also listed in the guide under forge

[hot thorn]

except both of them didnt work for me after format xD

and it was completely fine before

[ornate elk]

Can you try again setting it up?

Can help better if I know what's wrong and get some information

[hot thorn]

could you get on voice with me? so i could stream? i installed it myself several times now so im somehow well versed in it, just need to make sure i do everything correctly

[ornate elk]

Nope I'm on mobile rn

[hot thorn]

damn, ive been at it for few hours now, and no matter what I do i cant get it to work.
is there some specific software im missing? maybe something i didnt install after format?
i got gpu drivers, cuda, Git, python visual studio and c++, is there anything else?

[ornate elk]

You can make screenshots if you want to show me the steps

[hot thorn]

sure, thats an option

[ornate elk]

Python 3.10.11 64 bit is preferred

[hot thorn]

what was the line to check python version?

[ornate elk]

Python --version

[hot thorn]

3.10.0

[ornate elk]

Upgrade that to 3.10.11

[hot thorn]

i believe 3,10.0 is the version i used before. but sure.

[ornate elk]

I don't think so as every webui requires at least 3.10.6

Going lower or to high causes incompatibility

[hot thorn]

any other software outside of what I have that i need?

[ornate elk]

Nope

[hot thorn]

right, walk me through, maybe you shall fix the error of my wars

ways*

[real plinth]

Hello!

I’m experiencing issues with my Stable Diffusion setup, and I would greatly appreciate your help. Here's what's happening:

Initially, the program works fine, but after 30–60 minutes, it starts slowing down significantly.
My VRAM usage stays maxed out during image generation.
I’m using an RTX 3070, Ryzen 5 3600, and 64GB RAM.
Settings:
I start the user-bat with these arguments:
--xformers --medvram --no-half-vae

Recent Changes:
I recently upgraded my RAM to 64GB. While the default speed is 2400MHz, I’ve overclocked it to 3600MHz, which matches its rated speed.
Current Versions:
Stable Diffusion version: v1.10.1
Python: 3.10.6
Torch: 2.1.2+cu121
Xformers: 0.0.23.post1
Gradio: 3.41.2
Checkpoint: 821aa5537f
Yesterday, the performance was better, but something seems to have changed. Could the issue be related to my RAM settings, GPU, or perhaps something else?

Any insights or suggestions on how to fix this and improve performance would be greatly appreciated!

Thank you in advance!

[ornate elk]

Hey, change --medvram to --medvram-sdxl and try again.
Its needed for cards with less than 12gb of vram and sdxl/pony models

It should speed it up

[hot thorn]

also what does this mean? wwhere in file explorer? I believe last time i did this i used git, and did some cd /c/ai kind of thing

[ornate elk]

Yea you can do it both ways

The path bar is the top bar in the file explorer

When your in a folder you can just Click into the path bar, type cmd ans hit enter

So you dont need to use CD command

[hot thorn]

hm, didnt know that tbh, always used the start>cmd

anyway, doing the thing

and its finished

the --xformers --no-half-vae line, do i need space in between -- and =?

or just set COMMANDLINE_ARGS=--xformers --no-half-vae

[ornate elk]

Doesn't matter

[hot thorn]

k, i noticed inside the webui user, i see the line @REM set A1111_HOME=Your A1111 checkout dir, is it fine? since im not using automatic 1111

[ornate elk]

Yep that's for linking model directorys from Auto1111

[hot thorn]

and since my gpu is only 8gb, i should also use the --medvram-sdxl ye?

[ornate elk]

Reforge doesn't support that

[hot thorn]

so no need? got it

k, launching webui user, if it breaks again gonna let u know

and if it magically works this time, then i suppose it was the python version

[real plinth]

Hey! Thank you for your quick response! I've just tried to change the text and runed it again.
The specs are higher than before, but I'm seeing that the RAM and the VRAM have not changed, is there any way to make stable use more of my RAM for better and quicker performance, yesterday I've managed to generate one really high quality image with better specs and today the stable feels off.

[hot thorn]

@ornate elk ye, it went to poopoo

[ornate elk]

Connection error.
Bad WiFi, VPN, proxy, DNS can be the cause

[hot thorn]

my wifi is good, i dont use vpn i believe, not sure about others, as i mentioned, i just formatted this pc today, only got basic drivers etc

[ornate elk]

Is your Router near you?

[hot thorn]

its in a room next to me, im using a wifi usb

till this point, nothing happens

its getting everything just fine

[ornate elk]

Hey no problem. I see no issue here. It doesn't matter if you have 32gb or 64gb RAM when you use sdxl/pony models. They aren't huge and dont need so much.
The important part is the GPU vram.
As your card only has 8gb it will be slow when you generate at large resolutions. Better is to generate in lower resolution and use hires fix to upscale.
You can load the image you genned quicker yesterday, into the PNG info tab to see the settings you used.
Also make sure you dont run any Games or Wallpaper engine in the Background.
Also if you want faster image gen you can try out Forge Webui.
It has some tweaks for cards with less vram.

[ornate elk]

Now it looks better

[hot thorn]

why is the webui user suddenly downloading noob ai xl?

because i got nothing in models?>

[ornate elk]

Yep

[hot thorn]

ehh, do i just cancel this one, and put the one i want in? or is that gonna mess it up

[ornate elk]

You can close the cmd and put something in

Then relaunch
Folder is Models/stable-diffusion

And the file needs to be 2gb or larger

[hot thorn]

are those errors fine?

[ornate elk]

Yep

[hot thorn]

and it pooped

[ornate elk]

How much ram do you have?

[hot thorn]

same thing happened to me past few hours when i tried different things

16gb

[ornate elk]

Okay then you have to make your Windows Pagefile bigger

[hot thorn]

how so

[ornate elk]

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

Make sure its only enabled for the C drive and disabled for any other drive.
Then set it to custom for C.
Set it to 16000Min and 24000Max

Then save and reboot the PC

[hot thorn]

I have AI on D, is this setting specific to my system or to the disk with SD on it?

[ornate elk]

C should be preferred as its mostly an SSD

Doesn't matter where the ai is located

[hot thorn]

16000 min and 24000max for sure? ive never done this one so im not well verse in this

[real plinth]

Thank you very much! I forgot to turn off the wallpapers, I will try forge and if the problem with the slowing after some time come again I will let you know! Thanks again for your help and wish you a smooth and productive workday! 😊

[ornate elk]

No problem. Yea Wallpaper engine can slow down a lot and even crash the PC when used together with Stable diffusion. Even when WP is set to "paused" so better exit it completely for the gen time.

