🤝｜tech-support

[dusty mortar]

it is too much ?

[ornate elk]

No dont use these settings

[dusty mortar]

ooh ok ok

didn't know

[ornate elk]

Each loras behaves different

So you have to just change their number in the prompt at the end and test

[dusty mortar]

OOOHO IT S WAY BETTER

[ornate elk]

Or read what the creator wrote as info

[dusty mortar]

thx for this

not used to this website

but it seems way more easier than comfyui

but it means that i have to learn a new one again XD

[ornate elk]

NP, yes it's much easier

[dusty mortar]

do you maybe have good videos how to upscale inpaint etc etc ?

[ornate elk]

Nope but its not that difficult

But its to late here. I need some sleep xD

[dusty mortar]

no worry bro

you helped me well here

goat

[ornate elk]

For any questions about prompting/upscaling/inpaint feel free to ask in #📝｜prompting-help

[dusty mortar]

thx mate

[ornate elk]

NP 🙂

[tawny saddle]

thought i said sum in here todae

in any case

boop

[dusty mortar]

Heey, thanks again @ornate elk Know pictures are much better, i just would have ask you if you guys can advise me some of the best extansions to get, even for the user interface or something for generating etc

[ornate elk]

Np, get the booru tag autocompletion extension
Its a must have for prompting with illustrious based models

[dusty mortar]

thanks : ) And so i guess it's possible to download other detailer on civitai and put them in forge so?

[ornate elk]

Yes you can download adetailer models from civitai

For example for eyes or glasses etc

Then put them into models/adetailer

Also you can download add detail loras

Based on this screenshot it seems like you either use the old forge version or reforge?

[dusty mortar]

https://github.com/Panchovix/stable-diffusion-webui-reForge
this one so reforge i guess

[ornate elk]

Yep thats reforge

A very good extension to also install is:
https://github.com/Haoming02/sd-forge-couple

[dusty mortar]

Oooh thanks for this one mate

[turbid willow]

Okay, so I followed the instructions on the pin and installed forge, got it running and opened, I clicked generate and again, same error, "Connection errored out" Lol, why does it hate me? XD

Gonna update my drivers and switch to a studio driver rather than a gaming driver. Maybe that will help, idk.

Updating didn't help, same error. Gonna restart my computer.

[ornate elk]

no dont switch to studio

they are esentialy the same drivers only that studio supports more cad stuff. and gaming more games. but otherwise in ai the work the same

[ornate elk]

here is somethin very wrong

why is there a forge folder in appdata

[turbid willow]

Cus I named it forge

and put it there

[ornate elk]

dont put it there

that can make a lot of issues

[turbid willow]

When I first installed it, it asked where I wanted to put it, where should I put it?

[ornate elk]

then you didnt followed my guide

because a manuall install dont ask for that

[turbid willow]

I'll redo it, I must have done something wrong lol.

[ornate elk]

python needs to be installed in its default path

which is appdata

but no webui should be there

[turbid willow]

deleting it all, will re-install in a sec.

Does it matter where I extract it to?

That was what it was, when it asked where to extract, I sent it to a forge folder I made in app data.

Just putting it in a folder on my desktop, hope that's okay.

[ornate elk]

Desktop is okay if its not OneDrive synced

[turbid willow]

I don't even know how to use onedrive, so I'll assume it's not synced.

same error, "connection errored out"

When I asked how to fix the error on google, seems it's mostly unresolve-able. I find a lot of forum posts with people who have this error, but the only solution I saw posted in forms was to add the "--no-gradio-queue" line, though that didn't seem to fix it for me either.

I saw a video on youtube that said to delete your venv folder and claims that would fix it, but I tried that a while ago when I was still using matrix, and that didn't work either.

Weird that I am having the same issue with my old matrix, my new re-installed matrix, and with forge now.

Endless "Connection errored out"

Someone else on a forum suggested turning off my firewalls as they might be blocking the web ui, but that didn't work either.

[turbid willow]

Ugh, tried to have gpt help me for a bit, but everything it wanted me to do didn't work either.

missing, unexpected = model.load_state_dict(sd, strict=False)
File "C:\Users\Yeyian\Desktop\Stable\webui_forge_cu121_torch231\system\python\lib\site-packages\torch\nn\modules\module.py", line 2175, in load_state_dict
load(self, state_dict)
File "C:\Users\Yeyian\Desktop\Stable\webui_forge_cu121_torch231\system\python\lib\site-packages\torch\nn\modules\module.py", line 2163, in load
load(child, child_state_dict, child_prefix) # noqa: F821
File "C:\Users\Yeyian\Desktop\Stable\webui_forge_cu121_torch231\system\python\lib\site-packages\torch\nn\modules\module.py", line 2163, in load
load(child, child_state_dict, child_prefix) # noqa: F821
File "C:\Users\Yeyian\Desktop\Stable\webui_forge_cu121_torch231\system\python\lib\site-packages\torch\nn\modules\module.py", line 2163, in load
load(child, child_state_dict, child_prefix) # noqa: F821
[Previous line repeated 4 more times]
File "C:\Users\Yeyian\Desktop\Stable\webui_forge_cu121_torch231\system\python\lib\site-packages\torch\nn\modules\module.py", line 2157, in load
module._load_from_state_dict(
File "C:\Users\Yeyian\Desktop\Stable\webui_forge_cu121_torch231\webui\backend\operations.py", line 140, in _load_from_state_dict
self.weight = torch.nn.Parameter(state_dict[prefix + 'weight'].to(self.dummy))
torch.cuda.OutOfMemoryError: Allocation on device
Allocation on device
Loading Model: {'checkpoint_info': {'filename': 'C:\Users\Yeyian\Desktop\Stable\webui_forge_cu121_torch231\webui\models\Stable-diffusion\illustrij_v11.safetensors', 'hash': 'dd2f59a2'}, 'additional_modules': [], 'unet_storage_dtype': None}

This looked bad? Idk. This time it said I was out of memory and THEN said "connection errored out"

Reloaded, this time it felt like it almost crashed my computer, whole screen went black for 5 seconds.

Tried again, said out of memory and then crashed

[turbid willow]

I redownloaded my model because gpt said they thought it might have become corrupted. Now I'm back to "connection errored out"

[balmy plover]

would this explain why i'm getting duplciate faces?

gpu usage

[turbid willow]

I get duplicate faces when my cfg is too high when doing img2img, are you doing img2img?

I also get duplicate faces and other body related mutations when the image size is too high. Different modules prefer different sizes, but for me with Illustrious, if I make the resoultion higher than 1024x1024, I get multiple faces, body parts in wrong places, and all kinds of mutations.

[balmy plover]

i'm totally doing image to image, i'm also also high res tiff images i rang though topaz. was hoping they would help it. and i was using a 8 cfng scale

[ornate elk]

Can help you later but which model do you use?

Make sure your not running anything else than the webui

[turbid willow]

Illustrij v11 atm

GPT seems to think it knows what's wrong and is walking me through some stuff now, I'll let you know if I manage to fix it.

[dusty mortar]

Heey i tried to inpaint for the adetailer but it don't change really something, any ideas ?

[turbid willow]

increase cfg?

[dusty mortar]

here are the settings

[dusty mortar]

adetailer cfg ?

[cobalt agate]

@turbid willow You get double faces when you exceed any model's resolution by more than 50%. Typically, you can run SDXL as high 1280 native, but beyond that expect anomalies. If you add a control net with a guide image, you can go higher, but stages of upscale are the general approach.

[turbid willow]

Try increasing the denoising strength?

Honestly, I prefer just manually inpainting as I find it works a lot better.

[dusty mortar]

tbh i'm pretty new on webui so i tried with what i knew, but you say that manually inpaint is better ? i can try

but don't know how to make it

XD

[turbid willow]

I can't even get my difussion to work at all right now, so I can't offer much help sadly >_< lol

[ornate elk]

At which resolution do you get the out of memory issue?
What's your GPU and update your nvidia driver

[turbid willow]

I just updated the drivers, it's a 4070 gpu, and I was only trying to generate in image with a resolution 512x512

[sand dawn]

I’m adding support for “cancel” API endpoints in my discord bot. Comfy was simple, you just post nothing to “/interrupt” and it stops whatever is happening.

Now I go check out the A1111-like webuis and that endpoint is expecting a few inputs including “session_hash”. Anyone have experience working with this endpoint, could maybe point me the right direction before I start deep diving into it?

[turbid willow]

I'm not getting the out of memory issue anymore atm, just keep endlessly getting the "Connection errored out" message whenever I try to generate an image.

I tried a different module to see if that would help, nope, same message.

[proven bay]

Any comfyui users here? Im hoping for anyone to help with an inpainting/masking question

[ornate elk]

How much RAM do you have?

[ornate elk]

Hey checkout #🧣｜comfy-ui channel

[proven bay]

thank you!

[turbid willow]

16gigs I think. But I've never had a problem before, this is a new issue.

I used matrix like a month ago, zero problems, and I used it on and off for almost a year just fine.

I just decided to open and run matrix again yesterday and got hit with the "connection errored out" message.

So then you suggested i switch to manual forge, but i still get the same error. Idk what's up.

[ornate elk]

16gb can be to less for sdxl and illus models

You have to manually increase the windows pagefile

Also if you have only 15gb or less space left on C drive it will affect the performance too

But I still have a feeling something different is still wrong

[turbid willow]

I have like, 250 gigs left on my c drive, and it's a sdd card, and my pagefile has already been set to the recommended size.

[ornate elk]

Ah perfect

Then its something different

[turbid willow]

yeah, I just don't understand what's causing the issue and it's weird to me that despite forge giving me a clear error, "Connection errored out", it's not clear to me, gpt, you, or google what the problem actually is. It's a toughy! lol.

Gpt seemed to think that I had some incompatible versions of things like pillow and nubs or whatever those are called and so it had me install some older versions claiming that it would be more compatible with forge, but that didn't fix the issue either.

I tried verious command lines like no gradio, xforms, medium ram etc etc, but none of those worked either.

I've tried using a different model, still no good.

I've tried reinstalling the entire forge, still no good.

restarted my computer, updated my drives.

restarted a few more times for good measure XD

I've tried turning off my firewalls, no good.

What else can I do at this point?

It was suppose to be easy Cries in dispair lol.

should I try a different web ui?

Should I go back to matrix

[sand dawn]

derp

[turbid willow]

maybe I should try reforge, or an older forge?

I don't understand why it's so borked, it seems like such a simple set up, click the link, down load the file, extract it, run the update, run the bat, done.

Is there something I'm missing?

[ornate elk]

Can you try setup Auto1111

With my guide?

[turbid willow]

Yeah, I can give that a go.

