🤝｜tech-support

[sudden pike]

how can i limit it?

--medvram ?

or can i allocate general RAM to it?

like i have 96gb Ram in my PC

[vocal burrow]

1/ do what I said first.
2/ 16g should be enough. No need for medvram
3/ no you can't allocate ram

3 bis/ technically you can but it would be all or nothing. Can't mix and match ram and VRAM efficiently. And using ram + CPU ( can't use ram + GPU ) is sloooooooooooow

[sudden pike]

thanks for the help!

[analog cosmos]

in what folder shall i place the videocrafter models? (dynamiccrafter_512 t2v and i2v) in the text 2 video extension folder or in the normal models folder or? (automatic1111)

[dim tree]

i see a lot of setup guides/tutorials in this channel and on the net, but is there a text guide where i can learn about all the terminologies that stable diffusion (and other open weight models) use? i really want to know how the tools i'm using work 😔

there is a "stable diffusion wiki" on the net but its really hard to navigate and has the added bonus of feeling all over the place

😔

[sand tiger]

hey guys, does anyone know about lora training and kohya_ss ? i just can't manage to fix the error and i'm so desperate that i'm turning to you now, i hope someone can help and don't resit my head off because it's not more explicit stable diffuse content 😄

[ruby yoke]

start by going to youtube and searching on "how do diffusion models work" and also on "machine learning tutorials"

[dim tree]

I hate video tutorials honestly, can't focus through all the slow talk ;-;

but I'll check some of them out on 4x, thanks for the advice

[ruby yoke]

it's not just terminology - you're asking for a wiki that'll teach you the field of machine learning. almost all the terms are used because of that field of science, and if you really want to understand what stuff like latent space or vectors are, you need to study that field

and if you want to understand the papers, you'll need to at least learn caclulus notation

[dim tree]

that makes sense

I thought it'd be something along the lines of explaining what CFG scale, steps, etc are and what their respective function is. like even "low rank adaptation" is a mystery to anyone not in the know-how

guess we'll be doing a crash course on machine learning cuz I'm bored anyway

[ruby yoke]

cfg is classifier free guidance - so we can either say move the scale one decimal place at a time, render, and see what you think - lower numbers give you less details - or we can go into a long technical answer. depending on just how much information you want. but if you really want to understand what is going on, not just be able to use the tech to make pretty pictures, you need to study the science - this is a good place to start if you want to do that https://huggingface.co/blog/ml-for-games-1

[dim tree]

indeed, the long technical answer is what I'm looking for
"change scale and see result kaching" doesn't bode well for me

thank you again for this, I would've never found it myself

[ornate elk]

@dim tree If you using Automatic1111 or Forge/ReForge checkout this helpful overview of some features:
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features

[regal storm]

I have a problem with Mickmumpitz latest comfyui workflow video. specifically this inpainting workflow for whatever reason in the sampler area, the Differntial diffusion and all the samplers just won't run, have the red box around it, and it just won't work. Any advice? Is this the right channel for a question like this. https://mickmumpitz.ai/guides/basic-flux-inpainting-workflow-easily-replace-characters

I get this error in console "DifferentialDiffusion 41: - Required input is missing: model"

[ruby yoke]

this isn't mick's discord... however - that says your model is missing. see what is in your console log

[regal storm]

that is what the console log says, not sure what you mean

[grim arch]

does openpose work with illustrious on forge?

[regal storm]

So I connected the diffusion to the original load checkpoint, and it seems to work? they were using this anything everywhere node, which im not sure what that does

[lucid yoke]

If you use the illustrious controlnet models sure

Can be found on civitai

[grim arch]

my dumbass didnt realize there was one on civitai. thanks tho 👍

[ruby yoke]

i meant the entire console log, not just that one error. start at the beginning of it and read thorugh it line by line. i'm going to gues you've got comfy installed in the wrong place, and that when manager tried to install diffusers, it failed, and the node is now failing because of it

[analog cosmos]

Top Tip ⭐
if you want you automatic1111 as an app instead of using a tab in your browser instead then here is a solution. i use vivaldi browser (wich i recommend, very fast, fully customisable, chromium based so full support for extensions and everything else)

1. right click tab in vivaldi -> select install as app
2. edit webui.py to make the changes as shown in the image (comment out a section that opens new tab and add a small code to make it open the app instead)
   (edit the filepath and address and port to what is correct for your own)

In chrome you might need to make a shortcut of chrome instead and use the same command with filepath and --app function. should work in any browser really.

now when i run my automatic1111 shortcut from taskbar (shortcut for webui.bat) then it loads up and opens directly in the seperate windowed app instead of using a browser tab, enjoy 🫡

[fossil dew]

I came in the other day asking why Stable wont open the LocalHost when I run the cmd from batch file, it opened this time, but it keeps erroring out:

@ornate elk

[ornate elk]

can you show the full cmd log?

[fossil dew]

sure let me run the batch file again

its different this time 😐

could just be a memory issue as per usual

[ornate elk]

how much RAM do you have?

[fossil dew]

16g

i've literally almost never had issues with stable diffusion running, only time I had issues is after updating to XL 😐

[ornate elk]

then make sure you have 15-20gb free space on C drive and then recheck your windows pagefile settings

[fossil dew]

i have 51g free

how do I check windows pagefile settings?

my D drive is what has little to no space

[ornate elk]

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

[fossil dew]

ty

[ornate elk]

Make sure the pagefile is only enabled for the C drive and disabled for any other drive.
Make sure its set to custom for C at 16000min and 24000max

[fossil dew]

tas already set

[ornate elk]

okay then set 32000 instead of 24000

then reboot the PC

then it should work

[fossil dew]

rebooting, ill lyk if it worked

seems to be working now

thank you

[lyric tree]

sorry if ignorant question but completly new to this. I dont really get how loras work but Ive downloaded them, put them in the lora folder in my modles. got them into SD, then applied them to my prompt, and the picture it generates looks nothing like the lora of the person. Am i missing something?

[lucid yoke]

Did you use it's trigger words?

It should be on the lora page

[versed kernel]

Guys, I'm currently trying to use Forge UI but when I tried to run it, I bumped into this error where it says "RuntimeError: Your device does not support the current version of Torch/CUDA! Consider download another version: <a github link but I'm afraid sending it would violate the rules>". So I'd like to ask if there's a way to know whether my device is capable with the UI or not

[lucid yoke]

Are you running a AMD card?

[versed kernel]

U mean graphics card?

[lucid yoke]

Yes

[versed kernel]

I'm using Intel(R) UHD Graphics

[lucid yoke]

Oh

Yeah hmm that makes sense then

I'm not sure if Intel integrated graphics could run ai or if it can, anything beyond sd1.5

Since it would offload to ram fully

[ornate elk]

Maybe check the taskmanager if you really dont have a GPU in the performance tab other than Intel HD

[versed kernel]

Sooo what you're saying is that my device is incompatible to any AI image generator whatsoever? 😦

[versed kernel]

The performance section?

Is this the one?

[versed kernel]

U still there?

[lucid yoke]

Looks like theres no other gpu other then uhd

[ornate elk]

Yep no other GPU :/

[versed kernel]

Sooo what you're saying is that my device is incompatible to any AI image generator whatsoever?

[ornate elk]

To all local ai stuff yes

[lucid yoke]

Cloud programs are very affordable however

[versed kernel]

I'm somewhat technonologically illiterate so may I ask if this is the case for all laptops? Or more expensive ones like gaming laptops should do the trick?

[ornate elk]

Gaming laptops mostly have an dedicated GPU from Nvidia or AMD so they can work with Ai

But they are limited in vram most of the time

[versed kernel]

Which reduces the image generating speed, I presume?

[versed kernel]

Thanks a lot then!

[ornate elk]

Speed and capabilities

The more vram the GPU has the better its for ai stuff rn.
Nvidia GPUs are preferable

[versed kernel]

So you're saying PCs are the go-to choice when it comes to AI image generating?

[ornate elk]

Yes definitely

If you want to do it localy

Other options are renting a GPU in the cloud

Or using third party services

[versed kernel]

Sounds a little complicated...

Anyway, thanks for your answers

Have a nice day then~

[ornate elk]

NP, you too!

[terse hedge]

Getting this error every once in a while after Stable Diffusion is forcibly closed (due to electricity blackout/hardware):

Exception Code: 0xC0000005
 #0 0x00007ffac0646edb at::native::_rowwise_prune(class at::Tensor const &, class at::Tensor const &, enum c10::ScalarType) (B:\StableDiffusionUpdated\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\lib\torch_cpu.dll+0x6046edb)
 #1 0x00007ffac064771f at::native::_local_scalar_dense_cpu(class at::Tensor const &) (B:\StableDiffusionUpdated\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\lib\torch_cpu.dll+0x604771f)
 #2 0x00007ffac1214f8e at::cpu::where_outf(class at::Tensor const &, class at::Tensor const &, class at::Tensor const &, class at::Tensor &) (B:\StableDiffusionUpdated\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\lib\torch_cpu.dll+0x6c14f8e)
 #3 0x00007ffac11ec4c1 at::cpu::bucketize_outf(class at::Tensor const &, class at::Tensor const &, bool, bool, class at::Tensor &) (B:\StableDiffusionUpdated\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\lib\torch_cpu.dll+0x6bec4c1)
 #4 0x00007ffac0dda69f at::_ops::_local_scalar_dense::call(class at::Tensor const &) (B:\StableDiffusionUpdated\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\lib\torch_cpu.dll+0x67da69f)
 #5 0x00007ffac06478cf at::native::item(class at::Tensor const &) (B:\StableDiffusionUpdated\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\lib\torch_cpu.dll+0x60478cf)

The error appears immediately after trying to generate an image with any settings, resulting in crash

So far, the only solution I've found is to completely erase the Stable Diffusion program and reinstall in a different folder via your guide @ornate elk

Which is extremely frustrating to do on a weekly basis whenever my PC accidentally shuts down while using SD

I tried deleting the \venv folder, the error persists after the \venv folder is recreated, only a new installation makes the error go away

[fierce moat]

anyone know how to use lora in ltx 13B fp8

[ornate elk]

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

And how much RAM

[terse hedge]

7800XT

[terse hedge]

set COMMANDLINE_ARGS=--use-zluda --cuda-stream --attention-quad

[terse hedge]

24 GB

[ornate elk]

why does it shut down?

and when? only when generating images?