[real plinth]

That was out of my mind, I'm kinda new to SD and I've got 4K moving wallpaper along with one more 2K moving wallpaper, I'm assuming that killed the procces, you are savior! Thanks again!

[ornate elk]

Np, that definitly slowed down. And yea let me know if you tried Forge or still have some questions

[hot thorn]

@ornate elk ok, ive confirmed, it now works. Now im gonna try my older one, to see if the change in virtual memory fixed it as well

but so far its very big for me, thanks a lot mate

[ornate elk]

Np, if the other one doesn't work, you have to delete its venv folder.
So it can rebuild it with the updated python.

[hot thorn]

also

its ded

i was experimenting with many "fixes" on this old version and messed it up i think, do u think its savable? @ornate elk

[ornate elk]

How old is it?

Is it forge before the big changes?

I would say use reforge If you want the old forge stuff.
Or use Forge If you want the new stuff

Wouldn't rely on an old broken version

[hot thorn]

its also reforge, think i installed it about 2-3 weeks ago

is the one i just made gonna be fine as long as i just copy over all my lora/embeddings/vae?

[ornate elk]

Yep

[ornate elk]

It says forge in its name and not reforge

So it was the original forge

[hot thorn]

got it from here https://github.com/Panchovix/stable-diffusion-webui-reForge

the forge one i tried as well was https://github.com/lllyasviel/stable-diffusion-webui-forge?tab=readme-ov-file this one

but i didnt like the UI and some options

[ornate elk]

Ah okay

[hot thorn]

either way, big BIG thanks man. i was struggling with this for way too long

[ornate elk]

Np 🙂 and gn im off

[hot thorn]

someone ban this guy ^

stenmcommunity.com xD

[hot thorn]

in case youre not asleep yet. anything i can do to make generation a bit faster? before the format it took me about 4-5 minutes to make one pic with my settings, now its about 5-6 minutes

my settings

[loud pond]

whats your gpu?

[hot thorn]

gtx 1070

[broken epoch]

Hello guys I am trying to use https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/API but not work ın my project I got error can anyone help me?My terminal output like
✓ Compiled /my-followed-stores in 288ms
GET /my-followed-stores 200 in 328ms
✓ Compiled /api/generate-image in 324ms
Received request body: {
prompt: 'GamerGear, modern storefront design with professional appearance, professional product photography, high quality, detailed, 8k resolution, product showcase',
negative_prompt: 'blur, noise, distortion, watermark, text, low quality',
width: 512,
height: 512,
steps: 30,
cfg_scale: 7,
sampler_name: 'Euler a'
}
Sending payload to AUTOMATIC1111: {
prompt: 'GamerGear, modern storefront design with professional appearance, professional product photography, high quality, detailed, 8k resolution, product showcase',
negative_prompt: 'blur, noise, distortion, watermark, text, low quality',
steps: 30,
cfg_scale: 7,
width: 512,
height: 512,
sampler_name: 'Euler a',
batch_size: 1,
n_iter: 1,
seed: -1,
restore_faces: false,
tiling: false,
enable_hr: false,
denoising_strength: 0.7
}
Image generation error: {
message: 'Request failed with status code 404',
response: {
status: 404,
data: { detail: 'Not Found' },
url: 'http://127.0.0.1:7860/sdapi/v1/txt2img'
}
}
POST /api/generate-image 503 in 488ms
and google chrome F12/console error is Generation failed: "Request failed with status code 503" {}

[final vector]

Hello, I am getting the following error:
https://pastebin.com/wCH658Ne

This occurs when I upscale with the following settings:

[paper ridge]

What's the most recommended general purpose model that can be used for commercial use?

I've mostly experimented with DreamShaper XL Lightning, but it seems hard to get it to really follow detailed prompt instructions.

Would really appreciate any recommendations 🙂

[hot thorn]

can anyone help me out? my wildcards seem to not work. I have the extension installed, I have dynamic prompts enabled ticked and it just gets ignored...
is it possible that if too many lines exist in a wildcard it just prevents all of them from working?

[ornate elk]

try use this vae for sdxl/pony:
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/blob/main/sdxl.vae.safetensors
Then idk if reforge has the never oom setting at the bottom of txt2img but if not get the tiled vae extension, that imrpoves the speed too.
and you could try Forge to compare if its faster than reforge.

[broken epoch]

I solved the problem but its working too slow how to fast it?

[grim totem]

Is it possible to find out the seed if the image has been edited in photoshop and pnginfo can no longer provide generation metadata?

[lucid yoke]

no

unless you find the original image with metadata

no

[keen falcon]

Thanks, I got lost somewhere in the middle jumping between the awful Github pages and I don't know if I downloaded the right version of something and then it said I should skip it if my GPU is newer than 6800 or something... I probably need to wipe my system again and try to start over.

But I need to do that tomorrow, as that was all my neurospicy brain had in it for today.

[broken epoch]

Hello, I am using the stable diffusion model for an e-commerce website project. There are approximately 20,000 products coming from the backend on the web and I managed to run the model, but due to my computer features (1650ti/4gb vram) it is not running very fast. What can I do to speed it up? Will concurrent methods, threads etc. work?

[lucid yoke]

if you need 100x100 and 200x200 images why not downscale the existing products? sounds like you need a professional

or a machine upgrade. getting fast results on 4gb vram is not really a thing

[reef heron]

Thanks. I'm doing okay with LM Studio now. May try Zluda too then. Only things with LMS is I had to unplug my 1G Nvidia that I use for my 2nd screen as it was defaulting to use that instead of my AMD 6700XT. Got that working now though and its pretty fast up to about 7B models.

[loud pond]

Anyone know how i can install a good image to video local generator?

[ornate elk]

Yep LM-Studio and ollama work good with AMD

[ornate elk]

Yep start over and read it carefully.
And feel free to ask me if any step isnt clear.

[hot thorn]

u mean these? do i use both of these options?

[ornate elk]

Only for tiled vae

[hot thorn]

@ornate elk any idea why dynamic prompts might not work for me? I have 2 different ones in my prompt, like clothing and background and they just get ignored

[final vector]

Does anyone know the cause of this error...?

[ornate elk]

Nope sdnext does everything different then other webuis so no clue

[ornate elk]

Wildcards?

