🤝｜tech-support

[willow rapids]

aaah thanks thanks

is it pinned?

[ornate elk]

The first link

[willow rapids]

thx! <3

another question

[paper crag]

ok... so i ran cmd in my sd-webui folder, typed git pull, didnt work

[willow rapids]

which is the most optimised? and can run flawlessly

ill get that

[ornate elk]

Forge will work good for you

[willow rapids]

<3 thansk!

[ornate elk]

Yea the log from your screenshot shows that you have unsaved changes in the requirements_versions.txt
So you have to delete that file.
Then run git pull.
Then delete the venv folder and relaunch the webui-user.bat

Also you need to disable your AMD CPUs integrated GPU called Radeon TM Graphics.
Before relaunching it

[paper crag]

and we're back in business, thanks!

[primal pendant]

why would this suddenly start happening

OutOfMemoryError: CUDA out of memory. Tried to allocate 21.80 GiB. GPU

why trying to use so much

it don't matter what i try to create

[ornate elk]

Which settings do you use?

[primal pendant]

for what ?

do you mean in the ui or do you mean in the ini ?

[ornate elk]

For the generation

When you get the error

Like resolution etc

How much vram do you have?

[primal pendant]

i have 6700 xt so 12 gb

i mean i use to be just fine not used ina while went to use tonight and had never seen this before

i'm installing another

that was forge ui

hmmm seems forge broken automatic1111 works fine so weird

[ornate elk]

Hmm okay, forge worked for me too

But maybe its just a setting

[primal pendant]

well i used forge before been fine so it's bugged out using way more vram no idea how to solve it

i alread deleted and rebuilt the venv didn't help

i got no idea why it would suddenly double up on usage liek i can't create even a simple image

i guess i need to remove it fully and start over

as automatic works perfectly fine

[ornate elk]

Which txt2img settings did you used in forge?

[primal pendant]

it don't matter anything even putting a cat makes it use that amount

it's just using way more vram than it should

i removed forge now

[ornate elk]

ah okay

[primal pendant]

and automatic is handling with ease just like forge use to it's just weird why it would bug out like that

[ornate elk]

Yea sounds weird

[timber plank]

Has anyone had an issue trying to load the t5 text encoder gguf files?

I've placed it in clip, clip_gguf, t5, and none of them are detecting the q8 t5 text encoder.

I've opened the proper nodes as well, dualcliploader (GGUF)

It doesn't seem to be detecting it at all.

[golden fern]

Is there a solution to reduce my computer performance while generating images? My computer always freezes when generating images it hink that because SD takes 100% of my memory. I hope to limit my computer power, I don't mind if it takes longer to generate images.

[timber plank]

You can use --reservevram and the amount of memory you want to reserve for comfyui to prevent that from occuring

when launching comfyui*

Nevermind, a git issue has it here.

https://github.com/city96/ComfyUI-GGUF/issues/151

[honest quail]

anybody know where I can get euler a

for some reason I don't have it

[ruby yoke]

are you running comfy?

[honest quail]

yep

[ruby yoke]

unload comfy and kill the commandline. then go into /comfyui/update and run the update.py batch file. cause it comes with comfy and if it's not there, something's wrong with your install

also- it's called euler_ancestral, not euler-a

[honest quail]

well I do have that

does everyone call it the latter for short?

[ruby yoke]

that's what the a stands for. in all of them. ancestral

[honest quail]

ahhhh

nobody mentioned that on reddit lmao

[ruby yoke]

a lot do. and you'll see this: heunpp2, heun++2

same thing

[honest quail]

well dang

[ruby yoke]

link to post?

[honest quail]

I gotta get used to the terms

I've just looked through a lot of posts on recommened samplers

[ruby yoke]

in the ones that use pp it usually stands for plus plus - which is what ++ also stands for

[honest quail]

makes sense

what scheduler would you recommend

oh right since I found this server I wanted to mention an issue I've had running img2vid, but I'll quickly try and do that again first

[ruby yoke]

wellllll that depends on a whole lot of other things. first, i'm going to give you a link https://huggingface.co/docs/diffusers/en/api/schedulers/overview welcome to the world of math you probalby never wanted to enter. read through all the schedulers there on that page. there's an index on the left sidebar

[honest quail]

oh right I remember one thing

I could never get this node working

[ruby yoke]

there should be errors in the console log

[formal patrol]

i have two gpu's one with vram 12 gb on with 16 gb. is there any possibility to run stable diffusion video using these two. its would be a great help. i am new learner .

[hallow latch]

I am going to try and use the ComfyUI .. Do I need to install ComfyUI first and then SwarmUI ?

[tall geyser]

I am trying to create images like the following, but they always come out blurry. I don't have any problems with other images with all the same settings. I also sometimes get blurriness with photos of white sand beaches. Is it the brightness? What is it about this prompt? "flat white background, 3D Unreal Engine render of an old green wooden chair"

[ornate elk]

SwarmUI is Comfyui with a simpler interface.
You can either install Swarm and it will install comfyui. Or you can install just Comfyui

[ornate elk]

You can't combine their vram
Each GPU can only use their own instance

[hallow latch]

What would you recommend? 🙂

[ornate elk]

Idk, but as you have also a comfyui tab in swarm I would use swarm

[hallow latch]

Should be enough to just install the SDK right?

[median geode]

Hello! May I ask if running AI tasks for a long time puts special demands on GPU cooling? Should I choose a graphics card with better cooling performance? Also, is 32GB of RAM enough, or would I need to upgrade to 64GB?

[fervent shoal]

Was anyone already able to train SD 3.5 on KohyaSS ?
He released a new repo named 3-3.5-FLUX ?
Anyone ?

[ornate elk]

Yes only the SDK is needed

[rotund granite]

Hi, I would make an anime version of a picture of myself and use that as my profile picture. I'm new here and am a little overwhelmed with all the models. Which one would you recommend to me?

[hallow latch]

Okay nice :3

[hallow latch]

Hmm, any settings to turn off within SwarmUI to make it better?

They turn out like this .. with VAE "anything" I know there is some settings in auto 1111 to turn on / off but not in swarmUI xD

[iron lantern]

Are there any models like pony that will run in sd 1.5

[twin solstice]

Hi! I'm attempting to call pix2pix through an API using Python requests. The response I'm getting is ['{"status":"failed","id":"","messege":"Failed, Try Again","output":"","tip":"Get 20x faster image generation using enterprise plan. Click here : https:\\/\\/stablediffusionapi.com\\/enterprise"}']. Does anyone know how to fix it? The FAQ at ModelsLab hasn't been very helpful so far.

[hallow latch]

Any reason why this happens with SwarmUI?

[ornate elk]

xl model + 1.5 vae = this image

[hallow latch]

Okay that is understable, what about 😄

[hallow latch]

Auto 1111 compated to SwarmUI ( Auto 1111 ) -> ( SwarmUi ) They have same prompts, same vae and same negative and same Steps, model etc 😛

[true sphinx]

hey so ive been trying to install controlnet on automatic 1111 it seems to have installed correctly the folders r where they are supposed to be

but there is no box showing up on the generation tab

I do have the mikubill version so that is not it

[ornate elk]

Can you show the cmd log?

[true sphinx]

let me do a restart so its not clogged by 3 reloads

For somereason even with a basically fresh install im getting all these errors

around batch hijack and such

the errors are spooky lol

[iron lantern]

what gpu do i need to make pony models ?

[true sphinx]

like make a merge

what base model sdxl or 1.5

[vital void]

can anyone help me fix this error? is it a gpu problem?

[ornate elk]

What's your GPU?

[vital void]

gtx 1660ti

[ornate elk]

Then you need to have --xformers --medvram --no-half in your webui-user.bat

[ornate elk]

To use or to make?
Pony models need at least 6gb vram but better is more

[ornate elk]

Hmm Controlnet and mishaiou assistant get mentioned in the error

You could try disable mishoau assistant, apply, and then delete the venv folder and relaunch the webui-user.bat

[true sphinx]

thats really odd lol

[drifting tundra]

Does anyone know what causes this using facedetailer with SD 1.5 model for an image made from SDXL?