[terse hedge]

Random reasons, usually due to power blackouts or hardware issues or random crashes, unrelated to SD for the most part

[terse hedge]

Randomly, the issue comes with trying to relaunch SD after the shut down

[ornate elk]

can you show me a full cmd log when relaunching?

[terse hedge]

It seems that whenever SD unexpectedly closes the files get corrupted and the whole installation becomes unusable

[terse hedge]

In a bit

[ornate elk]

if you have power outages you could buy an UPS.
An better way would be to make backups of your files. Maybe onto an external drive or internal drive. That can be done with free software like:
Kopia: https://github.com/kopia/kopia
or VEEAM: https://www.veeam.com/de/products/free/microsoft-windows.html

[storm ridge]

hello!! im currently trying to generate with stable diffusion, but the generated image quality is always really really terrible, it seems to be the same no matter what .ckpt or .safetensors i use and i have no idea what the issue could be, any help would be great thanks :)

this is the kind of quality i get when trying to generate an anime character

[oblique robin]

Hey guys, I need help with webui. When generating an image, the image generates (and the preview as it is generating is available). However, as soon as the generation is complete, it completely goes back to the blank preview.

The image I generated is complete, and I can access it in my folder, but it disappears from my preview making it difficult to do anything with batches + hires.fix

If it helps, after each generation I get this message (although it says error, my image still generated and there are no issues with generation, just that preview being missing and thus not being able to run hires fix as a result):
**
OSError: [WinError 123] The filename, directory name, or volume label syntax is incorrect: 'outputs\txt2img-images\2025-05-08\00028-2696067343.png?1746758101.9361873'**

[ornate elk]

Hey, pls show your whole txt2img screen

[ornate elk]

Hey make sure the Webui is whitelisted in any browser adblocker

[oblique robin]

yeah I just did that and still the same issue

would it have anything to do with that error message I pasted maybe?

[ornate elk]

Have you edited anything in the stable diffusion settings? Maybe image name or save path?

[oblique robin]

no I downloaded it as is and it has been like this

[ornate elk]

Hmm, can you show the full cmd log?

[oblique robin]

okay I generated a random, extremely simple image using pony diffusion (although it happens no matter what checkpoint, lora, etc. is selected) and heres the cmd log

[ornate elk]

Wrong Python Version

Stable Diffusion needs Python 3.10.11 64bit

Uninstall python 3.9 then install 3.10.11 64bit
Then delete the venv folder and relaunch the webui-user.bat

You also should add --xformers to the commandline_args in the webui-user.bat to get a much better overall performance

If your GPU has 8gb vram then also add --medvram-sdxl
If it has 6gb or less add --medvram

[novel owl]

hello

i have a problem with comfyui

Warning, you are using an old pytorch version and some ckpt/pt files might be loaded unsafely. Upgrading to 2.4 or above is recommended.
Traceback (most recent call last):
File "D:\IA\SD-Zluda\ComfyUI\main.py", line 136, in <module>
import execution
File "D:\IA\SD-Zluda\ComfyUI\execution.py", line 13, in <module>
import nodes
File "D:\IA\SD-Zluda\ComfyUI\nodes.py", line 22, in <module>
import comfy.diffusers_load
File "D:\IA\SD-Zluda\ComfyUI\comfy\diffusers_load.py", line 3, in <module>
import comfy.sd
File "D:\IA\SD-Zluda\ComfyUI\comfy\sd.py", line 7, in <module>
from comfy import model_management
File "D:\IA\SD-Zluda\ComfyUI\comfy\model_management.py", line 195, in <module>
total_vram = get_total_memory(get_torch_device()) / (1024 * 1024)
File "D:\IA\SD-Zluda\ComfyUI\comfy\model_management.py", line 146, in get_torch_device
return torch.device(torch.cuda.current_device())
File "D:\IA\SD-Zluda\ComfyUI\venv\lib\site-packages\torch\cuda_init_.py", line 778, in current_device
lazy_init()
File "D:\IA\SD-Zluda\ComfyUI\venv\lib\site-packages\torch\cuda_init.py", line 293, in _lazy_init
torch._C._cuda_init()
RuntimeError: CUDA unknown error - this may be due to an incorrectly set up environment, e.g. changing env variable CUDA_VISIBLE_DEVICES after program start. Setting the available devices to be zero.

can you help me CS1o ? ^^

please

[ornate elk]

Hey, can you go into the Device manager and check if you have two AMD display drivers there?

If so, disable the AMD Radeon TM Graphics, then relaunch ComfyUI

And if you still get an error please provide the full CMD log

[novel owl]

where is AMD display driver please ?

this ?

with A1111 and Zluda it works

ah no

its the 25.5.1 driver update its incompatible

i return at 25.3.1

and it work

CUDA error: CUBLAS_STATUS_NOT_SUPPORTED when calling cublasSgemm( handle, opa, opb, m, n, k, &alpha, a, lda, b, ldb, &beta, c, ldc)

i have this now...

[ornate elk]

Oh that would make sense then

[ornate elk]

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

[novel owl]

okay where is the .zluda ?

[ornate elk]

In the same folder as venv

So in stable-diffusion-webui-amdgpu

[novel owl]

i dont have .zluda

im on comfy ui

a1111 works

[ornate elk]

Ah okay

Then you have to recopy the 3 DLL files into the torch/lib folder in the venv

[novel owl]

this ?

with the rename 64 or without ?

[ornate elk]

With the rename

The ones you marked/selected

AFK

[novel owl]

me ? no i try

it going to re compile ?

good thx^^

[ornate elk]

It should yes

[novel owl]

Very thanks for your help

[ornate elk]

Thank you for letting me know the driver caused it. 3 other User have the same Problem

Hope there will be a fix

[sage otter]

does anyone have a good understanding on inpainting, and how upscaling can take all your PCs processing power? I was inpainting, and sent it to gimp without downscaling.. then did that twice, now my generations are pegging my system to max VRAM and it freezes. I need help to see if i can back it down

[ornate elk]

whats your gpu?

and which webui di you use

[sage otter]

RTX3080 / 10GB VRAM. Forge UI

[ornate elk]

try add --cuda-stream to the webui-user.bat commandline_args
and if you use inpaint in img2img make sure to always set the Inpaint area to "only masked"

Also make sure to close other stuff in the background like Wallpaper Engine or Games

[sage otter]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --skip-torch-cuda-test --disable-safe-unpickle --no-download-sd-model --cuda-malloc

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

set PYTHONWARNINGS=ignore::UserWarning,ignore::FutureWarning
set HF_HOME=C:\AI Stuff\Forge UI\huggingface_cache
call webui.bat

This is what i had through working with Chat GPT to make my 86 sec load time turn into about 22 sec.

Do you disagree with anything?

[ornate elk]

remove --skip-torch-cuda-test --disable-safe-unpickle
and then relaunch the webui-user.bat

[sage otter]

relaunching

[ornate elk]

also this can be removed:
set PYTHONWARNINGS=ignore::UserWarning,ignore::FutureWarning

[sage otter]

re-relaunching

[ornate elk]

and if the huggingface_cache folder in forge is empty you can also remove this:
set HF_HOME=C:\AI Stuff\Forge UI\huggingface_cache

[sage otter]

wheres that one hiding?

[ornate elk]

if there is not such folder then you can remove that line

[sage otter]

but would it just be in the main C:\AI Stuff\Forge UI\webui folder? cause theres tons of folders to go through in here to confirm thats it is or isnt there

and removing what you said has this taking forever now

[ornate elk]

it would be in Forge UI folder

in your case

not in any subfolder

[sage otter]

got it so i can remove that line

ok maybe relaunching will make the next launch go quicker

its finally up

[ornate elk]

yea first time can take longer after changing stuff

[sage otter]

also removed the hugging face line. so may take long again? will launch twice

also... when is ther 5090 restocked, dammit?

bought dual 48gb ram sticks

just for this

[ornate elk]

the rtx3080 was one of nvidias biggest scams xD

with only 10gb of vram :/

[sage otter]

its so lame

it was never an issue until 2 weeks ago when i got into SD

https://civitai.com/user/TK503

Behold: My stuff

Plus this... learning inpainting with this one. needs more tweaking

[ornate elk]

nice aesthetics

[sage otter]

thanks

every generation after taking this in and out of gimp and back to inpainting without apparently downscaling before the export to gimp

[ornate elk]

40gb of memory used

[sage otter]

idk if i need to revert abck to a previous image before upscaling and redo work but make sure i downscale first

[ornate elk]

what are you running lol

[sage otter]

small form factor build

[ornate elk]

and what stuff takes up the 40gb of ram right now? xD

[ornate elk]

nice build

and lucky with the 12th gen

[sage otter]

that was last night. old screen shot. I was mid generation. was lucky to get that screenshot off cause my pc was frozen

also its taking forever to load still

think maybe i should reinstall forge and just get away from what GPT had me do?