[hot thorn]

ye

[ornate elk]

Idk, I dont use dynamic prompts.
I only use wildcards

And they work

Check the cmd for any errors

Also make sure dynamic prompts is enabled

[hot thorn]

i did, it kind of bricks when i used combinatorial generation.
is there a possibility that too many lines of text in 1 wildcard can cause issues?

because now, when im doing them without combinatorial generation it seems fine

it worked fine before format, but at night I added about 12 new parameters, which is about 30 lines of text

[lime adder]

Hello, I installed stable diffusion and everything works, however, when I try to generate a picture, it generates it too slowly, I open the task manager and see that it does not open the video at all, only part of the RAM but not all
I have a 1060 6GB video card and 32GB RAM. i also run the .bat file with
set COMMANDLINE_ARGS= --xformers --medvram --no-half-vae

[final vector]

Ty I posted in there as well, so I will focus my inquiries there from now on.

[lucid yoke]

@analog tinselwhats going on bruh

could be a child accidentally ruining their parents pc seeing the reaction lol

all id waste is my time (and im bored)

he added me to a group chat. mans mental lmao

tdlr: he wants to get banned from here asap

[solemn atlas]

Hello, does someone know why i cant generate anything?😅

[ornate elk]

Something wrong with the lora

[lime adder]

.

[ornate elk]

Video?

What are you trying to do exactly?

[lucid yoke]

vram is being used

[lime adder]

Iam trying to generate img 512x512

[ornate elk]

Vram is the important part

For your card using forge webui would be faster

[lime adder]

ok thanks

[cursive cobalt]

Automatic 1111. Missing Lora's in the "Lora" Tab.
Its pulling from my ComfyUI directory for Lora's. Comfy recognizes the Lora's, and displays them, Auto does not.
I have googled, but found the most common "Solution" was to ensure the 2 (scrnshot) was enabled. It does not fix the issue.

[ornate elk]

whats in your webui-user.bat?

[cursive cobalt]

But as mentioned, it reads 95% of the files correctly. Even as new as from today.

Which is where the "Visible Confusion" comes in

And you're probably wondering: If he uses ComfyUI; why is he playing with Auto1111 then?
Cause its a helluva lot easier, generating previews via Auto, than it is via Comfy.

[ornate elk]

and the missing loras are compatible with the selected model?

[cursive cobalt]

Ye

[ornate elk]

because they wont show up otherwise

[cursive cobalt]

They are fully compatible, and (surely?) should still show if the "Show incompatible" is ticked on, ye?

Anyhoo, to answer plainly, yes. Verified via ComfyUI.

[ornate elk]

have the missing loras something in common?

for a specific model type maybe?

for example auto1111 doesnt support flux or sd 3.5, so they wont show up

[cursive cobalt]

They're all (Including those showing) for Illustrious.

[ornate elk]

hmm okay

wha you could do is to move all loras to auto1111 and use the comfyui model path yaml to link them to comfyui

[cursive cobalt]

Hmm.... That's not gonna be an option I'm afraid. I dare not touch anything, now that I've got some large workflows working.

And finally got an Inpainter working too

[ornate elk]

you can right click and disable the node

[cursive cobalt]

CTRL + B, Bypasses a node.
CTRL + M, Mutes a node.

[low crescent]

does someone know why i keep getting this ? everytime i generate something it comes out like this

and it is really slow, I used to have SB before I reinstalled it but when I had it before I could just generate images and at a reasonable speed

[ornate elk]

can you show your settings and model you used?

[low crescent]

its a model that i have used in the past aswel https://civitai.com/models/833294/noobai-xl-nai-xl

[low crescent]

what kind of settings would you like me to show you ?

[ornate elk]

thats a v-pred model and wont work in auto1111 as noted in the description. You need to use Forge webui or reforge

but its a bad model anyway so just get a better illustrous model and you can use it in auto

[low crescent]

Well il try, I probally got somethings messed up in my head then haha i thought that this was the model i used in the past aswel, but thx 🙂

[ornate elk]

for example this:
https://civitai.com/models/376130/nova-anime-xl

[low crescent]

thx il try it

[low crescent]

I get something now but it comes out like this haha

[ornate elk]

which vae do you used?

[low crescent]

https://civitai.com/models/276082/vae-ft-mse-840000-ema-pruned-or-840000-or-840k-sd15-vae

[ornate elk]

this one is for 1.5 based models

for sdxl/pony/illust you need an sdxl one

like this: (direct download link)
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/blob/main/sdxl.vae.safetensors

[low crescent]

oh god, I feel so stupid

[broken epoch]

How can I speed up stable diffusion and can I save the images that generated ones?

[carmine ibex]

Using controlnet, I get this. How can I reduce so much noise in the background (ignore the character, I don't care about that). When I try to do img2img with a simple anime style, it adds so much extra stuff to the background because of that

[ornate elk]

Whats your GPU? And which webui do you use?

[ornate elk]

Lower the controlnet weight

Or set it to "prompt is more important"

[broken epoch]

Stable diffusıon DPM++ 2M my gpu is gtx 1650 ti 4ram

[carmine ibex]

i use comfy so I cant do prompt is more important. But ill try with weight

[ornate elk]

With that GPU you should try forge webui. It has some more optimisations for lower end GPUs

[broken epoch]

I use this https://github.com/AUTOMATIC1111/stable-diffusion-webui

Its okey if optimization is bad,main point I want to save the generated images for next run

Can I do that?

[ornate elk]

The images are saved automaticaly to the outputs folder

You can drag and drop then into the prompt field to get the settings back

If you press the blue arrow on the side it will auto apply them

You could also drag an image into the PNG-Info tab and there press "send to txt2img"

[broken epoch]

Can you dm me?

[ornate elk]

Why?

[broken epoch]

I want to ask you the details

[austere adder]

hey hey! 🙂 how doya'll get controlnet working in forge - deforum?

[gloomy bolt]

The program is tested to work with torch 2.3.1. To reinstall the desired version, run with commandline flag --reinstall-torch. Beware that this will cause a lot of large files to be downloaded, as well as there are reports of issues with training tab on the latest version.

do i just open a command prompt and type --reinstall-torch ?

or do i have to do it at the folder location of the webui?

oh ok, where i put the stuff like -- theme dark -- or -- half vae --?

awesome thxs

[gloomy bolt]

looks like it doesn't work?

oh wrong line actually lemme try again

ok it works now

sry bout the ping

[vocal burrow]

forgot a ^ at the second last line to concatenate the lines together

[civic nova]

Hi had stable diffusion for a long time now. Got a new PC and against my direct orders 3Xs I told them, they still reformatted my F drive even though it did not need it. As such I have lost all files and folders related to and for Stable Diffusion. I am not a programmer and need help in baby steps please on how to totally reinstall Stable diffusion, GIT and Python back onto a now totally empty F Drive. Any assistance is most welcomed.