[true sphinx]

ok

[true sphinx]

ok so disabling assistant worked

wtf

[vital void]

I did add those but now every time I do anything my pc is about to explode plus everything takes forever to load I let a image generate for 10 mins💀

[ornate elk]

can you show me the txt2img settings you used?

[ruby yoke]

you're trying to use 2 different base models.

[drifting tundra]

as long as I use 2 SD 1.5 models

or SDXL models

it will work

if I use 1.5 SD for generating and SDXL for the detailer or vice versa it will not work

[ruby yoke]

yea. that's what that error is telling you. you're using two different base models. you can not do that

generate an image with one, and then run that image into the workflow for the other

[drifting tundra]

something like image2image?

[ruby yoke]

not really, because with image 2 image, you're using the image as a prompt. but you're trying to use sdxl more as a refiner. however the way you have the workflow set up, you're asking it to do something it can't do

[drifting tundra]

Do you have a workflow for this refining?

I always used only 1 model for generating an image

like 1 sampler

I thought I read somewhere that refiners are no longer required?

[ruby yoke]

i don't have one, but you might ping @limber pier in the #🆕｜sd3 or in #💬｜general-chat and see what he has

[drifting tundra]

Thank you 🙂

[ornate elk]

hmm looks normal

are you trying to use an sdxl or pony model?

and how much ram do you have

[vital void]

pony

I have 16gigs

[ornate elk]

okay, then you need to increase the windows pagefile

Instructions here: https://www.tomshardware.com/news/how-to-manage-virtual-memory-pagefile-windows-10,36929.html
Enable it only for the C drive and disable it for any other drive.
Set it to customized: 16000 min and 24000max.
Then restart the PC and try to load the model again.

[vital void]

I downloaded mine on my second drive is it the same?

[ornate elk]

best is to have the webui on an ssd

but the pagefile should always be on C as its the fastes drive normaly

[vital void]

my 2nd drive is also an ssd faster even

[ornate elk]

then you could also set the pagefile to be on that drive

only set the pagefile to be on one drive

[vital void]

I did that it took me 15mins to generate an image and as it was about to get rendered it crashed💀

[ornate elk]

thats very strange, have you restared the PC before?
also make sure your not running any gpu heavy stuff like games or wallpaper engine

[vital void]

I don't have anything on I am going to restart again can my gpu even handle image generation?

[ornate elk]

with 6gb vram it could. but sdxl nad pony are heavy

maybe you could try forge webui as its a bit more efficient than auto for low and midrange gpus

[vital void]

I set the resolution to 200 x 200 it went much faster but it again crashed at the end

[ornate elk]

okay then you should try forge

[vital void]

I think forge was it I runs well for now

many thanks

[ornate elk]

nice no problem 👍

[distant narwhal]

Hello, Whenever I am using the Euler sampling method, my pictures always look decent through the steps, but then on the last one it gets blurry like this, almost like a horror movie lol, I am relatively new to this but I have looked and havnt found a proper answer to this

[onyx solar]

Some model types will not work properly with some sampler/scheduler combinations.

[distant narwhal]

oh alr, lemme try it with a different model then!

[ruby yoke]

what model are you using?

[onyx solar]

Note I'm specifically referring to model versions, not variations of the same version.

[distant narwhal]

i tried it initially with anything v3, then I just tried it with anylora

[distant narwhal]

and it did the same thing at the end

[ruby yoke]

what model, not what lora

sd 1.5, sdxl, sd3, pony, something else?

are you generating at home or on a website?

[distant narwhal]

at home, im using the webui rn

the first one was called anything v3

[ruby yoke]

that's a checkpoint. what's the base model it's for

[onyx solar]

Anything V3 looks like a 1.5 model, from what I see.

[distant narwhal]

idk how to check that sorry, like i said, im new.

[ruby yoke]

it probably is but i can't remember for sure

[onyx solar]

When you downloaded it, did you pick it up from CivitAI?

[distant narwhal]

yeah

[ruby yoke]

if you downloaded that from CivitAI - on the page you downloaded it from, on the right side, is a box with all that information in it

[distant narwhal]

it is SD 1.5

[distant narwhal]

just found it

ye

[ruby yoke]

okay, for sd1.5 use Euler for the sampler and normal for the scheduler

and make sure you've got the right VAE for 1.5 as well

[distant narwhal]

how would Ik i have the right vae?

[ruby yoke]

usually, if it doesn't have the VAE baked in, it'll tell you on that information box and tell you which one to get

[distant narwhal]

i dont see anything saying it so I believe it has a VAE baked in then?

[ruby yoke]

probably.

[distant narwhal]

it still comes back not like it should using Euler and Normal

looks like this

[ruby yoke]

wow. that's - that would actually make a realy cool window

[distant narwhal]

lol true, doesnt help im tryna make it work for a person

and from what ive seen, it only blurs this stuff with the Euler sampling methods

not the DPM++ 2M one

[ruby yoke]

i know you didn't say you're using comfyUI, but what are you using to run stable inside of?

[distant narwhal]

nvm now its doing it with this one too

[distant narwhal]

May I ask how I check that?

unless u mean what im starting it with

[ruby yoke]

yes. what are you starting it with?

[distant narwhal]

just the webui-user Bat file

[onyx solar]

Automatic1111

[ruby yoke]

okay. something seems set up wrong - but i don't use auto1111 so i'm not sure what settings to adjust

[distant narwhal]

it says Stable diffusion webui Automatic111

[ruby yoke]

@onyx solar do you use auto1111?

[onyx solar]

No, I'm a ComfyUI guy.

[ruby yoke]

me too

[distant narwhal]

okay, is comfyUi better or smth?

[ruby yoke]

um - in my opinion it is

[distant narwhal]

should I delete then install that or something?

fyi, i am using am amd gpu if that will affect anything

[ruby yoke]

actually, why don't you wait till the A1111 guys log on and see what they have to say

[onyx solar]

That's subjective. Better for some things, sure. Automatic1111 is good for a lot of beginners and people that don't want to get into node-based workflows.

[distant narwhal]

alr

[ruby yoke]

and you need to read through all the pinned posts in this channel for AMD stuff

[distant narwhal]

ykw, Ill just try installing the stuff from pinned messages, i installed this version using a yt video so Ill just try this and see if it works better

[ruby yoke]

you might want to clean it up and reboot first

[distant narwhal]

ye I will, thx

i didnt realize it was this easy for the stuff

I spent an hour yesterday tracking for a simple line of code to allow my gpu to be used

it sucked.

[ruby yoke]

if you wnat it to be really easy, then install swarmUI and let it deal with all the technical stuff - and just run comfyUI inside it

it'll auto detect your GPU, install what you need, and so on

[distant narwhal]

is there any downsides to that? or no?

[ruby yoke]

https://github.com/mcmonkeyprojects/SwarmUI

[ruby yoke]

well sure - if you really like battling with code and computer command line stuff, you don't get to do that

[distant narwhal]

thank god

[ruby yoke]

are you on windows?

[distant narwhal]

Yeah

Windows 11

[ruby yoke]

then go to that page, scroll down till you see the Installing on windows header, and follow instructions

if you get stuck, they have a discord

[distant narwhal]

Alrighty, sounds good

do I have to do anything specific to uninstall the stuff or no?

[ruby yoke]

i don't know. again, you should wait for the auto1111 guys and ask them - you might

how much space do you have on C:

[distant narwhal]

i use my D drive for Stable diffusion, they both are 1tb ssd's

[ruby yoke]

you need space on C: - certain files will need to go there no matter what AND that's where your swap space is.

how much do you have?

[distant narwhal]

368gb free on my C

[ruby yoke]

okay, that's plenty

you should be able to go ahead and install swarm and get stuff configured, and then ask them later about uninstalling auto1111 if you still want to. though i think swarm might run it as well as comfy

it does a lot of things

very useful things

[distant narwhal]

alright sounds good, and swarm is still like beginner friendly u would say right?

well swarm, then comfy inside of swarm

[ruby yoke]

yes. swarm is extremely friendly. and if you get stuck, we have a #🐝｜swarm-ui channel, and the dev has a discord. and is there a lot of the time

[distant narwhal]

alright then, thank you!

