🤝｜tech-support

[plain crypt]

Do I remove what's after the line, or do I add what I need to add?

[ornate elk]

At the line set Command line_args=

You have to add --cuda-stream

[plain crypt]

OK, next on the line after or before ^

?

Once that's done, what do I do next?

[ornate elk]

You launch the webui-user.bat

[ornate elk]

After the =

[plain crypt]

Ok it's done it has to work properly but I don't have a model to make images and videos

[ornate elk]

You get the models from civitai.com

[near rampart]

I want to create AI-generated images like this, featuring characters from HoYoverse games. What AI are they using? I see many AI artists on Pixiv using the same style.

[near rampart]

🥲

[ornate elk]

What's your GPU?

[near rampart]

RTX 5060 VRAM 8GB

[plain crypt]

OK thank you

[ornate elk]

Are you already using a local webui?

Or do you start from 0

[near rampart]

0

[ornate elk]

Ah ok, then you want to setup Forge Neo as your local webui to generate images.
You find the setup guide in the pinned messages of this channel.
After the install you can download anime models from Civitai.com
There you also find Lora's for Hoyo chars

[near rampart]

Pinned message, where?

[near rampart]

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

This one right? Thanks

[west socket]

Hello, I just did everything what the guide say and the error keep coming up with the CLIP installation

[vocal burrow]

which error

[west socket]

This

[vocal burrow]

post the whole log not just the end. Put it in a .txt and drop said file in here

[west socket]

like this?

[vocal burrow]

arf can t help you much with zluda. wait for @ornate elk they will know more about it

[west socket]

Ok thanks

I'll be wainting here then 😄

[ornate elk]

Its a new issue from openai/clip, can't fix it currently:
https://github.com/openai/CLIP/issues/528

[west socket]

I can't believe it XDDDDDD I swear I install it a few days ago, I didn't thought that an update will make it impossible to re install hahahah

Well I guess the only thing I can do is wait to that error being fixed in the next days right?

[ornate elk]

Yea or for a workaround

Someone there suggested downgrading some packages

You can do that in the requirements txt file by changing the version numbers mentioned there

Then delete the venv folder and relaunch

[west socket]

What do I have to change in that requirements txt file?

[ornate elk]

nothing there is a new fix now:
open up a cmd inside the stable-diffusion-webui folder and run these 3 one by one:
venv\Scripts\activate.bat
pip install "setuptools>=65.0.0,<81"
pip install "pip==25.0"
then relaunch the wbeui-user.bat

[west socket]

The error remains

[ornate elk]

can you show how you did the steps below?

with screenshots

[west socket]

Give me a moment to redo it

[celest delta]

how do i change model path like lora and checkpoints

this doens't help me at all since i have the same folder for another webui somewhere else, i want to use that

[ornate elk]

which webui do you use?

[celest delta]

forge neo

[ornate elk]

Linking Models, Loras, etc. from other Webui's or Folders to Forge Neo:
If you want to link all models from an other Webui you can do that by editing the webui-user.bat and add the following depending where your models are:
--forge-ref-a1111-home "X:\examplePathTo\Auto1111WebuiFolder"
--forge-ref-comfy-home "X:\examplePathTo\ComfyUIFolder"
--model-ref "X:\examplePathTo\ModelswithSubfolders"

[celest delta]

like editing with notepad?

thanks!

Also where is the upscaler esrgan4x+anime6B located in auto1111 folders?? i want to use it for forge neo but can't find it

and there's nothing in the esrgan folder

[west socket]

Did I miss something?

[ornate elk]

nope looks correct but i have an idea

delete the venv folder, then relaunc hthe wbeui-user.bat, wait 5 seconds and then close the cmd, then do the steps from above again, then relaunch the wbeui-user.bat

[west socket]

Okok

Well, now is this error, I guess I just have to put that command in red right?

[celest delta]

does forge neo updates automatically when launching it? If not how do i do it?

[ornate elk]

you have to add git pull to the webui-user,bat for auto updates

you can also manual update it by running git pull inside a cmd inside forge neo

[ornate elk]

hmm idk, i dont think it will work

[celest delta]

just adding the text git pull anywhere and it works?

[ornate elk]

in an empty line/row

[west socket]

🙁

There's something else I can do at this point?

[celest delta]

huuuuh, weird, right clicking on the .bat file, i can't see the option "open with" to modify the text inside with notepad

ooooh, i just have to click on edit

learning something new everyday

lol

[ornate elk]

at this point we should wait for an official fix

[celest delta]

didn't look like it updated or anything

am i doing this right?

i wanted to update cause forge neo doesn't seem to use my LORA

despite them being in the correct folder and being detected by the webui

[ornate elk]

add it in line 2
not at the end

[west socket]

Well, I'll be looking forward to it

[celest delta]

ooooh, like this?

thanks

[ornate elk]

you should see some green and red icons when relaunching in the cmd

or something like:
Already up to date

[celest delta]

hmmm

if i update it's gonna remove my models?

[ornate elk]

nope, just delete the config.json and the ui-config.json and relaunch

[celest delta]

pressing continue will update then? thanks

[ornate elk]

nope, just delete the 2 files and relaunch and its updated

[celest delta]

ah ok, thanks

[ornate elk]

no problem, you may need to change some settings if you had changed them in the past again

like folder paths for output images or something, or nothing at all

[west socket]

And thanks for your help btw

[celest delta]

said to me i need to update python, just downloading the latest version will update my current one i assume?

sorry for all the dumb questions, even if some seems logical

[ornate elk]

oh right, the forge neo dev switched to python 3.13
so you need to uninstall python 3.12 and install 3.13.12 and then delete the venv folder and relaunch the webui-user.bat

[vocal burrow]

dont follow the obvious scam link that poped just a few seconds ago

[celest delta]

doesn't update automatically then with the installer from the website??

also do i need that specific version? The last one on the website is 3.14

[ornate elk]

nope as its a mjaor version change

[ornate elk]

not compatible with ai stuff currently

[celest delta]

will that cause issue for my auto1111 webui btw? i know it's obsolete but it's what i always used and it's reliable for now, until i find a replacement

[ornate elk]

auto1111 mostly uses python 3.10 or 3.11, if you have that installed too then your fine

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

to see the versions installed

[celest delta]

when it said i had to change it said i had 3.11.9

i can install 3.13 and 3.11 will still work?

if i don't uninstall?

cool cool cool

thanks

i have to go, but last question

if my LORA still don't work on forge neo, any idea what i should do or check?

they work fine in auto1111 but they didn't with my forge neo right now

hopefully update will fix it

[ornate elk]

i think they will work after the update

i had no issues with them

but let me know

[celest delta]

ok, will talk to you later if i need to

thanks again

[chilly knot]

There are no errors when generating in the Wan, but videos with a black screen are created and they weigh very little, only 3 kilobytes. What could this mean?

What do I need to check?i use forge neo

[prisma moss]

i think theres plenty of black screen reported online and their solutions. Check your VAE version, and make sure you can actually run the safetensor/gguf models

[chilly knot]

i check

wan 2.2 14 b i2v q3 -VAE 2.1

[prisma moss]

also do you have use sage attention argument

[chilly knot]

i use --sage

[prisma moss]

try removing it

[chilly knot]

done

not help

[prisma moss]

screen shot what you've loaded so far

[prisma moss]

where is low noise

also try disabling fp16 and cuda stream arguments

[chilly knot]

low noise in refiner

[prisma moss]

.

[chilly knot]

i check it now

black screen

[prisma moss]

umt5_xxl_fp8_e4m3fn_scaled.safetensors try using that clip

or umt5-xxl-encoder-Q5_K_M.gguf

eh probably wont fit actually

[chilly knot]

i have umt5-xxl-encoder-Q5_K_M.gguf

its work with all model?

damn i know its must be work

i mean this gguf model

on my system

no out of memory

people run it on 8 gb

vram

[prisma moss]

idk try it

[chilly knot]

sad(

[ornate elk]

you neeed this:
umt5_xxl_fp8_e4m3fn_scaled.safetensors

the Q5 gives black outputs on forge neo it seems

some other guy posted the same issue

[prisma moss]

i was right at least on one thing xD

[chilly knot]

thx!

work!

looks bad but work XD

I wish you a lot of video memory, you are good guys.😎

[chilly knot]

how i clean ram after generation its take lot memory

[delicate citrus]

the a11 is showing this message when I started it, how can I fix it?