[vocal burrow]

check the pinned messages for guides.

[civic nova]

ok right now I am just still in a state of shock. I lost ALL of my NFT Work due t that for them to then tell me, But we improved it by installing windows 11 when I specifically told then DO NO do this and this. Then was told, well it is not our fault. %$/&§/&%$(/(=)(/&=&%)/(&%$%$ sorry just venting. Fortunately I had all my NFT work and art work backed up but not the original programs which I am now trying to reinstall form scratch. Will check the pinned stuff yet as stated I know nadda about programming. Thanks for the time and patience.

[vocal burrow]

no programming is required, just reading and patience.

[fair oxide]

When I select a checkpoint it automatically unselect it, I tried different models but the problem persists, is it because of my hardware or did I messed up something?

[ornate elk]

What's your GPU? Which webui do you use?

[fair oxide]

1660 super (but it did worked in the past with the same model), I'm using Anapnoe UI-UX Core

[midnight roost]

@ornate elk i am thinking ckpt might have memory leak with my automaticc1111? because if i used a safetensors theres no issues if i use x y z plot checkpoint, also what i forgot to mention is its meged checkpoints so i use ckpt format to merge checkpoints models which it wont let me merge as safetensors format

[ornate elk]

You should try forge uiux

Its better for your GPU

Also if you only have 16gb of ram you need to increase the windows pagefile

[civic nova]

ok got some side help, got it all installed and ran the Webui bat file and I get the following once it pulls up in the browser and I type in any word commands... "Connection Error Timed Out"

[ornate elk]

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

[civic nova]

it sends me to http://127.0.0.1:7860/

[ornate elk]

and how much RAM do you have

[civic nova]

GPU is a Nividia GTX 1080ti 11gb ram

96gb DDRAM

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --no-half-vae

call webui.bat

[ornate elk]

add --medvram-sdxl
then also make sure you whitelisted the webui url in any browser adblocker and dont use opera vpn etc

[civic nova]

ok for a total 65 year old novice to programming where do I add that please?

ACH wait that may be it, using strong vpn

nope vpn was not running

so where do I add the --medvram-sdxl

set COMMANDLINE_ARGS=--xformers --no-half-vae--medvram-sdxl ;;as such?

[ornate elk]

seperated with a space

[civic nova]

et COMMANDLINE_ARGS=--xformers --no-half-vae --medvram-sdxl

ok that add on did the trick thanks

[ornate elk]

no problem

its needed for gpus with less than 12gb vram

to run sdxl based models better

[civic nova]

eventually the plan is to upgrade that old GPU but with it being a totally new 14th Gen 10TB Ethernet ready system just could not afford it on my disability pay right now. Thanks again.

SO Once i put in the new GPU later on when I can afford it I remove that line addition, just to be sure?

[ornate elk]

yep you can then remove it and your good.

[civic nova]

copy that thanks and out

[loud pond]

@ornate elk just asking but would you have any idea on how to install minimax locally?

not the text to video one

the llm

this thing https://github.com/MiniMax-AI/MiniMax-01

[fair oxide]

this did the trick! thank you

[ornate elk]

you have to wait until a ui for llms supports it.
like LM-Studio or ollama

[loud pond]

oh so i cant download it for my pc yet?

[ornate elk]

if you know how to code and compile stuff you can run it now but i guess you better wait a bit

[loud pond]

oh ok

thank you very much

[ornate elk]

np

[keen falcon]

So I should skip the following if I have a AMD Radeon RX 6650 XT

[keen falcon]

As for this step, I should only download it, then what should I do with it?

[ornate elk]

Yep go one Step further

[ornate elk]

Comparing if the file is available for your GPUs gfx version. Then download it

[keen falcon]

Yes, I did and I've downloaded it, but then it doesn't say what to do with that download.

Instead it moved on to say I should download the AMD HIP and install that

[ornate elk]

Yea follow the guide

It will be mentioned

[keen falcon]

So it's not in chronological order?

[ornate elk]

It is

Afk, driving rn

[keen falcon]

When you have time, is this edit, the correct interpretation of the instructions?

[ornate elk]

Yes

[languid granite]

Hey, do you know what dynamic thresholding does

[lucid yoke]

https://github.com/mcmonkeyprojects/sd-dynamic-thresholding

it was litterally one google search man

[languid granite]

Thank u

[steady mountain]

Hey, anyone know why my ultralytics does not work?

grateful for any feedback from anyone who has encountered this and solved it. (no, in my case recreating the node does not fix the problem)

[lucid yoke]

Ah it looks like your missing a model

Null = none

[onyx solar]

You have no model chosen.

[iron lantern]

How can I get a wide shot .. it's allways giving me close ups

[slate ledge]

Trying to use flux in stable diffusion webui forge. Error - AssertionError: You do not have CLIP state dict!

[ornate elk]

which model do you try to use?

[crystal nacelle]

My Stable Diffusion A1111 web UI launcher has been like this for like 20min now. can i close it and try again later?

ah nvm, it justfinished

[wicked widget]

yeah amd compiling is gonna take a long time on first go

its normal

[crystal nacelle]

what do you mean "first go"? I've been using stable diffusion for like a year now 😅 . Is there like an update that i pulled off the git?

[wicked widget]

last time i used amd which was 2 yrs ago i needed to compile everytime so i just assumed. wasnt worth it so i upgraded to nvidia 🙂

[paper drum]

Still stuck

[wicked widget]

wdym

[brave pier]

3 day trial I signed up for $99 a month

I wanted to cancel but shows me no where to cancel help please

[round hedge]

HI, I'M NEW HERE. WHO CAN EXPLAIN TO ME HOW I CREATE THE IMAGES HERE? PLEASE HELP ME.

[crystal nacelle]

nah its not everytime now. its like once every few months i think its related to git pull updates

[onyx solar]

You don't unless you've paid for the service.

[crystal nacelle]

what web UI are you using? Most of the UIs have a big space for you to type stuff then a big orange button called GENERATE to make an image

[crystal nacelle]

stable diffusion is free bruv. you dont need to pay

[onyx solar]

Bruv, there is a paid service on this server that can be utilized and the person's question was how to "CREATE THE IMAGES HERE".

[wicked widget]

check the pins

[brave pier]

No one here knows about 3 day trial and a pain to cancel?