[distant narwhal]

Sorry quick thing, it says to click No when installing swarm in the github, but it has an amd specific version, so which one should I click for Zluda?

im just confused on the "No for default installation (not amd)" Like does this mean click no if I dont have amd? or click no even if I have amd

[ruby yoke]

the default installation is for non-amd gpus. if you have an amd gpu then say 'no, you don't want the default'

and as far as the rest of the question goes, that's a question for the #🐝｜swarm-ui channel

[distant narwhal]

so then click yes to install the amd version?

[primal wigeon]

Hi, I'm new here. Should I install ComfyUI using the portable standalone build for Windows or install it manually?

[primal wigeon]

Also, should I run on a gpu or a cpu?

[ornate elk]

What's your GPU?

[distant narwhal]

7900GRE

[ornate elk]

Ah nice, then I would recommend starting first with getting the Automatic1111 with ZLUDA version installed.
If that runs well you can always install comfyui/Swarmui, forge, with my guides too

[distant narwhal]

Alrighty, sounds good!

Thank you!!

[ornate elk]

Np, im on a 7900xtx and I made the AMD Guides from here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides
If you have any questions feel free to ask me.

[distant narwhal]

Will do!!

[untold flame]

Please for the love of god, does anybody know how to merge a LoRA into a transformers format model and save the output? I cannot find anything at all on how to do this. My project cannot advance until I find a solution

[ruby yoke]

um why?

[untold flame]

for training in AIT. It requires models to be in diffusers format, which is a pain in the ass, but it trains much better than the others I have used. I need to be able to train a lora, merge said LoRA back into the model, and then train from that model after. In order to do that, I need to be able to merge in a LoRA and save it somehow

[ruby yoke]

why does it sound like you're doing this

[untold flame]

I mean, its a pretty straight forward issue, with a pretty straight forward explaination

Train LoRA in AIT, merge LoRA into model, train new LoRA on merged model in AIT

[ruby yoke]

can't you just make several loras and merge them?

[untold flame]

no

I mean, you can, but thats not ideal at all

[ruby yoke]

what format is your lora in?

[untold flame]

I am honestly not sure, it just outputs as a single safetensor file. Its whatever AIt outputs in, but I can't find any information on that as of now

[ruby yoke]

i think the developers you really want to talk to are over on the banadoco server

[untold flame]

I figured I would try here first, since there are usually a lot of smart people here. Its a very simple problem, but unfortunately diffusers never does anything in a straight forward or logical way 😅

Its why I try to stay away from it as much as possible

[ruby yoke]

you could start by asking @limber pier and @fair oxide

[untold flame]

If either of you two lovely people happen to know, I would be GREATLY appreciative

[limber pier]

no idea, it would probably take a couple hours of digging through code to sort that out

haven't touched diffusers in a long itme

[untold flame]

it continues to perplex me why people still use diffusers when its such a messy pain in the ass 😭

[limber pier]

everything sucks tbh

[untold flame]

Yeah, thats fair honestly

[limber pier]

diffusers is at least documented

comfy's code has zero documentation whatsoever

and types aren't declared anywhere but wherever the object is created

so if you want to figure out wtf blah_object_name.sample() is actually doing or calling, you gotta debug

[ruby yoke]

i.e. standard coding practices

[limber pier]

so diffusers does have its advantages... they have actual fn documentation

it's pretty annoying to have to reverse engineer open source stuff at runtime

[ruby yoke]

better than having to reverse engineer your own stuff cause you forgot what you were doing

[limber pier]

lol

[untold flame]

Yeah, I do agree with that. It just seems any time I need to od something with diffusers, its impressive how little documentation there actually is lmao

[limber pier]

yea it's really just flavors of problems

everything's a mess

the field is moving so fast no one has time to stop and clean shit up

cuz it's onto the next thing

[ruby yoke]

it's slowed down some. there've been several tsunamis in the last couple years where advances were coming so fast that papers were just tweets on twitter with a release

[untold flame]

not being able to fix this is what is holding me back from being able to make a way better flux/lite tune so I can train SD3.5 off of it

I am potentially teaming up with a few friends to make a hopefully fantastic larger scale tune of 3.5

[ruby yoke]

well you PLEASE not do that? you're going to destroy all the work we put into 3.5 just to try to turn it into flux

flux is MASSIVELY overfit for several things

that's why you get the pretty pictures otu of it

it's a bloody LORA itself

[untold flame]

Said friend already has hands down the best Flux tune I have seen which gives SD3.5 a huge run for its money in creative/artistic stuff, but its not commercially viable, so we wanna try on SD3.5 medium/large and see what we can do

[ruby yoke]

flux is such a horrible joke it should be taken out, shot, and buried. do not mess 3.5 up with it

you want 3.5 to do what flux does, exactly like flux does then 1. scrape midjourney 2. over fit your data set with women, dogs, fantasy image,s and anime cat girls, and then after you train it on that, freeze it with DPO

[untold flame]

That's your opinion, and I respect it. I am still trying to do what I want however. Its not meant to be a tune to turn 3.5 into flux at all. Its just meant to help supplement gaps in data that we can't find for photographic subjects, which it does very good

[ruby yoke]

then it'll make really nice women in a very narrow range

[ruby yoke]

it's not my 'opinion' - it's based on hundreds of hours of reserch digging through flux one token at a time AND information from people that know wht was done

[untold flame]

ok man, again, I am not trying to get into this. You aren't even listening to me saying thats expressly not what I am trying to do 😭

[ruby yoke]

well that's what you're going to wind up doing.

[untold flame]

I don't want SD3.5 to be flux, I should have made that even clearer

[ruby yoke]

then don't train it on flux

[untold flame]

by having a few hundred supplemental flux gens mixed in with 10's of thousands of real photos? Thats gonna destroy the model?

[ruby yoke]

you don't need to train anything on flux for taht. just make images and put a data set together and train a normal lora

[untold flame]

this is something i am trying to do before we decide if we are gonna do a huge scale tune or not. I can't really work on the project until then

[ruby yoke]

and go geneate

better yet, hire a photographer, shoot real people, and train a lora on that

[untold flame]

I am a photographer, and not all things can be easily photographed, especially not things that don't exist, which is the problem I am trying to supplement

anyways, our goal is to make a general SD3.5 tune that can do photographic realism and also creative art stuff like this

[ruby yoke]

3.5 does that just fine without being tuned

[untold flame]

We were gonna od it on flux, but then we remembered we can't make money off of it, which is unfortunate, but not the biggest deal, so we are gonna dup much more compute into medium/large

[ruby yoke]

you should actually look at the nodes @limber pier has been creating - what he's got going might be more along what you're after

[untold flame]

My friend already made a flux tune that handily outclasses SD3.5 in artistic and photographic stuff from what I have seen, and he wants to see if he can do even better on SD3.5 given my help an dsome huge compute from a friend of mine. The goal is to do just that, if we can logistically figure a way out to do so

[ruby yoke]

we need to move this out of this channel and back into #🆕｜sd3

[untold flame]

oh yeah, good point

[misty dagger]

Hi everyone! I'm working with Fooocus on Google Colab, and I'm trying to replace one LoRA model with another, but I'm having trouble getting it to show up correctly in the Gradio interface. Could anyone give me a hand with this?

[visual bane]

I want to sent api request to automatic1111 from other private network but when i use --listen it is taking alot of gpu ram which stop generation can anyone tell other way ?

[misty dagger]

@visual bane To avoid using --listen (which consumes GPU RAM), you can set up a reverse proxy with Nginx to make the Automatic1111 interface accessible externally without increasing GPU usage. Here’s how:

Run Automatic1111 on the default local address (without --listen), just using 127.0.0.1:7860.

Install and configure Nginx on the server to route external requests to the local interface:

server {
listen 80;
server_name YOUR_IP_OR_DOMAIN;

location / {
    proxy_pass http://127.0.0.1:7860;
    proxy_set_header Host $host;
    proxy_set_header X-Real-IP $remote_addr;
    proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
    proxy_set_header X-Forwarded-Proto $scheme;
}

}
Restart Nginx to apply the changes (sudo systemctl restart nginx).

