🤝｜tech-support

[wild jungle]

That's sad, worked well for me on my 3060

Would it be better to go with the dev branch or the main branch? Does it really matter

[ornate elk]

would go with the stable

[wild jungle]

Alright

I haven't used SD in a while so idk if anything changed

[karmic crown]

that is the main problem, dev will be on a constant change, and it can stop working any time

[wild jungle]

Assuming not really

[karmic crown]

stable will not be updated soon

[ornate elk]

wa 3060 will work fine with auto1111

[wild jungle]

Yeah it works okay, it's just that Forge was a lot faster

[knotty turret]

@ornate elk

[karmic crown]

everyting depends on gpu and memory
on many systems forge is only slightly faster, or even slower than a111
on my m3 pro mac, a1111 works faster, unless you add some parameters to forge, than it works about 5% faster 🤷🏻‍♂️

[ornate elk]

hmm make sure you close every background programm and games or wallpaper engine

[wild jungle]

Memory as in VRAM or system RAM?

I only have 16GB RAM but 12GB VRAM

[ornate elk]

make sure you use --xformers --medvram-sdxl --no-half-vae in auto1111

[wild jungle]

Do I need medium VRAM for 12GB though?

[ornate elk]

oh forgot your on 3060

[wild jungle]

Yeah lol

It's one of the MSI ones that has 12GB

[knotty turret]

nothing open

[ornate elk]

nope, but if sdxl models are slow than add --medvram-sdxl

[knotty turret]

only stable diffusion

[ornate elk]

okay what you can try is to go into the settings and there enabled FP8 mode for SDXL models

then apply

and reload

[knotty turret]

FP8 what is this?

i select Enable for sdxl?

[ornate elk]

you select fp8 mode for sdxl

[knotty turret]

that way?

[ornate elk]

in the settings

[ornate elk]

yes

then hit apply

[knotty turret]

ok

[ornate elk]

and reload ui

[knotty turret]

trying to aply but cmd stopped here

[ornate elk]

close the cmd and relaunch

[knotty turret]

ok

with FP8 on worked the generation

🙏

but what i lost with this option ON?

and the generation its wayyy more faster than the webui forge

but not gonna lie the images still rlly good

[ornate elk]

Precision, but I cant describe it but if your image look good then its fine

[knotty turret]

i mean the images are good but i can see the images now are more simple

[ornate elk]

The quality is not affected, they just will look different

Like if you change the seed

[knotty turret]

hmm ok

thx man

[ornate elk]

No problem 🙂

[knotty turret]

btw embendings should put on positive or negative prompt?

[pale salmon]

settings wont apply im stuck ahaaaaaaaaaaaaaaaaaa

its my wifi ?

[ornate elk]

Depends on the embedding

[ornate elk]

Check the cmd

[pale salmon]

finally done

[tropic dew]

Hey there. Sorry for a less than helpful question or set of information. I'm new to this. And I'm noticing that, when I use SD, usually I get higher quality results. But, sometimes, I can go in with everything the same, prompt, seed, lora, checkpoint, steps, all of it, and I'll get much lower results. And I don't really understand what causes this or how to fix it. I'm not leading in with what the loras in question are, as they're NSFW, and I don't wish to go against the Discord rules. (Realized, yeah, since it happens across multiple loras and with all sorts of prompts, it's likely a tech issue. So moved it here)

[pale salmon]

sup thats normal ? @ornate elk

[pale salmon]

😡

[echo phoenix]

i cannot login? joining the always direct me to subscribe, I ALREADY SUBSCRIBED

[ornate elk]

Yes

[ornate elk]

What model did you tries to use?

[cursive hornet]

hey hi is there something like a style reference extension from mj where you can recreate a style by using just 1 image? thanks in advance

[compact sphinx]

What am i doing wrong when my images turn up like these?

I have some settings in comfy ui that work nicely, but quite often when i try a new workflow I end up with these results and I would like to know why its happening and what to do about it.

[ruby yoke]

change the scheduler and sampler you are using. SD 3 doesn't work with most of the samplers

[compact sphinx]

I am using a SDXL model

[compact sphinx]

Any thoughts on what scheduler and sampler to use?

[ruby yoke]

heunpp2 and normal or simple seems to work the best

[ruby yoke]

here

[compact sphinx]

No difference.. 😦

I solved it using my old workflow. But it happens from time to time

[ruby yoke]

when that starts happening to me, i just reboot. i use chrome, it likes to cache stuff and not turn loose of it

[silk mantle]

Just circling back to this in case anyone has any ideas 🙂

[ruby yoke]

it's not triggering for some reason

[silk mantle]

My embedding made in OneTrainer doesn't seem to be doing anything, the results are identical to not using the embedding

[ruby yoke]

what's your workflow?

[silk mantle]

In terms of generating the embedding, or using it?

Generated it using the embedding preset in OneTrainer and the HelloWorld SDXL model, using photos of me captioned using the inbuilt captioner in OneTrainer.

Using it in SD Automatic 1111, with the HelloWorld model, and the embedding that OneTrainer output.

[ruby yoke]

in terms of using it.

are you sure you're connecting it in correctly?

is the helloWorld model the model it's based on?

[silk mantle]

Not sure at all, but I think I am, from what I've read and what people have been saying on here

[silk mantle]

Yup

[ruby yoke]

okay. i use comfy, not Auto1111, because i prefer the interface. so - my workflow is - load the model into memory, access the prompt and then the lora so the lora modifies the information from in the prompt. And make sure that the keyword for the lora is stated in the prompt.

[silk mantle]

Though the file does seem to be quite small, in the low kb... Does that make sense?

[ruby yoke]

how many images were in yoru data set?

[silk mantle]

Right, though I have it saved as an embedding, not a lora. Should I be using it as a lora anyway?

[silk mantle]

About 30, kept the batch small

[ruby yoke]

only 30 so i wouldn't expect the file to be huge but i'd expect it to be several meg at least.

maybe retrain and see if the file comes out the same size

[silk mantle]

I trained it twice already, but might try it as a LoRa this time and see if anything different happens?

[ruby yoke]

won't hurt to try

[silk mantle]

Thanks for your help!

[ruby yoke]

welcome

[silk mantle]

And it should be saving as a Safetensors right? Not a checkpoint

[ruby yoke]

everything should save as safetensors

[silk mantle]

Oh, I think I've just realised something. Does OneTrainer require over 100 images for training? Maybe I've been having much too little input?

[ruby yoke]

here are the OneTrainer docs https://github.com/Nerogar/OneTrainer/blob/master/docs/QuickStartGuide.md

the link to the wiki is near the top

[silk mantle]

🙏🏽

[fair oxide]

my forge stopped working and now im trying to fresh reinstall it but i keep getting this error

[fair oxide]

I followed the instructions in the pins to a t and it still pops that up and says junk like that

[ornate elk]

Forge is outdated and not recommended anymore

Better use Auto1111

[fair oxide]

oh

well shiitttt

lets see if i can install that then

[fair oxide]

what do i do if this happens

[ornate elk]

There is already a folder with that name

In that folder

[fair oxide]

What

[ornate elk]

Follow the install guide from the pinned messages

[fair oxide]

i did

it brought me to this

[ornate elk]

Make a new folder and try again

Name it Ai

[fair oxide]

i did 😭

[ornate elk]

No

[fair oxide]

i have three folders named ai

[ornate elk]

But your not in that folder

[fair oxide]

ill make another

what do i do now

[ornate elk]

No wait

[fair oxide]

the folder is open

[ornate elk]

Your just not in the folder with cmd

[fair oxide]

???

[ornate elk]

You need to do exactly the steps from the Guide. Not just opening a cmd

[fair oxide]

wdym not just opening a cmd

[ornate elk]

You have to be inside the ai folder

[fair oxide]

i am 😭

[ornate elk]

Then click in the explorer bar
(Not searchbar, not Windows searchbar, not Internet explorer)

[fair oxide]

its open rn

[ornate elk]

Type cmd

And hit enter

Then screenshot if your cmd is open

[fair oxide]

like this???

ima bout to click it

its open now

[ornate elk]

Nooo

Just no

[fair oxide]

what did i dooo

[ornate elk]

Read the guide xD

Or my message from above

[fair oxide]

sob, i think im misunderstanding the guide

where did i mess up

[fair oxide]

whats the "explorer bar"

is that where i fucked up, theres so many bars man

[ornate elk]

Yes

[fair oxide]

show me

im not understandiing the guide 😔

[ornate elk]

I'm on phone rn. It has no Explorer bar xD

[fair oxide]

nooooooo

[ornate elk]

Its the thing where the path shows

[fair oxide]

where the path shows??