[ornate elk]

yea please dont listen to chat gpt, some other guy nearly destroyed his Windows by deleting Important Environment variables

[sage otter]

LOL

yea it confidentally steers you into the wrong direction. but it did make my load time a lot faster

[ornate elk]

yea because it got something right, --xformers xD

[sage otter]

the stuff you ahd me get rid of has it taking forever though. i wonder why. was slow cause i added a bunch of extensions. i forgot about that and realized removing them would fix it. but the changes it had me do were already set

[ornate elk]

does it still load`?

also make sure you dont have outdated extensions

scam bot nice

and hes gone, but i got his name xD he will get banned soon

[ornate elk]

if nothing works you can also do a clean reinstall of forge.

and then only add --xformers --cuda-malloc --cuda-stream to the commandline_args

[sage otter]

i just reinstalled it. gonna start moving all my models and what not over

i missed the scam bot

[ornate elk]

they are quick when they get messaged back, then they delete their initial mesage

[sage otter]

what was the message? for my awareness of the scam

[ornate elk]

Click here to get Support
-> scam link to a discord

There they pretend like a offical support and want to get your money/crypto or account data

[sage otter]

lol

wild

[ornate elk]

yep, mostly automated by bots, they often can be summoned when you type
help
i have a question
can someone help me?

help?!

[sage otter]

any of these known to cause slow downs? i noticed a few extensions were poorly optimized. but i really like the civitai helper, and tagcomplete ext. Detele the other foders but leave the excel sheets?

[sage otter]

lol it was probably me

[ornate elk]

nope they anserwed to my question "does it still load?" xD

[sage otter]

thoughts on those extensions though?

im ok with deleting all i didnt mention

[ornate elk]

controlnet is already included in forge so installing it again will break both.
Auto-Photoshop plugin is outdated since 2 years. so it doesnt work and breaks the webui too
the rest is okay.
Also install adetailer for auto face fix

[sage otter]

oh an di need sd webui controlnet

ok

[ornate elk]

and delete the venv folder after you removed the extensions. That fixes the webui and removes stuff left by the extensions

it will take a bit to reinstall the core files when recreating the clean venv

[sage otter]

ok delete facechain, auto photoshop, control net?

Also install adetailer for auto face fix

is this a specific detailer?

ok delete venv folder

[ornate elk]

adetailer uses automated inpaint to fix faces

[sage otter]

https://github.com/Bing-su/adetailer

what is facechain

no venv

[ornate elk]

it seems like a tool to train and convert face models

[sage otter]

keep or get rid of

[ornate elk]

if you want to test it you can, if you dont need it remove it

if it doesnt work, remove it

[sage otter]

thats the thing.. im not sure how to even use it since im still pretty new. so idk. guess ill leave it for now'

is this controlnet stuff outdated?

pretty sure i migrated it from A1111

so maybe like you said.. its not needed for forge? or is the included stuff that i i should carry over

[ornate elk]

these are the controlnet files used for 1.5 based models

if your work with 1.5 based models you need them

SDXL based stuff needs its own controlnet files

[sage otter]

no. I installed them with an A1111 tutorial when i was brand new but i like working with Illustrious Xl

I will leave it behind

[ornate elk]

for illustrious you need this for openpose:
https://civitai.com/models/1359846/illustrious-xl-controlnet-openpose

and these:
https://civitai.com/models/929685?modelVersionId=1239319

[sage otter]

--xformers --cuda-malloc --cuda-stream

looks right?

no space betwen = and -- ?

[ornate elk]

not needed

[sage otter]

ok. few more optimization questions if you dont mind

anything not needed? since im moving away for 1.5

this is for flux

any you recommend to get away from?

thoughts?

[ornate elk]

you can remove the 84000mse vae, its only for 1.5

[last flax]

hey ive made a comfy workflow which does img2img

i wanna add control nets

to it

[sage otter]

thoughts?

[last flax]

but idk where and how it shd be connected

can someone help me out w it

[sage otter]

ok thats it

[ornate elk]

hey, try ask in #🧣｜comfy-ui too, mostly they know how

[ornate elk]

you can also remove fluxVAEsft

leave the upscalers in place

all ok

[sage otter]

copy

then thats it. i will install the two links you gave me

[ornate elk]

you can find here more upscale models:
https://openmodeldb.info/

[sage otter]

how would i know which to choose?

[ornate elk]

Fatal Anime 50000 G is good for anime
Ultrasharp for realism
lollypop is nice

[sage otter]

remember the style i like? that semi realism anime ish type? whats good for that? fatal?

[ornate elk]

fatal or remacri

[sage otter]

i have remacri

downloading the list of xl CN's

also.. compressing 630GB of appdata backup files lol

i had two 2tb NVMes.. I replaced them with a single 8TB nvme. I transferred my entire appdata over in a backup folder just in case. I want to compress it to save some room cause that boy thicc

[ornate elk]

wow thats a lot

[sage otter]

yea i was surprised

also.... Sandisk is selling the Gen 4 western digital sn850x 8TB NVMe w/ heatsink for $630. If you get on sandisks chat support window, and say you are thinking about buying that NVMe, and are wondering if there are any promos available.. they may give you a 10% code that ends up being 20% like it did for me. I paid $540 after tax. instead of about $700

[cursive hazel]

I'm back, yet again! With more problems
But I think I know the issue this time. Just wanting to confirm my suspicions.

I've been trying to play around with LTX video (image to video, specifically). It works fast, which I appreciate! Buuut... The quality I'm getting is.. Sort of awful. I imagine this is because of the low resolution I'm using? Or the checkpoint I'm using (ltxv-2b-0.9.6-distilled-04-25).

My first thought was, increase the resolution for a better image. I'm doing 448x640. However, I cannot increase the resolution (or longer duration of the clip), because if I do, my entire computer locks and and black screens.
iirc, I tried higher quality models as well, and those don't work for me either. I believe I'm running out of VRAM, and then that's causing my crash?

I am using an AMD RX 9070 with Zluda. 16GB of VRAM. But from my understanding, the Zluda translation/conversion/whatever, uses some extra VRAM. Doing the workflow I have currently, I do hit like 14.5gb through most of it, then it hits 15.1gb or or so, when it gets to the VAE Decode part.

Attached is an image to video I have, as well as the original image.
On a side note, I also got Wan video to work, but it takes about 20 minutes, and also suffers from the same problems, more or less.

[sage otter]

Is that Harriet Potter?

Henrietta Potter maybe?

[cursive hazel]

Darn Potter for claiming round lens glasses as his brand!!!

[ornate elk]

nice good to know

[sage otter]

the glasses dont help but these are all the same people lol

[ornate elk]

yea you mostly run out of vram when increasing the resolution.
Wan 2.1 needs more vram than ltx so you have to use a smaller gguf model that may works

also make sure to close everything in the background and dont run Wallpaper Engine while genning

[strange horizon]

Hi! Is there someone that can explain to me the difference between main_device and offload_device on the load_device parameter on Wan model loader? Which one shoudl I choose?

[cursive hazel]

I thiiink I am using a gguf model, iirc. For Wan at least. From what I've discovered, I can use models that are around 6GB. Anything bigger doesn't work for me, and I'm still stuck at resolutions around 500x500-ish.

I do close everything. I actually give a fresh restart on my PC when I'm trying to play with video stuffs.

[ornate elk]

hey, i think the main device would be your gpu and offloading would be into RAM.
if you have enough vram then dont offload as that takes longer

[oblique robin]

that was it

appreciate it bro

[cursive hazel]

Oh, I was wrong. My Wan file is 10GB. wan2.1-i2v-14b-480p-Q4_1.gguf. I also have the K_S version of the same thing.

[ornate elk]

no problem

[strange horizon]

got it - so with 24VRAM I should be good to run on main device (for wan?)

[ornate elk]

yep

[strange horizon]

thank you

[ornate elk]

but dont run a 23gb model

it would be still to much

[sage otter]

omg 5090 when

fucking bots

[cursive hazel]

So, all in all, I'm feeling like I can't really do videos with my GPU? If I want better quality, I mean. Because my LTX one is just 6GB, but I can't get very high resolution at all, before I crash. And with Wan video, I'm using a 10GB gguf model, and have the same restriction on resolution (It's actually even smaller, at like 368x544)

[ornate elk]

10gb is to much

[ornate elk]

yea ltx is your best choice but just wait, every week some new video stuff drops. give it some months and we get something better to work.
Video stuff is currently not really worth on local devices with less than 24gb vram

[cursive hazel]

Ahh, ok. I've been seeing a lot of video content, and assumed it was do-able with 16GB or so. Then I thought, it was probably just because I have an AMD card that I'm having struggles

[ornate elk]

nope its just overall slow if the model sizes are too big

and the smal sizes have a bad quality

also dont update to adrenalin 25.5.1 driver because it breaks zluda

[cursive hazel]

Wan takes me like 20 minutes at that tiny resolution. Which, I'd be ok with!.. If it had a decent result. But the eyes just smudge into a blob, and.. Well, pretty much everything does.

Annd, good to know! I'm on 25.4.1, and perfectly happy with it.

Thanks as always! You're always such a help here.