[cosmic furnace]

any good news about the rock updates?)

[ornate elk]

What's your GPU?
Check the pinned messages of this channel for the updated setup guides.

[honest sand]

not tech support but is there a channel to discuss nsfw img generation?

[ornate elk]

Not on this server, its only sfw

[honest sand]

I see, thank you.

[ornate elk]

No problem

[ornate elk]

In comfyui sage attention works and flash attention

With the latest rocm stable builds

[cosmic furnace]

why i see in news that almost all PC parts are kinda rising in price =\ . Are they just looking at what happening with ram & trying to do same ? 😄 . Well for now only server CPU's are rising kinda... from what i get ... And also chineese & inter gonna produce ram now . But not helping yet it seems -_-

[paper dagger]

RAM is insane. Not sure whats worse in cost, top end GPU or Max RAM

Data storage too getting insane 😭

[prisma moss]

give it a year

hell maybe even less

[wary cloud]

also having this issue despite following guides to a t

[tawny saddle]

mmm

comfy..so cozy

wut comfy ui looks like ^

comfy init

[ornate elk]

Can you show the cmd log?

[wary cloud]

I can if you'd like but its the exact same as the previous persons

I also did some more digging and it looks like they removed the setup tools on February 8th and nobody had awork around

[ornate elk]

Ah that error

[ornate elk]

We have to wait for a fix or workaround:
https://github.com/openai/CLIP/issues/528

[plain crypt]

Hello, is it normal that when I put models in the files, they are not detected?

[ornate elk]

No, which files?

[wary cloud]

Thank you regardless

[plain crypt]

models that I downloaded from civitai

[ornate elk]

Yea but are they checkpoints, Lora's, embeddings?

[plain crypt]

uh no idea

[ornate elk]

Always check the info box on the models oaget

Check for base and type

[plain crypt]

ok thank you

[plain crypt]

How do I adjust the GPU weight?

[oblique ether]

I saw on the website that this model produces excellent realistic photos. I downloaded it and trained a LoRA based on it. I’m also using it as the checkpoint, but my images take a very long time to generate and come out really bad… What am I doing wrong? 😦

Which base model is best for training in order to generate photos like the ones in the previews — full body and that realistic? Please help, I’ve been racking my brain over this for four days straight, morning to night 😓

[tawny saddle]

i also would like to know how to train models and loras

but, imagine id prolly need a heckin beefy pc tho

prolly more powerful than one that generates images

cuz, prolly a lot more to training than generating

[ornate elk]

Can you show the setting you use with it?

Also do you have xformers in the launch args and --medvram-sdxl

Boath is needed for 8GB vram cards

[ornate elk]

You need vram at least 12-16gb

[plain crypt]

Can someone help me? I don't understand everything about making images and videos.

[tawny saddle]

oh my :o.

[delicate citrus]

GTX 1660. I will check it

[prisma moss]

yea u r cooked either way

[junior glacier]

Can someone tell me if I can train an SDXL LORA myself? And if yes, then how.
I've got 6 GB VRAM 4050, 16 gigs of RAM. I know it's low but... Is it possible?

[celest delta]

where are the upscaler on auto1111 located? there's a lot included by default but i can't find them, i want to use one of them for forge neo

r-esgan x4 +anime6b

ok i finally found it, weirdly they're all located in different folder

the one i said was in model/realersgan

thanks

[ornate elk]

sry i was confused, the upscalers are usually always in models/esrgan

[celest delta]

yeah but for auto1111 there wasn't any there

they were all in different folders

[ornate elk]

ah yea because in the past it downloaded them in different ones

[celest delta]

weird cause here it's called realesrgan4x + anime6b, but when i'm in the webui, there's no real in front, so i'm not sure if it's the same

remember our conversation from yesterday?

i installed python 3.13 but when launching the forgeneo webui.user, it says this :

you said i could have both installed at the same time

[ornate elk]

yes you can but you need to have boath added to the system path or specified the exact version i nthe webui-user.bat

open up a cmd and type
where python
then copy the path to python 3.13
then edit the webui-user.bat and paste it to the line
PYTHOn=
save, delete the venv and relaunch

[celest delta]

where is 3.13... i just installed it without issues

[ornate elk]

its not added to the path

[celest delta]

oooh sorry i'm dumb

[ornate elk]

relaunch the installer of it and select
Modify, next, add to environment variables

[celest delta]

just read what you said wrong

my bad

[celest delta]

working, but it's behaving weirldy, like i'll do a generation, it will go at a normal and acceptable speed

i launch a new one without anyt change, and suddenly it takes 30 minutes

also any chance to change the tiled VAE setting that are automatically included in forge neo?`(it' was an extension in auto1111, with more settings)

the current forge neo tiled VAE thing makes it take too much ressource, so if i watch a video at the same time, it's suttering

[ornate elk]

tiled vae doesnt do this, its for reducing the gpu load at the vae step and nope no other settings for it in forge neo

[celest delta]

yeah i'm sorry that's what i meant

with the auto1111 it reduced the load more allowing me to do stuff on the side easily

can i still install the extension like i did for auto1111 or will it not work?

man i'm conflicted, forge neo can do a lot of things, but the results aren't as good as i had with auto1111 (despite using the exact same settings

[ornate elk]

it wont work

[ornate elk]

the webui doesnt change the output quality, whats your gpu again?

[celest delta]

it's not about the quality, it just some slight change and the result while similar doesn't really have the same feeling

[ornate elk]

ok

[celest delta]

like some dumb exemple, the eyes are bigger in auto1111 vs forge neo (tried multiple generation to see if it wasn't just bad luck with the generation)

[ornate elk]

thats something the webui doesnt control

[celest delta]

well in any case my results are differents, and while i like forge neo diversity, the results just are just not as good, personal preference i guess

also it still does what i mentionned before

one generation goes all fine and fast

the enxt one, same setting no change as before, goes super slow

never had that in auto1111

[ornate elk]

whats your gpu?

and whats in your webui-user.bat

[celest delta]

3080

and only what you told me

[ornate elk]

left is auto, right is forge neo

there is only minimal change but not in the size of the eyes

[ornate elk]

can you show the settings you use when it starts to get slow? like steps, resolution etc screenshot the webui

[celest delta]

i have completely different results even with same seed, i thought it was just inevitable due to being a different webui

[ornate elk]

nope the key point is the rng setting

[celest delta]

?

[ornate elk]

its how the seed generates

cpu or gpu

if its different in one webui, the output will be not the same

so they have to be the same in the rng setting of each webui

best is to use rng cpu

[celest delta]

where is that setting?

[celest delta]

just tried it, it was on GPU on auto1111

changed to GPU just to see the result difference

it's still different

i don't know what other setting i should check

would the tiled VAE extension affect that?

using same seed, same upscaler, same sampling, same amount of steps, everything should be the same

also to answer your other question, as i said, i don't change the settings, one time it works great, the next gen it goes slow, no change in any settings at all

here's what it says :

don't have that in auto1111, and the webui-user has the same stuff in it

[ornate elk]

nope

[ornate elk]

can you show whats in your webui-user.bat? and a full cmd log?

[ornate elk]

i dont see any launch args there..
also your python path wont get used if you keep the : : infront of the python line

[celest delta]

oh so it doesn't use 3.13 still?

[ornate elk]

probably not

and without launch args it makes sense that it gets very slow in gen speed

[celest delta]

ah i see, in the auto1111 one, the :: aren't there

so the medvram is used, but not in the forge neo one

that's the issue i guess?

[vocal burrow]

those are comments

if it starts with :: it might as well not be in the file at all. It would not change anything, it's not gonna be interpreted.

same for REM those are comments (or REMarks)

[ornate elk]

try add:
--flash --xformers
at the line set COMMANDLINE_ARGS=

[celest delta]

these are the one that were used in auto1111 :

--xformers --no-half-vae --medvram-sdxl --disable-nan-check

should i use the same?

[ornate elk]

no use the ones i mentioned above

[celest delta]

that would still help with the slowdown? like medvram isn't relevant anymore?

also here's how my exemple looks :

left is auto1111, right is forge neo

[ornate elk]

compare the metadata of boath images in the PNG-Info tab

[celest delta]

i'll retry once i correctly updated with python 3.13

hmmmm, i deleted venv but it didn't do the update, i guess it already did but was still using python 11

i should just restore venv i deleted

yeah the metadata is basically the same, the only change is that forge neo adds a CFG scale for hires, but low or high, it doesn't really change the results that much

it's frustrating not knowing what's wrong lol (not your fault, you've been great help)

[ornate elk]

try copy and paste boath meta data into a text programm and check the exact lenght and stuff

best if they're over each other

easier to spot spaces or addittional , or something else

[faint horizon]

What do I do to fix this.