Now you’ll be able to access the interface at http://YOUR_IP_OR_DOMAIN with minimal impact on GPU RAM.

[visual bane]

Thank you so much for your help @misty dagger ! Your suggestion really helped me out a lot.

[misty dagger]

Happy for u !

[serene basin]

I got really comfortable with the generation and inpainting, settings, features etc in Automatic1111, but I've switched from Colab Pro to a local machine and so now it's back to the usual Automatic1111 python dependency nightmares. Is there anything worth trying for SD on Windows other than ComfyUI and Invoke? I'd much rather use A41 but I can't grow old trying to fix this thing all the time.

[hallow latch]

This is not an issue but question about VAE Extenstion

I know it was a person in here who said for me to run with 1536 x 96 on it.. and the default is 2048 x 128 but I don't know why I should run it with the first mentioned number. Is there a reason ?

[quartz musk]

is there a guide somewhere i can follow to upgrade to diffusion 3.5 as a stand alone?

all I get is this with 3.5 at the moment .. someone mentioned this webui doesnt handle 2.5 or something? its all totally new to me

[hallow latch]

I heard that this was because you are using a VAE that is for SD1.5 or other way around. A vae that is not suited for the same checkpoint.

Alright, then you are getting some help 🙂

[neon patio]

I’m running into a problem getting my app to work on devices outside my local computer, to use the API key it has to run through localhost port 5000 right?

[ornate elk]

Currently only comfyui supports sd3.5

[primal pendant]

hmmmmmmm

why would my gtx 1080 handle this but my 6700 xt won't same image

where do i reduce size? in comfy?

it's a sd 1.5

[ornate ruin]

@ornate elk

What's up

I know you have enough of me by now, I get that

But I figured I'd share something with you. - I finally made IP Face Adapter work.

We forgot that there must be the Face IP Lora inserted into the positive prompts 😭 well. It's working now.

[ornate elk]

What's your workflow?

[quartz musk]

scammers

[primal pendant]

i deleted comfyui now as that's ridiclious how my less vram card does with ease and yet 12gb vram won't create same image wanting more vram is nvidia really that much better where 8gb is better than 12gb exact same work load image

[quartz musk]

scammer

[ornate ruin]

My 4060 with 8GB works fine but it's slow. Takes about a minute, likely 2 minutes for a decent looking hires fixed image

[primal pendant]

yeah but what i'm saying is why is my gtx 1080 better than my 6700 xt lol

8 vs 12

and comfy wnated to use more than the 12 gb like leeching 18gb

and yet same exact workload works fine on the 1080 used just 8 no issue

[steady thistle]

Anybody want to help me with my OmniGen install? I got this error saying no available GPU

I used this web page, but I used a venv to install it. https://github.com/VectorSpaceLab/OmniGen

[steady thistle]

oh, got it. I had to do pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118

[loud pond]

Can the negative embeddings for SD1.5 be used on Pony?

[onyx solar]

No

These are 2 completely different architectures and, frankly, you don't need them.

[loud pond]

oh alright

thanks!

[onyx solar]

More info in case you want the really long answer:
https://civitai.com/articles/4859/the-incompatibility-of-sd-15-embeds-with-sdxlpdxl

[ruby yoke]

pony effectivly destroied the embeddings - it is its own animal

[loud pond]

so what should i use instead of embeddings? just normal lora?

[loud pond]

thank you!

[ornate elk]

Did you installed comfyui with ZLUDA ?

[ruby yoke]

are you training something to use with pony? then train whatever it is on pony. if you want to use pony as your base model, or a fine tuned version of it, then your loras must be trained for pony

your embeddings are numbers. they represent points in space. think of a map. if you want to go somewhere in Las Vegas, and i give you a map of new york, are you going to get where you're going?

Think of embeddings like map coordinates. If you want to find a location in Las Vegas, a New York map won't help. Similarly, AI models need the right 'map' (embeddings) to generate accurate images or understand text.

[primal pendant]

@ornate elkyeah ofc all is with zluda except the old pc with gtx 1080

[ornate elk]

Okay, idk what your workflow was, but comfyui is not as good as auto1111 or forge with ZLUDA.
It requires a lot more vram.

Also, did you used my comfyui zluda guide or the other repo ?

[primal pendant]

it was all your guides for all amd installs but as for nvidia that was something called Stability matrix and i was using comfyui and it used only the 8gb vram without issue but exact same workflow on amd setup and it wants to use more and does the error running out of vram on last part @ornate elk

[loud pond]

im using the negative embeddings to improve my image generations

thats what they r used for right?

[ruby yoke]

um, maybe? they're used to tell the AI what you don't want drawn. whether that improves it or not is a matter of opinion

[loud pond]

yeah, so i thought i could just use the embeddings in the negative section like easynegative, badhands etc..

but i couldnt find any for pony

and you explained how it doesnt work for pony if its from SD

[ruby yoke]

1. the AI has to have seen those labels attached to images and learned what they meant before it can do anything with them. how many images in its data training set do you think have the label "badhands"

2. if you put something in the negative, it'll affect the entire image, not just part of it

[loud pond]

right ok, so should i just use a lora?

but eitherway thanks, ill just avoid the embeddings i think then

[ruby yoke]

you can't use a lora without also using a model. a lora is just a sheet of instructions to the ai to update, add, or change information it has

[loud pond]

👍 , ill just use pony as my model

[ruby yoke]

ok

[loud pond]

again thank you!

[maiden stream]

When I run webui-user.bat I get this error, anyone know the solution?

[ruby yoke]

have you already got it running in another commandline window?

[maiden stream]

no

[ruby yoke]

how much diskspace on your system drive?

[maiden stream]

528 GB

[ruby yoke]

go into the c:\Stable_Diffusion\stable-diffusion-webui\venv\Scripts directory and see if you have a file called activate.bat

[maiden stream]

I do not, I only have python.exe and pythonw.exe

[ruby yoke]

do you normally run in a virtual environment?

[maiden stream]

I literally just followed a tutorial to download stable diffusion today so my knowledge of it is low. I merely followed a youtube tutorial and redid the entire process and got the same outcome.

[maiden stream]

Not sure what you mean

[ruby yoke]

oh. i missed that you were just now installing it. can i suggest maybe that webui isn't really the launcher you want to use to run stable diffusion and talk you into installing SwarmUI instead?

[maiden stream]

I suppose, if you think it is easier/better

[ruby yoke]

it's much eaiser. you can get it here https://github.com/mcmonkeyprojects/SwarmUI are you on windows or linux or...?

[maiden stream]

windows

[ruby yoke]

https://github.com/mcmonkeyprojects/SwarmUI#installing-on-windows

install instuctions are on that page but farther down

it will take care of all the technical stuff for you and if you get stuck, we have a #🐝｜swarm-ui channel

[steep gate]

A1111 does save clip skip info in PNG files but PNG info doesn't show it or copy it to txt2img. Is there a setting that enables this?

[iron lantern]

does this mean my pony model worked

[blissful imp]

Hi I've been trying this for hours and hours and would be super grateful for ur help. I basically have an image of mickey mouse in img2img inpaint. I inpaint the area of the hand I want regenerated, enable controlnet and upload a close up photo of a good mickey mouse hand. I run canny, select controlnet more important, and adjusted every setting possible but the new inpaint area doesn't seem to follow the controlnet (i.e. the good drawing of the hand). What do I do??

[ruby yoke]

if you're thinking that the AI is going to make a copy of the hand onto whatever you're inpainting, that's not how it works. what you're really wanting to do is composite the hand over the image - and that's photoshop

[worldly vine]

Whatsup guys, do you guys know how to implement webui_forge into my current Stable Diffusion

[ornate elk]

Forge is a seperate webui

You can either install it and link your models or you can switch the branches of Auto1111 to Forge

[serene basin]

Where do you rank Forge? Better than A41/Comfy/Invoke?

[distant narwhal]

hey quick question, when downloading comfyui for amd with zluda, on this step it says to not check the pro driver, but the only download I see for the needed version is a pro driver, and this is using the link it directed me to, so does this mean i do install it? i dont install it? or what?

[ruby yoke]

did you read the pinned guides in this channel, yet?

[distant narwhal]

Im using the github link for the installation guides rn, thats where thats located.

its for Zluda aswell, not directML.

[ruby yoke]

i know AMD pulled out of zluda, i wonder if that means they removed the drivers you need from their website

[distant narwhal]

it sends me to the link, its just the version that is being downloaded is forcefully a pro driver.

[ruby yoke]

i would stop where you are and wait till @ornate elk is back online, and discuss thsi with him

[distant narwhal]

Alright then, thanks

[ruby yoke]

welcome

[iron lantern]

Cs1o is the God I just wanna give a massive shout out they helped me so much and got me making nsfw pony models as they where not working on my laptop with python they advised me to try forgeui and from what I understand I got it working loaded in a prompt before heading out today from the cmd data I sent them they said it worked the excitement to head back home and see the final image 😀

[ornate elk]

You have to install the driver thats linked but in the install dont tick the pro driver option at the bottom

AMD named the install file pro, but its hip sdk

[distant narwhal]

ok, so after clicking on it, i dont see the pro driver option

is it on the next page then?

i just dont wanna mess up

pepe_cry

cuz it doesnt say next for a second page, just install

[ruby yoke]

if you don't click the install button, you won't have messed up. check and see what is on the next page

[distant narwhal]

i thought there was another page but it says install below, i dont see a second page

here is what it looks like

wait i may be dumb

Okay no, its just stupid amd is like this

i assume its this

i had to click Deselect all for it to say the pro driver name

before it just said the version of it :I

[ornate elk]

Yes this

Install all but not this one

[distant narwhal]

ye I did dw

thanks for the help

[ornate elk]

Np, now I'm at home, if you have more questions I can answer faster xD

[distant narwhal]

Sounds good!! I probably will have more questions but hopefully not lol

[ornate elk]

For me its over comfyui but behind Auto1111

[agile vector]

Hey I'm trying to figure out a workflow from a post made on civit, using ponyxl. As far as I can tell, everything comes in ok, I have all the nodes, and the same checkpoints and loras, but the only output I seem to be getting is stuff like this. I'm not sure what could be going on?

I assumed vae, but as far as I can tell I'm on the same VAE as the post

[onyx solar]

First thing would be your clip skip setting being wrong. Pony looks for a value of 2, not 1.

Second is that you should be using the SDXL VAE.

[agile vector]

believe thats at -1, it shouldn't be? Never had to touch that in the past

[onyx solar]

Same same. -1 = 1

[agile vector]

this is the vae i'm on, its worked for most things

[onyx solar]

Set to -2

That VAE is for SD 1.5 models.

[agile vector]

Damn, i was trying on some specifically xl vae's from civit and wasn't getting anything else than that. https://huggingface.co/stabilityai/sdxl-vae

[onyx solar]

Third is that your Princess XL LoRA would need to be the Pony version, not the v2 version.

[onyx solar]

The Pony model page has the VAE on it linked in the sidebar.

[agile vector]

Oh im a fool then, lol ty much

I think I had assumed importing the workflow would flag those differences if they were missing like it does for Loras etc

[onyx solar]

I'm not familiar with that expressive whatever LoRA, but I'd imagine that's not a Pony LoRA either.

So pretty much this whole workflow is fucked. 😄

But you now know what you need to do.

[agile vector]

Thats odd, again i just pulled it from a picture on civit so i guess it worked for somebody tho

[onyx solar]

Nope. Wouldn't work for anyone.

The VAE issue alone would stop it from working.

[agile vector]

Am I misunderstanding how comfy imports workflows then?

[onyx solar]

Depends on the file.

Bottom line is that the workflow in your image that you posted would not work. Period.

Not with those values.

[agile vector]

I'm not doubting that, I'm just trying to understand better how that could be, if the metdata is based on the generation of the image shouldnt it just import the workflow the image was generated with?

[onyx solar]

Not knowing how you got to where you were exactly, I can't tell you what occurred. Assuming an honest image with the actual workflow embedded that was used to generate it, Comfy should pick that right up. But do you trust everything you download? Embedded workflows are just text and are terribly simple to manipulate.

[agile vector]

I didn't think that would be common practice honestly, i get this on most of this person's images, it seemed more likely to be on my end

[onyx solar]

It very much could be something you did on your end...even accidentally. But again, not knowing what you did exactly makes that difficult. Either way, if you fix a few things in that workflow, you'd get images.

[agile vector]

I could post the original post, but its nsfw

[onyx solar]

Doesn't really matter...I've given you the stuff to fix above.

It does explain your output, though.

[agile vector]

Fair enough I was just kinda hoping to figure out where a workflow import could go wrong like that

[onyx solar]

I'm not saying it was the import, only that it could've been.

[agile vector]

I know, thats kinda what I'm trying to dig into, been my primary way of figuring out comfyui so if theres something I'm missing about it, would be pretty helpful

[onyx solar]

I suggest reimporting the original source image and before doing a single thing, look at the values in each of the model/LoRA/vae fields to see if those values are the ones you think or not.

[agile vector]

No I have been, thats where my confusion came from, the image I put in here was generated before messing with anything

[onyx solar]

You think so, but again, without re-importing the image you used the first time to refresh the workflow with that, you can't guarantee it.

[agile vector]

Ive done that a few times though, I even re-imported before making the tweaks you suggested

[onyx solar]

Did you run it after making changes?

[agile vector]

I did, i don't think I can post the result here, but it was definitely not similar or acceptable

[onyx solar]

Well, no matter what, your output will not be the same as the image on civit.

[agile vector]

it probably shouldn't be a bunch of vaguely nsfw blobs though

[onyx solar]

Depends...sometimes there are bad seeds.

But also, not seeing the new values in the workflow doesn't really help.

[agile vector]

I can post it spoilered? don't know whats against the rules

[onyx solar]

You could just save the workflow into a .json

Or screenshot the workflow

[ruby yoke]

you could dm it to him

[agile vector]

sure, let me get one together that I've been having more issues with, I got one of the pony ones working just now

I didn't want to assume he was cool with that

[ruby yoke]

for pony, you have to use only the models that are out for it, so make sure your vae, loras, etc are all for pony

[onyx solar]

It's fine.

[onyx solar]

The LoRA chain on this is longer than what you had posted above. But even so, they're mixing standard XL with Pony LoRAs, which means that some of those LoRAs will not have any or very little effect on the outcome...at least in the way they were hoping for.

Also, this is not made in Comfy, so Comfy is just translating what it sees into a workflow.

[agile vector]

I didn't know that was possible, i thought comfy would only import if the specific metadata was there

[onyx solar]

This was probably made in A1111.

It is importing the metadata, but it then turns that into a workflow.

[agile vector]

Damn I had pinpointed this since it was tagged as comfyui on civit, but I guess you can set that to whatever

[onyx solar]

Yup

Never trust what you see on civit as being the full data used to create the image. Period.

Images on there should generally just be used to get an idea behind a model/LoRA/etc.

[agile vector]

Damn, i was kinda using it to get a feel for how others were setting up workflows since its pretty overwhelming

[karmic crown]

i would just add that and even if everything correct and you are not have the same card, it might look different (nvidia vs others)

[onyx solar]

Yes, that too.

Basically, any single variance can change the output, even if just slightly.

[karmic crown]

even a space or comma

@onyx solar i know you know all this better than me, i am just adding to your answers so they know what i am talking about 🙂

[agile vector]

Ya i get it, just up till now dragging something in and hitting run got me something other than blobs, wasn't thinking about bad metadata

[distant narwhal]

@ornate elk I have a quick question

it says I have to add the paths, I have it open, but how would I add the two to the variables?

it shows a wikihow link but im still confused on this

[karmic crown]

@distant narwhal https://www.computerhope.com/issues/ch000549.htm

hope this will help

just find your windows version

[distant narwhal]

okay yes, but i cant find the %hip_path%bin

[ornate elk]

you have to add it, you dont need to find it

[distant narwhal]

so then like this?