[ornate elk]

no problem 🙂

@cursive hazel just saw that there is a gguf version of ltx 0.9.6 which is 2gb

can be found here:
https://huggingface.co/calcuis/ltxv-gguf/blob/main/ltxv-2b-0.9.6-distilled-fp32-q8_0.gguf

[cursive hazel]

Oh, very nice! thank you! I was actually looking around right now as well, and found an 8gb gguf one I was gonna try that. But I'll definitely try this 2gb one!

If I can continue to pick your brain... What's the big difference in model sizes? I imagine it's partly context and partly quality? Like, a 2gb file compared to the 8gb, the 8gb one is going to have better quality images, assuming they're at the same resolution. And I assume the 8gb file will have a bigger.. Data set? If I'm not mis-using that phrase. So if I specify like, "a fluffy white dog", it'd understand what I mean, but the 2gb one might not?

[ornate elk]

both would unserstand it. the smaller models (gguf ones) are just quantized. Means the precision would be less than the larger models. (A lower quality, but not in resolution)
The same goes for fp8 against fp16
You can see differences but under certain circumstances you dont care about a bit lower quality if it runs faster

like flux fp8 or nf4 for example

[cursive hazel]

Hmm. When you say precision, what does that mean? The clarity of the image?
And if I'm just being a nuisance, feel free to tell me to go read a book or something!

[sage otter]

on forge UI, and i click the XL UI, my gpu weight limit is auto set to 9215MB/10GB. Is this a good amount?

[ornate elk]

yep thats okay

[ornate elk]

it will look a bit different. for example it could create more deformations than the higher precision but mostly these are smaller changes that can be fixed with upscaling or other custom nodes / extensions.

[cursive hazel]

Gotchya! That makes sense. Thanks for helping me learn!

I'm someone who really enjoys teaching others, about the things I know and understand. I always appreciate when others share in that willingness to teach and help.

[ornate elk]

no problem, yea I also like to help people with my knowledge if i can

if you want to know more about float point precisions or quantization you can find a lot of papers on the web but its really technical stuff. Because its something that applies to much more than stable diffusion models

[cursive hazel]

Yeaaah, I did read a little about quantization stuff. At least the thing I was reading, was a bit too deep for me! Maybe one day I'll try diving head first into learning more, but right now I have enough going on

[sage otter]

@ornate elkreinstalled to the standards you said. loading takes a long time again :/

[ornate elk]

for what? generating or starting?

[sage otter]

starting

[ornate elk]

its done

[sage otter]

i know its done. takes like 2 min to load SD. Was startin gin 20 sec before is the issue

[ornate elk]

ah i see the issue

show me the models/stable-diffusion folder

i guess you have a large flux model in there, forge tries to load that everytime you start

[ornate elk]

whats in the two flux folders

[ornate elk]

yep

as i thought

it tries to load a 23gb flux model into your 10gb gpu

you can delete that and the 12gb model too, boath wont be usable on your gpu.

[sage otter]

ok. I will move Flux out of this folder

[ornate elk]

with 10gb vram you can try smaller flux gguf models

like nf4 ones

example:
https://civitai.com/models/141592?modelVersionId=992642

[sage otter]

i thought that was just the size of the model. didnt know it was also relating to how optimized it will be for your VRAm

[ornate elk]

the models get completely loaded into your gpus vram and then the gpu generates the image

if that cant be done it will offload into the ram or uses cpu to support (that makes it really really slow and laggy)

so everything smaller than your 10gb vram will run fast. but if you try to load like a 9.5 gb model it will be slow too because the inital image generation + resolution + loras, controlnet etc will also need vram to process

[sage otter]

interesting

gonna send you another screenshot in a sec

loading a sec time with flux models out of forge entirely

taking forever again

what do you think

[ornate elk]

your not launching the webui-user.bat it seems

the launch args got ignored in this screenshot

[sage otter]

im running a shortcut of the run.bat

[ornate elk]

can you show the webui-user.bat?

[sage otter]

oh, it didnt save your ARGS

can you re typoe those

[ornate elk]

--xformers --cuda-malloc --cuda-stream

[sage otter]

maybe its better to make a shortcut of webui-user.bat and launch off this instead?

added

k

install torch maybe

here is run.bat btw

[ornate elk]

wrong python version

[sage otter]

i just have what came installed

[ornate elk]

uninstall python 3.13, its not compatible with any ai stuff

[sage otter]

oof

[ornate elk]

then open up a cmd and type
python --version

Install python 3.10.11 64bit if you dont have any python

[sage otter]

standby'

[ornate elk]

forge doesnt ship with python 3.13

[sage otter]

it was installed for a separate thing i was doing. had no clue that SD was looking at my separately installed python

[ornate elk]

ah it did that because you launched the webui-user.bat instead the run.bat

the run.bat calls its extra environment

[sage otter]

so should i be running the run.bat or webui

.bat

[ornate elk]

run.bat then

[sage otter]

what about when i get python in order?

[ornate elk]

but still uninstall python 3.13

[sage otter]

stilllll slow. this is so weird. i dont see any errors.

I have pyton totally uninstalled

This is off of run.bat

[ornate elk]

can you generate one image with an illustrious or hasaku xl model in xl mode and then relaunch?

[sage otter]

yes

waiting for it to turn back on

to generate

[ornate elk]

did it worked?

[sage otter]

restarted casuse things were acting weird. regenning

restarting

acceptable to me

this is with run.bat

should i do webu.bat?

oh i have python uninstalled so it doesnt work

would it be theoretically faster if i had the proper python version installed>?

[ornate elk]

nope

[sage otter]

got it. ill leave it as is. thanks for your help!

[ornate elk]

no problem 🙂

[whole sparrow]

I have a problem with reinstalling comfyui

Via stability matrix

Idk if anyone is online to help me find a fix

[ornate elk]

Don't use stability matrix would be an easy fix

What's the issue exactly?

[whole sparrow]

Getting this error

I don't know how to fix it

[ornate elk]

What's your GPU?

[whole sparrow]

GPU?

Where is that

Oh nvidia

[ornate elk]

Nvm i think I found the issue

[whole sparrow]

What is it?

):

[ornate elk]

Reboot the PC. Not shutdown. Just hit restart.
Then open up a cmd and run
Pip cache purge
Then try to install Comfyui again

[whole sparrow]

Wait how does one open up a cmd?

Pip cache purge?

[ornate elk]

You click on the Windows logo and Type cmd and then hit enter or click on CMD

[whole sparrow]

Oh ok

Restarting my pc atm

Ok got cmd open

What next?

[ornate elk]

Type
Pip cache purge

Then hit enter

That removes temporary cached install files

[whole sparrow]

I got this

[ornate elk]

Ah okay

That just means you dont have python installed.
No issue because in your case Stability matrix does it

But then you have to do an other step to fix comfyui

[whole sparrow]

Ok

So what is the other step?

[ornate elk]

You need to go into:
C:\users\super\appdata\roaming\stabilitymatrix\packages\comfyui
And in that comfyui folder you need to delete the venv folder.

[whole sparrow]

Ok

But my packages folder is empty

I was just trying to reinstall

[ornate elk]

Wrong user folder

[whole sparrow]

Ok

[ornate elk]

Stability matrix searched in C:\Users\super\appdata

Im off now as its really late here.
If nothing works. Checkout the install Guides from the pinned messages.
There you find a simple ComfyUI install Guide.
Gn

[whole sparrow]

Darn

):

[ornate elk]

Empty thats strange

Then delete the stability matrix folder there

[whole sparrow]

Ok just delete it?

Ok should be deleted now

It says I need permission

Nvm

[ornate elk]

And then try to install Comfyui again

[whole sparrow]

It won’t let me delete the stability matrix folder. It says I need permission

[ornate elk]

Oh

Do you share the PC with someone else?

[whole sparrow]

It says I need permission from Msi/super

No

I don’t

[ornate elk]

Okay

Does it asks for a password when you want to delete it?

[whole sparrow]

No

[ornate elk]

Okay but you can't delete it?

[whole sparrow]

No

[ornate elk]

What shows up then?

[ornate elk]

Okay, do you have admin permissions ?

[whole sparrow]

Where do you enable them

[ornate elk]

In the User settings

Either in the settings/user or in control panel under manage users, edit user

[whole sparrow]

Never mind this is more complicated than I thought

[ornate elk]

This stability matrix folder is over a year old.
So if you get that deleted. The Installation should work again

[whole sparrow]

Ok

[ornate elk]

Need to go off now, let me know if you need help tomorrow, or if you fixed it

[whole sparrow]

Ok

[lime pivot]

hello, im new, i have 4 questions can i use this for my training image? i made the character big to make sure the main focus of the AI when training is the character not the background - should i make him smaller ? also my character is kinda blurry should i upscale him more or is this enough ? also should I crop him better ?

stabel diffusion 1.5

[bleak wagon]

Hello, I am unable to use Automatic1111 with my RTX 5080 card. How can I resolve this issue? " RuntimeError: CUDA error: no kernel image is available for execution on the device Compile with TORCH_USE_CUDA_DSA to enable device-side assertions." "

[ionic spire]

Hello, starting yesterday I started getting this error every time I try and generate an image. RuntimeError: mixed dtype (CPU): expect parameter to have scalar type of Float I already searched through past responses to this error and attempted what worked for them and I am still getting the error.

[sage otter]

first.... upgrade to forge ui. It is 1000000% more efficient, and better than A11. A11 is a dead program, where forge is still being updated. Plus it is modeled to look like A11. All your generations will generate so much faster too.