[sand dawn]

is supposed to be easy 😛

ReForge is probably best option if you don't care about Flux

[turbid willow]

The very first one on the top of the list?

[sand dawn]

Forge is if you want to use newer architectures like Flux, Chroma, SD3, etc

[turbid willow]

I wanted to do flux, but I was never able to get it to work so I stuck with illioustrious.

[sand dawn]

Flux has a few more things to know in order to use it... most of the finetunes do not have text encoder, VAE or CLIP baked in

[turbid willow]

yeah, I remember having to add my own vae.

seems like I was using an older version of python, updating now as I follow the guide for auto1111, maybe that was my issue?

I had .6, not .11

"In the first screen check "Add Python to Path"" - this was never an option made avaiable on the installer btw, at least not that I could see.

running the bat now, it's installing stuff.

it finished, the web ui opened, and shortly thereafter, "Connection errored out"

It haunts me, no matter what version I use, it's always the same error.

venv "C:\AI\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr 5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Version: v1.10.1
Commit hash: 82a973c04367123ae98bd9abdf80d9eda9b910e2
Launching Web UI with arguments: --xformers
C:\AI\stable-diffusion-webui\venv\lib\site-packages\timm\models\layers_init_.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers
warnings.warn(f"Importing from {name} is deprecated, please import via timm.layers", FutureWarning)
Calculating sha256 for C:\AI\stable-diffusion-webui\models\Stable-diffusion\illustrij_v11.safetensors: Running on local URL: http://127.0.0.1:7860

To create a public link, set share=True in launch().
Startup time: 20.8s (prepare environment: 4.5s, import torch: 8.0s, import gradio: 1.7s, setup paths: 1.4s, initialize shared: 0.3s, other imports: 0.8s, load scripts: 2.4s, create ui: 1.0s, gradio launch: 0.7s).
Calculating sha256 for C:\AI\stable-diffusion-webui\models\Stable-diffusion\illustrij_v11.safetensors: 6d37f828cd0b5df3e0b24dbcb17c4e05cb617ceffc97ad602c23ed95083e66ac
Loading weights [6d37f828cd] from C:\AI\stable-diffusion-webui\models\Stable-diffusion\illustrij_v11.safetensors
Press any key to continue . . .

If this helps./

[turbid willow]

Anyway, I didn't change a thing and now it doesn't wanna open the ui at all.

[ruby yoke]

switch to comfyUI

[turbid willow]

I might not be smart enough to use it, but I could try. though I'm affraid that I'll still just get the, "Connection errored out" message.

[ornate elk]

what browser do you use?

and can you try an other model file?

[turbid willow]

I've tried about 4 models now, I use google chrome.

[ornate elk]

any browser extensions installed on it?

[turbid willow]

nothing of consequence

trying one more time a fresh install.

[ornate elk]

i dont think it will change stuff

can you open up the cmd and run
python --version
and then
where python

then run
sfc /scannow
that fixes windows bugs

[turbid willow]

C:\Users\Yeyian>python --version
Python 3.10.11

C:\Users\Yeyian\AppData\Local\Programs\Python\Python310\python.exe
C:\Users\Yeyian\AppData\Local\Programs\Python\Python313\python.exe
C:\Users\Yeyian\AppData\Local\Microsoft\WindowsApps\python.exe

wont let me do sfc cus it says I need to be an admin

told it I was admin, running sfc now

Why is it that CMD says my version is .11, but when I run stable, it says .6?

[ornate elk]

ahh uninstall python 3.13

[ornate elk]

yes run cmd as admin then

[turbid willow]

Sure, how?

[ornate elk]

under apps

in the windows settings

[turbid willow]

Uninstalling, I figured it was something I'd do in the cmd window, gotcha.

okay, done.

it says .11 now in the bat window

it's doing a bunch of stuff

wow, taking a long time.

Still going, but it's thrown a bunch of errors so far it looks like.

ERROR: Exception in ASGI application

Says that like 5 times, I wonder if it's stuck in a loop.

TypeError: argument of type 'bool' is not iterable

ValueError: When localhost is not accessible, a shareable link must be created. Please set share=True or check your proxy settings to allow access to localhost.

[turbid willow]

I think it broke, it hasn't moved in a while.

I closed it and reclicked the bat

Doh, now it's saying I have .6 python again, weird.

It did open the ui this time though.

aaaand "connection errored out"

It's amazing how commited my computer is to not working XD

[ornate elk]

can you show me the cmd log again

[ornate elk]

you installed forge

[turbid willow]

Yes.

[ornate elk]

forge uses build in python 3.10.6

[turbid willow]

Yeah, gpt just explained that to me.

[ornate elk]

have you tried auto1111 ?

[turbid willow]

Yes, it does the same thing, connection errored out

Auto1111, forge, and stability matrix all give me the same "connection errored out" I even tried an older version of forge.

[ornate elk]

can you share me the cmd log of auto111 ?

[turbid willow]

I think I deleted that one already. I have so many versions installed my head is spinning XD

[ornate elk]

oh okay

[eager widget]

can anyone help? :/

[turbid willow]

Gonna start over, deleting everything, going with auto1111

[ornate elk]

if we get auto working again everything else then can be fixed too

[turbid willow]

Same error, "connection errored out"

I tried a second time and it just black screen crashed.

My whole computer is going ultra slow, I can barely move my mouse, and then the ui crashes.

I didn't even try to generate anything, didn't get that far.

[ornate elk]

can you recheck if your pagefile settings are still on 16000min and 24000max for C drive?

and disabled for any other drive

[turbid willow]

I only have one drive.

Let me check my page files.

[ornate elk]

and then make sure you dont run Wallpaper Engine or anything else in the background

[turbid willow]

Nope, I had nothing runing, I even closed discord.

It's set for 2920mb

Which is what it says is recommended.

set it to your values, trying again.

[sand dawn]

I don’t see where “face_swapper” is defined but at that line it is None

[ornate elk]

you have to restart the PC after setting these

[turbid willow]

Hot dog, that fixed it!

[ornate elk]

nice!

[turbid willow]

I actually generated an image for the first time lol.

That's so weird, so it was just my paging file the whole time?

[ornate elk]

yep

as you only had 16gb and tried an xl based model

[turbid willow]

I guess that makes sense, cus I lowered it a few weeks ago.

[ornate elk]

and pagefile of 3gb is too less

[turbid willow]

I lowered it cus it said it recommended I do so, silly computer.

But wow, thanks for sticking with me and for all the continued support ❤️

[ornate elk]

no problem ^^ i was already doubting myself as normaly always know whats the culprit when it comes to auto1111 or forge errors

[sand dawn]

CS1o has been on here helping ppl for plural years by now - they’ve seen it all

[turbid willow]

A legend

too bad I deleted all my stuff before I got to this point lol. I might switch to forge or back to matrix, cus auto1111 is missing a lot of features I'm use to.

[sand dawn]

A1111 is basically obsolete atm

[turbid willow]

Question, since upping my paging file fixed my problem, should I raise it even MOAR!?

[ornate elk]

you can increase it to 32000max but after that it wont make much of an impact

it also needs that much free space on C too

[tawny saddle]

infinity pagefile

wonder how gud itd work if u got an blank RAM pcb and instead of regular moduals, u installed nand flash, like youd have on an NMVe ssd instead

wit same controller for ram, like i mean, ig it would proly be slow ig xd, since it wouldnt be volatile

[tawny saddle]

keep in mind doe

SSDs typically have an limited read/write cycle

so, the more data blocks that are taken up by pagefile, the more of the ssd u ware out, and the more the ssd is accessed for memory/ram allocation services, the less itll last

SSDs typically last 5 to 10 years, but, more its used, esp in the case of a pagefile, the less itll last ;^

so, my recommendation is, use an SSD u dunt mind losing or it being in constant use by the system, that u dunt use for gaming, or storing important media

most SSDs firmware once the write cycle is met, u then can not write anymore, and it permanently swtiches to read only, so u can still read from it but cant write

however, by the time this happens, degradation of the oxide layers of the dyes will have lead to charge leakage and an increase in bit error rates, which will cause data loss and curruption to anything on the ssd xd

so, if u wanna use a massive pagefile, id use an SSD that u dunt mind losing, and only use it for pagefiling, and keep all u important data on another xd

if u dunt have avaliable SATA ports on ur main board, one thing u can do to prolong ssd usage whilst still using an pagefile is let Windows decide the size of the pagefile as its needed, so as to not take up unneeded data blocks of the NAND flash

which could extend the life of chu ssd :3

that is wut i do

you could also use an high density HDD for pagefile usage, since HDDs are consirably more faster now days than they used to be, they dunt have read/write cycles as SSDs do, they may not be as fast as an SSD, however they are cheaper and dunt have limitation, its just..theyre mechcanical, and if dropped like a centimeter and a half on a table, they'll prolly break or head crash xd, fragile lil things they are

still wouldnt recommend putting any thing important on it tho, not for the same reason i stated about SSDs, but bc theyre mechanical, and over a long time, they eventually suffer issues spinning up, or not spinning up fast enough to even initialize, or for wutever reason ur tower tips over, HDDs will prolly not like that, if u have power surges the read/write actuator may not make it to the park sector in time and scrape the platters, since they float on a cushion of air created by turbulant air of the platters spinning, remove the spin, u remove the cushion of air, actuator falls and leads to wut is known as a head crash, not gud nu

i simply dont trust HDDs for data storage at all tbh xd, but for pagefiling, is alr

[humble fable]

Anyone with extensive knowledge of stable diffusion and Ia creations? I'd like help.

[frigid topaz]

Does someone know how to start from sketch ? my results doesn't look like the pic i loaded

[untold flame]

Does anybody else get issues with comfy UI where it randomly freaks out and starts downloading 100's of GB's of different models, and the only fix it to completely nuke the install and redo the whole thing?

this is the second time its happened. The first time it downloaded over 250GB of useless models all over my PC. This time I caught it at just 34GB, but I have no idea how the hell to stop it without having to redo the whole install

[ornate elk]

Have you installed Comfyui trough stability matrix?
Screenshot your custom nodes folder

[untold flame]

I don't want to do any of that. I like having full control over my install of comfy, so I wanna keep it normal

I just nuked comfy manager and it seems to have stopped it (for now)

i remember that helping before as well

[ornate elk]

So it probably was a custom node

Comfyui by default doesnt do that

[untold flame]

I have no idea how, I only have these

Ok what the hell-

it just randomly started doing this out of nowhere

thousands and thousands of lines of that, WTF

Is somebody else using my install of comfy or something???

I did not just queue that

what in the actual hell is going on 😭

[ornate elk]

Close it down and remove the custom nodes

Then make a clean install of Comfyui + Manager.
Then install these extensions one by one to see which one breaks everything xD

[untold flame]

I just uninstalled all of the packages it installed itself in that freak out. Looks like this now

[ornate elk]

Custom nodes are always dangerous as they could give someone access to your PC or download malicious files.