[wicked widget]

this is a community server for the local usage of stable diffusion

check the site if theres a contact us or w.e for that specific issue cos honestly never heard of that

[ornate elk]

Yes it was an update for zluda

Zluda, torch and GPU driver updates can cause a recompile

[ornate elk]

Where do you stuck?
Have you followed the AMD Guide?

[steady mountain]

yes. This is what is strange, it should load bbox/ yolov models. And as far as I understand they should have downloaded automaticaly with the impact pack

[steady mountain]

they, thanks for your reply too. it is becausre there are none

[steady mountain]

Good people. I think I have solved it. I was a dumbass

[acoustic vortex]

https://huggingface.co/docs/diffusers/en/using-diffusers/merge_loras
Trying to figure out how follow this guide. I simply don't know what to do with this line of code? lol

[karmic minnow]

good day, I need some help, my SD forge is taking too much time to generate one image, to the point that it takes 15+ minutes or even my computer crash, that didn't happen a few days ago, but since I was trying to use inpaint, that happened, idk if I need to reinstall SD

[pale salmon]

sup, i can add more MB ?

[ornate elk]

What's your GPU?

[karmic minnow]

rx6600 8gb

[ornate elk]

Okay and when you try to generate an image? How long does it take?

[ornate elk]

What's your GPU and how much ram?

[karmic minnow]

more than 10 minutos

[ornate elk]

Can you show me the full cmd log?

[pale salmon]

RTX 4080 SUPER Corsair Vengeance RGB DDR5 2x16GB 6000C36

[ornate elk]

and what did you tried when you got the error?

[pale salmon]

tryed to upscale a picture by 2,5 i think

with hires fix

[ornate elk]

set the hires steps to 10 and try again

and set upscale by to 2

[pale salmon]

🙏 ok

btw i would like to fix this error

while using canny

[karmic minnow]

I was trying to take the cmd log, but my PC froze after generating

This is what I got

This what I got at the moment

I need to hard shut down it

My GPU is using 7164MB

I didn't want to change it cuz idk if more ram is more generation energy

[ornate elk]

That means the canny model and the checkpoint is not compatible

[pale salmon]

okay thx

[ornate elk]

How much RAM do you have?

[karmic minnow]

16gb

[ornate elk]

Then you need to increase the windoes pagefile

[karmic minnow]

where can I do that?

[ornate elk]

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

Make sure its only enabled for the C drive and not for any other drive.
Set it to custom.
16000min and 24000max

Then reboot the PC

[karmic minnow]

C drive only?

what if I have SD in another drive

[ornate elk]

doesnt matter

C is mostly the fastes drive you have

so better for the pagefile

[loud pond]

@ornate elk im still getting wonky face's and random generations with stable diffusion what checkpoint for pony would you reccomend for good images, (Note - ive also noticed that if i use less ( ), the images tend to be ALOT better)

[karmic minnow]

ok got it, let me do it and I'll let you know

[ornate elk]

i like:
novaAnimeXL_v70Happyhalloween
prefectPonyXL_v50
mistoonAnime_ponyAlpha

[loud pond]

alright cool ill try perfectpony then

[midnight roost]

if i wanna try the custom paging files would i just double the numbers 16000min and 24000max with 32 gb ram? its ddr5 if that matters

[ornate elk]

I would only double the 24000 but with 32gb you dont need the increase

[loud pond]

do i have to use the scores for prefectpony?

like score9, score8_up etc...

[ornate elk]

Yes

[wicked widget]

illustrious miaomiao is good too

[karmic minnow]

ok I done it, now what do I need to do?

[karmic minnow]

this is also something I see when I start generating

[ornate elk]

Which settings do you use?

[ornate elk]

Have you restarted the PC too?

[karmic minnow]

yes

[karmic minnow]

wydm?

this?

[ornate elk]

Looks normal

[karmic minnow]

yeah, looks like few hours ago, I tried once again, It was good now, taking like less than 7 minutes

sometimes taking 3 minutes, sometime 5 minutes

that depends idk why

[ornate elk]

Resolution, lora amount, prompt word amount, upscaling. Can all change the time

And model of course

[karmic minnow]

true

but anyway, it is good now, thanks

https://tenor.com/view/limbus-company-yi-sang-limbus-company-yi-sang-pou-yi-sang-pou-gif-8873119486740820408

[karmic minnow]

I think

It's still the same

This is the last thing I got from the cmd

Idk if it is my SD or my AMD drivers

[ornate elk]

Its SD

Which model did you used?

Also make sure you dont have any other programms open like Wallpaper engine or Games

[karmic minnow]

i actually close wallpaper engine if I use SD

[zenith seal]

hi

[karmic minnow]

i think is this one: https://civitai.com/models/288584?modelVersionId=379262

[zenith seal]

I have a prombem my ui don't work anymore

this morning when lauching the ui,a py torchmodel as been installed

no sure if this can cause the issue

C:\Stable Diffusion\SD-Zluda\stable-diffusion-webui-amdgpu\modules\safe.py:156: FutureWarning: You are using torch.load with weights_only=False (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for weights_only will be flipped to True. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via torch.serialization.add_safe_globals. We recommend you start setting weights_only=True for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.

now this

[ornate elk]

Okay its important you have Wallpaper engine not even running in the background

When using SD

"Paused" won't help too

[karmic minnow]

yup, I quit it

[ornate elk]

Okay do you noe get any issues?

[sonic hound]

Folks, i am unable to find the dreamroom to start creating? Can anyone guide me pls

[ornate elk]

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

[karmic minnow]

well still the same tbh

it takes too much time, and still frozing my pc

[ornate elk]

Can you try a lower resolution?

Or try without the loras

[karmic minnow]

let me try

[zenith seal]

6800 xt

set COMMANDLINE_ARGS= --use-zluda --update-check --skip-ort --precision full --no-half --ckpt-dir "D:\Stable-diffusion"

i just added --precision full --no-half

[karmic minnow]

it made it ok, saturated since has some lora info, but no the lora

let me try with one more without loras

[zenith seal]

launching without now --precision full --no-half

[karmic minnow]

same amount of time

[zenith seal]

are you crashing at the end of the generation?

[karmic minnow]

Yeah

Actually I tried with an image, the sames loras, the same seed

And got froze as well

Tbh idk what happened there, few days ago I was generating stuff with open sharing discord, and wallpaper engine in less than 5 minutes, and now I can't

[ornate elk]

Yes pls remove --no-half --precision full

And which model do you use

When you get the error?

[ornate elk]

With the same model?

Can you share a full cmd log when relaunching the webui ?