[ornate elk]

Like c:\cool\path\to\folder...

[fair oxide]

holy shit i think i did it??

??

[ornate elk]

Yessss

Now you can git clone it

[fair oxide]

im so sorry im so dumb

lets see if i can get through the rest

[fair oxide]

💀

this dont look good

[ornate elk]

Nope

[fair oxide]

what do i do

[ornate elk]

Are your from asia?

[fair oxide]

no

[ornate elk]

Hmm something is blocking the connection it seems

Do you use a VPN?

[fair oxide]

uhhh let me check

no

i dont have a vpn, not an active one at least

now what

[ornate elk]

Is your PC system time correct?

[fair oxide]

yes

[ornate elk]

Can you restart the PC and try again?

[fair oxide]

On my phone now

Pc is restarting

Once it's up I just try to run it again right?? Or do I redo the whole thing again

Restarted

[ornate elk]

You need to do the folder thing again (clicking in the explorer bar of the ai folder and type cmd)

[fair oxide]

okay

its doing stuff

i dont think it worked

[ornate elk]

Still connection error

[ornate elk]

Are you on WiFi or Ethernet?

[fair oxide]

sorry man, this must suck having to walk me through this

[fair oxide]

yeah

[ornate elk]

Yea what?

[fair oxide]

im on wifi

[ornate elk]

Can you try with Ethernet?

[fair oxide]

i dont have ethernet i dont think

[ornate elk]

Oh ok Is the wifi connection stable?

Or can you improve it?

[fair oxide]

uhhhh i dont think i can improve it 💀

and im not sure how good it is

what else can i do, is my issue really just that i have shit wifi 💀 thatd be kinda funny but

i still want to make images

[ornate elk]

This error is caused by a connection issue

[fair oxide]

[ornate elk]

So bad WiFi connection could be possible

[fair oxide]

is there any other way i could install a1111 or any other ai generation software that doesnt require the best wifi

[ornate elk]

There is an other option, let me find the link

[fair oxide]

omg thanks 😭

[ornate elk]

Here you can download the Auto1111 webui as Package.
Download the sd.webui.zip file.
Then unpack it into the folder you created (aifolder)
Then run the Update.bat
Then run the webui.bat

Here is the link:
https://github.com/AUTOMATIC1111/stable-diffusion-webui/releases/tag/v1.0.0-pre

[fair oxide]

bet, give me a sec

ill see if i fuck up again or not looool

[hot cloud]

Hi guys, I have a little problem with mov2mov extension (the extension itself works fine after m2m_ui.py replacement) The issue is that the either ReActor or Replacer cannot be enabled in img2img or mov2mov tab itself. But it works fine in txt2img. It looks like this in txt2img tab (I can select the check marks) and like this in the img2img/mov2mov (no checkmarks at all). ReActor/Replacer does not work when used without the check mark so in fact it seems to be not enabled in those tabs. I would appreciate any help with it as I am lost now

[fair oxide]

running what i think is the update.bat

its not doing anything

okay it did stuff

now im running the webui.bat

[ornate elk]

Wait its called run bat

[fair oxide]

now i do webui and not webui-user, right???

oop

okay i stopped what i was doing and im running that now

[ornate elk]

You can also run the webui-user.bat
But you have to edit it like in my guide First

[fair oxide]

run.bat will probably be easier though, right

[ornate elk]

Yea but first edit the webui-user.bat

To get the best performance

[fair oxide]

bet

okay i edited it

im starting it up

lets hope

[fair oxide]

i dont think it worked....

[fair oxide]

i tried running the webui-user.bat but this happened instead

idk why its doing this ive had a1111 before 😔

[ornate elk]

Same connection error

[fair oxide]

Is there a way to use a collection of images to basically do png > text2img to iterate through image generation details and load them in SD and run batches?

kind of like img2img batch, but for all PNG info in txt2img

I also don't seem to have a "save style" button to save prompt/neg as style

[fair oxide]

Does anyone know why this is happening? I am trying to launch Webui-user.bat but I keep getting this error

I tried doing research but so far nothing has helped

I reinstalled python and git

still nothing

[brave frigate]

I've got a question, can you run SD with a NVIDIA GeForce GT 730 4GB?

[pale salmon]

Autismxl pony

[elfin wyvern]

Hey! Quick question (sorry if it was answered already - it probably was): In ComfyUI, do I have to tag a lora in positive prompt if im using a lora loader node?

[brave frigate]

I've also got a quick question, how do you do the pip install and upgrade thing?

[stray tiger]

how can i make an image

[pale salmon]

@ornate elkout of memory is from my browser ?

[knotty turret]

@ornate elkman i getting out of memory with 512x768??? Something its rlly wrong right now🤣

[ornate elk]

Make sure your GPU driver is updated

[ornate elk]

What model did you tried ?

[pale salmon]

😿

ty

[severe sand]

My Lora doesn't seem to be showing up

[pale salmon]

1 — Go to the "Settings" menu.
2 — Click on the sub-menu "Extra Networks".
3 — Scroll down and click on the option "Always show all networks on the Lora page".
4 — Click on the "Apply Settings" button (at the top of the page).

[ornate elk]

Is the lora made for sdxl or 1.5 ?

Because it will only show with a compatible model selected

[severe sand]

I wish I knew
i didn't know compatibility was a question ( i don't really understand what a lora is, i thought a model is all i need)

[severe sand]

Thanks I think I summoned AI satan

lora is showing up now

[pale salmon]

i got the same bug lol just reload ui

[severe sand]

nah just ignore and keep going))

[ornate elk]

Checkpoints (models) are needed for generating images.
Loras are like dlcs that help the model get a specific style or character.
Loras and models need to be on the same base version to work

[severe sand]

Thanks that analogy helped

Downloading SDXL then

[ornate elk]

I wouldn't recommend activiting the setting you just did. It won't help you

[ornate elk]

Yep sdxl is needed

[severe sand]

what if i use 1.5?

[ornate elk]

It won't work

The model will ignore the lora our output an error

[knotty turret]

same as before pony xl

didnt change

but for some reason now cant even generate 512x768

I dont know if you still remember me yesterday

its using all my ram omg wtf is happening

[ornate elk]

Try restart the webui

[knotty turret]

did it nothing change

[ornate elk]

You also need to increase your windows pagefile

[knotty turret]

more?

32gb page

[ornate elk]

Hmm 32gb pagefile should be okay

[knotty turret]

itz crazy right now vram and memory at 100%

i think i cant use stable diffusion ;/

its rlly weird because I could generate very wide images easy with no problem

but now

have any idea what i can do?

[pale salmon]

opera gx or google for stabble ? or its not diff

[ornate elk]

Would use Firefox but you can use any browser.

[ornate elk]

Try using a lower resolution

Also check if every other programm is closed

[pale salmon]

@ornate elkdo u know what is this ?

sec i translate

looper 2 an exception of type 'system.outofmemoryexception' was thrown

[knotty turret]

lower? 512x768

[ornate elk]

Your out of RAM

How much do you have?

[pale salmon]

sec

got 16GO

[stark nymph]

Does stable diff have ROCm support on Linux

[ornate elk]

Yes

[ornate elk]

OK and when do you get the error?

[pale salmon]

first time i got, and it was when my picture was loading

[stark nymph]

Thanks. Is it included in the current build or a fork

[pale salmon]

but i watch videos in the same idk if its bad or not

[ornate elk]

Rocm needs to be installed manually.
But then you can install the normal Auto1111 webui

[stark nymph]

Great. Thanks

[pale salmon]

maybe bc i prompted a 4k pic 2048x2048 with upscaler lol

[ornate elk]

Yes, definitly thats the cause

[pale salmon]

i will buy a new pc soon with a 4080 GTX, 3000$ the pc hope it will work fine for upscaling and prompting lol

[ornate elk]

Yep a 4080 will work really good

[ornate elk]

What's your current GPU?

[fair oxide]

So does anyone know how to fix this or?

[ornate elk]

Lora node is enough

[ornate elk]

Nope as torch doesn't support that old gpus

[ornate elk]

That error means your drive is formated in a wrong format (MBR)
But needs to be (GPT)

[fair oxide]

And does this mean I gotta use a different drive or is there a way to change it?

[ornate elk]

you would need to reformat the whole drive (losing data if not saved)

[elfin wyvern]

Thx!

[pale salmon]

1080 GTX

[ornate elk]

Ah right

[knotty turret]

256 x 256 no prompt and using almost 11gb ram 512x560 not working

should i reinstall stable diffusion?

[ornate elk]

Try use an other model first

[solemn atlas]

Hi, I still cant generate 3 characters

it doesn't work, whatever i do

[knotty turret]

Ok

Other sdxl?

[fair oxide]

COWBOY SHOT

[knotty turret]

@ornate elk i made some crazy stuff on my stable now i can generate 4k images size easy i got it fixed!

Two extensions fixed it

Reduced vram and ram usage

[storm sail]

ok, i got https://github.com/openvinotoolkit/stable-diffusion-webui working for a minute . . . as i thought previously, i was getting no promised acceleration from openvino cuz i didnt have openvino on my machine. once i was able to pypi install openvino via https://docs.openvino.ai/2024/get-started/install-openvino.html?PACKAGE=OPENVINO_BASE&VERSION=v_2024_2_0&OP_SYSTEM=WINDOWS&DISTRIBUTION=PIP and check that my devices was enabled for openvino usage via following the extra "resources" -> "installation instructions" on said page, i was able to render images much faster on my integrated intel cpu/gpu . . .

only, then, idk what happened after rendering only a few images, the damned build broke saying something about the model broke or some stuff . . . then, so i tried same build of stable diffusion earlier today and was rendering with ok speed again (though not quite as fast as it was before) . . . also, i get errors about some bad keyword thing something about safety check cuz img2img refiner though i wasnt even trying to do img2img at that time but i changed refiner back to none and error disappeared

i get some other errors here and there, particularly annoying (and im aware that it may not be a stable diffusion issue but rather related to python and/or openvino) is something about npu

i dont have npu, im not trying to use npu - only using integrated intel hd graphics 620 gpu/intel core i7-7500U cpu

i keep getting this

wonder how i fix it, can anyone here help?

or, direct me to anywhere i might find someone who might help?

along with that error, there will be a lot of junk about "stable diffusion model failed to load"

would be pretty groovy if all the folk developing this stable diffusion builds could band together on the same page, currently stable diffusion not seeming too "stable" always (buggy) but if everyone come together in force on one united build stable diffusion would be "stable" and rip ass 😛

[stark nymph]

Between generations, let's say I generate 10 images of the same pose, I then switch to a face shot to generate. My generations stay stuck on the first pose images instead of going to the face shot image prompts. What do I do in this situation to solve this? It happens quite often when trying to generate good character images.

[storm sail]

u tried changing seeds?

no one else responding so i throw in my 2 cents tho i am newbie 😛

[stark nymph]

I have, I have entered random seeds and drastically changed the prompts

[storm sail]

wish i knew more about all this stuff too, apologies i can't be of more help 😦

u tried prompting help room?

damned difficult to find good support for stable diffusion anywhere, i am learning, think only one tech support person works this whole room (for instance) they are helpful where they can be but at same time they dont seem to know too much about stable diffusion, unfortunately - at least not too much about things like running on intel cpu/gpu with stuff like openvino

[stark nymph]

Yeah, there is so much info but at the same time...there isn't

[storm sail]

right?! call that a lot of confusion! lol 😛

[stark nymph]

reloading the entire ui fixes it but thats annoying...

[storm sail]

what might be happening, idk for sure but . . . if certain parameters/options are changed, webui wants to recompile model before it will run inference . . . damned stable diffusion is broken in lots of ways, im learning. maybe its glitching and not recompiling model, even not saying anything about it, until u reload whole ui/even closing then reopening whole server/service

again, i am newb and only venturing guess

possibility "stable" diffusion could be a kickass program but currently seems not anyone really knows much about how to bring it together and make it "stable", if it werent so promising i wouldn't have spent last two weeks trying to work around its many bugs and lack of cohesive community 😛

anyway, again today, looks like im not getting any hands on help, tech support is napping i reckon, guess i keep looking to solve my own stuff via google etc, haha, good luck and be well

[stark nymph]

Yeah I'm not too sure lol. I'm using forgeui so maybe there's a startup flag or setting I'm missing.

[fair oxide]

fixed it, you were right it was because my wifi was shit. so i went to mcdonalds and borrowed their wifi LOL

its up and running again

forgot how much space this takes though, is there anyway i could like, have a1111 pull my models/loras from a different drive with more space???

[stark nymph]

Are you on linux or windows?

[fair oxide]

Windows

[stark nymph]

iirc, there is a flag you can use to specify your chpk folder and others at startup

Lms if I can find it

[fair oxide]

:D

[stark nymph]

Which UI you using?

Linux it's easy, creat a symlink to external drive

[fair oxide]

UI???

Uhhh

Uhhhhhhh

[stark nymph]

Like, Comfy, WebUi, ForgeUI etc

[fair oxide]

Ohhh

A1111

[stark nymph]

Here is an example from the git: set COMMANDLINE_ARGS=--allow-code --xformers --skip-torch-cuda-test --no-half-vae --api --ckpt-dir A:\stable-diffusion-checkpoints

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings

In your web-user.bat, you will set these there

@fair oxide ^^^

[fair oxide]

:D

Omg thanks man

[stark nymph]

Np 🙂

[shy charm]

let's see how long it takes automatic1111 to output a single image!!

oh.

same parameters but using forge instead

why is a1111 like this

i want to move back to a1111 since forge is dead in the water, but i literally can't if it's going to take 25 minutes to generate a single image

i really don't think it should take that long on a 3070

it's a clean install, fully up to date on the main branch

[fair oxide]

hey is it possible to turn a preexisting 1.5 lora into a lora for pony/sdxl

[reef gazelle]

hhh

[unborn bramble]

How to run Lora trainer on Kaggle? Any reliable ways

[fair oxide]

trying to install dynamic prompts but, wtf is this

[ornate elk]

We can't see the settings you used.
But what helps in speed is to set the VAE decode to TAESD in auto1111

[shy charm]

no extras, using a ponydiffusion derivative that i believe has a baked-in vae, and i have 'none' selected in the vae dropdown

i used the same settings for both the a1111 speed test and the forge speed test

it takes an extremely long time to begin the hires steps

it begins "tiled upscale" steps around 4 minutes in, and finally starts on the hires steps around 21 minutes in

i really hope illyasviel reconsiders the direction they plan on taking forge in or someone else makes a fork and keeps updating it, because a1111 is actually unusable
i might seriously consider learning how to use comfyui at this rate

[fair oxide]

help 😭

what is thiiisssss

[ornate elk]

Seems like your Attwifi connection has an error or is blocking the download

[ornate elk]

What was your gpu again?

[shy charm]

RTX 3070

specifically an EVGA RTX 3070 FTW3 if that matters

[ornate elk]

Can you try the following for Auto1111?
Remove --no-half-vae from the webui-user.bat.
Install the tiled diffusion extension and enabled only Tiled VAE when enabling hires fix.
Then download the fp16 VAE fix and put it into the models/VAE folder and select it as VAE.
The file is sdxl.vae.safetensor
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main

[shy charm]

ok yeah this basically fixed it, but it's still around 25 seconds slower than forge

[ornate elk]

Last thing you can try is to set VAE decode to TAESD in the Settings

[fair oxide]

what do i do about this???

my "pydantic.fields" seem to be an issue??

[ornate elk]

Can you show the full error log?

[fair oxide]

how

this??

[shy charm]

are you using forge?

[fair oxide]

no, im using a1111, from what i understand i think its different than forge??

or are they the same??

i just know i didnt install "forge" this time because i heard its... dead

[shy charm]

if it's not forge then i have no idea how to go about fixing it other than try deleting your venv folder

[fair oxide]

i did

[fair oxide]

then it yelled this at me

so i put the venv back in

[fair oxide]

and it went back to this issue

[shy charm]

well ofc putting the broken venv back in isn't gonna fix it

if your wifi is blocking the download when it tries to create a new venv then idk what you can do

[lost crescent]

Are textual inversions/embeddings possible in A1111 any more?

My version of A1111 does not ask for an output folder/invert input photos/captioning ... ?

[ornate elk]

Did you installed Controlnet before?

[fair oxide]

yes but i deleted it

and it still wouldnt work

do i need to dig deeper and delete it from more than the extensions tab??

[nocturne turret]

What to put in Diffusion prompt?

[lost crescent]

SD3 and A1111 SD1.5 textual embeddings work well together!

[ornate elk]

You would need to delete the venv folder too

[fair oxide]

i did but then the yellow text showed up??

[ornate elk]

But your internet at home seems really bad configured so it won't download it correctly

[fair oxide]

hmm so delete the folder then restart the webui??

[ornate elk]

Yea but then you get the yellow text

[fair oxide]

yeah 😭

[idle pulsar]

#🏞｜general-with-images

How to generate images ?

[fierce hollow]

can you tell me which folder should be deleted from the SD in order to speed up the download of images? I was advised earlier, the venv folder, right?

[fair oxide]

ever run into this issue before??

[ornate elk]

Nope and I've seen a lot of SD errors xD

[fair oxide]

sorry 😔

i get a LOT of errors

[ornate elk]

I would advice to check your internet connection. Maybe change your DNS server in the router or PC

[fair oxide]

i just open it every day and get a new error

[fair oxide]

rn im trying on different wifi

[ornate elk]

Ok

[ornate elk]

Download of images?

[fair oxide]

anything else i can do in the meantime??

maybe if i told you how it happened??

[fierce hollow]

generation, I found your old answers, yes you advised to delete venv to speed up image generation)

[ornate elk]

Do you have multiple WiFi connections in your house?

[fair oxide]

no but im using a phones hotspot rn which uhh

is better than my wifi currently

but it all started when i moved extensions from my backup folder for a1111 files to my current install of a1111

it was workign fine

i moved the extensions into the extensions tab

and bam, it broke

i removoed the extensions and got the "pydantic" error

so i removed the venv and it started screaming at me in piss yellow text 💀

[lost crescent]

Symlink the extensions file?

[fair oxide]

???

[fair oxide]

wdym

[lost crescent]

Leave the extensions file where it is and any new installation symlink to it?

[fair oxide]

whats symlink??

im like, only good at making the images, idk any of the terms or files or what they do/how they work

[lost crescent]

mklink /D "New Extensions Folder" "Old Extensions Folder"

Give the folder names the correct addresses ...

[fair oxide]

....

you lost me bro

i do not understand your words magic man

[lost crescent]

You said you moved a folder (extensions) and it broke

[fair oxide]

yes

i had a spearate folder full of extensions to put into a1111 in case i ever installed it again

[lost crescent]

So - don't move the folder - just point to it in the new install

[fair oxide]

so, keep it where it is but make a1111 change where it draws my extensions from??

[lost crescent]

I'm trying to rescue an old favourite of mine - embedding in A1111 - cuz believe it or not - SD1.5 embeddings work great with SD3!!!

[fair oxide]

fr???

whats sd3??

[lost crescent]

It is the most advanced version of Stable Diffusion available ... beter than SDXL (some would say); and better than SD1.5 (some would also say!)

[fair oxide]

.....

is it like a1111??

like i can download it on my laptop and use it offline???

[lost crescent]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/tree/sd3

This is SD3 ready

[fair oxide]

i think i have this??

so is sd3 just like, models??

like how theres pony models and sdxl models??

I FIXED IT

[lost crescent]

SD3 is models AND encoders

[fair oxide]

whats encoders???

[lost crescent]

t5

[fair oxide]

t5???

[lost crescent]

Which is different from ClIP

[fair oxide]

assume i know nothing at all times

like, assume im starting at step NEGATIVE 42

[lost crescent]

OK, go out onto the interweb and Read A Lot!

[fair oxide]

noooooo 😭

google cant answer my dumb questions

[lost crescent]

Not being facetious - but Google is a true friend when using A111/ComfyUI/SD3 etc etc

[fair oxide]

ikik

i just cant understand it all

most of the tutorials/guides use words like venv, CLIP, stuff that you used

and it all immediately goes over my head

[valid kraken]

hi, is there a workflow for using more than 1 lora at once with pony?

[lost crescent]

OK, then check them out one by one

Yesterday, I learnt what Attention means

[valid kraken]

for comfyUI i mean

[fair oxide]

i can possibly try that

[lost crescent]

and I've been using ComfyUI for two years

[fair oxide]

and good sd3 models youd reccommend??

[lost crescent]

Good @fair oxide

SD3 medium

SD3 clip

[fair oxide]

[lost crescent]

amd sd3 clip t5xxlfp16

[fair oxide]

bet

[lost crescent]

https://huggingface.co/stabilityai/stable-diffusion-3-medium

[fair oxide]

oooo thanks

i was just about to look it up too LOL

[valid kraken]

is it even possible to load multiple loras at once with pony and comfyui?

[ornate elk]

Deleting venv and relaunch is a good advice from time to time to get a clean SD yes.

[ornate elk]

Yes

Just make more lora nodes

And connect them to the prompt or k diffuser

[ornate elk]

SD3 is currently in beta test for auto1111

[valid kraken]

should i also connect them to the model node?

sorry, i'm not too good at the node system

[ornate elk]

@fair oxide also does your webui works again?
Then please run the Update.bat
Then install controlnet fresh. Don't copy any extensions over

Controlnet had a bug in an old version which cause this pydantic error

[ornate elk]

I'm not at the PC rn to check. Better ask in #🧣｜comfy-ui

[valid kraken]

sure!

[fair oxide]

ye it works now, thanks for all the help. i wont be instlling controlnet again 💀

[ornate elk]

You can install it again. Just don't copy the broken version from your old install

[fair oxide]

oh nice

[crisp apex]

Hey all. I am having an issue with automatic1111 and Animateddiff. I have everything installed correctly and updated everything. But when I enable animateddiff, the generation is only a single image. I have watched a few YT videos, but I have not been able to find the issue. Has anyone else experience this at all?

[ornate elk]

Hey, can you show your settings?

[crisp apex]

Yes. When I get back to my computer I will send a screenshot! Thanks.

[brave frigate]

Can somebody help me with this? I just updated my AMD drivers and now this shows up when i start SD...

Already up to date. venv "C:\Users\Chef\Desktop\SD-Zluda\stable-diffusion-webui-directml\venv\Scripts\Python.exe" WARNING: ZLUDA works best with SD.Next. Please consider migrating to SD.Next. Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr 5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)] Version: v1.9.3-amd-28-g371f53ed Commit hash: 371f53ed7c926f9048ef95f45bc816cfbf37b564 Using ZLUDA in C:\Users\Chef\Desktop\SD-Zluda\stable-diffusion-webui-directml\.zluda Skipping onnxruntime installation. You are up to date with the most recent release. no module 'xformers'. Processing without... no module 'xformers'. Processing without... No module 'xformers'. Proceeding without it. C:\Users\Chef\Desktop\SD-Zluda\stable-diffusion-webui-directml\venv\lib\site-packages\pytorch_lightning\utilities\distributed.py:258: LightningDeprecationWarning: pytorch_lightning.utilities.distributed.rank_zero_onlyhas been deprecated in v1.8.1 and will be removed in v2.0.0. You can import it frompytorch_lightning.utilities` instead.
rank_zero_deprecation(
Launching Web UI with arguments: --use-zluda --medvram-sdxl --update-check --skip-ort

rocBLAS error: Cannot read C:\Program Files\AMD\ROCm\5.7\bin/rocblas/library/TensileLibrary.dat: No such file or directory for GPU arch : gfx1036

rocBLAS error: Could not initialize Tensile host:
regex_error(error_backref): The expression contained an invalid back reference.
Press any key to continue . . .`

[brave frigate]

for some reason it wants rocM stuff now, it didnt want that before the update, or maybe the update removed it from my pc... idk

[acoustic lynx]

I'm not sure if this is the right channel!
I need to generate advertisement images for the client products. These products are brand new hence any model does not knows about it. I cannot finetune as there not much images available. I tried using IPAdaperts and controlnet but they always miss some features.

Is there anythings like few-shot learning(similar to LLM) so that I can give my product images along with prompt and expect to generate images with that product.
OR there any better options ??

[ornate elk]

What's your GPU?

Oh wait I know that error

[brave frigate]

7900 XT

[ornate elk]

If you have an AMD cpu too, You have to disable your integrated gpu in the device manager (Radeon TM graphics)

Not the 7900xt

[brave frigate]

oh tf

it enabled it i see

ah yes it works again thx

[ornate elk]

No problem 🙂

[lament gulch]

Hi, can I prompt the API with text to video?

[covert lotus]

hey man u helped me a few weeks back with getting it working on my radeon card i come once more with a question

https://civitai.com/models/498835/ujac-artist-style-or-pony?modelVersionId=554503 how do i get this thing to work, just makes a grey square for me

is it cus its a dora not a lora

[ornate elk]

hey, what was you gpu again?

can you show the settings you used?

[covert lotus]

Rx6800, can’t access my exact settings atm sorry but I copied the exact parameters as the examples on the webpage

[wild jungle]

A1111 really doesn't like SD3 for some reason..

Loading weights [cc236278d2] from D:\AI\Stable Diffusion\models\Stable-diffusion\sd3_medium.safetensors Failed to load checkpoint, restoring previous Loading weights [6ce0161689] from D:\AI\Stable Diffusion\models\Stable-diffusion\v1-5-pruned-emaonly.safetensors Applying attention optimization: xformers... done. changing setting sd_model_checkpoint to sd3_medium.safetensors [cc236278d2]: RuntimeError Traceback (most recent call last): File "D:\AI\Stable Diffusion\modules\options.py", line 165, in set option.onchange() File "D:\AI\Stable Diffusion\modules\call_queue.py", line 13, in f res = func(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^ File "D:\AI\Stable Diffusion\modules\initialize_util.py", line 181, in <lambda> shared.opts.onchange("sd_model_checkpoint", wrap_queued_call(lambda: sd_models.reload_model_weights()), call=False) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:\AI\Stable Diffusion\modules\sd_models.py", line 883, in reload_model_weights load_model_weights(sd_model, checkpoint_info, state_dict, timer) File "D:\AI\Stable Diffusion\modules\sd_models.py", line 393, in load_model_weights model.load_state_dict(state_dict, strict=False) File "D:\AI\Stable Diffusion\venv\Lib\site-packages\torch\nn\modules\module.py", line 2152, in load_state_dict raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format( RuntimeError: Error(s) in loading state_dict for LatentDiffusion

[ornate elk]

yes its not support in the stable release rn, only in dev

[wild jungle]

Ah thank you

[glad fossil]

I'm trying to use Kohya on ubuntu 22.04. and whenever I run ./gui.sh, its not detecting cuda drivers so its not letting me use my nvidia GPU

i can see torch detected my gpu

[fiery sequoia]

Hey, im trying to install Stable Diffusion second time, and it again tells me to install idk "torch" and others stuff, how to install all of them at once?

can anyone help? and tell me what have i to do?

[fading heron]

Hi, I broke my Stable Diffusion and am hoping someone can help me. I am using an AMD RX 7800 XT on a windows 11 machine. I had SD running great using ZLUDA. The problem arouse when I added "--xformers" to the COMMANDLINE_ARGS= part of the webui-user.bat file. it appeared to have installed something that will now prevent me from using SD. It now crashes with a long list of errors relating to the xformers thing. Could anyone help undue to damage or tell me how to reinstall. Thanks for your time.

[native rampart]

I trained my own model, but for some reason I am not able to select it in the web UI

[fiery sequoia]

is "Easy Diffusion" good?and won't burn my PC? (GTX 1650 and i7-10400)

[lost crescent]

Go to pytorch.org - and if you have Windows (and an nVidia Graphics Card) use this d/load pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

[ornate elk]

Xformers is for nvidia only and won't work with AMD rn

Remove it from the webui-user.bat and then delete the venv folder.
Then relaunch

[cinder ermine]

accidentally added my folder containing about 30 img datasets to my infinite image browser extension, then tried removing the dir

now its going through the same folder over and over again

should i just let it run its course?

or can i stop it somehow

oh wait lol

idk what happened but kohya shit itself and nested the same folder in itself 10 times

Its like 40 gb no wonder my image browser was dying

[unborn bramble]

Please tell me how to train Lora with Kaggle

[vocal burrow]

That will install torch in their python's system. Depending of the SD client they re using it might be useless to do so and take disk space for nothing.

[lost crescent]

So install it locally in the windows_portable venv?

[vocal burrow]

Installing it in the venv is the proper way to go yes. But most likely if there s a venv it should install torch stuff by itself. The problem here is that we don t have enough informations to help

@fiery sequoia came saying "Hey I have a problem" but gave little to no informations about the issue. no log, no screenshot, etc... So like most of the time in those cases, people don t have time to ask for those so they can maybe help them ^^".

[fierce hollow]

Generation took a long time. Deleted the venv folder. I started the SD, downloaded the files - now this is it... help please

[ornate elk]

Your on a old webui version

1.7.0

The latest version is 1.9.4

You should update with git pull and then delete the venv folder again

[fierce hollow]

Please tell me where to download it) And it doesn't support SD3 yet?(

[ornate elk]

Currently its in testing

[fierce hollow]

Not for general use?

[ornate elk]

Go into the stable-diffusion-webui folder.
Click in the explorer bar (not searchbar)
There Type cmd and hit enter.
Then type
Git pull

Then delete the venv folder

[fierce hollow]

Yeah, I got it, thanks a lot!

[fierce hollow]

If I have already deleted the folder, is it okay? or download now and then update the version?

[ornate elk]

No problem 🙂

[ornate elk]

That's okay. But check the version after its done

You can see it in the first few lines in the cmd. Or in the browser at the bottom of the webui

[fierce hollow]

(

Before that, I directly entered git pull here like this

[ornate elk]

Okay hmmm

Then run the webui-user.bat again

And check if it says 1.7.0 or 1.9.4

[ornate elk]

Perfect

Looks good

[vocal perch]

Hi Everyone

Getting error when trying to generate image using inpaint mask using api
ValueError: images do not match

Can anyone provide suggestions how to fix this issue

[ornate elk]

Do you have a full screenshot of the cmd?

[ornate elk]

Your python version is not compatible

SD needs 3.10.11 64bit

You have 3.12

[fiery sequoia]

OOOH okey make sense i will uninstall, thanks <3

sorry im new i didnt know it matter

[ornate elk]

No problem, uninstall it, install 3.10.11, check "add python to path" when available.
Then delete the venv folder and relaunch the webui-user.bat

[ornate elk]

No problem. You can also checkout my install guides in the pinned messages of this channel.
There you also find the best settings for Max performance

[fiery sequoia]

installed new python, so now i have to delete venv and use webui-user?

i ask to be 100% sure

@ornate elk thanks, everything works, but is it normal that my GPU is using 100%?

[fiery sequoia]

how to repair?

[ornate elk]

What's your GPU?

I see 4gb vram. Then you need to edit the webui-user.bat and at the line Commandline_args= you need to add: --xformers --medvram --no-half-vae
Then save and relaunch

[fiery sequoia]

okey thanks

[fierce hollow]

@ornate elk I know you're already tired of me, but it's me again, I'm in trouble AGAIN.

1. The generation is still going on for a very long time.
2. Why does a blurry image appear when I turn on Hires. fix? (Explanation: I've been working with this particular model for a long time and before the update, such settings were always PERFECT)

[ornate elk]

you need so set hires steps to 10 and denois to 0.5 when enabling hires fix

[fierce hollow]

I repeat, I have always had these settings, you told me last time I remember telling them, I measured them myself, even without a function Hires. fix generation takes an extremely long time

[ornate elk]

Can you show the hires fix settings?

[fierce hollow]

?

[ornate elk]

looks okay, but latent can output blurry images

when used with the incorrect denois value

[fading heron]

This worked, thank you.

[fierce hollow]

Previously, I always used it, there were no blurring, what would you recommend to put?

[ornate elk]

drop an old image (where it worked) into the PNG-Info tab and show me the meta data

[ornate elk]

is that the blurry image?

i mean drop in an old image where you said "it worked before updating"

[fierce hollow]

yes, plus, I am not satisfied with the quality of such a model + with such a resolution, the quality should be better

[fierce hollow]

amm. okey

[fierce hollow]

Here's an example

[ornate elk]

ok, and as you see there you didnt used hires fix

[fierce hollow]

)))

[ornate elk]

okay

[ornate elk]

can you try to generate a new image with these settings?

[fierce hollow]

The same thing lol

the old version and the new one, the promt is the same

[ornate elk]

really strange

[fierce hollow]

I was most afraid to hear those words from you. Will you help me?...

It was all about the Schedule type.... In the previous version, it was not easy, I set everything up)

But the first question remains open, why can there be a longer generation?

[ornate elk]

how much longer?

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

[fierce hollow]

3070 ti

[ornate elk]

and your using --xformers --medvram-sdxl --no-half-vae ?

in the webui-user.bat

[fierce hollow]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--autolaunch --xformers --medvram-sdxl --no-half-vae

call webui.bat

[ornate elk]

try remove --no-half-vae

and test the speed again

[fierce hollow]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--autolaunch --xformers --medvram-sdxl

call webui.bat

?)

[ornate elk]

yes

[vocal burrow]

(iirc --autolaunch has been the default behavior for the past 6 months or so)

you updated from which version ?

[fierce hollow]

IT HELPED!!! Thank you, kind man!

[ornate elk]

nice np 🙂

[obtuse knoll]

I was wondering if someone python savvy can help me out, I just updated bmaltis khoya trainer and suddenly it wants a higher Python version(3.10.11). I'm on windows btw. I need 3.10.6 installed for A1111, and I have no idea how to install a different python version into the khoya venv. :<

[vocal burrow]

3.10.11 is compatible with a1111

it s safe to upgrade 3.10.6 to 3.10.11.

make sure to uninstall 3.10.6 and delete a111's venv folder afterwards

@obtuse knoll

[obtuse knoll]

Thanks! Was that always the case or did it get bumped up recently?

I remember reading that versions above 3.10.6 didn't support torch, but that maybe fell by the wayside when the upgrade to torch2 happened?

[vocal burrow]

that was pretty much always the case.

[past geyser]

When creating new generations, I like to keep the same prompt and seed and switch between all my checkpoints. Some checkpoints work just fine, but others produce this weird green overlay, does anyone know why?

[burnt totem]

how do i disable shared memory?[MSI A520M-A Pro]

[ornate elk]

Mostly caused by a wrong VAE

[onyx solar]

Are you choosing checkpoints/models that are mixed between SDXL and 1.5?

[past geyser]

Not to my knowledge, I've been specifically going for SDXL but I'm not reading the fine print

Think that might be it?

[onyx solar]

Possibly.

[fierce hollow]

Making changes is good, but ordinary people are waiting for the SD3 in automatic111 #📣｜announcements

[karmic crown]

even longer

[fair oxide]

you could run comfyui for sd3 right away, comfyui even has its pre built template which you can simply click and load, sd3 hugginpage site also lets you download a template for it

intalling comfyui on windows manually with few easy steps.

And if you have fair amount of storage space on your drive, there is no need to delete a1111, you could keep both a1111 and comfyui and share model paths w/o having to duplicate anything

[final mason]

why this is the most irrating bs to set up when you know nothing about this stuff

and all the videos are out of date crap

i installed python why is it not working

[ornate elk]

because you didnt checked "add python to path" or didnt deletet the venv folder before relaunching

the best install guides are in the pinned messages of this channel 🙂

[final mason]

it was the wrong version i wasnt on 10.06

[ornate elk]

3.10.6 - 3.10.11 will work

[final mason]

ya i was on the latest

so do i have to do that thing you said with the add python to path

[ornate elk]

only if it you get that error again or if its not loading

but checkout the guide from the pinned messages for the best performance settings, you need to set

[final mason]

ok ill try again i repaired it

ok ty

[ornate elk]

let me know if it works

[fair oxide]

one of the common discouraging approach with comfyui is the idea that you'd have to build a template everytime from the scratch with every new session for your images, when actually you simply build upon your template and save it and reuse with minor changes here and there depending on your intended outputs, once you get a fair understanding of that procedure its quite fun to use comfyui

here is a good resource for workflows that you can study and for the purpose of reusing them effectvely https://comfyworkflows.com/

[final mason]

im either blind or stupid cuz i dont see the box for add to python

is it called that or somthing else

[fair oxide]

open your eyes, its the first window at bottom

[ornate elk]

the box is only one time there, to get the same box you have to rerun the python installer, click on "modify", then on "next" then check "add python to system variables"

[final mason]

i didnt see that sorry

it seems to be doing something now

where do i put the model cuz i dont see stable diffusion folder for the models like that video had

[ornate elk]

your on AMD ?

whats your gpu?

the folder will be generated after the first launch

[sweet basin]

rror: Your local changes to the following files would be overwritten by merge:
webui-user.bat
Please commit your changes or stash them before you merge.
Aborting

You are not up to date with the most recent release. |
| Consider running git pull to update.

[ornate elk]

cut out the webui-user.bat from the folder. then git pull via cmd
then repaste the bat file

[paper mural]

hey friends, I'm using Juggernaut XL model with a1111 UI, and the results I'm getting look kinda strange, idk how to describe it, so here's a picture.
I remember trying it long time ago (a few months), and I had a similar result, dunno whu it's happening. Can anyone please help me to fix this?

[ornate elk]

hey, sdxl models require a resolution of 1024x1024 to look good

because they are trained on that resolution

[paper mural]

I tried this resolution too, and got a similar result, 1 sec

[onyx solar]

Additionally, for some of the best results out of A1111, be sure you're using the recommended samplers/schedulers/CFG/steps settings for the model(s) you use.

[paper mural]

here is 1024

[onyx solar]

Ultimately, because of how all of this works, you still may need to employ other steps such as inpainting to get the best quality images.

[onyx solar]

See my message above.

[fair oxide]

test it out with Euler A, steps 30, cfg 7

[sweet basin]

AttributeError: 'KDiffusionSampler' object has no attribute 'trajectory_enable'

[final mason]

yes was trying to fix the amd erorr but theres no videos at all

[paper mural]

I think it's kinda better? 🤔

[fair oxide]

still looks off to me, but those are the common settings for most models, i wonder about the vae you are using

and are any loras in use?

[regal kestrel]

feels like you have 2 loras fighting each other. one is geralt and the other about a monster

[paper mural]

I didn't explicitly choose any lora or vae, just downloaded the model and selected it in the select input field

[fair oxide]

ok just to be sure , make sure this vae is used sdxl_vae.safetensors since you are using sdxl model, also check with lora and what weight is recommended by the author

[ornate elk]

whats your gpu?

i have many guides for AMD

[final mason]

rx 570 i tried it one time and it gave me an error my buddy said theres a fix but idk if ill be able to figure it out

if its as long as that written guid is

[ornate elk]

you need to follow my directml guide

for amd

[paper mural]

looks like the main issue was in Sampling Steps and CFG value, when I set it to 6 and 2 the image started to look okay

[fair oxide]

sounds like you might have downloaded either lighting or turbo model which require cfg as low as 1-2

otherwise you'll over cook your images

typically low scale models like turbo, lighting will give you good results with steps between 8-10, cfg 1-2, but check the model page for exact specification

[sly spoke]

Just downloaded SD and at the end when I went to http://127.0.0.1:7860/ I got an error message that says refuse to connect. Maybe my firewall? I use Webroot.

[paper mural]

ah, it's called hyper , so I think it's like lighting or turbo

[fair oxide]

suggest you read the author page

[ornate elk]

can you show the whole cmd log?

[covert lotus]

bit late this is what i was rocking

[ornate elk]

And the hires fix?

[covert lotus]

yeah

[ornate elk]

Wrong settings. Always set hire steps to 10. and denois to 0.5 or below

[ebon sandal]

iam trying to install the comyUI manager but it gives this error in cmd: Microsoft Windows [Version 10.0.22631.3810]
(c) Microsoft Corporation. All rights reserved.

D:\Comfy\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\custom_nodes>git clone https://github.com/ltdrdata/ComfyUI-Manager.git
'git' is not recognized as an internal or external command,
operable program or batch file.

D:\Comfy\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\custom_nodes>

[ornate elk]

You need to Install git 64bit

[ebon sandal]

thx, youtube tutorial didnt mention it

[ebon sandal]

I feel like iam stupid or something, i followed the tutorial, ComfyUI is running, i have the correct models but when i press Queue prompt nothing happens

[ornate elk]

what does the cmd shows?

[ebon sandal]

Oh hold up, its just worked, it just took like 10 minutes

I didnt know my pc was this bad

[ornate elk]

wahts your gpu ^^

[ebon sandal]

2060 so thats probably why, also i used this for the first time on my pc at work with an 4090

[ornate elk]

ah well, yea a 2060 isnt that powerfull. best would be to avoid sdxl models

[ebon sandal]

hmm, is it possible to use my 4090 at work while at home? like a remote desktop?

[onyx solar]

Possible? Yes. Probable? Maybe.

There are a lot of questions to be answered that have nothing to do with ComfyUI in order for that to work.

Your job may block your access, for instance.
Or your connection at home may not allow inbound web requests due to a common thing called NAT.
There are several other things that would have to be known/understood beyond that.

[ebon sandal]

My job is an small team and my boss is my uncle that wants me to research using AI, so i think he wont mind using the work pc

[onyx solar]

Sure...but your ISP may not allow inbound server-like requests to your connection type, (or most commonly, not without paying for a static IP or business-like connection fee).

And then you get to the "how would you reach your machine". Would you serve it via a web service or would you use a private VPN or remote desktop-like connection?

[ebon sandal]

Well i can use AnyDesk to control my work pc

[onyx solar]

That's the other direction.

Businesses typically already have the connection requirements for that.

Home ISPs commonly restrict that in the other direction.

[ebon sandal]

Yeah i dont have much knowledge about this stuff

[onyx solar]

If you can make it happen in the direction of Work -> Home with what you have now, that's great...you're a lot closer to a Yes.

[ebon sandal]

I can control both my work pc and home pc from my phone, i can control my work pc from my home pc using anydesk

Iam not sure if i can control my home pc from my work pc but that isnt necessary

[onyx solar]

That's good. If you can use your phone to control your home PC while you're not on the same network at home, that's a positive step and you should try the control of your home PC while at work to confirm it.

At that point, the easiest way would be to simply use AnyDesk to remote in and use it as if you're sitting in front of it, rather than to set up the service like a web service.

[ebon sandal]

Isnt there something like anydesk that is like almost exactly the same as using phyiscal pc, or is that depended on the internet speed?

[onyx solar]

Apologies, but I'm not 100% sure what you're asking.

[ebon sandal]

Iam basically asking if there is an better alternative to anydesk, becuase i have used it before but it felt like there was some sort of delay

[onyx solar]

What is your home connection speed typically like? Remote desktop connections aren't typically too heavy on bandwidth, but they certainly can be laggy.

[rain egret]

@frigid forge & @ruby yoke

[ebon sandal]

938 mbps

[rain egret]

why is the LoRA field unaccessible is the current enigma (in Krita SD plugin)

[onyx solar]

So about a gig connection, which should be pretty good, typically. Even if you were connecting with something that had a bit of latency like satellite, you should be decent for a remote desktop connection.

[ebon sandal]

its not that bad but not as smooth as working from my own pc

[onyx solar]

Well...I'm not 100% with the backend workings of AnyDesk and if they're piping the connection through their servers or if it's allowing a direct connection, but if they are, that could explain some of that.

[ebon sandal]

i guess i just have to secretly switch pcs

[final mason]

where would that be?

[ornate elk]

Here, your GPU needs:
Automatic1111 with Directml
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides

[final mason]

k ty will that fix that error i was getting when trying to generate

[ornate elk]

Yes it should fix it

[final mason]

ok ty ill try it out

i was using monai until trashy ass google play shut down all nsfw FOR NO REASON AT ALL garbage

[final mason]

im getting 2 errors idk what i did wrong i got runtime error with lowram and a attribute error with no half

did i miss something?

[fading heron]

I just was able to get ComfyUI to work on my system and I like the look, but the processing time is really long. Does anyone know any ways to get it to run better. It is not using the video card, which is an AMD RX 7900 XT on windows 11. Their documentation talks about adding something using PIP, but I don't understand it and they haven't gotten back to me yet. Any help would be greatly appreciated.

[young bridge]

hi all

which xformers is compatible with pytorch 2.2.2

[worthy sapphire]

Can anyone help me with installation with an AMD Gpu? I followed a few guides but I haven't had luck yet.

[ruby yoke]

you mean adding the hardware to your machine?

[worthy sapphire]

no, just getting a version of stable diffusion to run locally.

[ruby yoke]

are you using one of the interfaces for SD - auto1111,comfy, something else?

[worthy sapphire]

I was trying to install auto1111 since that seemed to be a popular one. I already installed anaconda for it,

[ruby yoke]

you might need to wait till @ornate elk is back around, but if you search on his name, and read his previous posts, you might find information that'll help

[worthy sapphire]

I got the DirectML version to work from the guide they linked. ZLUDA version wouldnt work right.

[bitter pelican]

Why am i getting the "Error - ClientError: Received client error (400) from model. See the SageMaker Endpoint logs in your account for more information." error

[final mason]

can any one tell me how to fix this issue

[crude halo]

are there any inpaint controlnet models for SDXL that work with A1111? I tried many and none worked so far

[ruby yoke]

what are in the SageMaker endpoint logs it refers you to?

[final mason]

where is that

[ruby yoke]

what were you using that generated that error?

[final mason]

is this what you mean or

[ruby yoke]

sagemaker is an amazon web service item. were you trying to install something or just run a model?

[final mason]

model was trying to get rid of the stupid runtime error fixed i said it b4 but il say it again if i use --medvram it says runtime error and if i use --no-half i get attribute error idk what its exactly the other guy was helping me but i guess he got off

when i try and generate art

[ruby yoke]

is it your model or one you got from somewhere else?

[final mason]

cetus mix

but other ppl have the issue and it looks like theres a fix

[ruby yoke]

and are you running it at home or on a rented GPU on a website?

[final mason]

idk what tho

at home

[ruby yoke]

then i'm really confused why you're getting that error. you might have to wait till he logs back on

[final mason]

ya thats fine im pretty sure there is one i see other ppl fix it but idk how

[rustic roost]

Could someone help me with an error? I downloaded a few extensions to play around with and now my a1111 won't start. I tried deleting the extensions directly from the extension folder but that didn't help.

Traceback (most recent call last):
File "C:\SD\stable-diffusion-webui-directml\launch.py", line 48, in <module>
main()
File "C:\SD\stable-diffusion-webui-directml\launch.py", line 44, in main
start()
File "C:\SD\stable-diffusion-webui-directml\modules\launch_utils.py", line 689, in start
import webui
File "C:\SD\stable-diffusion-webui-directml\webui.py", line 13, in <module>
initialize.imports()
File "C:\SD\stable-diffusion-webui-directml\modules\initialize.py", line 35, in imports
from modules import shared_init
File "C:\SD\stable-diffusion-webui-directml\modules\shared_init.py", line 8, in <module>
from modules.zluda import initialize_zluda
File "C:\SD\stable-diffusion-webui-directml\modules\zluda.py", line 6, in <module>
import onnxruntime as ort
File "C:\SD\stable-diffusion-webui-directml\venv\lib\site-packages\onnxruntime_init_.py", line 57, in <module>
raise import_capi_exception
File "C:\SD\stable-diffusion-webui-directml\venv\lib\site-packages\onnxruntime_init_.py", line 23, in <module>
from onnxruntime.capi._pybind_state import ExecutionMode # noqa: F401
File "C:\SD\stable-diffusion-webui-directml\venv\lib\site-packages\onnxruntime\capi_pybind_state.py", line 32, in <module>
from .onnxruntime_pybind11_state import * # noqa
ImportError: DLL load failed while importing onnxruntime_pybind11_state: A dynamic link library (DLL) initialization routine failed.
Press any key to continue . . .

[sleek zephyr]

Hello everyone, I tried to write to technical support this morning. My name is Scerbo and I am an Italian digital artist, I recently released a project with expandedart.

The collection was minted on verisart and then claimed on opensea.

Now there is a scam with 14 fake pieces from the collection put up for sale on the polygon chain. I wish I could fix this, how can I do?

Thank you all

[ornate elk]

Hey, yes I made a guide to get comfyui working with Zluda. Checkout my guide:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides

[ornate elk]

Hey, what's your GPU?

For my AMD guides, checkout the link above.

[ornate elk]

Can you show me the full Cmd log when getting the error?

[ornate elk]

Hey, what's your GPU?
To fix the issue you have to delete the venv folder and relaunch

[final mason]

please tell me theres a fix

[ornate elk]

Try change --medvram to --lowvram

Then restart

Oh I see you don't have --medvram or --lowvram.
So add on of these but not both

Try --medvram first

[final mason]

i tryd medvram and i got the runtime error

b4

[ornate elk]

Then try --lowvram

And if that won't work, try using a model that's 2gb and not 4gb

[final mason]

and when i tried lo half it gave attribute error didnt try low yet

[ornate elk]

Remove --no-half if you don't get a NaN error

[final mason]

it seems to work not sure yet it worked once gunna do it again

Id send a pic but its nsfw LETS FKCING GOO ITS WORKING FINALLY

THANK YOU

[ornate elk]

No problem 🙂

[final mason]

it works fine without highres but with it i get this error is it not possible with my gpu? its fine if not im just glad i got it to work finnally theres just a little blur tto it

[final mason]

i changed the upscale to 1 and it worked but idk if thats even making it less blurry

[gilded canyon]

Lately I’ve been having memory errors when using stable diffusion. They aren’t VRAM errors I don’t think since my chrome browser will say “out of memory”.

Does regular RAM affect stable diffusion? I have 12GB vram but only 16GB of regular ram

should i upgrade, does it make a difference?

[ornate elk]

You need the tiled VAE extension. Maybe with that you can upscale with hires fix. (by 1.5 Max)
If that won't work you can always Upscale in img2img with the SD upscale script

[final mason]

ok ty ill give that a try

[ornate elk]

Sdxl models need a lot of RAM to load.
So switching models can fill up the RAM fast.
Best for you is to increase the Windows Pagefile. Enabled it only for C. Set the min to 16000 and Max to 24000 then restart the PC

[ornate elk]

I can later show you an example how to do it in img2img.
Also you should get the Adetailer extension

[final mason]

also does listing them like this mess anything up? i like to have it like this cuz im illiterate and have a hard time reading

[ornate elk]

No shouldn't mess it up but words need to be separated by ,

[final mason]

ok ty

[final mason]

ill do that if this doesnt work can you send link possibly please?

where do i get extensions

[ornate elk]

you get all extensions in the extensions tab

click on Available, then on "Load from" to get the list

[final mason]

ok so i installed xformers to speed it up and its like stuck on installing xformers

is it normal to take this long

[ornate elk]

xformers is for nvidia only and will break your webui xD

[final mason]

oh crud

[ornate elk]

close the cmd

remove --xformers

then delete the venv folder

then relaunch

[final mason]

ok deleting it now

so im guesing theres no speed up thing for amd

which is fine it does it in around 5 min

[ornate elk]

the arguments from my guide are already the best you can do for speedup :/ and --medvram will be faster than --lowvram

[final mason]

ok thx anyways its redownloading it i guess

[final mason]

?

[ornate elk]

Check the webui-user.bat

It needs to be --no-half-vae

[final mason]

oh i thought you said to get rid of it my bad

[ornate elk]

Then relaunch

Only of --no-half

[final mason]

so like this?

ok so where do i get the extensions

im ghuessing this

whic tag is it in

the vae thing

is this correct or do i get it on a dif site?

i got prompt history to work which is nice

[ornate elk]

Yes this

[final mason]

oh well i tried that one and it made like a weird gray tile

what settings?

are good

[ebon sandal]

anyone knows how to fix this?

I already tried the "Try fix" button and also tried reinstalling it

[final mason]

why so many bots popping up with scams?

also wondering if i need to have the 2 boxes checked or just one or does it even matter

[ornate elk]

For Hires Fix, The Important parts are the Resolution and the Hires Steps:
Always use 10-15 Hires Steps to not get out of vram. (10 for your gpu)
Not using the right Encoder/Decoder Resolution, or a uncommon Image Resolution will result in gray squares in the Image.
So if using 1.5 based models, set it to Encoder: 1024, Decoder: 128

[final mason]

wheres the encoder thing

like option to change it

and do you mean sample steps?

oh nvm i see hires steps

[final mason]

oh you mean the size thing

[ebon sandal]

why is this happening to the side of the image

[finite zenith]

Does anyone have SD3 running on A1111?

Their latest RC patch has SD3 support thought I might try it

[willow kettle]

Trying to get IP-Adapter faceid-plusv2 working in A1111 with an AMD GPU. I'm using @ornate elk's guide. When I try to run the preprocessor (ip_adapter_face_id_plus), it crashes with

[E:onnxruntime:, inference_session.cc:1981 onnxruntime::InferenceSession::Initialize::<lambda_b88fb06a047eb3fba81e04c489d52b7d>::operator ()] Exception during initialization: D:\a\_work\1\s\onnxruntime\core\providers\cuda\cuda_call.cc:121 onnxruntime::CudaCall D:\a\_work\1\s\onnxruntime\core\providers\cuda\cuda_call.cc:114 onnxruntime::CudaCall CUDNN failure 4: CUDNN_STATUS_INTERNAL_ERROR ; GPU=0 ; hostname=DESKTOP-E5QQI2O ; file=D:\a\_work\1\s\onnxruntime\core\providers\cuda\cuda_execution_provider.cc ; line=178 ; expr=cudnnSetStream(cudnn_handle_, stream);

I think cudnn is not supported by my AMD setup, so I think my first goal would be to see if I can just have the CPU run these functions?

[final mason]

can someone help me i dont have a vae folder like just vae i have the other vae folder

wanted to try grapefruit vae

[willow kettle]

I think you can just make a VAE folder and drop it in

[final mason]

ok ill give it a try

[finite zenith]

is this a VAE problem?

[shadow folio]

getting this error, can anyone help?

ValueError: Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined.```

[ornate elk]

I know this error. Its fixable. But you have to remove the Reactor extension

[rustic roost]

That fixed it. Thank you sir and or ma'am! You have saved me from a headache and losing what little sanity I have left. Thank you!

[final mason]

idk if i put this in prompt help or this help cuz my prompts worked b4 with little to know issue

not sure if its the vae

or something

i have vae on none and its still doing it

[willow kettle]

lol or (hear me out), I go into the controlnet scripts preprocessor setup py file and edit the line providers=['CUDAExecutionProvider', 'CPUExecutionProvider'], to remove the CUDA execution provider. I mean... it worked!

[final mason]

im losing my mind idk what we did but now im not getting any girls im only getting art like this

im like 2 seconds away from reinstalling every thing

i cant belive that it was working a second ago and i have no clue whats causing this

its not my prompts

[ruby yoke]

screen shot and post your settings

[final mason]

this? cuz i dont think its this it made a girl b4

now im just getting art

its pissing me off it was working fine b4

oh and i normally have it on this

and it still does the stupid art

pleeeeeeeeeeeeeeeeeeeease

[ornate elk]

Change the sampler

Euler a or DPM++2m

Then it will work

[final mason]

i have it to that

i swear its something else

something screwed something up

what if i delete the venv folder again will that reset stuff?

i swear its a setting or something else messing this up cuz it worked fine b4

its not the vae thing cuz i took it off completly and its still doing it

[ruby yoke]

okay, first let's clear out your cache files. reboot your machine, please. and i mean turn it off and leave it off for at least 10. seconds. then boot it back up, don't change anything, just render one time and post it

[final mason]

like delete this folder?

or just the inside

[ruby yoke]

no. memory cache. turn the machine off, leave it off for 10 seconds so all the electronics discharge, then reboot

[final mason]

nope still doing this

[ruby yoke]

alright. did you change anything at all between when it was working and now as far as your settings go?

[final mason]

thats part of the issue im not 100% sure what it could be cuz really the only thing i changed was the vae pop up ui and tryd it then thats when it started to break

then i took it off completly and its still doing it

also when i hit apply and the reload ui it stays stuck on reload and i have to close out completely and restart then the ui works

would deleting venv fix anything

[ruby yoke]

i don't think it's your environment. what model are you using?

[final mason]

its not the model trust me

[ruby yoke]

what model are you using?

[final mason]

it worked b4 cetus mix but its not the model

[ruby yoke]

please tell me what model you are using. i'm not saying it's the model

cetus mix. does that model have a VAE baked into it?

[final mason]

one sec where do i look

[ornate elk]

It worked?

[final mason]

https://civitai.com/images/1322390 idk this is the url

[ruby yoke]

yeah, i was just looking at that one. it's a 1.5 checkpoint

[final mason]

idk what that means but heres this really the only thing i messed with

i have a feeling its something with the vae

[willow kettle]

Hahaha yes, ugly hack but it works just fine. Out of curiosity, if I remove ReActor, would that fix it or would I need to change some settings?

[ruby yoke]

it very much is. did you put those settings back like they were when it was working?

[final mason]

honestly dont rem what they are default i thought it was automatic

trying it with this vae real quick however it was doing it with grape fruit when it was on\

[ruby yoke]

you have to use the right VAE for the model, that's not something you adjust

[final mason]

oh great didnt know that

ya its still doing it so what do i set it to

also wondering if theres a fix to this inf reloading

i have to turn it off and on again

[ruby yoke]

are you seeing anything in the logs?

[final mason]

where log lets just focus on the first thing the main issue

[ornate elk]

You would need to pip uninstall onnxruntime-gpu and install onnxruntime
The later one is for amd

[ruby yoke]

the log is going to tell you if there are errors that are causing the reloading issue

[willow kettle]

Oh. I should do that anyway then. But why would I need to remove ReActor? It has a setting to use CPU.

[final mason]

where they at?

[ornate elk]

If your hacky fix worked then you can use both. But for me IP-Adapter face Id works only with onnxruntime.
But Reactor installs onnxruntime-gpu that's why it needs to be removed

[final mason]

i removed the other ones i installed and set it to auto and its still doing it do i try and reinstall the model?

[ruby yoke]

@ornate elk any idea what might be causing his issues?

[final mason]

any way to reset it

[ruby yoke]

are any of your other models doing the reloading thing?