[ornate elk]

What's your GPU?

And can you show the complete cmd log?

[faint horizon]

rtx4070

[ornate elk]

Make sure your GPU driver is updated

[faint horizon]

well shoot, I'll do that and see if that does it lol

It only started doing this after I updated Comfyui itself, luckily I have a copy of the folder from before I updated

[ornate elk]

Have you tried the GPU driver update?

[faint horizon]

yup, still busted

[ornate elk]

You can try download the latest portable and test if that works

[faint horizon]

Like, the copy I have still runs

just not the original one I tried updating

[ornate elk]

Yea thats why you should try the latest portable

Comfyui switched to python 3.13

[faint horizon]

Oooooh, doy, sorry.

[ornate elk]

No problem, if it works then you can move your stuff over

[faint horizon]

well, it works now, but smz nodes is broken lol

[zealous root]

when i installed python, i didnt get a "path" option. so when i start stable diffusion it does detect pytorch or something

the setup was just a command window

I'm literally in python rn

oops wrong version

[prisma moss]

is this comfy manual or portable

[zealous root]

A1111

tried to uninstall python, now this message is stuck:

i keep clicking ok but message comes back instantly. Tried rebooting my computer but the message comes back right when windows starts

[prisma moss]

im sure a1111 (which you shouldnt even be using) doesnteven work with python 3.14

[zealous root]

I do use it. it's WAY easier

[prisma moss]

you should use forge at least

(same ui)

[zealous root]

the node crap is complicated af

[prisma moss]

thats comfy

[zealous root]

oh

but ya, its not compatible with 3.14

i have to use 3.10.6

[prisma moss]

x0.10.9 is fine

[zealous root]

ok

[prisma moss]

uninstall existing pythons

[zealous root]

I did. Now when I try to reinstall I get this

all well, guess this ai crap isnt for me

[unreal linden]

I've recently moved to Forge Classic, any advice for performance/speed? I've got a 4070 Super on an R7 8500X3D

[prisma moss]

downgrade to the 'abandonded' normal vanilla forge which is faster

[unreal linden]

alright, @prisma moss what's the proper python version?

[prisma moss]

3.10.x should work

[unreal linden]

excellent

[prisma moss]

Vanilla one is the one from illyasviel

[oblique ether]

Idk….

[oblique ether]

This setting

[prisma moss]

Medvram isnt needed but ok, i used 8gb card w sdxl

And it caused more issues than helping

[unreal linden]

With SD Forge I don't see any cross attention optimization options in the dropdown menu, is that normal? I only have automatic

[hidden night]

I fixed it thanks

Any suggestion for IMG-VID or TEXT-VID running locally?

[ornate elk]

--medvram-sdxl is needed for 8GB GPUs when using Auto1111.
Forge and its fork have it automatically enabled.
--medvram is for 6GB and less

[prisma moss]

Ah

[ornate elk]

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

[ornate elk]

Then check it xD. I've seen you installed stability matrix, there you can enable the launch args and see them

[oblique ether]

You're a genius!!! I found it!! One checkbox was actually disabled!))

[oblique ether]

Does it matter where in the prompt I insert the model token? At the beginning, at the end, or somewhere else?

[ornate elk]

Enable --medvram-sdxl too

[ornate elk]

You mean Lora trigger words?Doesn't matter that much, I would place it at the end.

[oblique ether]

Okey thanks, I’m trying now

Is this a good model to learn from if I want to create photos like this?

I just need to generate different "luxury" photos to post on Instagram with my model's face. But if I can also create NSFW photos, I'll be thrilled. Actually, I can take NSFW photos of my model myself with my camera. Luxury photos, however, are problematic.

I mean to train my Laura based on this model

[prisma moss]

Ppl moved on to flux/qwen/zimage/wan for realism

For starters i guess cyberpony is good if u cant afford local generation

Lustify is also a contender

[oblique ether]

Please tell me, is this all still installed locally or are these online sites?

[prisma moss]

Those 3 are open source, so they are both. And you should t be using more than like 40 steps for cyberpony

@ornate elk i think? Idk whats the usual step count for that model, definetly not 100+

[oblique ether]

In that case, I'll be looking for information about these resources now. Thanks for the advice!

[oblique ether]

I also found this on YouTube yesterday... I shouldn't waste my time on this either, right?

[prisma moss]

Comfyui is basically necessary for any serious work in those models, can be done in forge neo, but i wouldnt trust it

[oblique ether]

I read that Flux requires a 3090/4070 graphics card with 16GB or higher. I don't have one. But I have a friend who has a 5090. I could send him the dataset and ask him to train the model on his computer, and then he'll send me my model, and I'll just use my 3070 for photo generation. Will this plan work?))

[prisma moss]

Hardly lol, with 3070 u can only use those sdxl models

[ornate elk]

30 steps is the GoTo for illus/pony

[prisma moss]

Sure

[prisma moss]

Also dont forget to install studio drivers for nvidia, off to bed now

[oblique ether]

Okey..

[oblique ether]

Something tells me the best, fastest, and easiest option for me is to pay for Flux and generate photos online. Or is it better to rent computing power for a month?

[prisma moss]

That i wouldnt know

[oblique ether]

All this is so difficult

Damn

[unreal linden]

belatedly, I have a RTX 4070 Super

[ornate elk]

Do you use Forge or Forge Neo?

[unreal linden]

regular forge

[delicate citrus]

why?

[ornate elk]

Oh you should switch to forge neo then.
But on normal forge it already uses pytorch cross attention.
Forge neo supports it too and more stuff like sage and flash

[willow venture]

hey, can anyone please help me or tell me some tutorial on how to train a z image base lora locally with my rtx 3060 (12gb vram)

[pearl musk]

you can trin a LoRA locally on a 3060 12GB, easiest path is kohya_ss or sd_scripts

[willow venture]

but i heard that only ostris is on the newest level so far?

[pearl musk]

not only Ostris, but for the newer image model families like z-image or flux, ostris ai-toolkit is often the most up-to-date and easiest because it ships templates / ui for them and is actively maintained

[willow venture]

okay thank you but do you know what i should do now to train my lora? i got ostris on pinocio now and dont really know how to train it. could you help me or tell me where i can get that information?

[west socket]

I followed the comfyUI with DirectML guide and I did every single step, it runs only with my cpu instead my gpu, does someone know why?

[ornate elk]

whats your gpu?

[west socket]

RX 580 8GB VRAM

[ornate elk]

hmm okay

[west socket]

is that bad?

[ornate elk]

yea its difficult to get that working and probabyl not even fun to use with comfyui

better try setting up auto1111 or forge with directml

[west socket]

Well, when I tried with a1111 it kept throwing a CLIP installation error

I guess I should try forge instead

[ornate elk]

yea hopefully the clip error got fixed, its a very new issue

[west socket]

I read somewhere that is because they want pkg_resource separated from the setuptools main repository, but I'm not sure xD

Idk much about it

[ornate elk]

yea thats the issue

[celest delta]

so i'm about to build a new PC, my webUI are installed on a secondary SSD (not where windows is)

will they work on my new build as long as the python path is correct? (i'll also reinstall python anyway)

i dunno if the webui install anything on the main drive and if that will be wiped out when i install windows on my new build (using the same main SSD i'm using in my current computer)

[prisma moss]

@ornate elk

[celest delta]

@ornate elk sorry to bother you but you've been such a great help before with all my stuff!

and i'm starting to build today, so would be great if i knew beforehand, thanks!

[ornate elk]

yes they will work but you have to install the Python versions they use and Git

[celest delta]

thanks

it's been a lonng time i installed GIT, anything special i need to do or just installing will work?

[ornate elk]

just installing it will work

[celest delta]

btw, what's the file that keep all my prompt style saved?

i mean this ;

[ornate elk]

styles.csv

[rose apex]

is the pytorch version: 2.10.0+rocm7.12.0a20260208 not stable yet? It's giving this error.
HIP error: invalid kernel file
Search for hipErrorInvalidKernelFile' in https://rocm.docs.amd.com/projects/HIP/en/latest/index.html for more information. HIP kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing AMD_SERIALIZE_KERNEL=3 Compile with TORCH_USE_HIP_DSA` to enable device-side assertions.