[karmic minnow]

yup

[zenith seal]

i tried with 2 xl models

but i didn't get this issue yesterday

[ornate elk]

Try add --cuda-stream to the webui-user.bat
Then delete the .zluda folder
And relaunch the webui-user.bat

[zenith seal]

also this morning like i said,the cmd installed a new 10g torch model for some reason

[ornate elk]

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

[karmic minnow]

do I need to delete --use-zluda from my set COMMANDLINE_ARGS= ??

[ornate elk]

Nope

Its needed

[karmic minnow]

got it

should be like that right?

[zenith seal]

i already did :c

[karmic minnow]

Ok, now should I generate?

got it

gonna use same model, lora, and seed

[ornate elk]

It says torch not compiled with cuda. So something is wrong.

Can you try generate an image with an 1.5 based model and send the Cmd log?

[zenith seal]

yep

just need to install one

[ornate elk]

What you can try to is to delete the venv and the .zluda folder
Then relaunch

[karmic minnow]

Still the same

Let me check by deleting both folders

[ornate elk]

I see 3-4 models or loras mentioned here? Can you restart and just try meinamix?

Without any lora

[zenith seal]

mhh i only used meina mix

1girl

and no lora

but let me restart

[karmic minnow]

question, do I need to git pull on this SD?

I have fear that if I use that command, SD would be worse xd

[ornate elk]

Yes

Its an fork of forge for amd so it gets optimised and bug fixed

[zenith seal]

do you think i should reinstall my ui?

[karmic minnow]

Nope, still the same

[ornate elk]

Update your python to 3.10.11 64 bit then delete the venv and relaunch

If that won't fix it then a reinstall could help

[ornate elk]

Can relaunch and try a normal 1024x1024 with Euler a without any lora?

[zenith seal]

mhh i installed forge

i hope it will work

https://github.com/lshqqytiger/stable-diffusion-webui-amdgpu-forge

used this repo

[ornate elk]

Make sure you used my install guide for Forge

But as you already have set up auto1111 it should work

So if it won't work. Then redo from my guide

[zenith seal]

well the good news

it's that i don't lag anymore while genning

but it's seems to take a bit more time

maybe i can change some of the lauch option

i the .bat

[ornate elk]

You can try add --cuda-stream

Most others won't work as forge removed them

[thick owl]

hey guys

[zenith seal]

--use-zluda --update-check --skip-ort --ckpt-dir "D:\Stable-diffusion"

using this rn

[zenith seal]

is cude-stream

to get faster gens?

[mystic pumice]

I downloaded automatic 1111 and installed Dreambooth and rebooted and closed my UI and the dreambooth does not appear . It shows ticked but doesn’t appear

[cursive cobalt]

Anyone got some suggestions on how I can best clean my lora folder? xD Besides reinstalling all.
When i first began the venture into SD, i kinda just pressed "Download" and "Save", not editing the file name.
Like, is there a node in Comfy, that allows for Renaming? (Can't rename in Folder, cause its in-use by the WebUI)

[ornate elk]

Its broken and not recommended to use

[mystic pumice]

Gotcha thank you @ornate elk

[ornate elk]

Delete dreambooth from the extensions folder and then delete the venv folder.
That will fix your webui again.

Then get OneTrainer or Kohya_ss for training.
Both are seperate tools

[cursive cobalt]

Which is best for training of those 2? Sorry for barging in on the topic

[ornate elk]

I like OneTrainer more

Both do well

[cursive cobalt]

Huh. I heard of Kohya but that looked complex to install and use tbh. Got a link for the OneTrainer? (So i get the right one)

[ornate elk]

Sure:
https://github.com/Nerogar/OneTrainer

[onyx elk]

What is hiresfix?

I couldn't find any coherent enough source to say what it is

The closest I get is that it's upscale image in the extra tab and run it through img2img with the same prompt and seed

[cursive cobalt]

HiRes. fix: a simple way to upscale your images while they're being generated. Instead of first generating an image and then upscaling it through A1111's img2img or extras options, you can use HiRes. fix to increase resolution during the generation process. - Google (https://medium.com/rendernet/using-hires-fix-to-upscale-your-stable-diffusion-images-8d8e2826593e).

TL:DR it can help with details

[onyx elk]

Thank you, but can i do hires fix afterward ? I know it's during the process of generating but A1111 give me the option to hiresfix after an image is made

[cursive cobalt]

🤷 I moved on to ComfyUI tbh.

[onyx elk]

Oh, well thank you for the article link

[vocal burrow]

you can either regeneate with same parameters (+hiresfix) or simply upscale with img2img afterwards

[wild mirage]

Hi. I am having trouble with 'Use base prompt' in Regional Prompter.
When I use either Use common prompt or Use base prompt alone, it works fine.
However, when I use both options together, everything goes wrong in an instant.

I have been looking through many tutorials for a long time, but I can't figure out where the problem is.

The guide in the link below is very well explained and easy to understand.
https://civitai.com/models/339604?modelVersionId=610058
However, 'Use base prompt' just won't work.

Example prompt using both options (almost never works):

indoors, 2girl, seductive smile, looking at viewer BREAK
white t-shirt, white Skirt, long blue hair BREAK
black hoodie, jeans, short pink hair```