Most are probably fine, but there is still a chance

[ornate elk]

You have to delete them from the folder

[untold flame]

I know, I deleted them, and thats whats left

[ornate elk]

Oh ok

[untold flame]

I have comfy loaded, and I am watching it now

this is what happened last time. Some random issue happens when I am trying to connect. I open and close comfy multiple times, and then it starts downloading 100's and 100's of GB of models, packages and stuff, and then starts freaking out

I think specifically getting rid of Comfy API or whatever fixed it last time as well

I am keeping an eye on it in use. I have a dedicated monitor with it pulled up right now so I don't miss any activity

[ornate elk]

Yea API stuff gives other custom nodes even more control to do things.

That could have triggered it

[untold flame]

it seems ok... for now

[tawny saddle]

bru

that's nightmare fuel frfr

[vocal dragon]

Hey! Does anyone know why i receive good images only with hires? (i am not saying about image quality, without hires my images does not even match prompt, just random shit appears).

[ornate elk]

hey, can you show an example?

best with settings like resolution, model etc

[fresh crystal]

I've been getting errors installing SD Next. Would this be a place I could ask for help on that?

[ornate elk]

yep, but most of us use other webuis than sdnext, expecially when your on AMD i would recommend staying away from sdnext xD

[fresh crystal]

Alright, other recommendations? I am on AMD and automatic1111 tends to break if i mess with it even slightly

[ornate elk]

whats your GPU?

[fresh crystal]

6750

[ornate elk]

ah nice gpu, then i would recommend either Auto1111 with Zluda or Forge with Zluda

[fresh crystal]

RX 6750 XT if that's necessary

ill look into forge

[ornate elk]

make sure you follow the updated install guides:

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

[fresh crystal]

thanks man. it's so hard to find help on this stuff from someone who isnt an asshole about it the whole time

[ornate elk]

no problem at all, feel free to ask me if you have any questions, i made and tested all of the guides

[fresh crystal]

those guides are excellent. legend

[ornate elk]

thx, np 🙂 with 12gb vram you can use models up to sdlx/pony/illus and if really needed smaller flux or sd3 models.
Also when you want to upscale you have to make specific settings or you will run out of vram. But let me know if your at that point.

[vocal dragon]

model

everything etc is by default

tried to change sampling method

and scheduler

no success

[ornate elk]

Illustrious models are based on SDXL and so they are trained on a 1024x1024 base resolution

512x512 will look very bad on that as its made for 1.5 models

and then select **Euler a **as sampler.

[vocal dragon]

lemme try

[vocal dragon]

omg i was that dumb

the only problem is resolution

...

thank u so much❤️

[ornate elk]

no problem 🙂

[fresh crystal]

I'm getting the same problem installing Forge as I am with SD Next. It finishes installing requirements and before launching hits me with the exception: Exception Code: 0xC0000005. Nothing I've found online has helped.

[ornate elk]

Can you share a full cmd log?

Its very late here so I can help tomorrow again. But its fixable

[fresh crystal]

Alright thanks. I'll try and figure it out here for now but my dms are open whenever you're ready. Thank you

[dusty mortar]

hey guys, i feel after the face adetailer, it's slighlty "removing" the lora i used before, any ideas ?

on reforge

[green plank]

why cant i find the hi res settings for img2img on stable diffusion forge webui

[ornate elk]

At the bottom under scripts its SD upscale

[green plank]

ahhh how come whenever ii turn it on my img is broken into sections its weird

whats the tile overlap what should that be set to? plus some models have other hi res settings i cannot find there like this : Hires upscale: 1.5, Hires steps: 20, Hires upscaler: R-ESRGAN 4x+ Anime6B,Denoising strength: 0.35~0.5

there is no option to change the steps, or strenth in the sd upscaler

[proud terrace]

hey, stupid question maybe, but I am using Forge with Flux Dev and some Loras require "Clip Skip: 1" or "Clip Skip: 2"
I did not find a tutorial for that and only managed to get Clip Skip work in the UI for Stable Diffusion.... Is Clip Skip not compatible with Flux?

[plain ridge]

Hi everyone.
I am a filmmaker, I wanna build a pipeline that generates consistent images, ensures consistency for every character, wardrobe, prop, and location across the entire project.
How can I build?

[ornate elk]

I have time to look at the issue. Feel free to share some more info when you get the error and with which settings.

[ornate elk]

Hires upscale 1.5 and hires steps are only available in txt2img with hires fix

In img2img upscaling works a bit different

[fresh crystal]

Sorry for the hold up, its very early where I'm at. The error occurs just before it starts compiling. The Webui never even opens. I've attached the entire log but if you want me to send it another way I can.

[ornate elk]

Can you open up a cmd and run
Pip cache purge
Then delete the venv folder and relaunch the webui-user.bat?

[fresh crystal]

I'll give it a shot though I have done three fresh installs

Yeah, still getting the same issue

[ornate elk]

Have you copied the rocblas DLL into the rocblas folder?

In the log it seems that its in bin folder not rocblas

[fresh crystal]

Sure enough its in the bin folder. I did see that step but I was looking in the library folder

So do I copy it into the rocblas folder or the library folder?

[ornate elk]

Rocblas

[fresh crystal]

Got it.
For the first time in a while it's making progress. Got hit with a new error.

[ornate elk]

Can you relaunch again

And then share the full cmd log

[fresh crystal]

By relaunch you just mean the webui-user.bat file?

Here it is

[ornate elk]

Okay delete the .zluda folder and the venv folder and then restart the PC

Then relaunch the webui-user.bat

[fresh crystal]

Can do

[ornate elk]

Did it worked?

[fresh crystal]

Still installing requirements

[fresh crystal]

No it literally just now finished and it didnt work

Here's the log

[proud terrace]

I am using Forge with Flux Dev and some Loras require "Clip Skip: 1" or "Clip Skip: 2"
I did not find a tutorial for that and only managed to get Clip Skip work in the UI for Stable Diffusion.... Is Clip Skip not compatible with Flux?

[stable nova]

Hey fellas! I've been trying to fix an issue with the **Dreambooth **extension for Automatic1111 for 10 hours now (Win10, both OG WebUI and Forge).

Here's what I've found

Running this before installing Dreambooth:
print("torch:", torch.__version__, "→ cuda", torch.version.cuda, "GPU:", torch.cuda.is_available())

Result:
torch: 2.1.2+cu121 → cuda 12.1 GPU: True

Running it after installing Dreambooth:
Torch: 2.7.1+cpu CUDA: None

Dreambooth appears to be downgrading the torch version, and I'm not sure how to stop it. I'm sure deleting the requirements.txt would fix it, but then it wouldn't get any of its requirements.

The error I get from this behavior:
Traceback (most recent call last): File "K:\Stable Diffusion\sd.webui\webui\launch.py", line 48, in <module> main() File "K:\Stable Diffusion\sd.webui\webui\launch.py", line 39, in main prepare_environment() File "K:\Stable Diffusion\sd.webui\webui\modules\launch_utils.py", line 387, in prepare_environment raise RuntimeError( RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check

Any ideas?

[ornate elk]

Hey, dreambooth is outdated and shouldnt be used.
It breaks the Auto1111 and Forge installation.
To fix Auto1111 and Forge, delete the dreambooth extension from the extension folder.
Then delete the venv folder.
Remove --skip-torch-cuda-test from the webui-user.bat and instead add --xformers for Auto1111 and --cuda-stream for Forge.
Then relaunch them.

For lora training use standalone Tools like OneTrainer or Kohya_ss

[fresh crystal]

Alright, after one more fresh install it looks like I got it working. If I have any more problems I'll bring them up but for now I'm good. Thanks for the help.

[stable nova]

Oh man, really? So I've struggled for 10 hours to fix a broken extension, eh? Oh well, it is what it is.

I'll go ahead with Kohya_ss instead. Thank you for putting me out of my pain! 🤝

[ornate elk]

Perfect, no problem

[ornate elk]

Yes sorry, but now your on the right track ^^ np

[stable nova]

Yeah, that's all the matters! By the way, when you say that Dreambooth is outdated, is it only the extension that's outdated or just Dreambooth in general? Should I even consider using it in Kohya?

[ornate elk]

its outdated in general. I havent used it in Khoya but if you can i would avoid it there.

in OneTrainer its not integrated

aswell as in FluxGym

It seems that there are forks or scripts of Dreambooth which still get updated. But yea its not recommended for Lora training.

[stable nova]

Cool, LoRa it is - thank you!

[green plank]

I was able to install hires fix extension for img to img from within the forge ui actually and its working well now... most of the time haha

[ornate elk]

ah okay xD

[compact void]

i cant download stable diffusion

can someone help me

@light elm

[compact void]

can you help me?

[deft egret]

I need some help and im having trouble installing stable diffusion

[vocal burrow]

gonna need a bit more info for both of your problems.

what are you trying to install ? when does it fail ? what does it say ? get some logs, etc

[deft egret]

This

[vocal burrow]

most likely you did not install git

or if you did a1111 base model / assets hardcoded urls broke again.

You should probably not install a1111 anyways. It has not been updated in a year. Install forge or reforge instead, checked pinned message for a guide on that.

[hidden hamlet]

I have installed InstantID and am running it, but it produces completely different faces than the main model. In the video I watched, the person it produces has the same face as the model in the photo he uses. what am i doing wrong

I will train characters in Kohya, so I will create reference faces. I have set up InstantID for different facial expressions on same face , but I cannot get it to work.

[fierce light]

I'm really new to this, I'm currently trying to install comfyui-zluda for my windows10 and rx 7700xt gpu for generating some AI images but I get this error. I already made some clean installs and tried following chatgpt advices on how to fix it but it's just going around in circles so I came here to help, How to fix this and start using it? Thanks in advance

[barren walrus]

Is there a separate discord for DreamStudio? Or should I ask here?

[ornate elk]

I can't help with Comfyui-zluda.
For this you should open up an issue on its github.
If you want any other webui working then I can help

[deft egret]

is it possible to uninstall a1111?

[deft egret]

i actually couldnt find the files of a1111 because i think i deleted it

[ornate elk]

Just deleting its folder is enough

[fallow iris]

Guys has anybody used Deep Live Cam?

I need help 😭

[lament root]

ok because ive been having tons of issues trying to do this myself, i have come here for help. im using SD Web forge UI, on my 5070 its giving me all kinds of errors. numpy errors, not being allowed to alter my GPU Weights because i cant use the SDXL version due to CUDA and Python and pytorch version missmatchs...my version being 1.10.1 but the SDXL versions being like 2.0? Not being able to Access/read/install things due to Header numbers being wrong versions being miss-matched memory issues and just all around trouble. Can anyone help me, before i just cut my losses and move to a different AI art UI

im calling upon anyone, or whatever Outer Old Gods powers that be for any information or help. as google has me all over the place from forum posts to reddit threads to "yes do this" to "no dont do that, do this instead" and just all kinds of frustrated heh..

[lament root]

another issue using A detailer, just always ends up with only a Face image over everything , like this said image. and it refuses to NOT do this. , but when i have it NOT used, i get faces which look like Neegan from WD took lucille to them and the end up looking like horribly ground beef faces.

TLDR Adetailer is Cropping my images once it does it passes, with prompt/without prompts doesnt matter what Res i set or Upscaling i do or dont do, to be just faces. when its supposed to be full body images etc.,

[ornate elk]

Hey, can you share a full cmd log?

[lament root]

how do i find that for you.

does forge have a logs dump folder someplace?

[ornate elk]

Nope, just copy the whole cmd text

[lament root]

ok so i just load my Forge UI and copy verything it does while it loads...then paste it here?

Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.10.1-89-g2174ce5a
Commit hash: 2174ce5afea90ca489d222f539988dcef59f1027
Launching Web UI with arguments:
W0716 15:57:41.976587 38492 site-packages\torch\distributed\elastic\multiprocessing\redirects.py:29] NOTE: Redirects are currently not supported in Windows or MacOs.
H:\sd.webui\system\python\lib\site-packages\timm\models\layers_init_.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers
warnings.warn(f"Importing from {name} is deprecated, please import via timm.layers", FutureWarning)
no module 'xformers'. Processing without...
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.
[-] ADetailer initialized. version: 25.3.0, num models: 10
Loading weights [6701e33ec4] from H:\sd.webui\webui\models\Stable-diffusion\neroStyle_v10.safetensors

To create a public link, set share=True in launch().
Creating model from config: H:\sd.webui\webui\repositories\generative-models\configs\inference\sd_xl_base.yaml
Startup time: 11.1s (prepare environment: 2.3s, import torch: 4.3s, import gradio: 1.0s, setup paths: 0.8s, initialize shared: 0.2s, other imports: 0.4s, load scripts: 0.9s, create ui: 0.5s, gradio launch: 0.5s).
H:\sd.webui\system\python\lib\site-packages\huggingface_hub\file_download.py:943: 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: Doggettx... done.
Model loaded in 20.5s (load weights from disk: 0.9s, create model: 1.4s, apply weights to model: 17.8s, calculate empty prompt: 0.2s).

[ornate elk]

hmm looks normal

but you can add --cuda-stream to the launch args in the webui-user.bat for more performance

when do you get errors?

[lament root]

ok i just booted that up it says my python is incompatable. im using 3.13.5 it needs 3.10

[ornate elk]

?? your log says its using 3.10.6

what other messages do you get? and where?

[lament root]

in the webui---user.bat it says it is.

the reason being the 50 series install says i need to use the newest python pytorch and cuda so i got the newest ones

3.13.5 python

[ornate elk]

no

[lament root]

and 12.8/9 cuda

[ornate elk]

3.13 doesnt support any ai stuff

its not needed for rtx50

currently python 3.11.9 has the best compatiblity

please uninstall python 3.13 it just creates issues

[lament root]

:S oh okay.

what about the CUDA stuff tho?

[ornate elk]

2.8 and 2.9 will work but 2.7.1 is the latest stable which also supports xformers

so i would go with 2.7.1

[lament root]

well it says to use pytorch 2.7 i need to use 12.8/9 cuda..

but ...python does it support that version?

and i need those for a 50 series gpu do i not?

[ornate elk]

cuda 12.8

will do it

pytorch2.7.1 cu12.8

[lament root]

okay. so i guess i just download the... "quick EZ" ones from github Forges main page?

[ornate elk]

they dont come with rtx50 support

[lament root]

alright then. , one second ill remove python and get the lower version.

[ornate elk]

Here you can find my Forge install Guide which also features the steps needed for rtx 50:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[lament root]

sigh....this wouldve been so much more helpful if it was like easily found. the git hub stuff is just...filled with tons of other info and bloat. and i didnt know what i needed or where to look.

[ornate elk]

yea true, but we have that link here too in the pinned messages of this channel

[lament root]

sadly i never look in pinned messages anymore. just discords do them always have ppl askin the same questions. so it never helps lol

ok so i have python 3.11.9 now

time to just reinstall forge , god uh for the 17th time now lol

iver spent 2 weeks trying to work with forge and add things. only to haver it break and not work

then to reinstall and try differtent things or have ppl tell me to use X version or Do this X code line fix.

[ornate elk]

yea thats rough

50 series is still new and the proper torch support is needed, and it changes constantly as they update torch and cuda xD

[lament root]

this is what it gave me when i started doing the command line things in your guide

[ornate elk]

thats very strange

then try go into the python/lib/size-packages folder and there delete every torch and torchvision folder

[lament root]

okies then

this is just with the first command line prompt in the 50xx Extra steps

i deleted every folder with TORCH in its name

in the sites packages folder

left all the "loose" files that had torch as you said just folders.

[ornate elk]

can you open up a cmd and run python --version
and then where python

then screenshot

[lament root]

it had both versions of python still

i got rid of the 3.13 for real now

[ornate elk]

did you installed python from microsoft store?

[lament root]

no

[ornate elk]

okay

[lament root]

the python website

[ornate elk]

then you have to relaunch the python 3.11.9 installer again

then click on Modify

next

and select "add python to environment variables"

[lament root]

thats the only one?

[ornate elk]

leave the other stuff at it is

[lament root]

apparently no torch audio?

anyways the rest seemsto be good

now isympy.exe issue?

[ornate elk]

okay and when you now launch it?

[lament root]

i havent i did what it said and ran that code to update

and it said the EXE was alrdy found and it was a warning.

but it installed

[ornate elk]

which exe?

[ornate elk]

yea ignore that

[lament root]

ok

alright...now it looks like others SD forge uis

do i need to do anything else? like in settings or extensions?

[ornate elk]

Adetailer and boorutag autocompletion are very good extensions.
Specific settings are not needed
Only If you want to adjust the file output paths or image names

[lament root]

okay

time to see if i still gfet weird issues or messed up faces. or iuf adetailer still crops the images to just faces like beforew

[ornate elk]

It shouldnt but let me know.
Adetailer only crops it for the inpaint process but you should still get the full image after its run trough

[lament root]

seemns to have worked and come out not cxropped

thank god finally.

[ornate elk]

Perfect

[green plank]

Can someone help me please, i cannot activate control net, i downloaded the lastest version from the gitub and installed it sucessfully but everytime goto extensions and click the checkbox to enable it and click reload ui it does nothing, when i get back in its not there and when i goto extensions the checkbox is unchecked again.

[lime pivot]

Hello, guys, I have a question. Is there a way to get Clip Skip 2 in SDXL Forge without going to the 'All' UI in forge? In A1111, I just changed it in settings, but I don't think that setting exists in Forge.
The reason is, when I go to the 'All' UI and generate an image with the same settings and prompts I used to generate an image in SDXL, the results and style vary so much. So if possible, I don't wanna use the 'All' UI and I want to just change Clip Skip 2 in the XL UI without switching.

[ornate elk]

Sounds like your using forge. Which already has Controlnet integrated and doesnt allow it to be there two times.

[random night]

hey, i just create a lora based on a real person, however when i tried to use it, it generate the same outfit every time, with how many prompt and clothes lora i tried. Some said i need to do captioning, but how can i do it though, and is there any better way to change the outfit of the character in my lora?

i trained it for sd 1.5 model

[upbeat atlas]

Hi, I have a question. Which one is better/faster to run on AMD 6900 XT windows, ZLUDA or Olive?
I've been using ZLUDA for a while now, it's good but I'm wondering if there is anyway to make the generations any faster. I'm currently getting 1.4 to 1.6 it/s on sdxl models with 832 by 1216 resolution. I've been wondering if there are any tricks to make it go faster.

[stray flare]

Hey bro i also want to face train a lora
And i generated a face, i want to generate consistent image(same face) with diff expression and angles

What is the best way bro
Pls help me if you can

I know A1111 and a little bit comfy ui

I did img to img also but im not satisfied

[random night]

i still pretty much a newbie for lora training, so i could not be any much help for you

[ornate elk]

Onnx could be a bit faster but it comes with a few disadvantages like needed model conversion and some extensions not beeing supported.
Zluda still offers the best in case of performance and usability.
But soon AMD will release TheRock which will be faster than zluda.

[stray flare]

@ornate elk BRO HELP ME 😁

[tawny saddle]

Tyler

I dare u to uninstall ur graphics drivers and reboot

uvu

[ruby egret]

nah i'm good. the tech gods are on my side

[tawny saddle]

xd

[ruby egret]

ROCm agent (my gpu) not found but SOMEHOW it is now using it to generate very successfully

i have no fucking clue. i'm not going to ask questions.

[tawny saddle]

ou that's gud, can call it magic

[zealous fog]

How do I install a custom upscaler to my stable diffusion?

[ruby egret]

click buttons before generating and it will automatically download the respective scaler after generating the image requested.then it will compile it and finally use it.

this is SD Next

generally under 'refine' category

[zealous fog]

I'm using the amdgpu one

[ruby egret]

SD next is better in general for ZLUDA, currently running 9070xt on it better than ever

fewer issues than ever

but you can download a refiner just fine on webui too

[ornate elk]

Download them and put them into models/Resrgan folder

[ornate elk]

Sdnext ui and settings is still horror xD

[zealous fog]

Mine is called RealESRGAN, is that fine?

[ornate elk]

Have you used hires fix before?

With an Resrgan based upscaler?

[zealous fog]

No I'm like brand new to this.

[ornate elk]

Ah okay then you have to use it one time so it will create the esrgan folder

Set hires fix to:
Upscale by 2
Hires steps: 10
Upscaler: Esrgan4x+
Denois: 0.5

Then generate an image

And then you should have the esrgan folder

[zealous fog]

the one I was gonna use is the 4x_fatal_anime_500000_G

[ornate elk]

Yea you can then put that into the folder and select it for the next image.
A restart is required for the upscaler to show up

[zealous fog]

So the one I showed the 4x anime is a esrfan file right

I don't have a ESRGAN4x+, would ESRGAN_4X work?

[ornate elk]

yep

[zealous fog]

Ok, I want to make real good anime realism

[zealous fog]

Sorry I had to shower, but thank you so much, it appears in the list now

[ornate elk]

perfect np

[prisma moss]

@random night u can dme

[random night]

Umm, do you call him or call me?

[prisma moss]

both

[prisma moss]

dm me

[stray flare]

i just installed controlnet and ipadapter
but it taking so much time for single image
GPU-rtx 4060

[vocal burrow]

laptop ?

[stray flare]

desktop

[vocal burrow]

what does the log says ?

what settings are you using ?

[stray flare]

can i dm u ?

[vocal burrow]

sure but I d rather do it here if possible. If you want to dm because of nsfw stuff I won t be able to answer in dm either as I m at work rn.

[stray flare]

i asked chat gpt

idk but something is broken.
someone recommend me to use forge
so im switching to forge now

[vocal burrow]

cant see much here... there s no input picture for your controlnet.
I can t see the prompt, the model, etc...
no logs either so I can t tell what s going on.

but it s probably a good idea to switch to forge / reforge anyways as A1111 is pretty much on life support by now.

[stray flare]

im going to try same settings on forge and tell u if it is working properly or not

[weary nacelle]

hi, i'm trying to get webui set up but it's being really difficult and giving me an error every time i try to run it. i can't seem to find any videos going over how to fix it that have worked, and i'm not nearly smart enough to figure out how to fix this by myself, and i'm really new to this stuff
is anyone free to help out

[prisma moss]

whats the error

[cursive hazel]

Hey, so I know the AMD RX 9070 and 9070 XT can't run SD with the newer AMD drivers. At least 25.6.1, I tested and it didn't work. I just saw there's a 25.6.3 though.

Is there a way I can tell if it'll work or not, without going through the hassle of installing and probably uninstalling drivers again?

[ornate elk]

The latest drivers work with SD, checkout the first link of the pinned messages for the install guide.

Go for "Forge with Zluda"

[cursive hazel]

No, no, I have Forge with Zluda installed and working! I'm on 25.5.1 right now, and that works. But last time I Tried 25.6.1, it wouldn't work. And from what I heard, that was just "these drivers don't work with Zluda, tough luck" kinda thing.

I just saw 25.6.3 was out though, and haven't tried that.

[ornate elk]

Hmm okay

Normaly zluda 3.9.5 supports 25.5.1 and above

[neat saffron]

trying to figure out how to fix this

[zealous fog]

I'd like to know, what's the prompt to make sure ai art doesn't make human ears

[karmic niche]

can someone help me fix python not being detected in the path when i try to run webui user

[karmic niche]

yeah

[zealous fog]

I am active, but I wanna know if its possible to get stable diffusion to remove human ears from your art?

[neat saffron]

yes, got taken down a rabbit hole by a scammer in this server

[karmic niche]

ive tried setting the python to the file location of my python.exe but it wont work

all of us?

or for the scammer

[neat saffron]

that discord link isnt legit i believe

[karmic niche]

yeah thats what i assumed

lol

why cant u js give us live support here lmfao

[zealous fog]

I clicked it by accident.

[neat saffron]

i know lol, its weird they dont wanna

[karmic niche]

its not no ip grabber

dw

[neat saffron]

just leave immediately

[zealous fog]

I left

[neat saffron]

they get you to make a ticket

then ask if you have a connected wallet

its so jank lol

[karmic niche]

400000 people in the server and its this dead 😭

[neat saffron]

i know lol

its terribel

i just wanted help setting up stable diffusion lol

[karmic niche]

same

[neat saffron]

mine used to work

but now that im on a new pc, its not working with python

[karmic niche]

mine wont even detect python for some weird reason

[neat saffron]

same

im tryna remember what someone said t ofix it lol

but clearly this place isnt useful lol

[karmic niche]

u getting this too?

[neat saffron]

yup

exact same thi g

[karmic niche]

i think i fixed it

chatgpt helped out

[neat saffron]

oh?

what did it say?

[stray flare]

Yo bro i need help your help

I generated a face, i want to generate that same face with different expressions and angles for TRAINING a Face LoRA

[chilly trail]

Just setting up Forge webui after a long time hiatus. I got an rtx 4080 16GB. Is there any commandlines args that I should use for better performance?

[ornate elk]

Uninstall every python version, then install python 3.10.11 64bit
And check "add python to path" kN the first screen

[ornate elk]

It seems you need python 3.11.9 64bit, so uninstall 3.10

[ornate elk]

Yep there is --cuda-malloc --cuda-stream

[chilly trail]

Thank you 🙂

[limpid marten]

when you try to generate it, this pops up. how to fix it?

[half island]

Hoi, I need a smol bit of assistance.

Decided to try run comfyui on my steam deck, and I'm one step away from running it, but I hit a small snag

How do I add python to path on Linux? Got the zipped python 312 downloaded and extracted where I want, but when I do export path to it, even pythonpath both when in regular konsole as well as venv for comfyui, then do the command to show paths, it only shows .local/bin something and not home/deck/python312 that I added.

[empty pollen]

any idea? @half island

[half island]

If I had an idea, I wouldn't ask lol

[thin vapor]

Help me bro, A message like this appears

[half island]

Oh, gotcha. What's the
Is channel for specifically?

LoL. No thanks.

And answers here wouldn't be? Rather ask here than some shady random users "helpdesk"

@eternal elbow might need to clean up that shady user acting like staff stating you can't ask for support here, but his 2000 member random "helpdesk" server instead.

[gleaming lodge]

HELP IM HAVING PROBLEM IN AUTOMATIC1111

MY PC SPEC
I5 12400 F
16GB RAM
RTX 4060(8GB)

the image rendering time is so f#ucking slow like generating single image taking 12 minutes
help i did everything i could 😭

using base model - epic realism / i also tried cyberrealistic still the same issue

[weary nacelle]

i've got a 7800xt if that helps, and on windows 11

[languid coral]

venv "C:\forge\webui\venv\Scripts\Python.exe"

INCOMPATIBLE PYTHON VERSION

This program is tested with 3.10.6 Python, but you have 3.11.9.
If you encounter an error with "RuntimeError: Couldn't install torch." message,
or any other error regarding unsuccessful package (library) installation,
please downgrade (or upgrade) to the latest version of 3.10 Python
and delete current Python and "venv" folder in WebUI's directory.

You can download 3.10 Python from here: https://www.python.org/downloads/release/python-3106/

Alternatively, use a binary release of WebUI: https://github.com/AUTOMATIC1111/stable-diffusion-webui/releases/tag/v1.0.0-pre

Use --skip-python-version-check to suppress this warning.

Python 3.11.9 (tags/v3.11.9:de54cf5, Apr 2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)]
Version: f2.0.1v1.10.1-previous-669-gdfdcbab6
Commit hash: dfdcbab685e57677014f05a3309b48cc87383167
Installing torch and torchvision
Looking in indexes: https://pypi.org/simple, https://download.pytorch.org/whl/cu121
Collecting torch==2.3.1
Downloading https://download.pytorch.org/whl/cu121/torch-2.3.1%2Bcu121-cp311-cp311-win_amd64.whl (2423.5 MB)
---------------------------------------- 2.4/2.4 GB 7.1 MB/s eta 0:00:00
Collecting torchvision==0.18.1
Downloading https://download.pytorch.org/whl/cu121/torchvision-0.18.1%2Bcu121-cp311-cp311-win_amd64.whl (5.7 MB)
---------------------------------------- 5.7/5.7 MB 7.9 MB/s eta 0:00:00
Collecting filelock (from torch==2.3.1)
Downloading filelock-3.18.0-py3-none-any.whl.metadata (2.9 kB)
Collecting typing-extensions>=4.8.0 (from torch==2.3.1)

Downloading filelock-3.18.0-py3-none-any.whl.metadata (2.9 kB)
Collecting typing-extensions>=4.8.0 (from torch==2.3.1)
Downloading typing_extensions-4.14.0-py3-none-any.whl.metadata (3.0 kB)
Collecting sympy (from torch==2.3.1)
Downloading sympy-1.14.0-py3-none-any.whl.metadata (12 kB)
Collecting networkx (from torch==2.3.1)
Downloading networkx-3.5-py3-none-any.whl.metadata (6.3 kB)
Collecting jinja2 (from torch==2.3.1)
Downloading jinja2-3.1.6-py3-none-any.whl.metadata (2.9 kB)
Collecting fsspec (from torch==2.3.1)
Downloading fsspec-2025.5.1-py3-none-any.whl.metadata (11 kB)
Collecting mkl<=2021.4.0,>=2021.1.1 (from torch==2.3.1)
Downloading https://download.pytorch.org/whl/mkl-2021.4.0-py2.py3-none-win_amd64.whl (228.5 MB)
---------------------------------------- 228.5/228.5 MB 7.8 MB/s

[stray flare]

bro install forge

[stray flare]

A1111 is trash

i have same specs like you

[weary nacelle]

does anyone know how to fix this
trying to run webui

[vocal burrow]

Please read the first three lines of what you just copy pasted

[vocal burrow]

Consistency without lora or very very precise prompt is not a thing. However, you can try to train a lora from a single image. Google "one image lora training"

[vocal burrow]

Don't use directml. Check pinned guide and look for zluda.

[ornate elk]

You need --xformers --medvram-sdxl in your webui-user.bat

[stray flare]

after 1 image training " i can lock the face consistency and then i can generate 30+ images in diff angle and expressions , for proper LoRA training right ? bro

[neat saffron]

Oki

[primal barn]

Hello everyone, I want to train LORA. What tutorials should I watch, or where can I get some information?

[weary nacelle]

everything seemed fine after running (trying to get sdnext) but i got this error

not sure what to do

[lime pivot]

@primal barn not4talent !

[neat saffron]

so now its using 3.10

[neat saffron]

this is the error i keep getting, installing 3.10.6 doesnt work, neither does updating my gpu

hey, im getting stable diffusion installed, however it looks like its only supported for other gpus, and not mine, howe can i fix this? I have a 5070 btw, so i think its too new

[vocal burrow]

A1111 does not work with rtx5000 series. Try forge / reforge / comfyui / etc

[neat saffron]

how would i get that? i assume i check the channel for comfyai?

[ornate elk]

Check the pinned messages first link.
There are Guides for auto1111 and forge for rtx50 series

[haughty pilot]

what, why? I was about to get 5070 ti (of 9070 xt) this month

[zealous fog]

So I want to get rid of this human ear from my cow, and I heard that inpainting could help, is there something I need to remove the ear from the art?

[prisma moss]

try putting extra ears in negative

[zealous fog]

what does extra ears do?

[prisma moss]

generate human ears on animal

at least in theory

[zealous fog]

do I still need to put in human ears i negative?

[prisma moss]

try without it

[zealous fog]

In inprinting, what's the prompt to remove human ears and replace it with hair?

[loud pond]

could someone help me with this error my control net doesnt work im using Illustrious. the error is below:
2025-07-20 00:19:21,248 - ControlNet - INFO - unit_separate = False, style_align = False 39/40 [02:27<00:03, 3.33s/it]
*** Error running process: E:\AI\stable-diffusion-webui\extensions\sd-webui-controlnet\scripts\controlnet.py
Traceback (most recent call last):
File "E:\AI\stable-diffusion-webui\modules\scripts.py", line 832, in process
script.process(p, *script_args)
File "E:\AI\stable-diffusion-webui\extensions\sd-webui-controlnet\scripts\controlnet.py", line 1228, in process
self.controlnet_hack(p)
File "E:\AI\stable-diffusion-webui\extensions\sd-webui-controlnet\scripts\controlnet.py", line 1213, in controlnet_hack
self.controlnet_main_entry(p)
File "E:\AI\stable-diffusion-webui\extensions\sd-webui-controlnet\scripts\controlnet.py", line 898, in controlnet_main_entry
Script.check_sd_version_compatible(unit)
File "E:\AI\stable-diffusion-webui\extensions\sd-webui-controlnet\scripts\controlnet.py", line 827, in check_sd_version_compatible
raise Exception(f"ControlNet model {unit.model}({cnet_sd_version}) is not compatible with sd model({sd_version})")
Exception: ControlNet model control_sd15_openpose fef5e48e is not compatible with sd model(StableDiffusionVersion.SDXL)

any help would me greatly appreciated thanks in advance

also please @me if you manage to find out whats wrong thank you

[ornate elk]

Your trying to use an Controlnet model which is trained for SD 1.5 to use with an SDXL bases model

That won't work.
There are seperate sdxl based controlnet models needed

[loud pond]

where can i find this?

ive just been using the same thing for pony too

and for pony it was working

[ornate elk]

Here:
https://civitai.com/models/136070?modelVersionId=267493

[loud pond]

thank you!

[ornate elk]

No problem

[loud pond]

there are multiple open poses which one should i get?

[ornate elk]

Depends on the model you try to use it with

There are seperate openpose models for Noob-ai and illustrious too

[loud pond]

is there a general one?

[loud pond]

because im just using illustrious

[ornate elk]

Yea Controlnet Union+ Promax

[ornate elk]

This one is for illustrious:
https://civitai.com/models/1359846/illustrious-xl-controlnet-openpose

[loud pond]

thank you, ill download this and see if it works now

[weary nacelle]

trying to install sdnext but get an error when I try to run it

mb thanks

[stray flare]

yo bro im getting vram issue in kohya

is there any way to fix this ?

[livid jetty]

when i use comfyui with zluda, i get a "reconnecting" error and then the command window closes, any ideas whats going on?

[dense tapir]

Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)] Version: f2.0.1v1.10.1-previous-669-gdfdcbab6 Commit hash: dfdcbab685e57677014f05a3309b48cc87383167 Traceback (most recent call last): File "C:\Users\vence\Downloads\webui_forge_cu121_torch231\webui\launch.py", line 54, in <module> main() File "C:\Users\vence\Downloads\webui_forge_cu121_torch231\webui\launch.py", line 42, in main prepare_environment() File "C:\Users\vence\Downloads\webui_forge_cu121_torch231\webui\modules\launch_utils.py", line 434, in prepare_environment raise RuntimeError( RuntimeError: Your device does not support the current version of Torch/CUDA! Consider download another version: https://github.com/lllyasviel/stable-diffusion-webui-forge/releases/tag/latest Press any key to continue . . .

How can I fix this?
I have a 3050 in my laptop and an AMD Radeon (TM) Graphics, CPU is AMD Ryzen 5 and 16 gigs of RAM (Windows 11)

[ornate elk]

Update the graphics card driver should fix it

[dense tapir]

Thank you so much 🐐

[weary nacelle]

is anyone able to help me set up sd.next properly?
every time i follow any guide there's always something that goes wrong at the very end so i was wondering if someone could walk me through it and help me fix anything that goes wrong
i've got an amd rx 7800 xt and a ryzen 5 7600x

[ornate elk]

Hey, dont use sdnext, better use Forge with ZLUDA for AMD.
You find a setup guide on the pinned messages first link

If you have any questions feel free to ask

[weary nacelle]

forge's setup looks devilish but i'll give it a shot

what's wrong with sdnext though?

[ornate elk]

Sdnext does a lot different then other webuis (Auto1111, forge, Comfyui), so you find less help if you have any issues, questions etc.
Also its UI and settings panel is one of the worst ive seen xD

[weary nacelle]

oh damn ok

ty for letting me know

do you reckon if i was having problems setting up sdnext i'd still have issues with forge?

[ornate elk]

Depends on the issue

I don't think you will have any issues setting up forge with ZLUDA. But there can be some stuff like third party antivirus or a broken windows update which can cause issues

But that can be figured out if you run into an issue. Just post it here then I'll take a look

[weary nacelle]

alright, tysm legend

[ornate elk]

Did it worked?

[weary nacelle]

yeah, i've got the ui up

just looking for a model now to see if it actually uses my gpu

praying

[half island]

Meanwhile, i got comfyui to work on my steam deck, but blasted thing crashes it's desktop environment as soon as i hit generate lol. If anyone got any tips, i'm all ears :P Using the launch command for amd 7000 series, as well as --low-ram

[summer socket]

hey, so i got to this step when installing zluda in the forge with zluda for amd webui guide:

Launch the webui-user.bat. It will install everything and open in your default Browser.
Noteworthy: Zluda will show Compilation please wait.... The first image generation can take from 15 up to 40 minutes. Compilation is needed for zluda to work and will appear again when Zluda, Torch or Driver Versions change.

when i launch the .bat the last couple lines of cmd say this

File "C:\Users\myname\SD-Zluda\venv\lib\site-packages\torch\cuda_init_.py", line 363, in _lazy_init
raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled
Press any key to continue . . .

so i click a key and nothing happens

what is the issue.

chatgpt says this but i just wanna check here first

Open a command prompt inside your SD-Zluda folder
(Type cmd in the File Explorer bar at the top)

Activate the virtual environment
Run:

venv\Scripts\activate

Uninstall the existing PyTorch
Run:

pip uninstall torch

(Repeat if it says more than one torch install)

Install a CUDA-enabled version of PyTorch
Run:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

This installs PyTorch with CUDA 11.8, which is the version most compatible with ZLUDA.

[ornate elk]

Can you share the full cmd log?

[haughty pilot]

which UI would you recommend for 9070 xt and which one for 5070 ti?

[ornate elk]

Forge for boath.
For video gen on nvidia and AMD, Comfyui

[haughty pilot]

my friend has 9070 xt, was trying to set up zluda but it doesn't detect graphic card, do you perhaps have a solution?

[ornate elk]

Yea the solution are my AMD Guides. 😄
First link of the pinned messages in this channel.
If he get an error feel free to ask me here

[ornate elk]

Make sure the AMD driver is updated and the latest zluda files get used and that HIP SDK 6.2 is installed + the custom gfx files for gfx1201 are needed

[haughty pilot]

he did get an error with something not detected on the last step, will let you know when he do that again what exactly the error was

btw. is 5070 ti much better choice for all that? i myself need to choose between 9070 xt and 5070 ti (which is unfortunately almost $200 more)

[summer socket]

venv "C:\Users\jenso\SD-Zluda\venv\Scripts\Python.exe"
WARNING: ZLUDA works best with SD.Next. Please consider migrating to SD.Next.
fatal: No names found, cannot describe anything.
Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr 5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Version: f2.0.1v1.10.1-1.10.1
Commit hash: ef7f1aff1005f4d166c7af3ec5c05c40fe47feab
ROCm: agents=['gfx1036', 'gfx1101']
ROCm: version=6.2, using agent gfx1101
ZLUDA support: experimental
ZLUDA load: path='C:\Users\jenso\SD-Zluda.zluda' nightly=False
Launching Web UI with arguments: --use-zluda --cuda-stream --attention-quad --skip-python-version-check
Traceback (most recent call last):
File "C:\Users\jenso\SD-Zluda\launch.py", line 54, in <module>
main()
File "C:\Users\jenso\SD-Zluda\launch.py", line 50, in main
start()
File "C:\Users\jenso\SD-Zluda\modules\launch_utils.py", line 675, in start
import webui
File "C:\Users\jenso\SD-Zluda\webui.py", line 21, in <module>
initialize_forge()
File "C:\Users\jenso\SD-Zluda\modules_forge\initialization.py", line 58, in initialize_forge
from backend import memory_management
File "C:\Users\jenso\SD-Zluda\backend\memory_management.py", line 133, in <module>
total_vram = get_total_memory(get_torch_device()) / (1024 * 1024)
File "C:\Users\jenso\SD-Zluda\backend\memory_management.py", line 100, in get_torch_device
return torch.device(torch.cuda.current_device())
File "C:\Users\jenso\SD-Zluda\venv\lib\site-packages\torch\cuda_init_.py", line 1026, in current_device
lazy_init()
File "C:\Users\jenso\SD-Zluda\venv\lib\site-packages\torch\cuda_init.py", line 363, in _lazy_init
raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled
Press any key to continue . . .

[ornate elk]

Nvidia is currently better for ai stuff, AMD works okay too with ZLUDA and will work better when AMD releases the next HIP SDK 6.4 with native pytorch support. So zluda won't be needed.

If youe going mainly for Ai stuff go with Nvidia. Also if you plan to do business stuff with ai.
For gaming boath GPUs are mostly equal. Some games run better on Nvidia and some run better on AMD.
If you like Linux or in Future may want to switch to Linux, then your much better going with AMD as their drivers are Open Source and run native on Linux.

[ornate elk]

Delete the venv and the .zluda folder from the stable-diffusion-forge folder.
Then relaunch the webui-user.bat

[summer socket]

when you say the stable-diffusion-forge folder, you mean the sd-zluda folder right?

[ornate elk]

Nope

[summer socket]

where can i find this folder then

[ornate elk]

The stable-diffusion-webui-forge-amdgpu folder

Inside the SD-Zluda folder

[ornate elk]

Oh now I see it

You git cloned it wrong

[summer socket]

that is my sd-zluda folder

[ornate elk]

Okay delete everything in the SD-Zluda folder then please redo step 2 and 3, 4, 5

[summer socket]

ok lol

including the webuibat?

[ornate elk]

You then have forges folder inside the SD-Zluda folder.

[summer socket]

do i need to redo the forge install thing aswell

so i just go from the beginning?

[ornate elk]

Yea but if you done the zluda steps correctly you dont need to redo these

[summer socket]

ok thank you, i hope i dont mess up this time!

[ornate elk]

No problem these steps shouldn't take long

Let me know how it goes

[summer socket]

TypeError: argument of type 'bool' is not iterable
Exception in thread Thread-28 (webui_worker):
Traceback (most recent call last):
File "C:\Users\jenso\AppData\Local\Programs\Python\Python310\lib\threading.py", line 1016, in _bootstrap_inner
self.run()
File "C:\Users\jenso\AppData\Local\Programs\Python\Python310\lib\threading.py", line 953, in run
self._target(*self._args, **self._kwargs)
File "C:\Users\jenso\SD-Zluda\stable-diffusion-webui-amdgpu-forge\webui.py", line 101, in webui_worker
app, local_url, share_url = shared.demo.launch(
File "C:\Users\jenso\SD-Zluda\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\gradio\blocks.py", line 2446, in launch
raise ValueError(
ValueError: When localhost is not accessible, a shareable link must be created. Please set share=True or check your proxy settings to allow access to localhost.

this is the last couple lines after clicking the webuserbat for the first time in this step

Launch the webui-user.bat. It will now install everything needed and when it's done the Webui opens in your default Browser.

@ornate elk

[ornate elk]

can you copy and share the full cmd log?

also relaunch the webui-user.bat once

[summer socket]

ive just relaunched it

WOAH it worked now

the web ui opened, now ill do the zluda part

which one to click

oh im silly, the one that says download

[ornate elk]

no not the one

oh wait

this one:

i have to edit the guide to make that clearer thanks

[summer socket]

ok, thanks

@ornate elk what does it mean by the zluda folder on the c drive

Unzip the ZLUDA-windows-rocm6-amd64.zip and move the zluda folder onto the C: Drive and rename it to ZLUDA

[ornate elk]

you unzip the .zip file and you get a zluda folder

move that to the C drive main directory

[summer socket]

does it need to be capitalised

[ornate elk]

doesnt matter if you add the same name to the path later

[summer socket]

i cant restart my pc right now, but i created the paths earlier

when i restart it do i just click on the webuserbat to open the webui

[ornate elk]

you need to rename the folder here

because you didnt renamed it before on C

make ZLUDA to zluda

[summer socket]

thank you so much for all the help!

also do i need to install models to start generating images

because ive used ollama before and you needed to install models for that

[ornate elk]

yep SD webui needs also models to work with

you can get them on Civitai.com the largest database for image gen models

[summer socket]

how do i know which one to use, ive heard sdxl is best

[ornate elk]

yea 1.5 models are outdated and sdxl based ones are much better if the gpu cna handle it (yours can)

for anime go for Illustrious based models

[summer socket]

@ornate elk how do i start generating

are there any settings i need to do before starting

[ornate elk]

you have to select XL at the top left

and put the models (checkpoints) into model/stable-diffusion folder

[summer socket]

so do i just click the download button next to the create button

[ornate elk]

exactly

[summer socket]

ive done that

[summer socket]

so how do i generate stuff now, are there any settings needed?

[ornate elk]

you can test and write something into the prompt field

then hit generate

in the negative prompt field you write stuf fyou dont want in the image

[livid jetty]

is there a simpler comfy ui? i get a little overwhelmed with it

[ornate elk]

there is SwarmUI which adds a UI on top of comfyui. its maybe easier to use

[livid jetty]

just did it, it looks way better but it says i have no backends and comfyuiseflstarting has an error

[ornate elk]

you have to checkout the guide for AMD how to set it up (pinned messages first link)

[storm knoll]

Hey can anyone give me a ballpark number on how much faster a 5090 w 32gb vram is compared to a 4070 super 12gb? Specifically at generating/detailing/upscaling images. Thanks

[brittle lantern]

Does anyone know about external drives that I can use to store models and loras an stuff and run them off I would want one for my MacBook and just in general to use on my devices when doing AI stuff idk where else to ask this

[hybrid swift]

Well i need help regarding stability ai api, i am developing a discord bot i made a commnad (powered by Stability AI API) isnt working even after following the documentation

Uh

Is this official

?

[ornate elk]

Wait

Thats scam

[hybrid swift]

I see

I already figured why he sended discord app protocol

[hybrid swift]

.

[ornate elk]

Can't help much with the api sadly

[hybrid swift]

Hmm so is their any way i can get support regarding the api?

[ornate elk]

Stability has a support email on their website regarding their paid products

[hybrid swift]

Well API is free

[ornate elk]

Ah found it:
https://kb.stability.ai/knowledge-base/kb-tickets/new

Worth a try to submit a ticket there.
I guess they answer for the free API too

[hybrid swift]

Ok

[ornate elk]

This discord is just a community discord so you won't get answered by stability ai staff here.

[hybrid swift]

Ok ok

[ornate elk]

You can go for normal external USB Hard Drives like WD Elements if you need to store 2-4TB for cheap.
Or you can go for external SSDs like the Samsung T5, T7, T9, Sandisk Extreme.

Or you can buy an M2 SSD Enclosure and any cheap M2 SSD and build yourself a superfast External Drive.

[ruby egret]

is there any way to train lora with sd.next?

[ornate elk]

Nope but you can use OneTrainer

[zenith geyser]

whats causing this? its doing it on multiple models

[ornate elk]

Can you share the settings you used?

[zenith geyser]

another example:

prompt, using:

1girl, petite, adult, bratty, sassy, black hair, makeup, ponytail, attractive face, cute face, cheeky smile, smirk, teasing, penthouse background, black crop top, short black shorts,

No negative

Euler A samping, clipskip 1, upscale latent. 25 sampling steps

its something to do with the upscale

[ornate elk]

Resolution, model and hires fix settings would be interesting

[ornate elk]

Ah okay "latent" upscaler can cause this.
Try latent bicubic and denois on 0.5
Hires steps 10

[brittle lantern]

What would I recommend for speed/value wise something that isn’t to slow but something that isn’t expensive

[ornate elk]

Depends on the size you want. I would go for a 2TB SSD.

[weary nacelle]

can i use any lora with any model?

[cobalt agate]

No

[green plank]

Hey guys I'm in dire need of help! I've been using stable diffusion for over a week now and I've developed a crippling addiction to it.

It was working fine, it would crash every now and then prob 2 - 4 hours apart but i didint think anything of it, and i could tolerate that.

Then last night idk what happened but tried to generate a picture using a new model i just downloaded and it crashed and the screen went black for a moment. Then my desktop disappeared with all its icons, although I could still run everything else fine I just couldint use my desktop. I restarted and my desktop came back and i tried again with the same model, it crashed again, but the desktop didn't dissappear that time. I tried again with another model, then another model, then I went back to the first model I download that I used the most surely that would work.... but it didn't. Nothing works... I might get one picture to generate, but not often it just keeps crashing. Please do anybody have any idea what might be going on and how I can fix it? Would a simple reinstall work? Please any help would be greatly appreciated 🙏

[haughty pilot]

he managed to do forge with zluda but its very slow, comparable to my rtx 2080, is it supposed to be that bad currently?

set COMMANDLINE_ARGS=--use-zluda --cuda-stream --attention-quad --skip-python-version-check

those are his options

[ornate elk]

As the gfx files are not optimised cause they are for HIP SDK 6.2 its not as fast as possible.
The 90xx series need the new HIP SDK 6.4 which will be released for Windows in August. Then it will be much faster as no custom gfx files are needed and AMDs native TheRock support will be available too.

How much it/s does he get on a 1024x1024, 30 steps, sdxl, Euler a, image without upscaling?

[ornate elk]

Hey, which webui, which model and what GPU are you using and how much normal RAM do you have?

[haughty pilot]

20 steps instead of 30 but everything else matches

[ornate elk]

Thats to slow. Can you show the full cmd log?

[ornate elk]

Thats not the full log and also is power shell

[ornate elk]

The webui should run in cmd

Also make sure he has 32GB RAM and not running Wallpaper engine or games in the background

If he has only 16gb of RAM he needs to increase the windows pagefile

[brittle lantern]

https://amzn.eu/d/isELCTk

this?

[ornate elk]

Yes this is a good one

If you want to go cheaper go for the sandisk one or buy an m2 SSD enclosure and a m2 drive from WD, Kingston etc

[ornate elk]

Okay so I compared it with a user on a 9070 non xt.
And he gets nearly the same speed currently.
Like I said thats caused by the missing hip SDK 6.4 support and the need of the custom non optimised gfx files.

What your friend can try if he want a better speed is to try the Experimental TheRock Guide. Also on my Github.
With that he should get a better speed.
But yea the full power currently is only available on Linux as there is already rocm 6.4 support and later on windows when hip SDK 6.4 releases.

[brittle lantern]

What is the 9070 performance it can compare to which nvidia GPU?

[ornate elk]

in gaming the non xt maybe to a 5070, the 9070xt to a 5070ti

[brittle lantern]

Got it, in image gen?

[ornate elk]

dont know as i have to less data

[brittle lantern]

Where would u even find the data people in the server or is there a doc or website for that?

[ornate elk]

people on the server or on github

[brittle lantern]

Understood whenever we get more info on it I’ll consider the 9070 XT or 9070 for now I’m considering spending £700 instead of £600 for the 5070 TI over 9070 XT

[haughty pilot]

same, I think I will just grab the 5070 ti

[brittle lantern]

I mean cheapest 9070 XT vs cheapest 5070 TI £75 GBP ish difference isn’t all to bad

[haughty pilot]

it's almost $200 for me unfortunately

[brittle lantern]

USD?

[haughty pilot]

180ish

yes

[brittle lantern]

Yh u US mf can’t compete with UK market rn

[haughty pilot]

im in EU, not US

[brittle lantern]

Why use USD then

[haughty pilot]

it's easier

[brittle lantern]

I mean just use ur own currency I use pounds that just makes it harder for me

Cuz now I’ve just said the wrong thing

[haughty pilot]

im used to typing in usd since most people won't understand any other currency 😂

[brittle lantern]

Understandable I’m used to pc build help servers and stuff and they are used to literally every currency to man XD that’s why it got me confused when u didn’t use native currency

[green plank]

Im using the forge ui, I have a Asus TUF Gaming A17 (2024) Gaming Laptop, 17.3” FHD 144Hz Display, GeForce RTX 4060, AMD Ryzen 7 7735HS, 16GB DDR5, 512GB PCIe 4.0 SSD, Wi-Fi 6, Windows 11, FA707NV-DS71-CA... id rather use my desk top but it has a amd gpu and cpu so supposedly it won't work very well atall

@ornate elk

[ornate elk]

Okay then make sure the models are no flux based models. They wont work good on 8gb vram

also as you only have 16gb of ram you need to increase your windows pagefile.
https://www.tomshardware.com/news/how-to-manage-virtual-memory-pagefile-windows-10,36929.html
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.
Then click OK and reboot the PC.

[green plank]

No ive only been using illustrious, pony, and sdxl, models, which are all based on sdxl right? So they should be okay than?

[ornate elk]

Yep thats okay but still needs the pagefile tweak to not freeze your system

[green plank]

🙂 thank you

I really appreciate you bro

[ornate elk]

Np, also what's your amd gpu on the PC?

[green plank]

ASRock > AMD Radeon™ RX 6750 XT Challenger Pro 12GB OC https://share.google/rZ8ELxYpwugHuDiMi

Thats the card, does cpu matter atall?

Is there anything else I can do to ensure it runs well? Like any small optimizations or tweaks i can make to the base model other than you just shown me?

[ornate elk]

CPU doesnt matter. The GPU can work good with stable diffusion too.
On the first link of the pinned messages you find the Guides for AMD.
Go for "Forge with ZLUDA"

If you want to set it up and test

[green plank]

Yeah I probally will give that ago soon whats the odds of that working better than my laptop?

Is there any other tweaks i can make to the forge one for my laptop that will make it run smoother? Like Minecraft arguments or any extensions or something?

[haughty pilot]

Does resolution matter for SDXL or it just has to be above 1024x1024? Something like 1800x1200, would 1600x1000 be alright? The results seems okayish I guess

[hot thorn]

@ornate elk Hi mate, need your help again. Could you tell me how to go about making short animated videos with stable diffuion? Im using reforge. Do i need a specific checkpoint to do it or do most checkpoints work? I'd like to use an illustrious one
I got deforum but no idea how to tackle it

[ornate elk]

yep would work too

[ornate elk]

--cuda-stream for forge

[ornate elk]

i guess it could be faster as the laptop

[ornate elk]

Can't help with deforum sry

[hot thorn]

know anybody who can?

or could you tell me how to properly install controlnet?
After installing it its just turned off automatically.
i turn it on, reset webui, and its back to being off

[haughty pilot]

What schedule type should I select? I just realized there's an option like that, was automatic all the time. Not even sure what the option does.
sampling method is euler A

[hot thorn]

@ornate elk what about AnimateDiff, do you know how to use that one?

[green plank]

Where do i put that? In the run batch file?

[ornate elk]

Forge has it already integrated so you can't install it

[ornate elk]

Yes but its not that great

[ornate elk]

Webui-user.bat

Commandline_args

[hot thorn]

what should I use for animation then? got any recommendations?

[ornate elk]

depends on what you want to do

and what your gpu vram is

[hot thorn]

just 8gb vram.
i want some anime animations, few secs long

[ornate elk]

ah okay

animatediff could work but is very slow

[hot thorn]

from what i learnt, animate diff doesnt support sdxl

[ornate elk]

it does

it has an sdxl module

[hot thorn]

? everywhere i looked they say it doesnt. tho those could be old

so what would you recommend me? or should i try animate diff

[ornate elk]

AnimateDiff doesnt work with forge and reforge.
Maybe try deforum

[hot thorn]

deforum constantly gives me a message i dont have controlnet installed...

i installed it through a link, and it always turns itself off in installed extensions

[ornate elk]

yea because forge has controlnet included already

[hot thorn]

so why it tells me to install it? i cant even test deforum

[ornate elk]

hmm maybe its not updated or broken

[hot thorn]

i just get this when i try to make anything with deforum

[ornate elk]

oh okay

[hot thorn]

@ornate elk im asking chat gpt to help out, but to no avail

[ornate elk]

i guess deforum doesnt work on reforge then

if you want to make smooth morphing animations try out loopback wavescript

[hot thorn]

where do i get it?

[ornate elk]

https://github.com/FizzleDorf/Loopback-Wave-for-A1111-Webui

[hot thorn]

and how do i use it?

[ornate elk]

here its explained:
https://rentry.org/AnimAnon-LoopbackWave

[tepid crypt]

Anyone can recommend any way to generate ai online? im mostly looking for a tool that allows you to use Stable diffusion forge the same as if doing in local, if possible free or really cheap. Most tools i found are very limited, doesn't allow to use extensions or other tools, some even giving really slow and crappy results.
Pd: I have a very average pc, can't afford a new one. Also, looking for something without censor, allowing NSFW

[ornate elk]

What's your GPU?

[tepid crypt]

gtx 750 ti

[ornate elk]

Ah well

Maybe Leonardo ai has the features

It was very popular in the beginning of SD

[tepid crypt]

also looking for one without censor

[ornate elk]

Then you might need to use Vast.ai or runpod.io to rent a cloud GPU and setup forge on that yourself

[tepid crypt]

Hmmm, i've seen videos of vast.ai, but not sure about their billing, i rather prefer paying a service a single time than introduce my card in a service online.

anyone has any experience related to the billing on vast.ai?

[haughty pilot]

from what I have seen 7900 xtx is faster than 5070 ti in SD, doesn't that mean 9070 xt is going to have a better performance when it's fully supported? I'm really torn between getting 9070 xt or paying extra $150 for 5070 ti (even tho both have the same performance in games)

wouldn't feel great about paying extra 150 for 5070 ti just to turn out 9070 xt is going to be superior in a near future

[ornate elk]

I have the 7900xtx and with TheRock experimental support its crazy fast.
With ZLUDA its very fast too but idk the it/s of a 5070ti

I guess the 9070xt will be much faster when its fully supported

[haughty pilot]

from what i have seen it's around 5 it/s on 1024x1024 euler a

the 5070 ti

[ornate elk]

here is a speed chart benchmark data by an extension of the SDnext dev:
https://vladmandic.github.io/sd-extension-system-info/pages/benchmark.html

[haughty pilot]

this is quite hard to read for a newbie like me
but even now 9070 xt would be a decent upgrade over my 2080 but ~twice as slow as 5070 ti at the moment and possibly same speed/faster after it gets fully supported (this year)?
I might just go for 9070 xt then, spending additional $150 just to have a better performance for no longer than half a year might not be a great idea.

[ornate elk]

yea the chart is a bit confusing

the 9070xt is a great card and will be really good when its fully supported (probably august).

[livid jetty]

when i use Comfyui it will wait until the percentae for image generation is 29%, it will say reconnecting, and then it will crash the cmd tab. any help?

[quiet matrix]

Black image when using hires fix, this only happens after I switched to reforge. I tried setting the hires sampling and schedule type to match what i used but still black image. I currently dont have anything on my "set COMMANDLINE_ARGS", if that is the problem, what should i put?

[ornate elk]

What's the workflow?

[ornate elk]

What's your GPU? You probably want to add --cuda-stream to reforge

[quiet matrix]

GTX 1660 Ti

[ornate elk]

Ah okay that explains the black images, let me check which command arg might help

[ornate elk]

Okay so add --cuda-stream --all-in-fp32
To the commandline_args=
Then try again

[livid jetty]

im sorry i dont know what that is

[ornate elk]

In Comfyui that means the whole node stuff together

So model, sampler, etc

[quiet matrix]

hmm still black images

[ornate elk]

Okay then change the 32 to 16

So --all-in-fp32 to --all-in-fp16

[ornate elk]

What does the cmd shows?

[livid jetty]

nothing, it just closes when it says "reconnecting"

[ornate elk]

I mean before that

The whole cmd log

[livid jetty]

oh

[quiet matrix]

now it generates black images without hires fix

[ornate elk]

Oh did you got them only with hires fix enabled before?

[quiet matrix]

yep

[ornate elk]

Interesting

Try remove --cuda-stream then

And test again

[ornate elk]

You have the numpy error

Its written in my Guide how to get it fixed. Then it will work

At the end of the ComfyUI zluda guide

@livid jetty also I see in your log that comfyui tries to use your CPUs igpu (Radeon TM Graphics)
Better disable it in the Device Manager if you have a dedicated GPU

[quiet matrix]

still black image with 16, trying it with 32 rn

[livid jetty]

it says to open manager, but i dont see an app like that in the folder

[ornate elk]

If you dont have installed Comfyui manager you need to do this first

Then you can fix the numpy error.
The guide lists also how to install Comfyui manager

[livid jetty]

let me try again, i swae i downloaded it

[quiet matrix]

still not working

it was normal image at first but it becomes black at the 75% mark

[ornate elk]

Ah okay then try add --vae-in-fp32

And maybe --force-upcast-attention

[livid jetty]

ok i did it, but it still crashes when i try to generate somthing

[livid jetty]

also does it matter alot?

[ornate elk]

Yes

It won't work with an igpu

[livid jetty]

alright

could you walk me through it or give me a guide, i dont wanna screw anything up and leave my pc unable to function

[quiet matrix]

Tried both, didnt work. Is it a model issue or upscaler issue? I'm using Remacri as the upscaler.

I did switch to R-ESRGAN but still black image

[ornate elk]

Can you try latent bicubic?

[ornate elk]

What's your GPU?

[livid jetty]

AMD Radeon RX 7900 XT

[ornate elk]

Right click on the Windows logo.
Select Device Manager.
There go to Display Adapters.
And there right click and disable only Radeon TM Graphics.

[livid jetty]

oh that was my second monitor

[ornate elk]

Oh, plug that into your GPU then, not into the Mainboard xD

[quiet matrix]

holy, it really is an upscaler problem lol, it's working now

guess i cant use upscalers i download now

[ornate elk]

Strange so its only with Esrgan based upscalers

What you can try is to setup Forge and test it there or in Auto1111

[quiet matrix]

I used to use Forge, and Remacri(along with other upscalers) work, never experienced black images issue until I switch to reforge

[ornate elk]

Ah okay

Them reforge does something different in applying esrgan based upscalers It seems

[livid jetty]

that might actually help me with screen tearing

alright im gonna try to use comfyui again, i dont know if i mentioned this, but i plan on usong swarmui does that change anything?

[ornate elk]

Nope

But I need to go off now.
Let me know if it worked, if not I can help tomorrow again

[quiet matrix]

thx for the help!

[turbid wing]

i am having trouble reinstalling stable diffusion

can someone help ?

[ornate elk]

Sure what's the issue?

[turbid wing]

so i was reinstalling stable diffusion but it is giving me error when i put download link in cmd that command is not reconised as internal or external

[ornate elk]

Do you have Git 64bit installed?

[turbid wing]

yes

i had python and git installed

[ornate elk]

Had ? Or still have?

[turbid wing]

have

[ornate elk]

Okay good

Can you screenshot the cmd with the message?

[turbid wing]

i am installing right now from the manual download link

and running cmd

[ornate elk]

What's your GPU and which webui do you try to install?

[turbid wing]

rtx 2060 and automatic a11111

i installed it before and no problem

[ornate elk]

Ah okay Im just asking because Auto1111 is pretty much outdated and Forge webui might be better for GPUs with less than 8gb vram

[turbid wing]

so how can i install that one ?

can u help me ?

[ornate elk]

Check the first link of the pinned messages here, there are all the install Guides