[ornate elk]

which command did you used to install it?

[rose apex]

It was from a reddit comment which I can't find anymore 😑

[ornate elk]

this command should work: its for RDNA4
pip install torch torchvision torchaudio --index-url https://rocm.nightlies.amd.com/v2/gfx120X-all/

[rose apex]

I have a 9060 XT

[rose apex]

This works. Thanks.

[paper ridge]

does anyone know how to solve an issue where CyberRealistic XL is only producing solid black images? I assumed it was due to no VAE, but apparently that's baked in

[ornate elk]

make sure you use negative tags

in forge neo its always needed

[paper ridge]

Im using comfyui, but i didn't try adding any negatives yet 😮

[ornate elk]

oh ok whats your gpu?

[paper ridge]

rx 7900 xtx

[ornate elk]

then make sure you use the fp16 xl vae

https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main

it fixes black outputs for all sdxl based models

[paper ridge]

thank you! oddly enough that didn't work but I noticed making stop at clip layer -2 worked, even tho most workflows i referenced had it at -1

[ornate elk]

ah yea, for sdxl based stuff, you dont even need to set a clip skip

[paper ridge]

ah i didn't realize that!

[bitter terrace]

Can someone please please please help

[ornate elk]

whats your gpu?

auto1111 has known issues rn

[bitter terrace]

Is there an alternative to auto111

[prisma moss]

Forge

[bitter terrace]

Is it better?

[prisma moss]

On intel uhd u might as well throw it in the bin

[bitter terrace]

Oh so I cant get it😭

[prisma moss]

U can, but it aint gonna be pleasant experience

[bitter terrace]

So nothing else?

[prisma moss]

It doesnt matterz u cant generate anything with uhd

Or its gonna take half an hour per image if not more

[bitter terrace]

Ah I see so even with Index tts2?

[prisma moss]

Idk what that is

[bitter terrace]

Voice closer basically

[prisma moss]

Wouldnt know never meddled wkth those

[bitter terrace]

I see well i thank you so much for stopping my unnecessary struggle here I need to sleep

[rose apex]

Not much difference in Wan generations for me but in SDXL image generations where the upscale with model node used to take majority of the generation time (still does), it has gotten significantly faster with this nightly build.

Generation time went down from 110s to 79s

[ornate elk]

that doesnt shows your gpu

click on the nvidia one

[rose apex]

Only the GTX 1070 and 1080 has 8 gb vrams I think

[ornate elk]

you cant see the vram of the nvidia one

the screenshot shows intel igpu shared memory

[ornate elk]

sounds good!

[rose apex]

Oh I see.

[plain crypt]

Hello, when I create an image, I get a message saying I need to reduce GPU usage. How do I do that?

[rose apex]

By lowering the resolution ig. Whats your GPU and what resolution are you using>

[plain crypt]

I have an Nvidia GeForce GTX 1650 and I don't know my resolution.

[rose apex]

What app are you using? Send screenshot of the error

[grizzled gull]

I get this error and web UI ddoesn't launch

[ornate elk]

What's your GPU?, also automatic1111 has known issues rn so can't be installed

[grizzled gull]

RTX 3060 TI

[ornate elk]

Then setup Forge Classic or Forge Neo instead.
Boath guides are in the pinned messages

[grizzled gull]

what's the ddifference between those?

Also my python is 3.10.10 do I need to update for Forge Neo?

[ornate elk]

Yes you need to update it.
The difference is that forge classic only supports models up to sdxl which work good on 8GB vram, and forge neo supports more and larger models too like flux, qwen, zit etc

[grizzled gull]

Also do I have to use 3.11.9 or is 3.11.14 fine too?

[ornate elk]

You have to uninstall python and make sure to install python with its own installer and not trough the Microsoft store.
Forge classic needs 3.11.9
And Neo needs 3.13.10

[grizzled gull]

Yeah I don't use microsoft store if I can avoid it

I think I'm gonna use Neo, though I don't plan on using heavy models

the guide says I need 3.11.9, I'm assuming this is old information?

[ornate elk]

Oh yes I have to update it

Will do later. 3.13.10 is the go to

[grizzled gull]

Also the instalation of 3.13.10 has frozen for some reason

it's been like that for about 10 minutes

[ornate elk]

Hmm strange

[grizzled gull]

should I restart it?

[ornate elk]

Wait a bit more and then restart if it does nothing

And open up a cmd as admin and run
SFC /scannow
That fixes windows system issues

[grizzled gull]

i tried cancelling it but that's stuck now too

[ornate elk]

Maybe reboot the PC and try again

[grizzled gull]

yeah that fixed it

The commad lines are already in the webui do I delete them?

the ones not for my GPU that is

[ornate elk]

no they are just placeholders

you have to type the correct ones at the line set COMMANDLINE_ARGS=

[grizzled gull]

Now this happened

it can't find python for some reason?

[ornate elk]

you have to rerun the python installer and click Modify, Next, and then check "add python to environment variables"

then relaunch ther webui-user.bat

[grizzled gull]

still doesn't work it was already enabled

[ornate elk]

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

[grizzled gull]

C:\Users\Admin\AppData\Roaming\Microsoft\Windows\Start Menu\Programs\Python 3.13

not sure why it ended up there

[ornate elk]

this is bad

that means it was installed from the microsoft store

[grizzled gull]

I used the installer

[ornate elk]

ah true it was confused, its just in start menu

[grizzled gull]

maybe i need to change the default location?

[ornate elk]

but its not installed correctly

[ornate elk]

dont mess with the install path of it

better reinstall it and make sure the "add to pyth" is checked

[grizzled gull]

should I be careful with the install path

for small programs I usually let it go to default

[ornate elk]

default is recommended

[grizzled gull]

Thanks a lot! I finally got it working also when I chose custom installation it puts python is C while with the express it puts it in app data

[ornate elk]

Appdata is the default path and okay, but C Programms is okay too

[wary cloud]

Will the issues with auto1111 carry over to forge?

[ornate elk]

you have to use forge classic or forge neo

[knotty cradle]

Hmmm

Whats happening herrre

[wary cloud]

Okay thanks

[quasi pilot]

not sure why it can't download torch but can i get a hand here

[ornate elk]

Wrong python version

3.14 is not supported

[quasi pilot]

so do i have to uninstall and reinstall the right version

[crystal nacelle]

im reintalling Zluda Forge from scratch and getting this error when running webUI - 32Gb ram Intel i5 AMD RX7600 8GB

tried deleting venv file, no help

[quasi pilot]

okay so now im getting this

[ornate elk]

Delete the venv folder and rerun the webui-user.bat

[ornate elk]

Thats a known issue rn and breaks auto1111 and forge rn.
Its caused by a library (PKG_resources) that was removed from setup tools
https://github.com/pypa/setuptools/issues/5174
You can try downgrade the setup tools version by editing the requirements_versions.txt
And there change setuptools to setuptools==81.0.0
Then save and delete the venv folder and relaunch the webui-user.bat

[quasi pilot]

okay something completely different this time after deleting venv, i'll try the other method

[crystal nacelle]

ill try that. my requirements_versions.txt file says "setuptools==69.5.1 # temp fix for compatibility with some old packages" so perhaps it already has an older version? i will change it to 81.0.0 any try deleting the venv file now

[ornate elk]

wait, with a 7600xt you dont need zluda at all, follow the Forge Neo with ROCm guide instead

[vocal burrow]

I haven't followed your whole story but this specific error looks like the usual "woopsie clip repo owner messed up" error. Basically someone deleted a file on github and most stable-diffusion-webui thingies's scripts expect it to be there sooooo....
You can try :
1/ open a cmd in webui's directory
2/ run venv\Scripts\python.exe -m pip install https://github.com/openai/CLIP/archive/d50d76daa670286dd6cacf3bcd80b5e4823fc8e1.zip --prefer-binary --no-build-isolation
3/ close it and run webui-user.bat , it should work after that

that's if you ever see an error like that when installing forge neo + rocm

[drifting prairie]

If your planning to run Z Image Turbo which is the easiest to use program that supports this model??

[onyx solar]

Easiest to use is relative.

[drifting prairie]

Which program would you recommend??