Example using only 'Use common prompt' (almost always works):
```score_9, score_8_up,score_7_up, source_anime, indoors, 2girl, seductive smile, looking at viewer BREAK
white t-shirt, white Skirt, long blue hair BREAK
black hoodie, jeans, short pink hair```

Some additional info:
DPM++2M, Karras, Sampling steps 30, 512*768
No lora or negative prompts used.

[ornate elk]

If you enabled common and base you have to add another BREAK

Oh I see you tried that in the first one

Dont use just score tags as the first thing. You dont know If it works

Try add something the other definitly won't generate. So you can check if it works

[wild mirage]

indoors, 2girl BREAK
white t-shirt, white Skirt, long blue hair BREAK
black hoodie, jeans, short pink hair```
Hmm... for some reason it still doesn't work.

batch count is 4.

[lavish void]

i cant find anything about AV Face detailer anywhere

i need help

how do i get this custom node

i use comfy ui Manager and it cant find AV Face detailer

[ornate elk]

Try using a wide image format 768x512

[wild mirage]

Good idea. I'll give it a try.

It's getting better, but it's still unstable.

2girl, smile  BREAK
white t-shirt, white Skirt, blue hair BREAK
black hoodie, jeans, pink hair```

I feel like I'm watching a Bob Ross drawing show.
It looks so easy, but it's a series of bewilderment.

[cursive cobalt]

Install missing nodes?

[wild mirage]

Hopefully I'm not the only one who can't do this.

[cursive cobalt]

U got the Impact Pack installed?

[cursive cobalt]

Hmm... C1so might provide better help, but could it be that its an Art Venture node u ain't got? idk tbh. Might be worth reinstalling the node? Again, idk. My knowledge stops here

[lavish void]

the Artventure stuff has it

tried to install

but thanks for help

[cursive cobalt]

See if u can do a git pull? 😄 https://github.com/sipherxyz/comfyui-art-venture

[lavish void]

i managed to install it now and restarted everything but nothing shows up

no nodes from it

as if its not installed

dispite saying it is

nvm some things are showing up

but not the face detailer AV

[tepid rose]

Hello. I have problem. I can't launch Stable Diffusion. I have already done a lot to fix it, but the program does not start, here is the log from the terminal

The system gives an error that it does not find the file when running the .bat file

[ornate elk]

Now try with only base or common prompt

[ornate elk]

Hey download and install python 3.10.11 64bit
Then delete the venv folder inside the stable-diffusion-webui folder.
Then open up a cmd and run
Pip cache purge
Then launch the webui-user.bat

[wild mirage]

Then it works fine. But I need both options to use the Mask function.

[ornate elk]

Hmmm

Will do some tests later

[wild mirage]

thx

[lavish void]

@ornate elk do you perhaps have a idea how to fix my problem?

[ornate elk]

Make sure your Comfyui is updated and the nodes too.
Then install the art venture again

Also check the cmd log

If you see any errors

[lavish void]

alr ill check

[wicked widget]

delete the custom node and install it manually instead

reboot comfy completely after

[tepid rose]

Thx.Especially for the tips on the Python version. But I still solved my problem. It turned out that the system can't find .bat files. But I fixed it.

I have another question about the program. Is it normal that sometimes the program eats up a lot of space on the system disk, being installed on another disk? I just have a system one, it's SSD and I don't want the writing cycle to end.

[ornate elk]

It downloads install files that can be deletet afterwards.
Open up a cmd and run
Pip cache purge
To get 5-10gb back

[patent holly]

Hi. Anyone from stability who has a view above the API, I need help tracing down some billing/request issues and since the interface doesn't provide anything other then the current credits we need your help to figure this out?

Credits are going down, even after I have rotated the key (the new key is not used anywhere)

[zenith seal]

sometimes i get that ahaha

[loud pond]

@ornate elk whats a good way to increase the quality of image generations (just in general)? are there any things i can do to improve prompts? whats the best way to use hires? should i use a bunch of detailer loras?

[loud pond]

is it possible to change the lora files? one of the lora names has '__' in it which causes it to be detected as a wild card so none of my wild cards are working? Does anyone know if its possible to fix this?

[ornate elk]

Thats a question for #📝｜prompting-help
But yea hires fix and ad detail loras can help

[loud pond]

oh ok ill try using a detailer, would you also mind helping me fix this?

[loud pond]

this is the lora i am having a trouble with ' lora:Sssir8_Style__Pony_XL:1' I dont know how to change the 'double underscore' , since it is ruining the wild cards

[ornate elk]

Had the same problem too. Didn't found a fix for it other than disabling the wildcard extension when using the lora

[loud pond]

if i disable the wild card will i still be able to use the wild cards i have?

[ornate elk]

If you disable the wildcard extension, they won't work

[loud pond]

oh right, ok thanks a bunch

have a good night!

[ornate elk]

Np gn!

[calm river]

Hi. I'm kinda curious, would it be possible/practical to train an embedding on a bunch of the quality tags I use pretty often?

That is, creating a bunch of images using my tags like score_9, score_8_up, etc, and using all those the train that embedding.
The idea being, I'd only need that embedding instead of all that other stuff

[clear pewter]

i have a stupid q!

how to create a lora how to create a LoRa Concept from one picture?

https://cdn.discordapp.com/attachments/1154385802936201226/1294982019478061086/mikus_cute_-cry.gif

[onyx solar]

You generally don't create a LoRA from just one picture.

[zealous cargo]

reactor dosent exist anymore?

i cant download it over extensions tab in a1111

[neat slate]

[onyx solar]

Short answer: Yes.

Long answer:
Yeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeesssssssssssssssssssssssssssssssssssssssss.

[zealous cargo]

wow

thats lame

so what is being used rn?

[karmic minnow]

hey there, I was in school, I'll do it rn, let me check

[signal jasper]

Hi all, has anybody found a fix for this? I guess I am doing something wrong because I am following a tutorial and he is not having any issues with it. I have tried the "Try Fix" button and the switch version button has no other options. Ty

[wicked widget]

delete and try manually installing

[karmic minnow]

i did it, and it takes the same amount as 832 x 1216, and also makes my browser froze a little bit

[signal jasper]

Oh that worked, TY

[wicked widget]

np

[karmic minnow]

i think ig my only resolution is reinstalling SD xddd

[ornate elk]

Worth a try

[karmic minnow]

can you send me the forge link please?

ah nvm

i got it

thx

[karmic minnow]

Well, now I can finally said. After installing SD once again. It's not the software

Something f#ked up my gpu xdd

Idk if it's any drivers or windows update

[ornate elk]

Oh why ?

[karmic minnow]

I tried to generate the first image, which I already knew would take 15 to 40 minutes, but in the middle of that, my pc froze

[ornate elk]

Completely?

So you had to restart the PC?

[brazen veldt]

Hiya, I've installed everything needed for sd3, but when I run my webui-user.bat to install it cant seem to latch onto my python.

During setup I had to reinstall python as I was using an older version (3.6) and now it wont detect python 3.10.6 even though it seems to be installed properly.

I set the route for python inside webui-user.bat but it says No Python at 'C:\Python38\python.exe. I also doublechecked i installed the path with python but no avail. Any help is appreciated, I'm still new.

[brazen veldt]

finally got it to latch by copying the python data to the python38 folder and the python folder in appdata, now it is just struggling to install torch

[brazen veldt]

fixed by deleting the venv folder python 3.6 created inside of sd. reinstalled python 3.10.6 after deleting venv.

[ornate elk]

Hey, better uninstall Python 3.8 as 3.10.6 or better 3.10.11 is the recommended version for webui stuff

[brazen veldt]

ah ok! thank you!

[ornate elk]

And as you found out everytime you change something with Python versions you have to delete the venv folder.
So at a relaunch it will use the new Version.

[brazen veldt]

sweet. I think I'm getting the hang of it! Not as difficult and complicated as i expected.

[karmic minnow]

Yeah

[regal mason]

Is there a successful method to train LoRA models for SDXL with an AMD GPU on Windows 11?

[ornate elk]

You can follow the OneTrainer Zluda Guide:
https://github.com/CS1o/Stable-Diffusion-Info

But you need a GPU preferable with more than 8gb of vram

[regal mason]

Thanks. Have you gotten this working yourself?

[ornate elk]

Yep works on my 7900XTX

[regal mason]

I have a RX 6800 XT (16GB VRAM)

[ornate elk]

That should work too.
But you need to lower the batch size for the training

Batch size 4 for sdxl lora used 20gb vram

If you have any questions regarding the setup, let me know

[regal mason]

Thanks dude. Getting comfyui working with my setup already was quite the process, though I ended up getting it running eventually

[ornate elk]

I made guides for every webui also comfyui. To get it working with Zluda.
Which is much faster than directml

Feel free to check them out

Same link

[regal mason]

yeah, I probably used one of your guides tbh. It's using ZLUDA and not directml

[regal mason]

okay.. question already.

To get my comfyui working, I specifically had to install AMD HIP 5.7 (instead of the latest version) for it to finally work.

This guide is mentioning HIP 6.2.1 ... will it work with 5.7?"

[ornate elk]

Yes it can work with 5.7

[loud pond]

@ornate elk do you have a good tutorial on how to download deepseek v3?

[loud pond]

preferably with a good UI, i'd like to train an LLM to make perfect prompts and i thought deepseek since local would be the best for this

[ornate elk]

i would wait until ollama or lm-studio has support for it

a good ui would be AnythingLLM

[loud pond]

i thought ollama has support for it now?

[ornate elk]

it should but they dont provice an easy download for v3

only for r1

[loud pond]

would v1 provide the same memory features as v3?

[ornate elk]

https://docsbot.ai/models/compare/deepseek-v3/deepseek-r1

a comparison

seems like r1 is even better

[loud pond]

yeah r1 seems to have a larger output windo

would r1 be able to run on ollama?

[ornate elk]

yep

downloading it rn

it has all size variations

8b is 5gb

[loud pond]

oh great then, would you mind sharing a tutorial or something on how to download it via ollama?

[ornate elk]

not today im going off now

but you can dm me and i can help you then tomorrow

its not difficult

[loud pond]

ooh ok, have a nice day, sure ill just dm you tmrw then if your free

[ornate elk]

install ollama:
https://ollama.com/
select a model and size:
https://ollama.com/search
then copy its install command and paste it into a cmd.
i can then give you a bat for quick access ollama

[loud pond]

dw abt it, ill just msg you tmrw have a nice day!

[calm river]

Anyone here familiar with training SD embeddings? (not Lora)

[placid lagoon]

hello, what am I doing wrong for it to generate such ugly images? https://civitai.com/images/47914666

help pls 😦

[wicked widget]

its 512 resolution. use the lora with a pony or illustrious XL model

gen at 1024 res

[placid lagoon]

I am using the lora am I not?

[wicked widget]

use the lora WITH a checkpoint, loras arent checkpoints

the standard one nowadays are PonyXL or Illustrious XL

You will place models in your Stable Diffusion/checkpoint folder/
Loras in the Lora folder

[placid lagoon]

almost done downloading illustrious and I'll try that

[remote charm]

I am having such a headache with an error that no one, anywhere has apparently had, according to my searches.

Recently decided to attempt to make the jump to ReForge from A1111. Did a clean install of the thing but I am now unable to generate anything due to the error:

'Options' object has no attribute 's_churn'

TypeError: 'NoneType' object is not iterable```

Any thoughts on what the cause and/or solution might be?

[vocal burrow]

Why do I get different results from civit.ai/my friend/alien lifeform examples :

1/ --xformers (and some other flags but mainly this one) depending of its version can make stable-diffusion non deterministic. Meaning that for the same prompt+settings running on different hardwares you'll get different results. The output should still correspond to whatever the prompt is.... It might just be slightly to very different if you're unlucky. So yeah, some command line arguments can alter the outputs of your prompts and your hardware might also do the difference.
2/ Do you have all the loras, embeddings mentioned in the prompt ?
3/ Are you really using the same settings ? Copy pasting manually what's inside the "Generation Data" boxes is not enough. Civitai is hiding a bunch of data there. If you want to get the exact same settings you'll have to click the "Copy Generation Data" button at the bottom of the image page, paste its content into Auto1111's prompt field and then click the blue arrow under generate to automatically apply each value to the correct field. Hidden values can be; clip_skip, ENSD, token merging ratio, etc
4/ Did you have any overrides set beforehand ? They should show up at the very bottom of the page (some will probably show up too if you use data from civitai like I mentioned in 3/)
5/ Maybe it's using some extensions that does not record its settings in the metadata
6/ Maybe you're using different versions of some extensions/auto1111/lora/models/etc that yield different results
7/ Are you using the same random number generator ? GPU / CPU / NV (Settings->Stable Diffusion->"Random number generator source")
8/ Maybe you're using a newer version than OP. One that includes a "seed breaking change", cf https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Seed-breaking-changes

[vocal burrow]

can't help without a full log

[remote charm]

Then the full log shall be provided without reservation!

https://pastebin.com/5eQkd5vy

[zealous cargo]

anyone know why this suddenly pops up on forge ui? this did not happen yesterday

theres actually more than enough mb left

[vocal burrow]

apart from missing options for potential speedup

Hint: your device supports --pin-shared-memory for potential speed improvements.
Hint: your device supports --cuda-malloc for potential speed improvements.
Hint: your device supports --cuda-stream for potential speed improvements.

and some warning (dev's fault)
nothing wrong shows up on log, can you screenshot the webui so we see the settings you're using

[remote charm]

Update: Panchovix (the owner and dev of ReForge) got back to me and put up a new commit that fixed the issue. XD

Gotta give the guy props for working quickly!

[placid lagoon]

yeah I was not understanding what the difference between loas, checkpoint and base model was

im struggling with something else now thou, i realized sd was using my cpu and not my gpu but I have a 6700xt so I've been trying to do the setup for amd, unsuccessfully so far

[placid lagoon]

yo that tip about copying the entire settings was really useful

appreciated

[placid lagoon]

thanks for clarifying this

[wicked widget]

yw

[placid lagoon]

I think i got pretty close

[wicked widget]

model is general knowledge and style and lora is the additional guidance filter in layman terms.

So find a model with the style you enjoy the most, I recommens the SDXL models at 1024x1024 resolution