[ornate elk]

no dont

[distant narwhal]

ok

[ornate elk]

look here:
https://github.com/CS1o/Stable-Diffusion-Info/issues/7

[distant narwhal]

ok so like this then?

[ornate elk]

yes

[distant narwhal]

thank you

I feel dumb asf rn ngl

[ornate elk]

no need to, thats the step i get asked the most to

[distant narwhal]

so now I make the copy of the files, and then just put them back in the same folder?

it seems thats what it says but im just double checking

I assume that

[ornate elk]

yep

and then rename these copys

[distant narwhal]

1. Copy the files needed, then repaste them
2. Rename those copies, copy them again and put them into the folder and replace them

[ornate elk]

thats how it should look like then

[distant narwhal]

so do I rename the copies first or second?

for repastying

[ornate elk]

after you copied them

so you still have the original

[distant narwhal]

I see, so copy it, rename, paste

?

[ornate elk]

copy, paste, rename

[distant narwhal]

Okie

[ornate elk]

so you have a cublas.dll, and make a copy of that and then rename this copy to cublas64_11.dll

[noble goblet]

Hello everyone, I'm trying to use controlnet, but I must have made a mistake somewhere, in model I always have none, so I can't put canny... Moreover when I ask to generate the image , it gives me an image that has nothing to do with it... Some advice please

[ornate elk]

hey, did you downloaded the controlnet models ?

[karmic crown]

you need to select correct model here

but as CS1o said, you need to download them first

[noble goblet]

probably not good... when I put the canny links directly in stable diffusion, it tells me that I already have it. But maybe I'm forgetting something

[karmic crown]

controlent models needs to be in proper folder

not in the same folder with normal models

[ornate elk]

you need to download the canny model first to use it:
https://civitai.com/models/38784?modelVersionId=44716

[ornate elk]

goes into models/controlnet

[noble goblet]

Thanks guys for your responsiveness, I'll try 😉

[noble goblet]

Thanks guys for your responsiveness, I'll try 😉

[distant narwhal]

Hey so, I did the rest of the steps, but its not launching in the browser, it opens cmd for a second then just closes the window @ornate elk

[remote nebula]

hey guys, i just got done with installing comfyui with zluda on my PC, and im generating my first image. I saw that it can take up to 40 mins for the first one, but wanted to make sure that it's not stuck. the percentage means that it is generating the image, correct?

[ornate elk]

yep, also you can check the cmd of comfyui

[distant narwhal]

should I still just put this in the Swarmui backend folder? or no?

[ornate elk]

which webui are you trying to install ?

[distant narwhal]

I installed swarm, and I needed this as the backend

comfyui specifically

[remote nebula]

thanks a lot! 😄 this is what i got in CMD, does the torch entry mean anything bad? apologies for the dumb questions, ive only experimented with automatic1111's webUI so far

[ornate elk]

nope looks good so far. that warning can be ignored

[remote nebula]

great, thank you! its up to 57% so far, so its def doing something 😄

[ornate elk]

i guess you didnt setup auto1111 yesterday as i recommended to do first xD

did you downloaded my comfyui-zluda bat ?

[distant narwhal]

i may be dumb, okay, is it bad i did all this first?

[ornate elk]

no but its harder to get it working first try and more difficult for me to trace back errors

but we can fix comfyui

have you replaced the torch files?

[distant narwhal]

The ones that had to be pasted into the torch/lib folder?

cuz yes, I did

all 3 asked to be replaced and i replaced them as prompted

[ornate elk]

what is your gpu?

[distant narwhal]

7900GRE

[ornate elk]

ah okay, have you restarted the PC after installing hip sdk?

[distant narwhal]

no

should I?

[ornate elk]

yep

[distant narwhal]

okie, uno momento

[ornate elk]

its needed

[ornate elk]

whats your gpu btw? just curious

[distant narwhal]

Okay, restarted

should I try opening it again?

[ornate elk]

yea but i dont think it will launch

[distant narwhal]

correct lol

[ornate elk]

you have to try delete the venv folder and then relaunch the start-comfyui bat

[distant narwhal]

alr

btw this is the closest i can get to

for a pic

it immediatly closes after that

[distant narwhal]

doing that rn

alr, it finished, it closed the cmd promt

[remote nebula]

7700xt

With 12gigs of vram

[ornate elk]

nice card!

it will work in comfyui, but keep your expectations and workflows normal.
In case of upscaling your far better with auto1111 or forge than with comfyui.

[remote nebula]

Thanks! I already got automatic installed, but wanted to give comfyui a try, might check out forge too, I heard good things about it

[distant narwhal]

@ornate elk so after redownloading venv, it doesnt close the cmd promt right away anymore, it does a few lines of code then closes, doesnt open the webui still tho

[ornate elk]

no problem, if you need some basic workflow for comfyui let me know, (sdxl, 1.5, flux)
forge is nice because you can use flux there too. (idk if it will work on 12gb vram, but maybe with a smaller gguf version of flux)

[distant narwhal]

This is what is says then closes

im assuming now its looking for the torch files?

[ornate elk]

looks okay, you need to replace the 3 files in torch again

[distant narwhal]

sick lemme try that now

[remote nebula]

Will do, thanks a lot, and thanks for the great guide! Comfyui seems pretty confusing at the start, ngl 🤣

[ornate elk]

yea its not as comfy as auto haha 😄

[distant narwhal]

@ornate elk its back to not doing any lines of code rly

after I pasted them in

[ornate elk]

did you used the renamed files or the normal ones?

[distant narwhal]

the renamed ones

[ornate elk]

okay

best would be to setup auto1111 first and if that works, comfyui will work too, thats also how Wayne did it rn.

[distant narwhal]

okay, so install auto1111 then?

do I need to delete anything or no?

[ornate elk]

nope

you can also skip the zluda steps

you only need to git clone the auto1111 amd fork

and do the first steps of the guide

[distant narwhal]

okay, sounds good

so many Sd-zluda folders lmao

Now i need another one :I

[ornate elk]

wait xD

whats in SD-Zluda ?

and whats in SD

[distant narwhal]

Sd is the first Stable diffusion i tried to install

from a video a bit ago

but it wasnt working properly

so then I went here

[ornate elk]

ah okay

[distant narwhal]

sdzuda is the swarmUI

[ornate elk]

but not the one from my guide right?

you installed it before

[distant narwhal]

SwamUI is from your guide, then it told me to download comfyUi, which i was doing

[ornate elk]

ahh okay

[distant narwhal]

but the SD was not from your guide no

am I good to delete that folder u think?

[ornate elk]

you can delete SD and SD-Zluda

[distant narwhal]

okay, so the swarm and SD?

[ornate elk]

the thing is my guides always start with create a SD-Zluda folder so you can place all of the webuis in there if you want to

i like having all in one place

[distant narwhal]

alr, so now I should just rename comfyui to SD-Zluda then?

[ornate elk]

nope

[distant narwhal]

i mean like the first foldr

[ornate elk]

dont rename any folder after you git cloned a webui

[distant narwhal]

ye i meant the folder its in

mb

so this folder

not this one

[ornate elk]

ahh okay

you can do that

then you can git clone auto1111 in SD-Zluda too

so you have both in one folder

[distant narwhal]

Oh

so clone it here?

bet lol

again, i apologize for being dumb asf rn

[ornate elk]

WD black is nice

[distant narwhal]

i appreciate the help

[distant narwhal]

Ikr, this pc is beautiful lmao

im smart enough to build my own pc, but cant organize my folders properly

its creating the venv folder rn

@ornate elk okay, so it launched now

the auto1111

what should I do next?

[ornate elk]

put a model into the models/stable-diffusion folder

and then try to generate an image

[distant narwhal]

alr

[ornate elk]

it can take 15-40 minutes (sometimes longer), for the first image as it needs to compile stuff

[distant narwhal]

alr, sounds good

this is ok right?

[ornate elk]

yep looks okay

[distant narwhal]

alr bet, Ill just wait a while and then ill lyk what happens

[ornate elk]

if after 40 minutes its still compiling, just close it and relaunch the webui-user.bat and try to generate an image again

then it should work

[distant narwhal]

kk

@ornate elk it created the image

now, what should I do about comfyui?