[onyx solar]

I recommend ComfyUI. There are example workflow templates that make simple image generation very accessible.

[celest delta]

@ornate elk hi, i'm on my new build, and i have issue reusing my webui

i instlled gitbash, the same python as we discussed last time too

put in the correct path, but it doesn't work

[ornate elk]

You can use ComfyUI or if you want an easier UI, then go for Forge Neo

[ornate elk]

Delete the venv folder and relaunch

[celest delta]

oooh, i forgot about that part! thanks!

[ornate elk]

Also open up a cmd and run
Python --version
And then
Where python

[celest delta]

yeah i already did put the new path

also now that i'm on a new build (had 10 gb vram, now 16), anything i should change or should i keep the settings as is?

[ornate elk]

What's your GPU?

And which webui

[celest delta]

5080

also webui1111 and forge neo

[ornate elk]

For auto1111 use:
--xformers
For Forge Neo:
--cuda-stream --sage --fast-fp16

You also need to reinstall torch for auto1111 to work with the rtx50 GPU.
You can find the steps in my guides

[quasi pilot]

I tried that and this is what I got, any ideas?

[ornate elk]

What's your GPU?

[quasi pilot]

4070ti

[ornate elk]

Then install Forge Neo instead. It doesn't have that clip error and is better with your GPU than auto1111 anyway

Guide is in the pinned messages first link

[elfin minnow]

can anyone just help me get this working PLEASE, instant $50 minimum for human help!

[celest delta]

still not working, same error as before

i checked and it's where python is

[ornate elk]

Run the commands I told you

[celest delta]

i did

[ornate elk]

And what do they show?

[celest delta]

i tried both path too

[ornate elk]

Okay looks good, and you deletet the venv?

[celest delta]

yeah

[elfin minnow]

where can I get some help if not here anyone please...?

[ornate elk]

Have you tried testing the example workflows?
That one doesn't look like one.
You can find them in the settings under templates or workflows

[vocal burrow]

:/ might have been some incompabilities with the python version you were using (fun times...) I only validated the version I gave you for python 3.10. Best is probably to wait for clip to fix their github repo and others to point to the updated url...

[ornate elk]

Can you try put the path into " "

[celest delta]

same

i guess i really have to reinstall the full thing... unless you find one last thing to try, :p

thanks nyway

[ornate elk]

Okay, nope then reinstall :/

But currently installing auto1111 isn't possible because of the clip issue

[celest delta]

it's forge neo here

that i was retrying

[ornate elk]

Ah okay. That will work

[celest delta]

does the auto1111 style.csv works on forge-neo?

cool cool cool

but man, i wish it would have worked wth my new build...

dunno what's the issue

[ornate elk]

Don't know but you can easily copy the stuff over, or make a separate models folder and link it to the webui

[celest delta]

oh wait... i launched webui.bat and it seems it's doing something... i was launching webuser.bat before

it works!

and wow the generation is so fast now, lol

sorry it must have been obvious that i should have been launching the normal .bat instead of user, my bad

[deft owl]

Hi I think actually fixed the clip issue ive been having along with everyone else and now im getting this: RuntimeError: Couldn't clone Stable Diffusion.
Command: "git" clone --config core.filemode=false "https://github.com/Stability-AI/stablediffusion.git" "F:\AUTOMATIC1111\stable-diffusion-webui-master\stable-diffusion-webui-master\repositories\stable-diffusion-stability-ai"
Error code: 128

[ornate elk]

Its important to always and only launch the webui-user.bat and not the webui.bat

[celest delta]

yeah but launching webui.bat is what fixed it...

i usually only launch user

[ornate elk]

Remove the python path from the webui-user.bat and it will also work...

Using the webui.bat won't use any launch args so its not good

[celest delta]

user still doesn't work

uh...

[ornate elk]

Then your bat is corrupt or has a typo

[celest delta]

it's the same as before

[ornate elk]

Can you show it?

[ornate elk]

What's your GPU?

[deft owl]

4090

[ornate elk]

And how did you fixed the clip error?

[deft owl]

you know thats a good question, i just did what chatgpt told me to lol and it worked

it said to delete a folder and put some in command prompt to reinstall and it worked

[ornate elk]

Then don't use Auto1111. Its outdated.
A high end GPU should use Forge Neo instead.
Its updated and has more models supported.
A setup guide is in the pinned messages of this channel

[deft owl]

oh ok

[celest delta]

i didn't modify anything in the folders when changing my PC

[deft owl]

im just trying to follow this to get novelai to run locallyhttps://aituts.com/run-novelai-image-generator-locally/

if forge can do that then thats great

[ornate elk]

Yes forge neo can do all of that and more

[deft owl]

alr

any good guides you reccomend

[ornate elk]

Also that guide is outdated

[deft owl]

or is it just a simple install

oh

october 2022 lol i see

[ornate elk]

The first link of the pinned messages of this channel is all you need for setups

[deft owl]

alr thanks bro

[ornate elk]

No problem!

[ornate elk]

Can you show the content ?

[deft owl]

oh and one last thing, which one am i supposed to install? i did the one that says click here to download

[ornate elk]

Windows 64x bit setup

[deft owl]

ok got it

should i uninstall the one i installed

[ornate elk]

Right click edit...

[ornate elk]

Yep

You can delete it

[deft owl]

alr

[celest delta]

yeah? That's what i've been doing from start

[ornate elk]

I just want to see what's in that file pls don't make it so difficult xD

Edit it with any text editor

[celest delta]

oooh my bad, i thought you were telling me to do that like i didn't do it before, sorry, xD

the path written, but you told me to remove it

[ornate elk]

Can you add
:: before the python= then save and relaunch and then show the cmd log you get?

[celest delta]

same as before

[quasi pilot]

Hey so got mine working, but it says no models found. Am I right in assuming I need to find some to download

[ornate elk]

Yep, you can download them from civitai.com

Any recommendation? Anime, real, semi realism?

[quasi pilot]

Not sure yet kinda wanna muck around I guess, maybe real

[ornate elk]

Okay, then go for sdxl models or z-image-turbo or flux klein

[celest delta]

how did you fix the clip issue?

[ornate elk]

Can you delete the venv folder again and relaunch?

[celest delta]

same as before

[deft owl]

Like I said above I just asked chat gpt, I got a different error later so I'm not even gonna touch that anymore

[celest delta]

not fucking with chat GPT, lol

thanks anyway

hope it will be fixed soon

i still have some use for A1111

[ornate elk]

Hmm, then I would recommend a reinstall and test if that works

[celest delta]

it's weird since webui.bat works perfectly

[ornate elk]

If it doesn't work then something is wrong with python

Yea very strange

[celest delta]

also the webui-user of aut1111 seemed to work too

it just stopped at the clip error you said earlier

how will that be fixed though? I though a1111 was obsolete and they didn't do stuff with it anymore

[ornate elk]

The clip error is not an auto1111 error, its a library that got updated and had something removed

The devs need to add it back to the library then it will work again.
The 128 error from stability ai is already fixed in the dev branch

[deft owl]

hi

now im getting this

forge neo

[celest delta]

cool, thanks!

[ornate elk]

Can you show the complete cmd log?

[deft owl]

new one

i updated python and added it to path

still doesnt show 3.13.12 tho for some reason

[ornate elk]

Delete the venv folder and relaunch

Also why do you don't follow the Forge neo guide from the pinned messages?