Also im new too so idk how to help with the rest. But what I said is good advice

also I cannot recommend that you get civitai-helper extension enough

it fills in all the checkpoints and loras automatically

[ornate elk]

Hey, what's, your GPU?

[solemn atlas]

Idk why my SD doesnt work, maybe soemone can help me please

[vocal burrow]

it's telling you everything here (and above)

basically don't copy paste git stuff from one pc to another blindly

[solemn atlas]

i didnt do that?

or wait, i downloaded krita yesterady, but i cant remember that i copied somethiing

how to make it work again?

[vocal burrow]

what's the full log

[solemn atlas]

it's this

[lucid yoke]

Did you install it as administrator/ cmd opened with admin?

[solemn atlas]

do you mean krita?

[vocal burrow]

no sdwebui

[lucid yoke]

Also i see you had it in backup

And now its not in backup

[solemn atlas]

i must did that, but i installed it if i remember last year

[lucid yoke]

Or the ownership changed somehow

I just recommend reinstalling and moving the models/loras over tbh

[solemn atlas]

you mean sdwbui? How do i reinstall it?

no how do i dealte that first

[vocal burrow]

you just copy paste whatever you want to save somewhere else (models, loras, etc), delete the whole stable-diffusion-webui, and then reinstall it

[lucid yoke]

And if you use sdxl i recommend forge or swarm anyways as a1111 is outdated as hell

[solemn atlas]

I already saved my models/lora etc on my SSD. So I just delate the whole sdwebui right?

[lucid yoke]

The folder where the start file is, delete that one

[solemn atlas]

Sure I would love to do that but forge or swarm is the same as a1111? I also use comfyui sometimes

[lucid yoke]

Forge is similar to a1111 in looks

Swarm is more advanced

Follow one of the guides in the pinned messages

[solemn atlas]

is there a way to deinstall ir or do i just delate the folder?

[lucid yoke]

Delete

[solemn atlas]

Ok, then I'll go to swarm I guess

[lucid yoke]

Swarm is getting some used to but you have much more model support

But the install is easier

https://swarmui.net/

[ornate elk]

Hes on AMD, so don't use the normal installer

[lucid yoke]

Ahhh amd

Yeah then follow the Cs1o install guide in the pinned messages

And probably forge

[ornate elk]

Yea Forge or Auto1111 will do it

[lucid yoke]

Cant wait till pytorch is officially supported for amd lol

[solemn atlas]

okay so no swarm.

but forge

[lucid yoke]

Yes, pinned messages, forge

[ornate elk]

Forge with Zluda

But save your outputs from the old webui first before deleting it

[solemn atlas]

yes i realized that just after i wanted to delate it, now i need to hope that it's still here

[ornate elk]

You can restore it if its in the trash bin

And then save the outputs folder

[solemn atlas]

yes I jsut did that. Need to wait a bit, it really takes a long time to search SD. idk why its so slow always

[ornate elk]

You dont need to use the search

The outputs folder is in the same folder where the webui-user.bat is

[solemn atlas]

ok, now got it. Now I'll delate webui again

it's the first one right?

stupid question but do i need to download git and python again?

[vocal burrow]

no

good luck with that... It might work with small sd1.5 models but it will be sloooow.

[ornate elk]

no chance as pytorch doesnt support 1gb vram gpus as they are to old

[solemn atlas]

--use-zluda --cuda-stream --attention-quad

I'll ad that here right? (right-click then editor)

do i need to delate everything that is in the editor?

[ornate elk]

it mostly uses cpu and ram not gpu in that case then

[ornate elk]

no

[vocal burrow]

in theory tho

[solemn atlas]

like this? Sorry i ask often before I do wrong things again

[ornate elk]

yea in theory xD

[ornate elk]

nope

[solemn atlas]

ohhhhh xD

i have 4070s that means i use this one right?

[ornate elk]

wait your on nvidia?

then forget the zluda guide

sry

then delete the forge folder and just follow the forge nvidia guide

i thought you were still on amd xD

with nvidia its a much easier and faster setup

[solemn atlas]

No, I never had amd 😄 I always had nividia. Ok, then I'll delate everything

[ornate elk]

oh well xD yep delete it

[solemn atlas]

xD

I guess its this one

[solemn atlas]

it does take long to extract it

[ornate elk]

its paused

click the play button

[solemn atlas]

yes I paused it becaut it took an hour for just one 1% is that normal

[ornate elk]

nope

you need to have enough free space on the drive you extract it to and then also make sure its not on a hdd

also 7zip can extract it faster:
https://7-zip.de/download.html

[solemn atlas]

i used windows extracter, ok i will download 7-zip

hopefully that works

ok it worked but if i want to run the update.bat i get this

[ornate elk]

Click on weitere informationen and then ausführen

[solemn atlas]

okay it worked, i run it but it says know this. i think this is ok right?

wow it worked , but it says i dont have any model

[ornate elk]

you have to link them from your models folder

in the webui-user.bat

you can add --ckpt-dir "D:\PathToYourModels" and --lora-dir "..." etc

[solemn atlas]

you mean edit this in tis one

[ornate elk]

no

[solemn atlas]

ah sorry

webui.user.bat

this one right

so this one needs to be removed?

--ckpt-dir %A1111_HOME%/models/Stable-diffusion ^

[ornate elk]

in the commandline_args= also add --cuda-malloc --cuda-stream

for the best performance

[ornate elk]

can you show your folder where your modls are in?

[solemn atlas]

sure, its here

[ornate elk]

then its easy

add --models-dir "C:\Stable diffusion\models" to the other args

should look like this

[solemn atlas]

where to add it? To psython, git, venv_ dir?

[ornate elk]

dont delete the other stuff from the bat lol

[solemn atlas]

i didnt do that haha

[ornate elk]

screenshot pls xD

[solemn atlas]

haha

[ornate elk]

looks okay

then install python 3.10.11 64bit
https://www.python.org/ftp/python/3.10.11/python-3.10.11-amd64.exe
and check "ad python to path" in the first screen.
Then relaunch the run.bat

[velvet lynx]

can i ask questions about controlnet here

[solemn atlas]

like this

[ornate elk]

no

when you launch the python installer

there is a checkbox

[ornate elk]

click install now

[solemn atlas]

ok

i think now everything is ok

i can see my models and checkpoints, is there a chance to change the color to black? the white one is killing my eyes xd

[ornate elk]

if you change your windows theme to dark it will also change in the browser

or you can add --theme dark to the webui-user.bat args

[solemn atlas]

in the commandline?

like this?

[ornate elk]

only --theme dark

[solemn atlas]

did not work

[hybrid magnet]

i tried millions of times installing and upadate , use chat gpt steps. still i cant generate images . showing some error .... what is the solution ?

[solemn atlas]

so run.bat works and SD opens but webui.user.bat does not work

[hybrid magnet]

how to raise tickets ?

[ornate elk]

its a scam

dont click on their discord link, dont give away personal information

[ornate elk]

yes only use the run.bat for forge

[ornate elk]

whats your gpu? what does the cmd shows?

[solemn atlas]

ohh okay, thats why

thank you for your help :))

are there tutorials for this version? like making 2 characters or stuff like that on yt you would say its good?

[bleak wagon]

Hey thank you for helping, forge worked directly.

[lime pivot]

is it possible for my character to do the open pose skeletal images using controlnet ? if so may i know the settings ? (specifically controlnet type, preprocessor and model :D)

stable diffusion 1.5

[lucid yoke]

If you have a lora for said character, then just controlnet openpose, no preproccor sinse you have the skeleton already

[lime pivot]

no lora yet unfortunately

its just one image. im trying to make a lora but i need a character sheet, which is my biggest problem right now xD

[lucid yoke]

Could try looking into 1 image loras, no guarantee though

[lime pivot]

alright ty will try that 😄

[whole sparrow]

Hey I fixed it,but I was getting the same error. Decided to install stable diffusion classic on stability matrix instead and it works fine now

[ornate elk]

alright 👍

[ionic spire]

Its an AMD 7900 XT.

[ornate elk]

ah okay then the issue is the new AMD 25.5.1 Adrenalin Driver

it doesnt work with Zluda

so you have to revert back to 25.4.1 or 25.3.1

[ionic spire]

Thats odd, I dont recall updating anything, must have done it on its own

Wait, no I updated it like a week ago, so why is it only now causing issues. I know for certain I updated it to that version and it was working fine

[ornate elk]

can you show the full cmd log?

[ionic spire]

I just woke up so brain not working, how do I do that?

[ornate elk]

just launch the webui and copy everything fro mthe cmd

and paste it here

[ionic spire]

Gotcha

So when starting up everything is fine

But yea once I try and do anything it gives that error so give me a min to get this all ready for you

This is just the error

and this is just startup

I recall that 2 days ago when I initially opened it up it shot out an error at start up but I didnt think much of it so I just closed it and reopened it and everything seemed fine but then it started giving that error.

[ornate elk]

can you check your AMD Adrenalin Driver Version?

[ionic spire]

It is the 25.5.1

[ornate elk]

yea then thats the issue

[ionic spire]

Alright, il see about reverting it

Alright, reverted to previous version and its acting differently at startup. Since I deleted my venv folder since I saw that was a potential solution its currently compiling which means its probably fine now.

[ornate elk]

yep thats fine

[ionic spire]

Yep, feels just like when I first set it up so it should be fine, thanks once again for your help.

[ornate elk]

no problem 🙂

[ionic spire]

Yep, everything working fine now

[olive topaz]

I am new to ComfyUI and was trying to use someone's workflow to make a video but I got this issue and dont know what to do.

[cobalt agate]

None type error message usually indicates a node can't find an external file like a model or a lora. Make sure you have downloaded the models the workflow references and placed them in the appropriate folders. Check out the many youtube offerings on how to use Comfy from step one.

[olive topaz]

thank you

[solemn atlas]

I have a question

When I started using it, it generated it it in just a few secounds which was CRAZY

but then all of a sudden it took minutes + my pc freezed again

scam?

[olive topaz]

I have a new error

actually I think jumping into someones workflow wasnt my best idea. Im gonna look up a guide so i have all the files in the correct place and start from there

[ornate elk]

Can you show the txt2img settings?

[olive topaz]

Hey can anyone help me figure out where/which file im missing? I am refering specifically to the LOAD CLIP section on the left, "type" should be wan but I dont have it on my list and not sure where/what to add

[upper linden]

now i get crashes when i make a photo (768x512) (7800xt AMD)

[upper linden]

so we have to revert the drivers for amd, also what version did you revert to?

[ionic spire]

I reverted back to the previous version, 25.4.1

I was talking with my friend about it and im pretty sure there was an update but the number didnt change because we both updated to 25.5.1 over a week ago but it says it was released on the 7th of this month.

[fervent oracle]

Reactor ka working url de sakta hai koi
Any one have working url of reactor?

[cobalt coral]

how can i add a power lora loader?

[fervent oracle]

any one give prompts dictionary

[frozen flower]

My suggestion is to ask any chat AI for the prompt you want to generate. AI will elaborate on your initial idea with an appropriate prompt. If you prefer doing it yourself, just remember generative AIs get confused with negatives, so only use positive. What I mean instead of telling the AI to do not do something, tell it what to do, example do not say in the prompt "not dark" instead say "sunny day" etc

[tepid crypt]

trying to generate her without horns, even though the lora has it as part of the character, i wanted to know if there's any way to remove them from the generation, i tried everything, using horns:0.01 and {horns}

[lucid yoke]

tried adding horns to the negative prompt?

never seen {horns} though, what Ui are you using?

[open fox]

did you tried with adding "wearing a large sombrero hat" in the positive prompt?

[tepid crypt]

yes

[tepid crypt]

i dont want her to have a headwear

[lucid yoke]

Its probably the lora then not labeling it as horns, try inpaint to remove it

[open fox]

when chosing a lora for a specific character which is trained by images of the character with horns you will have to inpaint or chose a anime girl with brown hair which does not have horns...

[olive topaz]

can anyone help? Ive looked at alot of guides but dont no one metiontions this. I think I am missing a "WAN" file in my ComfyUI/comfy_extras folder but cant find it

[open fox]

It sounds like you would like to use a wan model for video generation. If you use a current version of comfyui you should find a file called nodes_wan.py in the comfy_extras folder

[olive topaz]

I found someones nodes_wan.py file online and put it in the correct folder but it still doesnt show. I think this issue could be linked with my comfyui not able to update.

stashing current changes
nothing to stash
creating backup branch: backup_branch_2025-05-11_07_00_43
checking out master branch
pulling latest changes
Traceback (most recent call last):
File "D:\Comfyui\ComfyUI_windows_portable\update\update.py", line 76, in <module>
pull(repo)
File "D:\Comfyui\ComfyUI_windows_portable\update\update.py", line 19, in pull
repo.checkout_tree(repo.get(remote_master_id))
File "D:\Comfyui\ComfyUI_windows_portable\python_embeded\Lib\site-packages\pygit2\repository.py", line 286, in checkout_tree
payload.check_error(err)
File "D:\Comfyui\ComfyUI_windows_portable\python_embeded\Lib\site-packages\pygit2\callbacks.py", line 99, in check_error
check_error(error_code)
File "D:\Comfyui\ComfyUI_windows_portable\python_embeded\Lib\site-packages\pygit2\errors.py", line 66, in check_error
raise GitError(message)
_pygit2.GitError: 1 conflict prevents checkout
Press any key to continue . . .

[ebon frost]

Hey everyone! 👋

I’ve recently seen some examples of Gemini Flash 2.0 Experimental being used for image generation inside tools like ComfyUI, possibly through custom nodes or API integrations.
Does anyone know if there’s an official API, beta access, or any documentation on how to set this up?
Or maybe how to connect it via Vertex AI as part of a programmatic workflow?

Would really appreciate any insights or experiences you can share. 🙏
Thanks in advance!

[open fox]

Something like this:

https://github.com/ShmuelRonen/ComfyUI-Gemini_Flash_2.0_Exp

[silent jackal]

hi, I am new, and I have a problem that I don't undertand,I download new check points, but when i try them it only does random dots,i put them in the stable difusion folder that is in the model folder,what is my mistake?

[open fox]

guess the new models are based on pony or sdxl and you might need another VAE file

[silent jackal]

were can I get those, there is one you would asvise or I just picl one?

[open fox]

What for models did you download?

[silent jackal]

v1-5-pruned-emaonly (is the one it comed with the stable)

flux_dev
flux1CheckpointEasyTo_v10
i was tring

[open fox]

Flux needs another VAE File

[open fox]

https://huggingface.co/black-forest-labs/FLUX.1-dev/tree/main/vae

[silent jackal]

thanks,question did you know this by experience or is a was to know it? so in the future I can check for my self insted to bother someone else

[open fox]

By experience been here a while and it is far easier to keep up-to-date if you stick around. So you just need to learn the new delta from the latest models, controlnets etc.

[silent jackal]

ok,thank again

[olive topaz]

I still cant figure it out, I cant update, I cant find my node_wan for LOAD CLIP type even though its in the folder

could it be about this? Your ComfyUI isn't git repo. I see this on my ComfyUI Manager menu

[lime pivot]

guys when i upscale the image the face becomes weird. im using this as my character sheet for training lora, can i just ignore it or should i fix it ? does anyone have a solution to the weird face ?

first image: upscaled, 2nd image: original

[lime pivot]

can someone check my settings idk if inpaint face doesnt work with multiple characters in 1 image or im doing something wrong. its my first time using inpaint so idk

[magic trout]

Half-human, half-dead robot

[unborn moth]

Just one help needed please,
I was trying to train on custom sdxl model abd it would take 2hrs to just transfer the files to server, i tried gdrive method and in five minutes it was downlosded in oneTrainer/module/model folder, but when i start procees on my local device, it would again start to download entire thing at 100kbps, what am i doing wrong?
Where should i put it?

[signal sigil]

Hey guys. I was wondering if anyone could help me with this issue.
I'm trying to get my 5090 running on a LoRa training script gui, but despite my Cuda ver being 12.8, and my Pyth being 3.10.6 The GUI just says this

"NVIDIA GeForce RTX 5090 with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 sm_90.
If you want to use the NVIDIA GeForce RTX 5090 GPU with PyTorch, please check the instructions at Adress"

I wa wondering if there is a way to update it so that it may work? Does anyone here know?
Thx.

[main oyster]

is there any way tp consistently generate a person with the same features but in different poses and places? im very new to all this but i want to create some selfie-style pictures

[lucid yoke]

Best check in the onetrainer discord for detailed instructions tbh

[half wave]

Hi all! I'm very new to this space and just experimenting a bit – but I’d like to start my own long-term workflow for an AI character project.

Before diving in fully, I’d love to have one or two LoRA models trained by someone with more experience – just to see how far I can push the quality and learn what’s possible. I already have 30–50 images (generated via Astria AI) – same face, consistent style.

Happy to pay for the work (PayPal/Fiverr/etc.). If anyone’s open for it or knows someone, feel free to DM me. Would mean a lot. Thanks! I am from Germany and my „Girl“ will be a realistic Charakter“

[lime pivot]

@half wave unrelated, but may i ask if this "Astria ai" can generate consistent anime styled characters too? im looking in google and seems to be all realistic

[inland portal]

CLIP_stop_at_last_layers option is nowhere to be found. Is it changed?

[lime pivot]

sd 1.5 automatic 1111

i found a solution: just inpaint them individually, so its not a big problem anymore 😄

[half wave]

I think so but you have to Pay

[lime pivot]

@half wave you still looking for a Lora trainer ?

im in a sd discord and theres a lot of good people there that might be up for your job

[strong oar]

Hello, does exist a package ready to install with all plugin and models, lorss etc for windows nvidia intel ?

[lucid yoke]

All models and loras? Your going to need to have a few petabytes at that rate

But a easy install check the pinned messages

Models&loras on civitAI

[reef pond]

is there any chat here where I can find a specialist who works professionally with stable diffusion to involve in working on a game project?

[lucid yoke]

You can cast a line in #🌶｜off-topic & General chats but i recommend being clear in what you want and expected compensation since most won't bother otherwise

But if you want some guides on how to do it your self, most people would be willing to share experiences

[half wave]

Yes:)

[sacred depot]

Hi guys, sorry for bothering,

I have a 5070ti but i can't run stablediffusion UI
https://github.com/AUTOMATIC1111/stable-diffusion-webui

I tried following the guide but i am stuck with

NVIDIA GeForce RTX 5070 Ti with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 sm_90.
If you want to use the NVIDIA GeForce RTX 5070 Ti GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

But i can run sm_128

[lethal void]

Is there a way to get stable diffusion to work with intel arc b580 without using gimp?

[ornate elk]

Hey, try to install Forge Webui.
It should work with the 50 series

[lime pivot]

in img2img "image window" my window seems small, is this normal ? i want the window to fill the whole window (red square), right now the window is small (blue square)

sd 1.5 automatic1111

[daring reef]

is installing comfyui from pinokio the same as from the github?

[ornate elk]

Nope, better install it the normal way from github

[ruby yoke]

nope

[polar hemlock]

hey!, I just joined and I am new here but I was wondering if anybody can help me out with stable diffusion, my pc is very capable for generation but things like hands and feet are still a bit wonky, just send me a dm if you are able to help me out please!

[gusty tide]

Im trying to load flux 1-dev in cmfyui but whys it using my ram instead of vram?

[strong oar]

Hello, there is a way to use img2img to duplicate the character of an image ? I just try with training a model but my result is 99% a clone, I want the same face expression, same lips, etc and whatever specific prompt I wrote I dont get it 100%

[sudden pike]

Hi...
idk why but i just wanted to use Stable diffusion but it somehow uses my CPU

also i get "RuntimeError: mixed dtype (CPU): expect parameter to have scalar type of Float"

[ornate elk]

What's your GPU?

[sudden pike]

RX 7800xt

it worked days ago

[ornate elk]

Ah okay

[sudden pike]

was there a update to anything that fcked it up?

[ornate elk]

Yea the issue is because AMD Adrenalin 25.5.1 doesn't work with Zluda

[sudden pike]

ah

so should i downgrade?

[ornate elk]

If you want to use any stuff with zluda yes.
Or you wait until the Zluda dev makes a fix.
He is currently working on one

[sudden pike]

ah k

i will just downgrade

and maybe i update adrenalin in the next weeks

[ornate elk]

Yep, I'm still on 25.3.1 and waiting

25.3.1 and 25.4.1 should both work normaly

[sudden pike]

where can i download a older version?

ah there

thanks for your help ill try it out 😄

[ornate elk]

No problem 🙂

[sudden pike]

also u know what's up with the "(whatever:0.7)" stuff in the prompts?

#📝｜prompting-help

[ornate elk]

Yes thats the words strength

You can give it more importance or less.
word:1.3 or word:0.7
It goes from :-2 to :2

[olive topaz]

Hey I want to us Comfyui to make cartoon characters fight, but it isnt making a good art style for the characters. Is there a way to use a checkpoint or lora like StableDiffusion to control that?

perferably text to video but I can do Image to video. I tried Wan2.1 but it kept chaning the image

[viscid gate]

i justswitched to forge ui and its being a lot slower on my 4050 laptop, ive tried optimization methods but mostly it just loads low vram mode and moving model and its overall not as fast as a11111

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

whenerver i use euler a for some reason

[ornate elk]

Try add --xformers --cuda-malloc --cuda-stream to the webui-user.bat and relaunch.
Also if you have 16gb if ram, you should increase the Windows Pagefile

[viscid gate]

wow, asked a question here a year ago and u responded here u are a year later responding first again 😂 , I added that and what should i change the pagefile t

@ornate elk when i launch with those arguments, it no longer launches webui anymore

launch.py: error: unrecognized arguments: --cuda-malloc --cuda-stream

[ornate elk]

Hey, yea I'm still here haha 😁
Can you show me the webui-user.bat file? And is your forge updated?

[scenic sundial]

have the same gpu and faced the same issue, thats why using comfyui currently

[unborn moth]

Whats ideal epoch, repetition, network dim/alpha, unet lr and txt encoder lr for 200 photos?
I sought help from chat gpt twice and both the times it ruined it, one lora produces great results but result does not resemble the person in data set, second time it burned, all i can see is orange burn on image.
Chatgpt and grok both suggest something vastly different, i am running on cloud gpu, so cannot learn by trial and error a lot.

Any help is highly appreciated

Using OneTrainer cloud

[tired grove]

Hi, can anyone help me get this file? animatediff_motion_module_v1.safetensors

[ornate elk]

Hey, here:
https://civitai.com/models/239419?modelVersionId=270322

[maiden frost]

Having issues— Whenever i use IPAdapter with a face, the output is always a closeup portrait of face/head-and-shoulders, even if the prompt specifies otherwise. I have tried using negative prompt, putting weights on positive prompt, etc. Anybody else have this issue?

Same issue with different models too - juggernautXL, RealVis 50 Lightning, SDXL 1.0

[tired grove]

thank you so much !!!

[viscid gate]

yes its latest i just downloaded it yesterday

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --cuda-malloc --cuda-stream

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

call webui.bat

i mostly fixed the speed but whenever i use euler or euler a i get RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

[rigid stone]

i just found out about prompt from file or textbox script. is there no way to add cooldown between job if using that script?

[gusty tide]

ControlNetFlux.forward() missing 1 required positional argument: 'y'

anyone know how to fix this

[ornate elk]

Okay, then remove --cuda-malloc --cuda-stream
Then relaunch and show me the full cmd log.

[barren lily]

Does any1 here run forge on a AMD gpu?

[viscid gate]

launches normnal

from .dynamic_modules_utils import get_class_from_dynamic_module
File "C:\stable-diffusion-webui-forge-latest\venv\lib\site-packages\diffusers\utils\dynamic_modules_utils.py", line 28, in <module>
from huggingface_hub import cached_download, hf_hub_download, model_info
ImportError: cannot import name 'cached_download' from 'huggingface_hub' (C:\stable-diffusion-webui-forge-latest\venv\lib\site-packages\huggingface_hub_init_.py)

---

To create a public link, set share=True in launch().
Startup time: 71.4s (prepare environment: 30.9s, import torch: 22.2s, import gradio: 6.3s, setup paths: 3.1s, initialize shared: 0.4s, other imports: 1.8s, load scripts: 4.3s, initialize extra networks: 0.1s, create ui: 1.7s, gradio launch: 1.2s).
model_type EPS
UNet ADM Dimension 2816

[lime pivot]

does tiled upscaling for stable diffusion 1.5 cause more deformation when upscaling ? im forced to use tiled upscaling bc of low vram, so ive never tried norml upscaling and want to know the experience of other people

[sour lark]

I've been trying to resolve a bug in the FluxRegionAttention custom node for a few weeks now. I've looked everywhere for information, updated nodes and comfyUI, changed the code type to bring attn_mask to dtype, but it still doesn't work. Can anyone help?

[ornate elk]

Can you provide an example? With the settings

[potent sun]

enableHr":true,"hrUpscaler":"R-ESRGAN 4x+","hrSecondPassSteps":10,"hrResizeX":1024,"hrResizeY":1535,"denoisingStrength":0.5,"sdVae":"Automatic","sdxl":{},"ksamplerName":"dpmpp_sde_gpu","schedule":"karras","guidance":3.5} does anyone know how to put this settings in forge?

hrfix seems limited there

[formal glade]

Has anyone tried running framepack here?

[compact salmon]

hey gang. a few months back when i got my AMD 9070xt i did unofficial rocm 6.3 on it to use zluda and stable diffusion/comfyui. do i need to revert this if i no longer am interested in using these capabilities? idk if this stuff causes more instability and just want full stability back

ive had a few gpu driver crashes so looking to just go back to stock

[sterile shell]

Hello can someone help me with the download of stable diffusion. i watched multiple tutorials but none seem to work. i could use a hand.

[ornate elk]

Hey, rocm and zluda doesn't affect anything related to stability with driver or games. Its only used for ai stuff.

But if you want to revert the stuff you can just uninstall rocm and delete the zluda files. Thats all.

[ornate elk]

Hey, checkout the first link of the pinned messages in this channel.

There you find all Guides

[sterile shell]

aight thanks g

[ornate elk]

@pure tusk

[compact salmon]

thank you for confirming this!!! guess its just a couple bugs in th 9070xt firmware need to get ironed out. appreciate you cs1o youre a real one always

wouldnt be suprised if my -15% power limit could be causing a driver crash but damn is this oc/uv balance you can get on the 9070xt impressive if you can get it to not crash lol

runs at same performance as stock but with like a 80w lowered difference

[dull sleet]

Guys, where can I ask for help in this situation? A few months ago, my main discord was hacked and they started sending spam on the servers I was on. They banned me from the server here at Stabledifusion. I've tried to contact them through the official website to get me unbanned and I've never received a response. I'm using it here through the secondary profile. I've already resolved all the security issues on my main account with discord months ago, but I can't log into the server.

[lucid yoke]

Youd have the most luck pinging maxfield/fruit if thats the case though youd want to copy your user id from your main account so they can find yours easier if you do

The long number line

[dull sleet]

Thanks, I'll try to contact them.

[ornate elk]

NP, yep that card is awesome for undervolting!

[main rune]

@ornate elk not sure what to make of this
RuntimeError: CUDA unknown error - this may be due to an incorrectly set up environment, e.g. changing env variable CUDA_VISIBLE_DEVICES after program start. Setting the available devices to be zero. Deactivating virtual environment...

Running Zluda and it utterly failed one day, and this pops up

[viral crystal]

hello tech bros, my lora tab disappeared today and i can't figure out how to re-enable it, send help 😭

[ornate elk]

Hey, if your on Adrenalin 25.5.1 you need to download the latest Zluda version to make it work

Or either downgrade to 25.4.1

[ornate elk]

Hey, do you have any extensions installed?

[slate walrus]

Hi all, I’m using Illustrious XL for image generation, but I’m having a problem. Every time I create a character, they fill the whole frame — their head almost touches the top, and feet the bottom, even when I change the canvas size (like 1080x1950). It feels like the model always stretches the character to fit the full height. How can I get the character to appear smaller and more “inside” the scene, without filling the whole image? Thanks!

[chilly jolt]

Hello
I would like to get a refund for the payment error in the stable diffusion API credit, but the email inquiry is too slow to respond, so I am leaving a message here
I am in a hurry, so can you process it quickly?

[short hill]

Hey, I bought credits but it didn’t get applied to my account after charging me

Not looking for a refund, but am hoping to resolve this quickly. My business relies on it and will run out of credits very soon

[ruby yoke]

you'll need to contact customer support. there's nothing that this discord can do about that

[main rune]

I got it I think, but it still does this...do I need to remove the previous Zluda in some way first before that?

[viscid gate]

I have an issue where euler a and euler a just doesn’t work and I get runtime error expected all tensors to be on same device but found 2 devices cuda:0 and cpu

[lucid yoke]

Community server, no real staff here

Scam

[lucid yoke]

No real staff here, community server, i recommend reaching out thru official methods

[main rune]

Huh, glad my irresolute desire to not have too many social connections got me to evade that, was wondering why I needed to join a different discord

LOL

[lucid yoke]

Yeah its a common scam here that they will try to get you to leave

Sadly due server size its hard to moderate

[main rune]

For what exactly, most here are just free users of an open platform
But I guess malicious be what be

[lucid yoke]

Eh a user here fell for it once

They try to get you to link a bitcoin wallet

[main rune]

pfft

[lucid yoke]

Any problem you have is a "oh its a common issue, you need to link your btc adress"

[hybrid moat]

is it normal for most results to be really ugly/distorted? im using Automatic1111 and ive followed a bunch of youtube tutorials, like i'll be using the same checkpoint and loras and whatnot but their results are always gorgeous and mine are really ugly

and by ugly i mean lots of smudgy distortion mess

i've used the same prompts but i get ugly results - am i doing snth wrong?

[lusty fractal]

Hi Jinglebell. Im havin the same issue. i watched some tutorial vids on youtube and improved the quality but its still quite unrefined.

[ornate elk]

Hey, can you show an example?

[hybrid moat]

yes pls give me a min to boot up my computer

[lusty fractal]

So i hope im not disturbing you but can i get in on this topic? only if im not bothering you that is.

[ornate elk]

Sure you can also show an example with the settings you used

[lusty fractal]

Thanks

the thing is im gettin ok pictures. but im pretty lost when it comes to loras. im trying to create a velvette picture but its quite bad.

[ornate elk]

Which model (checkpoint) do you use?

[hybrid moat]

i am using this model https://civitai.com/models/1485631/bleach-style-tite-kubo and heres a result just now, i copied everything off https://civitai.com/images/70943739

ngl this is one of the better ones AHAHAH but the faces are still pretty uncanny and much worse than the displayed img from civitai

[lusty fractal]

this ones made by dreamshaper8. ive also tried pony xl,rev animated.this ones actually not that bad compared to the others

[hybrid moat]

im just wondering how these ppl get such gorgeous results even tho we r using the same prompts and whatnot

[ornate elk]

Which base model did you used? Because the one you linked is only a lora

And which resolution are these images

[hybrid moat]

im using https://civitai.com/models/827184/wai-nsfw-illustrious-sdxl?modelVersionId=1367422 same as the example i linked

[hybrid moat]

i also tried usign the same seed and i got weird results too

[lusty fractal]

512x512

[hybrid moat]

^ me too

[ornate elk]

Okay then an easy fix is to set the resolution to 1024x1024
Wainsfw is an illustrious model.
These are based on SDXL which was trained on 1024x1024.
So using 512 will look ugly

[lusty fractal]

ok thanks ill try that

[ornate elk]

Dreamshaper v8 is 1.5 based so its trained on 512x512

[hybrid moat]

OMFG THANK YOU

[ornate elk]

Try Nova Anime XL with 1024x1024 and it should look much better

[hybrid moat]

results are so much better king

🙏

[lusty fractal]

by the way could you show us an image you made that shows off diffusions capability?

[ornate elk]

Sure, here are some:

[hybrid moat]

YOO THATS SICK

btw how mang imgs do u need to train ur own lora?

[ornate elk]

I think like at least 20-40 more is better but also not to much. Less than 100 I would say.

[hybrid moat]

thanks sir 🙏

u rly helped me a lot

im genuinely so impressed with this

[ornate elk]

No problem, also a tip for illustrious models is to use Euler a as sampler.
Makes the image look cleaner imo

[hybrid moat]

THE FINGERS LOOK RIGHT????

[hybrid moat]

thanks o7

[lusty fractal]

is it normal that the images you guys posted are still loading?

[hybrid moat]

nah they loaded fine for me

[ornate elk]

Maybe restart discord

[lusty fractal]

the first ones from jinglebell loaded fine but now nothing....already did that.

[main rune]

I'm not sure why zluda doesn't seem to be working for me, it worked before then crapped out suddenly
Now I just get cuda related errors and device not recognized

[ornate elk]

Seems like some big services have connection problems right now

[lusty fractal]

well too bad...but im gonna try these models you recommended. thanks for the advice. i have to go to work now. may the force be with you 🙂

[hybrid moat]

bye dude

[ornate elk]

Alright cya

[main rune]

now I'm in the process of resetting my PC in the hopes it clears the issue, maybe for some reason my rx6600 xt is dying or some other stuff, but man the sudden fail blindsided me

[ornate elk]

Have you tried downgrading to Adrenalin 25.4.1 ?

[main rune]

I'll try that

[main rune]

Thanks, it worked
Probably should've left it to not auto update

[viral crystal]

i do not

nvm my lora is working again, i had "disable all extensions" checked for some reason lol

[main rune]

haha....I accidentally deleted my models while trying to move them, not sure how that happened 🥲

[main rune]

crap, my journey is not yet over

CUDA error: CUBLAS_STATUS_NOT_SUPPORTED when calling cublasSgemm( handle, opa, opb, m, n, k, &alpha, a, lda, b, ldb, &beta, c, ldc)

[potent sun]

does anyone know how to add eta noise into comfyUI?

ima lready in their server

[vocal burrow]

time to leave it then

this is a scam discord

freaking clown

[potent sun]

I didnt even checked the link I thought he was linking the comfyUI server lol

[thick pollen]

@potent sun always be careful

[ornate elk]

Which webui do you try to use?

[main rune]

comfy

[ornate elk]

Then you have to replace the 3 zluda DLLs in Comfyui

The one you renamed in C:\zluda

Copy them to venv/lib/site-packages/torch/lib

Ans overwrite if asked

[potent sun]

{"class_type": "KSampler", "inputs": {"cfg": 7.0, "denoise": 1.0, "ensd": 31337, "latent_image": ["10011", 0], "model": ["10012", 0],......... "sampler_name": "dpmpp_sde_gpu", "scheduler": "karras", "seed": 2354585644, "seed_mode": "A1111", "steps": 25}, "_properties": null},
Im trying to recreate an image and its seems its trying to use the a1111 workflow
but I cant see these ptions in the ksampler
ensd, seed mode
eta noise seed delta and seed mode
I have to add these values to the ksampler but I dont know how and I dont know if there are more custom ksamplers I could use

[potent sun]

is there a way you can generate the same image in comfy and in a1111? you can change how the noise works right?

are there any custom nodes?

[main rune]

Thanks CS1o

have a taco eating elf

[open fox]

the prompt could also be: "elf vomiting on a taco" 🙂

[main rune]

I blame the model not knowing what a taco is

[open fox]

Hey did you tried the screenshot function of your operating system? For a 1:1 copy it is fast and easy to use...

Just thought if you have time to ask in different discords even with a push every hour it seems important for you to get exact the same result. And as it seems that every feedback you get (like CS1o explaining Seeds etc.) does trigger your learning allergy. You could have created like 1.000 of own generations and optimizing the prompt to your needs in the same time. So i suggested an easy to use solution 🙂

[placid zinc]

Hello

not an idiot, I promise

I got SD working ( cpu only ) on an old laptop.

I resurrected an old mobo with an rtx3060

all it says is cannot load cuda

I deleted everything and now I'm at automatic1111 sd webui .. again

should sd.webui be in my user directory?

[hybrid moat]

is it possible to get to generate very niche and unique character designs? for instance i have ocs with specific scars, hats, etc that i designed and the results im getting are generally pretty generic. my im2im using my own art gives generic results as well, not the specific designs i want

[placid zinc]

c:\users\TFRTard\sd.webui

@hybrid moat if you have a favorite artist you can direct it " in the style of MoStAwSomePeEp"

in general you can use decently known references and it should pick it up

@hybrid moat after re-reading your post. I think you need to build a lora next

you need to train the system now

[hybrid moat]

ahhh

[placid zinc]

that's like, your only hope

[hybrid moat]

train it on my art? 🤔

[placid zinc]

yup

it's just images

[hybrid moat]

bet thanks

[placid zinc]

that's how the trainings work

[hybrid moat]

my art style is very amateurish 😭 if i combine loras (with a prettier art style) or just use a base model with a prettier art style will it keep the art style

like, i want it to use my oc designs but not my amateur art style

[placid zinc]

I cannot say, I haven't had the system set up long enough to test.. just lots of reading

so far my average gen has been 3hrs

I read an article that referenced something like 100 images to 150 images as a start

but more, more, more is always better

the numbers max for popular concepts.. too many images is a thing but you have to go pretty large I think

This new win10 build is weird with it's admin permissions, Maybe I should have gone ltsb

[dense tapir]

Deforum won't show up [Forge]

[main rune]

I have a feeling this isn't right