[ornate elk]

okay good so everything is setup correct

[distant narwhal]

Here is what it looks like, it looks perfectly fine rn

[ornate elk]

make sure you downloaded the start-comfyui.bat and not the comfyui-directml bat

[distant narwhal]

yeah, it says start-comfyui.bat

should I maybe try just redownloading the whole thing?

like the comfyui stuff?

since I already installed Zluda, and it seemed to work well on Auto1111, it shouldnt be too complicated

or if u have any ideas ill be happy to try them

[ornate elk]

yea could be the best option

[distant narwhal]

alr, ill try redownloading it

@ornate elk its doing the same thing again

I redownloaded it and copy and pasted the renamed folders into the file and replaced the destinations

I did everything properly

atleast i believe I did

[fair oxide]

Heyy, someone know how to solve this error :
{error: 'cannot identify image file <_io.BytesIO object at 0x7fa4140879a0>'}

[primal wigeon]

Where can I find a good tutorial for comfyUI and generation in general

I have next door to zero experience with this stuff

[ornate elk]

Can you recheck the path settings of zluda?

[karmic crown]

in that case i would recommend to temporary switch to automatic1111, and then when you understand what is what, to try comfy again

a1111 is much easier to use

[distant narwhal]

sure

[karmic crown]

but if you are familiar with some other node based tool (blender for example), its fine to start with comfy

[primal wigeon]

I heard, I figured if I just used the more advanced tools I'd eventually catch up

[karmic crown]

when you understand a1111, its easy to apply the same tings to comfy

[primal wigeon]

Plus a1111 doesn't support pony iirc

[karmic crown]

i do not use it, but i am pretty sure it suports everything except 3.5

so that includes pony

[distant narwhal]

@ornate elk

[karmic crown]

i am using lightning mostly

since they require 8 steps only

less steps = "faster" generation
actually each steps takes the same
but since i use 5 times less steps, i get image 5 times faster then with 40 steps

[primal wigeon]

i'm ignorant, what is a step exactly

[karmic crown]

https://stable-diffusion-art.com/automatic1111/

start here, basic staff aplies to comfy too

[primal wigeon]

ty

[karmic crown]

https://stable-diffusion-art.com/comfyui/

also this

[ornate elk]

Okay looks correct

But there is a space in the folder path

To mitigate any point of failures you should move the zluda folder into an other directory without a space in the folder name

And then edit the path entry too

[distant narwhal]

okay so move this folder somewhere else?

alr

[ornate elk]

Wait

[distant narwhal]

?

[ornate elk]

Where are the cublas64_11 and the other renamed files?

[distant narwhal]

cublas64_11.dll
cusparse64_11.dll
nvrtc64_112_0.dll

[ornate elk]

Yea keep a copy of them in the zluda folder

[distant narwhal]

okay

so like this?

[ornate elk]

Yep

[distant narwhal]

should I also just move it to a folder without a space?

[ornate elk]

Ysp

And then edit the path in the environment variable's too

[distant narwhal]

that should work for that right?

[ornate elk]

Then delete the venv folder of comfyui and relaunch and show me the cmd log

After its done and before replacing the torch files

[distant narwhal]

alr

ill see if it works, because ut usually closes itself after its done

like ill try and send it to u, if it doesnt close itself

[distant narwhal]

Oh i see mb

Im dumb, ur asking to see if before I replce the files, mb xD

it closed itself

[distant narwhal]

but there is what it says mostly

[ornate elk]

That doesn't help me

I need the end of the log without the replaced torch files

So delete the venv again and relaunch please

[distant narwhal]

ok, but like i said once it finishes it automatically closes the cmd prompt

ill relaunch it again

[ornate elk]

Oh okay thats strange

Can you try git clone comfyui into a folder without a space too?

I remember people with the same problem and I guess it was related to the install path

[distant narwhal]

alr sure, once this finishes ill try again and put it in a path with no space aswell

[ornate elk]

And dont move it just git cloned it again

[distant narwhal]

kk

k, I cloned it to just the D drive

should I just do the same steps but without replacing the torch files?

[ornate elk]

Launch it, then after its done replace the 3 torch files and relaunch

[distant narwhal]

alright

[ornate elk]

And if that won't work we have to troubleshoot tomorrow because its 00:30 here for me😴

[distant narwhal]

oof sorry for keeping u up late

[ornate elk]

Np your not the first with that comfyui issue (closing cmd after launch)
But its so rare, always a bit of troubleshooting steps needed to get it working

You can also try to record the cmd and pause the video to see where it failed. That can give us a hint

[distant narwhal]

sure

btw should I keep this in the zluda folder or just move it to the lib file?

ik im supposed to move it

[ornate elk]

Keep it and copy it to the torch lib

[distant narwhal]

alr

it fails there

lemme make it an mp4

@ornate elk

[ornate elk]

Ah thanks!

The caffee error

[distant narwhal]

so what does that mean?

[ornate elk]

It means something is wrong with the replaced files

Or it can't find the replaced files

[distant narwhal]

there are the files here

in the lib folder

[ornate elk]

Have you downloaded the zluda files for Rocm 6 or 5 ?

[distant narwhal]

Rocm6

i also reinstalled the zip earlier

Those are the three renamed files

not cublaslt sorry

Thats the renamed file

[ornate elk]

Okay

Can you open up a cmd and type
Pip cache purge
And then delete the venv folder again?

[distant narwhal]

sure

did that

should I reinstall now?

not reinstall

Like relaunch?

[ornate elk]

Yep relaunch

Also what's your Python Version?

[distant narwhal]

I believe 3.10.11

but lemme check

[ornate elk]

Open up a CMD and type
Python --Version

[distant narwhal]

yep 3.10.11

[ornate elk]

Okay good

[distant narwhal]

alr it finished installing

now what

put the torch files back in again I presume?

[ornate elk]

Yep

Also can you open up a CMD and run
Where Python

[ornate elk]

Ahh thats the issue

Python from ms store

[distant narwhal]

i think thats because i tried it for ther old SD to work

lemme try uninstalling that

[ornate elk]

Yep please uninstall it

And every other older Python

And conda or minoconda etc

[distant narwhal]

should I just uninstall all 3 of these?

[ornate elk]

Yes

And then use the normal python 3.10.11 64bit installer

[distant narwhal]

alr

[ornate elk]

Also check "add Python to path" if available

[distant narwhal]

this?

[ornate elk]

Install now

[distant narwhal]

or customize installation?

alr

alr now what?

[ornate elk]

Open up a CMD and run
Where Python
And
Python --version

Now it should show the correct path to appdata

[ornate elk]

And 3.10.11

Okay uninstall all Python versions again

[distant narwhal]

should I turn these off?

in the Manage app execution aliases?

[ornate elk]

Not needed but uninstall everything related to Python

[distant narwhal]

alr, so same thing in Add or remove programs right?

[ornate elk]

And then reboot the PC

[distant narwhal]

alr

[ornate elk]

After the reboot download this file:
https://www.python.org/ftp/python/3.10.11/python-3.10.11-amd64.exe

And before installing run
Where python
In a CMD again
It shouldn't show a path

[distant narwhal]

kk

lemme reboot

Should I go to that path?

[ornate elk]

Okay nope

Install the python with the file you downloaded

The 3.10.11 64bit exe installer

[distant narwhal]

k

Im here

[ornate elk]

Perfecr

Checkbox the bottom

Add Python to path

[distant narwhal]

now what

just install now?

[ornate elk]

Yep

After its done, again cmd
Where python

[distant narwhal]

should i allow it to bypass the wherepath limit or no?

[ornate elk]

Ah disable

[distant narwhal]

so yes? disable it?

[ornate elk]

Helps too

[ornate elk]

Okay good

So now you delete the venv folder of comfyui and relaunch. Let it install and after its done copy the 3 files into torch lib again

[distant narwhal]

kk

[ornate elk]

I'm 85% sure it will work now xD

[distant narwhal]

Ok cool xD

installing rn

it shows a lot more progress bars thats for sure

[ornate elk]

And if this works I will ad a big warning to my comfyui guide so that people don't have MS Store Python installed before trying

But I need sleep now fr xD

[distant narwhal]

its taking a lot longer

IT LAUNCHED BBY

Now lemme put the torch files in

[ornate elk]

Okay. Also if auto1111 webui doesn't launch you also need to delete the venv folder one time.