[deft owl]

i did, but i got the error

[ornate elk]

Your trying to manually create a venv which is not listed in the guides for Nvidia forge neo

[deft owl]

yeah ill stop asking chatgpt

[ornate elk]

Yes please

[deft owl]

if this doesnt work then what

[ornate elk]

I wrote the guides so better ask me xD

Use the guide to install forge neo and if you get an error post it here

[deft owl]

oh and kinda unrelated but my pc wont restart and just gets stuck on a black screen

power button on and all

[deft owl]

alr

ok heres the error same one i think

[ornate elk]

Open up a cmd as admin and run
SFC /scannow
That can fix windows system errors
Then reboot

[ornate elk]

Complete log please

From start to end

[deft owl]

ok ill send in 2 parts

[ornate elk]

Wrong python version

3.10 is not supported by forge neo

You need 3.13.12

[deft owl]

yeah i updated but when i check its still is on 3.10

[ornate elk]

Delete the venv folder and relaunch

And make sure you uninstalled python 3.10

[deft owl]

oh i havnt done that yet

i did delete and reinstall tho

[ornate elk]

The venv always needs to be deleted after a python version change

[deft owl]

alr uninstalled

deleting and starting it again

it worked

thank you so much

[ornate elk]

Perfect, no problem

[ornate elk]

Models can be found on civitai.com
For anime go for illustrious models like Nova Anime XL

[deft owl]

do you know how i can get stuff the quality of novelai or if you know anything better

k

[balmy shore]

is running DPM++ 2m karras 40 steps for sd1.5 still the standard? I asked Claude for help fine tuning settings and this was kind of what it suggested

for final renders anyway. Euler a for quick unscaled

[balmy shore]

I'm working on improving quality today for sd1.5, if that's what ur asking about

[deft owl]

oh im asking what models would match the quality of what novelai makes, im new to this so idk what you really mean

[balmy shore]

if u have the vram i think sdxl models almost always produce better results than sd1.5 across the board btu im not positive

i dont so ive only ever used 1.5 really but yeah some models are definitely better than others in general

currently trying to figure out adetailer cuz thats supposed to help for things like faces and hands but

my best results have come from generazting a starting image at like 512px, then upscaling it in stages like 1.25 increments

dont have my glasses on though

[ornate elk]

whats your gpu that your on 1.5 ? 😮

[balmy shore]

gtx1060

6gb

some say i could fit butits so tight

[ornate elk]

with 6gb it will work,
if you use auto111 make sure you have --xformers --medvram in the webui-user.bat
Or try Forge it should work there too

and if you have 16gb or less of Ram you should increase the windows pagefile, that helps in performance

and then try a resolution of 768x1024 for example, euler a, 30 steps

[deft owl]

what should i be running

[ornate elk]

You can run anything with that. It depends on what you want to create, for Anime there is nothing better than Illustrious based models (6gb)
If you want to make realism images then try out Z-Image-Turbo or Flux, or flux klein or Qwen Image.

[balmy shore]

ya

[deft owl]

would it be ok if i upload a pic with the style that i want

[ornate elk]

if its not nsfw then ok

[deft owl]

ok what do you consider nsfw before i send anything

[ornate elk]

nudity, overly exposed erotic scenery, stuff you wouldnt show your mom or boss 😛

[deft owl]

hmm

theres no nudity but alot of cleavage

[ornate elk]

you can dm it to me if your not sure

[deft owl]

ok ill do that

[narrow obsidian]

"stuff you wouldn't show your mum or boss" is the best definition of nsfw I've seen in a long time 😄

[celest delta]

@ornate elk tried checking around the internet for my previous issue, and i found this : "I solve it in the ide by just replace all the word "Administrator" in the entire project to my windows user name"
but there's not further detail, any idea how to do this? what's the "ide" ?

[ornate elk]

Have you tried the reinstall of forge neo?

[crystal nacelle]

okay, im running neo forge for first time now. will let you know if i get issues

[ornate elk]

Alright 👍

[celest delta]

not yet cause i really wanted to find a soluton before doing so... i remember it being annoying to do, lol

how should i proceed to reinstall, since i already have it,

what to delete, what to keep, and so on

also for the clip issue for auto1111, i guess all i need to do is retry to launch it everyday? until they fixed it?

[ornate elk]

Keep the models folder and the extensions folder and the outputs folder

And the styles csv

[celest delta]

yeah so basically everything and restart from scratch, thanks (i would have kept those regardless)

[delicate citrus]

I did what the link says almost install in the end show me this message, do you know how can I fix it?

[ornate elk]

hey whats your gpu?
the Automatic1111 and forge guide are currently broken due a library error. will update the guide.
Boath are outdated anyway.
If you have a RTX GPU then go for the Forge Neo guide.
If you have a GTX then go for Forge classic

[delicate citrus]

gtx 1610

so I must install forge classic?

nicee

@ornate elk show me this error now with forge classic, its possible to fix it?

[ornate elk]

let me check something real quick

[delicate citrus]

okaayy 🙂

[ornate elk]

yep okay, it seems the dev updated classic to need Python 3.13 instead of 3.11 , last week

so you need to uninstall python 3.11
then install Python 3.13.12 from here: https://www.python.org/downloads/release/python-31312/
then delete the venv folder and relaunch the webui-user.bat

ill update the guide real quick

[delicate citrus]

ah ok, I will install it, can be an overlapping installation?

[ornate elk]

nope

[delicate citrus]

can I remove from windows remover tool?

[ornate elk]

yea just uninstall python with the windows system tools

[delicate citrus]

thanks cs!

[ornate elk]

np, also keep an eye on step 7 from the guide, as the older cards need a different torch version

[delicate citrus]

humm ok

[gleaming stirrup]

i'm trying to run the docker compose command but i am receiving this error message, any thoughts?

0.245 Initialized empty Git repository in /repositories/stable-diffusion-stability-ai/.git/
0.245 + git remote add origin https://github.com/Stability-AI/stablediffusion.git
0.249 + git fetch origin cf1d67a6fd5ea1aa600c4df58e5b47da45f6bdbf '--depth=1'
0.429 fatal: could not read Username for 'https://github.com': No such device or address

[urban pollen]

looks like it's trying to ask for a login but can't, maybe try cloning it manually first

[gleaming stirrup]

ok im doing git clone again and trying to build

[urban pollen]

cool yeah, if you're on windows just make sure you're using wsl2 for that build

[gleaming stirrup]

im on ubuntu atm

it was weird because my git config outside the container has those values for email / name

[urban pollen]

yeah the docker build is isolated so it won't see your host git config

[gleaming stirrup]

if i clone it again and download are you thinking the post download -> auto should work ?

[urban pollen]

yeah that should work, usually fixes those weird credential issues

[gleaming stirrup]

same error again?

[urban pollen]

try adding ENV GIT_TERMINAL_PROMPT=0 to the dockerfile, might show what's actually wrong

[gleaming stirrup]

sorry where would that go? this is what i'm seeing on auto section

auto: &automatic
    <<: *base_service
    profiles: ["auto"]
    build: ./services/AUTOMATIC1111
    image: sd-auto:78
    environment:
      - CLI_ARGS=--allow-code --medvram --xformers --enable-insecure-extension-access --api

[urban pollen]

you can just add - GIT_TERMINAL_PROMPT=0 right under the CLI_ARGS line

[gleaming stirrup]

sorry but that appears to be the same, the entire block is this

[urban pollen]

did the rest of that block get cut off? also try checking your docker dns

[gleaming stirrup]

that's everything there the next line is just cake@cakeubuntu:~/stablediffusion/stable-diffusion-webui-docker$

[gleaming stirrup]

i don't seem to have a username

user.email=elem######[e-mail]
user.name=#######
core.repositoryformatversion=0
core.filemode=true
core.bare=false
core.logallrefupdates=true
remote.origin.url=https://github.com/AbdBarho/stable-diffusion-webui-docker.git
remote.origin.fetch=+refs/heads/*:refs/remotes/origin/*
branch.master.remote=origin
branch.master.merge=refs/heads/master

do i need a Username there?

[urban pollen]

nah you don't need a username, it's probably just failing to reach github

[gleaming stirrup]

this line continues to grow in download size the more times i run this => => sha256:0660c54152977c71466dc720dcd377f9349b50ad1f29d0271eceb5d666d58133 668.99MB / 3.69GB

it's climbing

it started at 400MB / 3.69GB

[urban pollen]

no worries, those layers are just huge because of the ai stuff, it'll finish eventually

[gleaming stirrup]

#!/bin/bash

while true; do
  sudo docker compose --profile auto up --build
  sleep 2
done

[urban pollen]

lol brute forcing it usually works if the connection is just being flaky

[gleaming stirrup]

i will have cats wearing astronaut suits tonight. this is happening

[urban pollen]

haha nicee, definitely gotta see those space cats when they're done

[gleaming stirrup]

it failed later on, should i pass git config into the compose?

[urban pollen]

nah that wont help, try cloning the missing repos into the repositories folder manually first

[gleaming stirrup]

that immediately requires a username..

[urban pollen]

wait thats super weird, public repos shouldnt ask for a login at all

[gleaming stirrup]

https://github.com/Stability-AI/stablediffusion.git

[urban pollen]

maybe check git config --global -l for any weird insteadof redirects

[gleaming stirrup]

i don't see anything, does this require a login from your end?

sudo git clone https://github.com/Stability-AI/stablediffusion.git

[urban pollen]

nah it's public, maybe try cloning it without sudo

[gleaming stirrup]

asking for username again

[urban pollen]

that is so cursed, can you clone any other public repo right now?

[gleaming stirrup]

cake@cakeubuntu:~/stablediffusion/stable-diffusion-webui-docker/services/AUTOMATIC1111$ git clone https://github.com/zpqrtbnk/test-repo.git
Cloning into 'test-repo'...
remote: Enumerating objects: 1736, done.
remote: Counting objects: 100% (43/43), done.
remote: Compressing objects: 100% (38/38), done.
remote: Total 1736 (delta 19), reused 1 (delta 0), pack-reused 1693 (from 3)
Receiving objects: 100% (1736/1736), 274.47 KiB | 2.39 MiB/s, done.
Resolving deltas: 100% (896/896), done.

[urban pollen]

okay so git is definitely working, try cloning it in your home folder instead of inside that docker tree

[gleaming stirrup]

nope

is the repo having issues?

[urban pollen]

that is so cursed, try the ssh link instead of https

[gleaming stirrup]

that's beyond my skills

[urban pollen]

the repo is just dead, try cloning the w-e-w mirror instead

[gleaming stirrup]

sorry the what?

[urban pollen]

its just a working copy, use this link: https://github.com/w-e-w/stablediffusion.git

[gleaming stirrup]

i cloned that repo and reading the documentation i'm a bit lost

any pointers?

[urban pollen]

just move that folder into repositories/stable-diffusion-stability-ai and try building again

[gleaming stirrup]

i tried to locate repositories dir with

/home/cake/StableDiffusion/stable-diffusion-webui-docker/.git
/home/cake/StableDiffusion/stablediffusionwew/.git
/home/cake/Documents/testscripts/testscripts/.git

but i don't see a repositories directory

[urban pollen]

it won't exist yet, just create the repositories folder yourself in the main directory

[gleaming stirrup]

cake@cakeubuntu:~/StableDiffusion/stable-diffusion-webui-docker$ ls
data docker-compose.yml LICENSE output README.md repositories services

still complaining about the stabilityai

=> ERROR [download 4/9] RUN . /clone.sh stable-diffusion-stability-ai https://github.com/Stability-AI/stablediffusion.git cf1d67a6fd5ea1aa600c4df58e5b47da45f6bdbf && rm -rf assets data//*.png data//.jpg data/**/.gif

i can't even click that link and download or see the repo by hand

[urban pollen]

yeah it's totally gone, just swap the link in the dockerfile for the mirror

[gleaming stirrup]

got it

let me try

[urban pollen]

nicee, did that fix the stabilityai error or is it still acting up

[urban pollen]

for sure, fingers crossed it finally builds for you

[gleaming stirrup]

stil an issue but i'm going to try to swap the docker file git link

[urban pollen]

look in services/AUTOMATIC1111/Dockerfile and swap the dead link for the w-e-w mirror

[gleaming stirrup]

done, trying to download now

[urban pollen]

nicee, fingers crossed, let me know if the build throws another fit

[gleaming stirrup]

new issues

[urban pollen]

rip what's it saying now? paste the error log if you can

[gleaming stirrup]

oh wait i think it needs a SHA

how can i see the SHA on github?

[urban pollen]

click the commits link on the main page and grab the id next to the top one

[gleaming stirrup]

gasp, its running

waiting for success message

[urban pollen]

nice, just gotta let it cook now, first run is always a slog

[gleaming stirrup]

failed around that stability-ai area i think

[urban pollen]

damn, did you make sure the url in the clone script was updated too

[gleaming stirrup]

yes i did

COPY clone.sh /clone.sh

RUN . /clone.sh stable-diffusion-webui-assets https://github.com/AUTOMATIC1111/stable-diffusion-webui-assets.git 6f7db241d2f8ba7457bac5ca9753331f0c266917

RUN . /clone.sh stable-diffusion-stability-ai https://github.com/w-e-w/stablediffusion.git add7234e806807d8c153050b0fb10e66fb0d4775 \
  && rm -rf assets data/**/*.png data/**/*.jpg data/**/*.gif

RUN . /clone.sh BLIP https://github.com/salesforce/BLIP.git 48211a1594f1321b00f14c9f7a5b4813144b2fb9
RUN . /clone.sh k-diffusion https://github.com/crowsonkb/k-diffusion.git ab527a9a6d347f364e3d185ba6d714e22d80cb3c
RUN . /clone.sh clip-interrogator https://github.com/pharmapsychotic/clip-interrogator 2cf03aaf6e704197fd0dae7c7f96aa59cf1b11c9
RUN . /clone.sh generative-models https://github.com/w-e-w/stablediffusion.git add7234e806807d8c153050b0fb10e66fb0d4775
RUN . /clone.sh stable-diffusion-webui-assets https://github.com/AUTOMATIC1111/stable-diffusion-webui-assets 6f7db241d2f8ba7457bac5ca9753331f0c266917

[urban pollen]

yo check that generative models line, you used the same link as the sd one

[gleaming stirrup]

there were two places where the stability-AI was there

i swapped them both

[urban pollen]

yeah but those are separate repos, generative models needs its own link

[gleaming stirrup]

oh crud, do you have that link?

wait i can get it

[urban pollen]

yeah it's https://github.com/w-e-w/generative-models.git

[urban pollen]

bet, let me know if it actually finishes the build this time

[gleaming stirrup]

ok we're going again

[urban pollen]

fingers crossed, that should be the last hurdle

[ornate elk]

Normal automatic1111 is broken since the repo of stability ai got removed from github.
So you all you have to do is instead git clone the dev branch as there the URLs got updated.
I see you already got it working, just want to let you know the easy way too ^^

[urban pollen]

true, dev branch fixes it, but forge is way less of a headache lately

[ornate elk]

Forge is also outdated.
Forge Neo should be used instead

[urban pollen]

yeah for sure, the memory management on the new version is way better

[gleaming stirrup]

have you tried hooking up openwebui to SD?

[urban pollen]

yeah it's sweet, usually just works through the api once it's up

[gleaming stirrup]

i am trying to get stats on different things and use SD to create charts

[urban pollen]

sd is actually kinda bad at charts and text, it usually just hallucinates the data

[gleaming stirrup]

any recommendations that work well?

[urban pollen]

try flux if you have the vram, it handles text and structure way better

[gleaming stirrup]

flux is free and operates on my gpu?

[urban pollen]

yeah totally free to run locally, just needs a lot of vram to be fast

[gleaming stirrup]

is flux just straight up better than SD? i'm lost now in the web

[urban pollen]

for text and structure it's way ahead, try flux schnell to save time

[celest delta]

man that clip issue still not fixed, hopefully this week-end!

[urban pollen]

yeah it's a mess since the repo move, hopefully they patch it soon

[crystal nacelle]

tried running Neo Forge for the first time and i got this error in less than a minute. this is after successfully installing all the recquired stuff

btw the install guide says to install this on the C drive but I installed it on my E drive instead, is that okay?

[ornate elk]

Hey can you try these steps:
While in the forge neo folder click in the File Explorer bar (not searchbar) and type cmd then press enter.

Now run these three commands one by one:

venv\Scripts\activate.bat

pip uninstall torch torchvision torchaudio rocm rocm-sdk-core rocm-sdk-devel rocm-sdk-libraries-custom -y

pip install --index-url https://rocm.nightlies.amd.com/v2/gfx110X-all/ torch torchaudio torchvision

Then relaunch the webui-user.bat

Let me know if that worked. Then I have to edit the guide.

[sudden pike]

yo

sorry for interrupting.
i wanna use SD again (havent installed it) and did a pause using it
but now its just completly broken

i try to "redo" your tutorial but Zluda does not install / work

or no.