[distant narwhal]

kk

It launched with the files

We got it bois

[ornate elk]

Nice GG

[distant narwhal]

Thank you SI MUYCH

Ur an absolute chad

[ornate elk]

Don't forget to install the Comfyui Manager

[distant narwhal]

for staying up this late and helping

yes, I will, tmmr

lol

[ornate elk]

No problem, thanks for the patience and help to finally get the solution to this error

[distant narwhal]

No worries, GGs

Now go sleep

u deserve it :D

[ornate elk]

I'm off cya tomorrow 🙂 🫡

[distant narwhal]

Cya!

cat_salute

[jovial summit]

So I'm on webui reForge and this is my first time trying out controlnet. I'm using PonyXL and xinsir's Union promax for the controlnet model
but it seems no matter what I do I can't get openpose to work right. What's bizzare is that the openpose skeleton just got passed through the final image without being recognized as a skeleton by the controlnet or something 😅

Anyone have any idea what I'm doing wrong? Sorry I'm a complete beginner to all of this

(previous attempt)

[silent stream]

Whenever I try to run Kohya-ss to train SDXL lora, I get hit by returned non-zero exit status 1

[primal wigeon]

are people still using a1111 or have they moved onto forge

[hardy blaze]

Paint a picture of sunset and lone ducks flying, autumn water together in the sky

[ruby yoke]

plenty still using it

[ruby yoke]

stop posting this in every single channel

[chilly frigate]

Hey, got a question related to CUDA, haven't updated from a some time and I'm getting this issue

`2024-11-05 22:46:54.4124535 [E:onnxruntime:Default, provider_bridge_ort.cc:1548 onnxruntime::TryGetProviderInfo_CUDA] D:\a_work\1\s\onnxruntime\core\session\provider_bridge_ort.cc:1209 onnxruntime::ProviderLibrary::Get [ONNXRuntimeError] : 1 : FAIL : LoadLibrary failed with error 126 "" when trying to load "G:\Projects\AI Image\stable-diffusion-webui\venv\lib\site-packages\onnxruntime\capi\onnxruntime_providers_cuda.dll"

*************** EP Error ***************
EP Error D:\a_work\1\s\onnxruntime\python\onnxruntime_pybind_state.cc:857 onnxruntime::python::CreateExecutionProviderInstance CUDA_PATH is set but CUDA wasnt able to be loaded. Please install the correct version of CUDA andcuDNN as mentioned in the GPU requirements page (https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html#requirements), make sure they're in the PATH, and that your GPU is supported.
when using ['CUDAExecutionProvider', 'CPUExecutionProvider']
Falling back to ['CUDAExecutionProvider', 'CPUExecutionProvider'] and retrying.`

I already updated my drivers and installed CUDA 12.6(For some reason webui-user file says it still uses 12.1), I'm not sure if I'm still using forge but according to git, it is using the main branch from the stable-diffusion-webui

Everything works smootly, however. I wonder if I fix that issue it might improve my generation time hehe.

[umbral bramble]

Question: I want to resolve a problem regarding downloading stuff from objaverse for my dataset, but it seems like most of the data i want to download(from github) isn't being cloned and thingiverse is not allowing us to download anything through the script provided. So far, i'm trying to collect human-like video game cahracters such as villagers of animal crossing and etc.I appreciate any guidance. Please let me know if i'm asking at the right place.

[ruby yoke]

what are objaverse and thingiverse?

[umbral bramble]

Objaverse is a dataset where SV3D was trained on. and Thingiverse is one of the websites that objaverse subscribes from to download 3d models?

[ruby yoke]

okay well, um, not sure you're going to find anyone here that uses that. but maybe in a few hours when people log back on.

[umbral bramble]

Understood. I'll wait patiently. Thank you very much!

[ruby yoke]

welcome. i hope someone can help

[umbral bramble]

Thank you!

[autumn stag]

Hello Is this possible to run cogvideoX-5b on a rtx 3060 12gb, when i try the inference take an eternity (10mn/it ) is this normal ?

[ornate elk]

Sounds normal as Cogvideo is made for 4090, A100 and H100

I guess it can work on smaller cards but will take significantly longer

[autumn stag]

Ha ok thanks

Is a lcm/hyper lora possible for a video generation model ?

[iron lantern]

How hard is it to make a gif in forge ui ??)

With pony xl model

[deft hemlock]

activating extra network lora with arguments [<modules.extra_networks.ExtraNetworkParams object at 0x000001C0AF1E1F00>]: ValueError
Traceback (most recent call last):
File "F:\ai\sd.webui\webui\modules\extra_networks.py", line 135, in activate
extra_network.activate(p, extra_network_args)
File "F:\ai\sd.webui\webui\extensions-builtin\Lora\extra_networks_lora.py", line 32, in activate
te_multiplier = float(params.positional[1]) if len(params.positional) > 1 else 1.0
ValueError: could not convert string to float: ''

i think this doesnt allow me to use loras

ok sd now crashed

[ornate elk]

Can you show your extensions?

[ornate elk]

Is your webui updated?

[normal tartan]

does anyone knows how to fix this error when running stable diffusion webui-user?

[ruby yoke]

what gpu do you have?

what model are you using?

[iron lantern]

any one know why this is happening ???

[karmic crown]

wrong vae or too high cfg

[iron lantern]

how do i fix it ?

[ruby yoke]

use the correct VAE

[karmic crown]

as i already told you, you need to use model and lora of same type. the same applies for vae

https://civitai.com/models/660673/pony-standard-vae

since you mentioned pony models and loras

if i remember correctly

[iron lantern]

just want to clean up the models so the colour looks better

[karmic crown]

you need to use compatible things

generate an image and send this part below generated image

[normal tartan]

gpu is intel iris Xe Graphics

Model is VAE

[karmic crown]

vae is vae, model is model, you cant use vae as model

[ruby yoke]

VAE stands for Variational AutoEncoder. your model has to use it, but your model is not it.

[ornate elk]

How much vram ?

[normal tartan]

16gb of ram

[ruby yoke]

none. i just went over this with ikku. they're on a laptop with integrated gpu

[ornate elk]

I mean vram, GPU memory

Ohh

CPU only mode then

[ruby yoke]

yes

[dapper creek]

on stability.ai /stable-image/generate endpoints - is there a progress update somewhere? (It's pretty fast already, just looking for per-second approx % updates)

if not I'll just fake it heh

current time usage is ~15 secs in ultra

real    0m15.729s
user    0m5.227s
sys    0m0.147s

thing is, I'm running approx 40+ in parallel and it would be great to know approximate ETA

[distant narwhal]

I keep getting this error, my gpu has more then enough vram to generate it, can anyone help?

[formal trout]

when trying to load a model

hi, I'm agetting this error on Automatic1111

weird, I reinstalled Ubuntu to solve this but it is still there

[ruby yoke]

any errors before this?

[formal trout]

no that I notice

[formal trout]

but it is really strange, I though something was broken so reinstalling the whole system, rocm, pytorch, etc would solve it

[ruby yoke]

changing setting sd_model_checkpoint to v1-5-pruned-emaonly.safetensors: NotImplimentedError

that's the first error in this. what is cauing that

[formal trout]

no idea!

[formal trout]

As I said I though it was something that was broken from software but after a full reinstall I don't know what's going on

[deft hemlock]

Ye

[formal trout]

using Automatic venv for Comfy gets me a similar error

I've uninstalled torch and torchaudio and installed this instead as before

well I just used the torch it came with Automatic and now it is working (at least SD 1.5)

[formal trout]

well I just used the torch it came with Automatic and now it is working, it seems

[distant narwhal]

can anyone answer this please?

[ruby yoke]

i think you might want to wait for @ornate elk to log back on

[formal trout]

Any idea about this error?

[ruby yoke]

what's in the console log?

[balmy shore]

does anyone use the ux ui and tried the "spaces" tab or know what they are