CLIP does not install

[sudden pike]

venv "C:\Users\Kunde\Desktop\AI\stable-diffusion-webui-amdgpu\venv\Scripts\Python.exe"
HIP Library Path: C:\WINDOWS\System32\amdhip64_7.dll
WARNING: ZLUDA works best with SD.Next. Please consider migrating to SD.Next.
Python 3.11.1 (tags/v3.11.1:a7a450f, Dec  6 2022, 19:58:39) [MSC v.1934 64 bit (AMD64)]
Version: v1.10.1-amd-53-g32f147b7e
Commit hash: 32f147b7eb59be991db6619b9fa63874d59e1ed3
ROCm: agents=['gfx1101']
ROCm: version=6.4, using agent gfx1101
ZLUDA support: experimental
ZLUDA load: path='C:\Users\Kunde\Desktop\AI\stable-diffusion-webui-amdgpu\.zluda' nightly=False
Skipping onnxruntime installation.
ControlNet init warning: Unable to install insightface automatically. Please try run `pip install insightface` manually.
You are up to date with the most recent release.
W0219 08:28:53.794000 27284 venv\Lib\site-packages\torch\distributed\elastic\multiprocessing\redirects.py:29] NOTE: Redirects are currently not supported in Windows or MacOs.
no module 'xformers'. Processing without...
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.
C:\Users\Kunde\Desktop\AI\stable-diffusion-webui-amdgpu\venv\Lib\site-packages\pytorch_lightning\utilities\distributed.py:258: LightningDeprecationWarning: `pytorch_lightning.utilities.distributed.rank_zero_only` has been deprecated in v1.8.1 and will be removed in v2.0.0. You can import it from `pytorch_lightning.utilities` instead.
  rank_zero_deprecation(
Launching Web UI with arguments: --use-zluda --update-check --skip-ort

rocBLAS error: Cannot read C:\Program Files\AMD\ROCm\6.4\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.
Drücken Sie eine beliebige Taste . . .

[ornate elk]

Hey, it seems that zluda found your CPUs iGPU.
Go into the device manager and there disable the Radeon TM Graphics under display adapters.
Then relaunch the webui-user.bat

[torpid sapphire]

So im using forge ui with wai illust, whenever i use this hire fix, the image generation will always stuck on 98% for an extra 10 minutes while the terminal says it's completed

Is it something to do with ram allocation or something else?

Yes it does eventually complete the generation, but it stays at 98% for an awfully long time, didn't behave like this 3 weeks ago

[ornate elk]

Hey, what's your GPU?

[torpid sapphire]

RTX3070 TI

16GB

[ornate elk]

Okay, the issue is that your on a too high base resolution and then want to upscale.
The solution is to enable the Tiled VAE setting under NeverOOM at the bottom of txt2img.
Also lower the resolution for upscaling to 823x1216.

But a 3070ti has only 8GB of vram and with 16GB of ram you also need to increase the windows pagefile for the best performance.
Also you should use forge neo instead of forge ui.

All that will make it run much faster

[torpid sapphire]

Oh i see i see, yeah was a bit too lazy to switch to neo forge, thanks, will try neo forge when i get to it

[ornate elk]

Yea its much better optimised and supports the latest torch and stuff.
Use the setup guide from the pinned messages of this channel to set it up. It includes setting the correct launch args too.
But the pagefile increase needs to be done too for Max performance

And then always enable Tiled VAE when using hires fix

[torpid sapphire]

Thanks

[ornate elk]

No problem, let me know if you have any questions

[torpid sapphire]

Right, uh, installed neo forge, but it appears im missing... things? No clip speed, no "all" UI preset, and gpu weight, some upscaler is missing as well... I dont remember what i did to my old forge back then to get those, was it GGUF? sorry for so many questions...

[urban pollen]

just change that ui preset from sd to all and everything shows up

[torpid sapphire]

yeah doesnt have that option

[urban pollen]

ah my bad, try picking 'xl' since you're on illustrious

[torpid sapphire]

Nope, not seeing it even after chanigng to xl

[urban pollen]

it's right there under sd, picking xl should show those settings

[torpid sapphire]

imma just install gguf and things that used to be on my old forge..

[ornate elk]

Upscalers can be downloaded into models/esrgan folder

[urban pollen]

yeah for those you definitely want the gguf extension installed

[ornate elk]

Clip skip isn't needed but can be activated in the settings

[torpid sapphire]

alright

[ornate elk]

No need for a gguf extension

Forge neo supports gguf natively

Just drop the models into the models/stables-diffusion folder

Clip models go into models/text-encoders

[torpid sapphire]

right

[torpid sapphire]

So what about gpu weight slider

[ornate elk]

Mostly not needed but can be activated in the settings. Under User Interface Quicksettings

Im not at the PC rn to check

[torpid sapphire]

i see, man what did i do back then to make them show up on top

[ornate elk]

That's the user interface quicksettings setting

There you can add stuff you want to see at the top

[torpid sapphire]

i see

[urban pollen]

ah yeah my bad, you'll love the speed once it's set up

[torpid sapphire]

Yeah, it's definitely faster, but not enjoying this setting up process

[torpid sapphire]

Hi sorry to bother again...
After some testing with forge neo... Is there any other way to control the memory usage(still can't replicate the memory slider like the old one), it's really resources intensive right now, but it is fast I'll give it that, I found the settings(kinda) and I'm not sure what's going on there

[urban pollen]

check the integrated tab for the swap slider, helps with the heavy usage

[urban pollen]

try pulling that gpu weights slider down to 0.5 or 0.6 to save vram

[torpid sapphire]

alright

[torpid sapphire]

Sorry where is this located?

[urban pollen]

just scroll down to the integrated section at the bottom

[torpid sapphire]

sorry not there, or you meant in specific section?

[urban pollen]

ah my bad, gpu weights is the one, just move that slider

[torpid sapphire]

oh, so the one just one, alright

[torpid sapphire]

should i turn this on as well?

Alright it's not taking up all my resources now, thanks guys

[urban pollen]

nice one, glad it's finally behaving for you

[sudden pike]

it worked...
thanks

[ornate elk]

perfect np 🙂

[ornate elk]

but wait your on a 7700XT, you dont even need Zluda

[sudden pike]

7800xt

[ornate elk]

you can follow the ROCm Guides

[ornate elk]

even better

[sudden pike]

i dont want this SDnext thing

[urban pollen]

lol fair, forge or comfy work great with rocm if you want simpler setups

[ornate elk]

no you can use Forge Neo with ROCm or comfyui, no need for sdnext

forge neo is better than forge

[sudden pike]

different UI 🙁

[ornate elk]

same ui

[sudden pike]

oh 😄

what's the performance difference

[sudden pike]

im on windows

still works?

[ornate elk]

im using forge neo + rocm with my 7900xtx on windows rn

[sudden pike]

idk what the big difference is

my Vram is fcked up

will it get better with forge neo?

[ornate elk]

yep

[sudden pike]

ok good

[ornate elk]

it works pretty good and has a better memory optimisation than forge

[sudden pike]

i use the standard automatic1111

i can just copy paste Lora etc right?

[ornate elk]

yes you can even just link them with a launch arg, so no copies needed

[urban pollen]

yeah symlinks or the lora-dir arg save so much space

[sudden pike]

doesnt it need fast internet?

[urban pollen]

nah, it's all local once you have the files downloaded

[sudden pike]

SystemError: Please update your GPU driver to support cu130 ; or manually install older PyTorch

[urban pollen]

that error is cuz it's looking for nvidia stuff, you need the amd or directml version instead

[ornate elk]

which step from the guide?

[sudden pike]

nvm. i didnt read until end cuz im stoopid

im installing the 7XXX pip thing now

[urban pollen]

nice one that should get you sorted for the 7800xt

[sudden pike]

ok damn

VRam is optimized by like 200% or so DAMN

[ornate elk]

also a tip, if you gonna use hires fix for upscaling, make sure to enable the Tiled VAE option at the NeverOOM setting at the bottom of txt2img
That helps a lot and reduces freezes

[sudden pike]

its so much faster and i can generate WAY higher resolutions LOL
without tunning into my system RAM

[ornate elk]

sounds good 🙂

[sudden pike]

what software optimization can do...

[urban pollen]

yeah software is half the battle, forge is just built different for vram

[ornate elk]

yea this is now native AMD rocm instead of emulating cuda + optimisations of forge neo

[sudden pike]

ah so ROCm works now on Windows

[limber flax]

Hi all! I've just installed ComfyUI and the "ace step 1.5" music gen. template, but the input field seems to need Google Gemini credits to use it. From what I read and saw, I thought that music could be generated completely locally without any API or credits. Can anyone help or am I missing something? 👀

[sudden pike]

i can do fatter Batch sizes ❤️

[celest delta]

auto1111 didn't have a cgf scale for hires, but forge neo does, what cfg scale should i use to get the closest result as auto1111 was? was it just using the same amount?

[ornate elk]

hey, ou probably loaded the wrong workflow in.
here is one without any cloud stuff: