🤝｜tech-support

[night orchid]

it was all I could afford 😉

@karmic crown Deforum is an extension?

[karmic crown]

yes

https://github.com/deforum-art/sd-webui-deforum

but be carefull if you install it

[night orchid]

ok so all I have to do is install from url ?

[karmic crown]

there are 2 versions

[night orchid]

https://github.com/deforum-art/sd-webui-deforum.git

[karmic crown]

one for a1111, and the other for forge

the one you get from load button is for a1111

https://github.com/deforum-art/sd-forge-deforum

this one is for forge

i have no idea if forge version is fixed or not

i just made some changes for a1111 version

but some changes are not merged yet (from today)

[night orchid]

ok have you done a push? or is it only changes to a file ?

[karmic crown]

https://github.com/deforum-art/sd-webui-deforum/pull/960
i made a PR, but is not mergded yet

[night orchid]

https://github.com/deforum-art/sd-webui-deforum.git this one is for 1111 ?

[karmic crown]

yes

[night orchid]

just to make sure

so you " corrected" Deforum for 1111?

not both?

[karmic crown]

only "normal" one

[night orchid]

ok 🙂

ok so how can I test ?

I sould see a tut I think

[karmic crown]

install this one https://github.com/viking1304/sd-webui-deforum

remove the original one first

it is the same thing, just with some changes

just go to Deforum and click Generate, it will create video for you with some test script

click button with extreamly long caption "Click here.... to show video"

[night orchid]

ok nice

[karmic crown]

then try interpolation

[night orchid]

I am watching a tut on Deforum

[karmic crown]

use FILM and set slow-mo leave rest to default, that will create slowmotion video

and last RIFE insted of film

I cant test FILM properly, because that one crash on any Mac with or without my changes 😦

[night orchid]

wow it is very advanced

what is not working in the orginal one?

[karmic crown]

i just fixed few small things to make it work on clean install of a1111 1.8.0

[night orchid]

is it some kind of morphing?

Deforum

[karmic crown]

{
"0": "tiny cute bunny, vibrant diffraction, highly detailed, intricate, ultra hd, sharp photo, crepuscular rays, in focus",
"30": "anthropomorphic clean cat, surrounded by fractals, epic angle and pose, symmetrical, 3d, depth of field",
"60": "a beautiful coconut --neg photo, realistic",
"90": "a beautiful durian, award winning photography"
}

you make a prompt like this one

[night orchid]

ok so it morph 4 times

[karmic crown]

actually more, but yes

#🤝｜tech-support message <- click here to see the video

[night orchid]

and now for a very stupid question, why go to such a great lenght to make a extension that morphs between animations?

btw the original one works fine

[karmic crown]

if you upgrade from 1.7.0 than yes

[night orchid]

I have 1.8

[karmic crown]

if you made a clean install of 1.8 and used original, "normal" generate will work

[night orchid]

ah ok

[karmic crown]

but not FILM or RIFE interpolation

[night orchid]

so sorry I could not test your solution

[karmic crown]

if you upgraded from 1.7.0 all will work

you can

remove the original extension (remove the folder deforum-for-automatic1111-webui)

install from url https://github.com/viking1304/sd-webui-deforum

[night orchid]

ok I will to that tomorrow 🙂 late in my part of the world 🙂

[karmic crown]

then remove folder again, and and restore the original one

[karmic crown]

i hope changes will be merged by then, so you will not have do anything else except to update

[night orchid]

ok 🙂

thanks for help today

[old parrot]

What are ya'lls favorite sampling method and steps amount?

[karmic crown]

LCM 8 🙂

[noble belfry]

i've a few rules of thumb i use. for every 5 extra words over 10 in a prompt, crank the cfg up by 0.5. (with standard models, especially sd1.5) if im upscaling, half the number of steps used originally and 0.4 denoise.

the upscaling rule of thumb works for lightning models too, so if im doing a 5 step, i dont need more than 2 or 3 upscale steps

seeing hires steps left at 0 makes me weep.

[old parrot]

interesting, I'ma try

[old parrot]

What's your fav sample method?

[karmic crown]

always start with what author of the model recommends

than experiment

[old parrot]

I rather just experiment, working great so far

[karmic crown]

some models just do not work properly with some samplers

[old parrot]

I've only been using euler a

[karmic crown]

so always read recommandations for sampler, steps and clip skip

[old parrot]

and dpm 2 sd karras or whatever

[noble belfry]

depends, euler a is fine for sd1.5 models, generally use dpm 2 sde, but i'd be lying if i said i could pick any of the different flavours out of a lineup.

i might try that in a script

[old parrot]

What do you mean by flavours?

[noble belfry]

SDE SD karras, SDE SD 2 ++, SDE DVD XYZ with pineapple.

[brazen trout]

Guys i got this error everytime i run a render RuntimeError: CUDA error: unknown error
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

any solvation ?

[cursive acorn]

quick question, when im on txt2img i can use a hires but how would i replicate that on img2img? just generate @ source image's resolution and then separately latent upscale with the same prompt?

[noble belfry]

down in scripts there should be an option for SD upscale

[wintry folio]

Heya! I recently got personal access to an instance on a local computing cluster, and it would give me access to two Tesla V100 GPUs. The instance would be running Linux and I can access it remotely via shell and I even get a VPN so I can e.g. host a website on that server and access it from home.
Now for the question: What would be the best way to make use of that for stable diffusion? As far as I know webUIs like A1111 do not support multiple GPUs. Are there alternatives? And are V100's even good for Stable Diffusion in the first place? (idk if they're 16GB or 32GB models, I'm assuming the former)

Also I'm not 100% sure whether I'll actually be able to connect to any local websites being hosted on that cluster. It seems likely going by the description of the VPN but they don't clearly state it

Ah, just found out that the V100s are definitely the 16GB ones, but I also discovered that we can use H100 GPUs with 24GB/GPU

[quick forge]

ok, so I had something go very wrong, lockup and crash my computer and require me to reinstall linux, and all of the stable diffusion stuff....
But now I get this error when i run webui.sh

RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check

Keeping in mind, it was woking just fine beforehand with my AMD card without any command line arguments....so what did I miss when I reinstalled everything?

oh for some reason there is no video and render group

[noble belfry]

was the thing that went very wrong a hardware issue?

[nimble cargo]

i need help i think

i reinstalled my installation of Forgeui, was redownloading all my extensions, however CivitaiBrowser+ keeps giving me an error everytime i click on literally anymodel

CivitAI Browser+: Error: 500 Server Error: Internal Server Error for url:
Traceback (most recent call last):
File "C:\WebUI\stable-diffusion-webui-forge\venv\lib\site-packages\gradio\routes.py", line 488, in run_predict
output = await app.get_blocks().process_api(
File "C:\WebUI\stable-diffusion-webui-forge\venv\lib\site-packages\gradio\blocks.py", line 1431, in process_api
result = await self.call_function(
File "C:\WebUI\stable-diffusion-webui-forge\venv\lib\site-packages\gradio\blocks.py", line 1103, in call_function
prediction = await anyio.to_thread.run_sync(
File "C:\WebUI\stable-diffusion-webui-forge\venv\lib\site-packages\anyio\to_thread.py", line 33, in run_sync
return await get_asynclib().run_sync_in_worker_thread(
File "C:\WebUI\stable-diffusion-webui-forge\venv\lib\site-packages\anyio_backends_asyncio.py", line 877, in run_sync_in_worker_thread
return await future
File "C:\WebUI\stable-diffusion-webui-forge\venv\lib\site-packages\anyio_backends_asyncio.py", line 807, in run
result = context.run(func, *args)
File "C:\WebUI\stable-diffusion-webui-forge\venv\lib\site-packages\gradio\utils.py", line 707, in wrapper
response = f(*args, **kwargs)
File "C:\WebUI\stable-diffusion-webui-forge\extensions\sd-civitai-browser-plus\scripts\civitai_gui.py", line 449, in update_models_dropdown
(html, tags, base_mdl, DwnButton, SaveImages, DelButton, filelist, filename, dl_url, id, current_sha256, install_path, sub_folder) = _api.update_model_info(model_string, model_versions.get('value'))
File "C:\WebUI\stable-diffusion-webui-forge\extensions\sd-civitai-browser-plus\scripts\civitai_api.py", line 789, in update_model_info
for item in model_images['items']:
TypeError: string indices must be integers

[deep heron]

        attn_output = torch.bmm(attn_probs, value_states)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    RuntimeError: expected scalar type Half but found Float

i get this after using easynegative in negative of txt2img

[fair oxide]

Is there any app for iPhone ?

[fair oxide]

Theres an app for Android?

[wintry folio]

btw if anyone has an answer for this I'd greatly appreciate if you pinged me 😄

[surreal sentinel]

Is it possible to move sd webui off C drive? I've been downloading a lot of models, loras, etc, and im at over 40gb already lol

[surreal sentinel]

found a reddit thread with 'mklink' and i'm gonna test that in a sec

[drifting prairie]

Which of these models gives the best results Waifu2x Real-ESRGAN Real-CUGAN or UltraYandere

[void cove]

im playing around with streamdiffusion and there are some values which im trying to understand better so i know what it is that im changing and why i get the errors i do.

            model_id_or_path=base_model,
            use_tiny_vae=args.taesd,
            device=device,
            dtype=torch_dtype,
            t_index_list=[0, 4],
            frame_buffer_size=1,
            width=params.width,
            height=params.height,
            use_lcm_lora=True,
            output_type="pil",
            warmup=10,
            vae_id=None,
            acceleration=args.acceleration,
            mode="img2img",
            use_denoising_batch=True,
            cfg_type="full",
            use_safety_checker=args.safety_checker,
            # enable_similar_image_filter=True,
            # similar_image_filter_threshold=0.98,
            engine_dir=args.engine_dir,
        )

        self.last_prompt = default_prompt
        self.stream.prepare(
            prompt=default_prompt,
            negative_prompt=default_negative_prompt,
            num_inference_steps=4,
            guidance_scale=1.5
        )

does anyone know of some good resources in relation to t_index_list, cfg_type, num_inference_steps, guidance_scale, they all seem to rely on eachother but not 100% sure how or why.

[surreal sentinel]

hmm... moving my models, loras, etc, from one ssd to another ssd and symlinking them has slowed down generation significantly

[ornate elk]

That doesn't make sense as the models and loras are loaded into the RAM and then into GPU vram before generating

[surreal sentinel]

not sure, i have 64gb system ram and 8gb on the 3070

[ornate elk]

Make sure you use --xformers --medvram-sdxl --no-half-vae

In your webui-user.bat

[surreal sentinel]

yep, those are in

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --no-half-vae --medvram-sdxl --skip-python-version-check --ckpt-dir "G:\Stable Diffusion - MODELS\models\Stable-diffusion" --lora-dir "G:\Stable Diffusion - MODELS\models\Lora" --vae-dir "G:\Stable Diffusion - MODELS\models\VAE"

call webui.bat

full file

[ornate elk]

Why skip python version ?

[surreal sentinel]

that was the default

even args was blank

[ornate elk]

Thats not default

[surreal sentinel]

it was for me

[ornate elk]

Then you didn't installed automatica1111 the default way

[surreal sentinel]

I just hit 'webui-user.bat' and let it do its thing. didn't change a thing until after it was done

[ornate elk]

Hmm and how did you installed it in the first place?

[surreal sentinel]

i do have python 3.11 installed, but that's cause i was fooling around with extracting game textures earlier

[ornate elk]

Did you git cloned the webui?

Python 3.11 is not supported by SD an can cause issues

[surreal sentinel]

can i have both .11 and the recc version at same time?

[ornate elk]

Yes

You need to install python 3.10.11 64bit
And then edit the webui-user.bat
At the line PYTHON=
You set the path to the python.exe of appdata.

Then remove the --skips pythons version
And then you need to delete the venv folder

[surreal sentinel]

C:\Users\kenso\AppData\Local\Programs\Python\Python310
This seems to be where .10.6 is

i just put in base directory or a specific one

[ornate elk]

Default

[surreal sentinel]

set PYTHON=C:\Users\kenso\AppData\Local\Programs\Python\Python310
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --no-half-vae --medvram-sdxl --skip-python-version-check --ckpt-dir "G:\Stable Diffusion - MODELS\models\Stable-diffusion" --lora-dir "G:\Stable Diffusion - MODELS\models\Lora" --vae-dir "G:\Stable Diffusion - MODELS\models\VAE"

call webui.bat
```this is what i have rn

dunno if right

[ornate elk]

Oh you already have it in there

[surreal sentinel]

no, just added it now

[ornate elk]

Ah alright, then remove the --skip-python-version

And then delete the venv folder

[surreal sentinel]

and rerun it?

[ornate elk]

When you relaunch the webui-user.bat
It will recreate the venv using python 3.10

[surreal sentinel]

set PYTHON=C:\Users\kenso\AppData\Local\Programs\Python\Python310
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --no-half-vae --medvram-sdxl --ckpt-dir "G:\Stable Diffusion - MODELS\models\Stable-diffusion" --lora-dir "G:\Stable Diffusion - MODELS\models\Lora" --vae-dir "G:\Stable Diffusion - MODELS\models\VAE"

call webui.bat
```so this?

[ornate elk]

Oh wait

You need to add \python.exe
At the end of the path

[karmic crown]

I have no idea about V100 and H100, but I saw this myself on NVIDIA A100 80GB:

Steps: 20, Sampler: Euler, Size: 1024x1024
Time taken: 3.7 sec.

[surreal sentinel]

set PYTHON=C:\Users\kenso\AppData\Local\Programs\Python\Python310\python.exe
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --no-half-vae --medvram-sdxl --ckpt-dir "G:\Stable Diffusion - MODELS\models\Stable-diffusion" --lora-dir "G:\Stable Diffusion - MODELS\models\Lora" --vae-dir "G:\Stable Diffusion - MODELS\models\VAE"

call webui.bat

venv is still in C. Just moved the large files like models/lora/etc

[ornate elk]

Thats okay

Venv shouldn't be moved away

[ornate elk]

Okay thats strange.
Would try to install 3.10.11 64bit and then rerun the webui-user.bat

[surreal sentinel]

this one? https://www.python.org/downloads/release/python-31011/

[karmic crown]

and as you already wrote, you will be able to use only one of those 😦

[ornate elk]

Yep

It will overwrite the 3.10.6

[surreal sentinel]

moment of truth

so far so good

[ornate elk]

Looks okay, but you didn't used git clone at your first install

[surreal sentinel]

https://github.com/AUTOMATIC1111/stable-diffusion-webui
I used this to install it.

[ornate elk]

Have you downloaded it as zip file perhaps?

[surreal sentinel]

then 'download zip'

[ornate elk]

Yea okay, thats not how you "install" it

[surreal sentinel]

do I use one of these?

[ornate elk]

Yes, normaly you use git in a Console window

[surreal sentinel]

Hm, so del entire folder and redo it with the git thingy

[ornate elk]

If you want to easily update the webui in future, then yes ^^

[surreal sentinel]

have the reddit thread up to do the symlink again after

[ornate elk]

Why Symlink? You already set custom paths in your webui-user.bat

[surreal sentinel]

i can leave the default folders in with that?

[ornate elk]

I dont get the question sry

[surreal sentinel]

i have no cmd prompts open, how is this still showing

when trying to del folder to redo

[ornate elk]

Close everything and then try again

If not you need a PC restart

[surreal sentinel]

hm, i did a restart of Explorer.exe even

restarting pc ig

[ornate elk]

@surreal sentinel
Here is my complete install guide for auto1111:
#🤝｜tech-support message

With git clone and every step needed

[surreal sentinel]

hmm, would installing this on an nvme make much of a diff? (base webui, store models on ssd with more space)

I use reg SSD for os, but nvme for open-world games

[ornate elk]

Would be faster in loading models mostly

But not a speed improve for generating images

[surreal sentinel]

ok so just put on reg ssd then, only takes like 10s to load some of the larger models like 5gb

weird, I can't right click > edit on the bat. guess ill open up N++

alright those two 'git' lines are gone

[ornate elk]

Looks good. Now the webui knows where to look if you want to update it

[surreal sentinel]

and to update i just run the git command again?

[ornate elk]

To update you run
Git pull

[surreal sentinel]

so navigate to the webui folder, open cmd and simply run git pull?

at least that's easy

[ornate elk]

You can also add --update-check
To the webui-user.bat
So when you launch it it will tell you if there is an Update available

[surreal sentinel]

ill add that in, so next time I run it, it'll do that. does it matter where on the cmdline?

[ornate elk]

Nope doesn't matter

[surreal sentinel]

ill put at start before xformer then

should i worry bout this?

[ornate elk]

No that can be ignored

If you want to you can run the green command it tells you

But if you dont have any issues, you can ignore it

[surreal sentinel]

i ran it in another cmd tab real quick for next time, since once its installed and tested i'm off to bed, 3am

any recommended extensions?

UI is loading, but this appeared in cmd

[inland zephyr]

Does Hires fix have the same effect on an image it doesn't upscale the size of?

[ornate elk]

Controlnet, Adetailer, Boorutag Autocomplete,

[ornate elk]

Did you tried to install an extension?

[surreal sentinel]

yea, i sorted by most stars and picked one

[ornate elk]

There is old - photo - restoration that gives an error

[surreal sentinel]

yea that's one of em

is that a problematic extension?

[vocal burrow]

it's a "difficult" one to install

but hey it's telling you what to do if you really want it

[surreal sentinel]

when trying controlnet

[vocal burrow]

would need to see the full log to understand that error

[vocal burrow]

In your webui-user.bat replace --ckpt-dir G:\Stable Diffusion - MODELS\models\Stable-diffusion with --ckpt-dir "G:\Stable Diffusion - MODELS\models\Stable-diffusion", same for lora--dir and vae--dir
Or better yet remove any spaces and weird character from your SD path.

[surreal sentinel]

this is a fresh install via cs's guide

weird, it removed the quotes

actually, they are in ```@echo off

set PYTHON=C:\Users\kenso\AppData\Local\Programs\Python\Python310\python.exe
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--update-check --xformers --no-half-vae --medvram-sdxl --ckpt-dir "G:\Stable Diffusion - MODELS\models\Stable-diffusion" --lora-dir "G:\Stable Diffusion - MODELS\models\Lora" --vae-dir "G:\Stable Diffusion - MODELS\models\VAE"

call webui.bat```

[vocal burrow]

ok then

[surreal sentinel]

removed spaces and edited bat

Ok, RIGHT after fnishing boot up and tab opening

[vocal burrow]

remove that sd-webui-old-photo-restoration extensions btw if you don't intent to use it/fix it. By deleting the "venv/sd-webui-old-photo-restoration" folder

[surreal sentinel]

folder doesn't exist

[vocal burrow]

my bad

just woke up

"extensions/sd-webui-old-photo-restoration"

[surreal sentinel]

alright, removed, restarting webui

17s startup time? that seems fast

[vocal burrow]

don't worry it will get slower once you installed a crap ton of extensions 😄

[surreal sentinel]

should I not do controlnet if its tricky?

[vocal burrow]

controlnet should work, it's easy to install and relatively easy to use.

[surreal sentinel]

CS recommended this but its not showing up: Adetailer

[vocal burrow]

For controlnet, just don't mix up controlnet model version and sd model version. By that I mean if you're using an SDXL sd model you'll have to use SDXL controlnet modules, same for sd1.5, sd2.1 and incoming sd3 (very probably)

[surreal sentinel]

what does it even do, was just recommended

[vocal burrow]

Does it not ? It should be under the generation settings. cf screenshot

(time for me to go afk and commute to work)

[surreal sentinel]

nope

[ornate elk]

In the Extensions tab its called !After Detailer

When you want to install it
Click on extensions tab, then on Available, then on "Load from"
And then search for !After Detailer
This extension is perfect for fixing faces.

[ornate elk]

With Controlnet you can do amazing stuff.
You can define for example the Pose of an character or recreate an image in an other style.
It basicly converts input images to alter the output image in different ways.

[surreal sentinel]

You can try to install dependencies manually by activating venv and installing requirements.txt or requirements-gpu.txt Hmmm not seeing a requirements.txt file in venv/

ah its in root

[ornate elk]

What are trying to do?

[surreal sentinel]

i figured it out: py -m pip install -r requirements.txt in root

a bunch of ones weren't installed

[obtuse hare]

Got an error when training a lora

[vocal burrow]

That's not how it works. You just installed a bunch of stuff for your system instead of installing it in SdWebui's venv

[obtuse hare]

help

[vocal burrow]

try asking in #🔧｜finetune

[paper crag]

agghhhhhhhhhhh

again with this shit

'OSError: [WinError 126] The specified module could not be found. Error loading "C:\SD-Zluda\stable-diffusion-webui-directml\venv\lib\site-packages\torch\lib\caffe2_nvrtc.dll" or one of its dependencies.
Press any key to continue . . .'

[vocal burrow]

at this point, just stop updating once you get it back working again

[paper crag]

i'd do that if i could make it working again

the method i used to fix it last isn't working this time...

[vocal burrow]

and what s the full log ? (cs1o is the one using zluda, I don't need it)

[atomic vapor]

ho

[paper crag]

guess i have to wait cs1o, im using zluda

[vocal burrow]

you can still share the full log XD

gaining some times and who knows, I might read something interresting in it

[vocal burrow]

ok. That s not much to work with indeed

love that warning message at the top xd

if you activate the venv and check what package is installed pip list, what do you get ?

[ornate elk]

Idk how you always manage to get that error lol xD

Did you installed an extension?

Or did you updated anything?

[vocal burrow]

my bet is git pull is in their webui-user.bat

[paper crag]

i don't know, yesterday i fixed it.
it might be because i have --update-all-extensions --update-check in webui, maybe something got update when i opened it. I only reopened it today
Other than that, the only modifcation i did to my pc in general was enable back my hibarnate mode becuase it was gone for some reason.

[vocal burrow]

--update-all-extensions => On startup, it pulls the latest updates for all extensions you have installed. Therefore yes it can install new dependencies and f** up everything

and do you have git pull in your webui-user.bat ?

[vocal burrow]

ok then yes I'd only remove --update-all-extensions to be safe later on

I m still interrested to see what pip packages are installed in your venv btw

.\venv\Scripts\activate.bat
pip list

[paper crag]

do i just run this command in cmd?

[vocal burrow]

open a cmd in your stable-diffusion-master folder and run those yes

[vocal burrow]

@ornate elk shouldn't there be cudatoolkit in there (even for zluda) ?

[viral fox]

Hey, I am reading through #✍🏼｜rules-and-tos and #1080946152318443610 and i keep coming across a channel marked as no-access. Being the inquisitive sort, can you tell me if this is by design or have i missed something hence I dont have access to these resources.?

Below is snip from #1080946152318443610 Community Section showing th no-access i am seeing there.

[vocal burrow]

Yes I just noticed too. We often move channels around, so links can break up. I've notified the person in charge of those. Hopefuly they'll be updated shortly

[viral fox]

Ta

[ornate elk]

Idk, can't verify that rn

[ornate elk]

Why driver booster?

[paper crag]

i recall having some issues getting my drivers, so i used it

or something, im not sure

[ornate elk]

Okay, but not the best practice for installing drivers.

[paper crag]

probably, i did help a bit
i used it on my laptop and i had over 30 things that needed update

[ornate elk]

Yea it finds alot of stuff but not every driver in there "needs" an Update

But anyway, what's the error you get now on zluda?

[ornate elk]

Okay and when you delete the venv and relaunch?

[paper crag]

4th time today...
gimme a minute to do it again

[ornate elk]

Make sure driver booster isnt in autostart after a PC restart.
Did you installed the latest AMD Driver Adrenalin by Hand?

[paper crag]

the adrenalin i got ir from the amd official site and installed it myself

[vocal burrow]

if you have any extensions try moving them to another directory for now, delete venv and try again

[paper crag]

moved all extensions, did the venv again, same error

il purge my pip cache and do it again

[ornate elk]

Do you have multiple python installations?

[paper crag]

i don't think i do

[vocal burrow]

driverbooster and avast, I can hear cs1o crying from here (it s not related to your problem don't worry)

[ornate elk]

Dont forget advanced system care 🥲

And the old unsecure winrar version

[paper crag]

okay, im not exactly in touch with the current stuff

[ornate elk]

Yea np, but avast driver updater won't help if you already have driver booster

And update your winrar Version if you have time. Current version is 7.00

There was a big security risk in an old version

[paper crag]

at this point, won't it be better if change to 7-zip?

okay....

i dont know which was it

but after restarting my pc, now it opened

while im glad its fixed, im curious what the source really was so it wont happen again

[fair oxide]

Do XL models still not support image2image? Every time i try a Xl model it gives me a error

[karmic crown]

when you reach my level of laziness, venv\scripts\pip is enough 🙂
no need to activate venv first if you do it like this 🙂

[ornate elk]

I would uninstall "avast driver installer" for sure and "advanced system care" they are bloatware.
(Personal opinion, avast is a bloated antivirus and windows defender is enough protection if your using and browser adblocker like ublock origin)

Also winrar 7.00 is safe to use.
But you can also use 7zip if you want.

[ornate elk]

It does. What's the error and your settings?

[limber prairie]

Please help a friend out 😪

[fair oxide]

XL models didn't work with image2image since day one for me

[karmic crown]

looks like it trying to download something and that fails, whatever that is

most likely some model for DWPreprocessor

but as i said, i never used that

[ornate elk]

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

[fair oxide]

I had the issue on both my machines with a 3090 and a 4070 ti

[limber prairie]

It's same for this too

[ornate elk]

On boath since day one?

How did you installed the webui?

[fair oxide]

I don't remember anymore.

Just how it says it on the github

[ornate elk]

Can you relaunch the webui-user.bat and show me the cmd log?

[ornate elk]

Okay torch and xformers can be updated.
Delete the venv folder.
Then relaunch the webui-user.bat
After its done load an sdxl model into img2img and just type something in the prompt and hit generate

[fair oxide]

Deleting the venv folder updates it?

Or do i have to upgrade the 2 first?

[ornate elk]

Launching after you deletet the venv will update it

[vestal laurel]

i can't find the user.bat

webui.user.bat

pls help

[ornate elk]

The second from the bottom

[vestal laurel]

webui.user?

[ornate elk]

That says Windows batch file as file type

[vestal laurel]

yes ys

it's the one?

i've been struggling with this i hope it works

thanks

[ornate elk]

No problem.
Here is my install guide. So you can recheck if you did everything correctly:
#🤝｜tech-support message

[vestal laurel]

is this it?

[karmic crown]

i had to do it 🤣

[ornate elk]

Yes its installing

[vestal laurel]

so i don't have to do it manulally?

[ornate elk]

Thats awesome haha!

[vestal laurel]

wow

[ornate elk]

Nope, when its done it should open in browser automatically

[vestal laurel]

https://tenor.com/view/lel-teeth-man-laugh-gif-24159653

dude we got One piece fans here???

[karmic crown]

just an extremely quick edit of an original image in PS 🙈

[vestal laurel]

W

[karmic crown]

Death Note for me, please! 🙂

[fair oxide]

Hmm..am i disabled

[vestal laurel]

Death note is a good anime, like really good, you should give Code geass a try if you liked death note

[ornate elk]

Why you have open 2 cmds ?

[karmic crown]

one is python

[ornate elk]

Oh

Right

[karmic crown]

do not run python as itself

[fair oxide]

One was the window to update. But well i did update and its still the same

[karmic crown]

just use CMD

[fair oxide]

On cmd i also got error

[karmic crown]

i need get back to work now

cu later

[ornate elk]

Is that with the 4070 or the 3090?

[fair oxide]

3090

Everything else works fine, just not image2image with xl models

[ornate elk]

Can you show the settings you used?

I see controlnet stuff in there

[fair oxide]

Just the standard settings

[ornate elk]

Any loras or embeddings used?

[fair oxide]

I just loaded a random pic and typed "test"

[ornate elk]

Okay

Then add --no-half-vae to the webui-user.bat

And try again

[fair oxide]

It works with non Xl models

[fair oxide]

Ok that works

[vestal laurel]

@ornate elk

[fair oxide]

Thanks

[vestal laurel]

😰

[ornate elk]

Perfect np

Add that to the 4070 PC too

[ornate elk]

Is your GPU AMD ?

[vestal laurel]

i think yes

so it's not compatible?

[ornate elk]

You installed the nvidia version of the webui

[fair oxide]

If i like playing around with SD should i keep my 3090 or 4070 ti pc for myself, and give my gf the other?

[vestal laurel]

oh

[ornate elk]

@vestal laurel What's your GPU model?

[vestal laurel]

i'll send you

wait

[ornate elk]

Def, keep the 3090 as more vram is better for any Ai stuff

[ornate elk]

Ohhh

[vestal laurel]

😰

[ornate elk]

You dont have any dedicated GPU :/

[vestal laurel]

oh man

rip

so only pc's have that I pressume?

[ornate elk]

Laptops can have that too

[vestal laurel]

mine doesn't, rip

[ornate elk]

Yea, it would still work on CPU mode, but thats very very slow. Like 1 image in 20-30 minutes

[vestal laurel]

Jesus

but thanks for all you help, appreciate it

[ornate elk]

No problem, better you checkout some Websites where you can generate for free
Like StableHorde or checkout Leonardoai (I think they have some free credits but not sure)

[vestal laurel]

oh that yes

i did try leaonardo

i was trying to make colouring books

but well, stable diffusion seems way more efficient to use

when I saw yt videos

but guess imma have to stick with leonardo

for now

[ornate elk]

Yea with Stable Diffusion you have the most freedom and tools to generate what you want

[vestal laurel]

not just that it doesn't mess up hands and other body parts like other ai tools

tried to make a rabbit turned out really well, but the feets, well...no need for words 💀

*leonardo

[fair oxide]

Do we already know how hard it will be to run sd3?

[vocal burrow]

not really considering the moment it will release the community is gonna be hard at work for a couple of days to optimize its usage :p

[trim raven]

can stable cascade be used in SD where it's really something special?

[vocal burrow]

stable cascade is an SD model, yes it can be used with comfyUI, auto1111 (with a specific extension), etc

[trim raven]

ok thanks 🙂

[surreal sentinel]

srry, was turning pc off at the time. Doesn't hurt anything right?

[vocal burrow]

not necessarily but you just installed a bunch of "useless" stuff in your system

[surreal sentinel]

i have over 12tb total storage, but only about half that is internal or SSDs. Other half is backup/external

so not really an issue here, tbh

[karmic crown]

it is usually not the problem, but it potentially might be a problem in the future if you need to have a newer version of some Python module in venv but you have a old version in the system

[surreal sentinel]

I could just run the update requirements command again tho, right?

[karmic crown]

just do the opposite of what you did

pip uninstall -r requirements.txt

[vocal burrow]

doing that might get rid of """too much""" if you're using python somewhere else, but you can always reinstall stuff if needed

[karmic crown]

i saw here a few times some unexplained things, until we figured out that some old libraries installed globally were making a mess

[surreal sentinel]

wouldn't uninstalling reqs also uninstall stuff needed for SD?

[ornate elk]

yes

if you dont get anny errors, your fine and dont need to do anything.
if you get any weird erros, delete the venv folder

[vocal burrow]

they meant uninstall reqs on their system.

because they ran pip install -t requirements.txt outside of venv

[ornate elk]

ohh outside of venv

[vocal burrow]

ok so short introduction to venv.
@surreal sentinel venv is sort of a virtual machine for your python. venv, virtual environment.
So basically you have your "system's python" which is installed in c:/program, appdata or wherever. This is the python that is gonna be called by most program whenever they need to run python stuff.
On top of that "system's python" you can install "modules" using "pip". eg : a module to handle face recognition, a module to do optimized matrices computations,a module to download files from the internet, etc....
The problem is that python (like most software) come in different version (3.10.6, 3.10.11, 3.11.0, etc) and so do each of python's module..... With each version, little things change in the way all of those interface with each others. Therefore whenever you try to run a very complex python program which needs many modules, those little changes add up pretty fast.
And obviously you can't let a single program dictate which version should be installed and which shouldn't. Otherwise you'd be uninstalling/installing different version of everything everytime you run different python programs. Ok maybe you don't run that many python programs, but some people do.

That's why. Venv !
Venv is basically a copy paste of your "system's python" in a different folder (stable-diffusion-webui/venv in our case) dedicated to one program (Stable Diffusion). That way, said program has its own python and can mess with it as needs be. It won't impact the "system's python", it won't impact the other python programs.

So when you ran py -m pip install -r requirements.txt (or pip install -r requirements.txt) you installed a bunch of "useless" stuff for your "system's python". Which can cause issues because depending of how the venv are configured they might actually use their python modules AND system ones.
Conclusion, you need to clean your room.

[noble belfry]

This is worth a pin

Its a universal problem solver for almost anything using torch

[surreal sentinel]

so tldr, I should uninstall reqs file and then recreate venv for optimal?

[vocal burrow]

pretty much

is it gonna fix your problem ? Honetly by now I forgot what it was XD

[surreal sentinel]

can't seem to figure out how to get this installed, so if I re-install venv, how would we fix that?

[vocal burrow]

oh it's roop again

[surreal sentinel]

dunno what could be causing it. I run win 11 pro, with an encrypted c drive

[vocal burrow]

faceswap, reactor, roop... basically anything """"deepfake"""" related is noteworthy for being bad at managing their dependencies.

at least it seems like that.

starting with a clean system python and fresh venv is a good start to isolate the problem.

[vocal burrow]

doesn't matter much here

[surreal sentinel]

hm, one of my games uses panda3d, which I think uses python as a base, would that require a reinstall of both?

[ornate elk]

you need to delete the roop extension

its outdated

better use Reactor for faceswap

but read its install guide on gtihub

[surreal sentinel]

i had all these from the cmdline version, then webui version. Remove all but 3.10.11?

[ornate elk]

yes

[surreal sentinel]

should I add in py launcher, or is that ok to leave uninstalled?

[vocal burrow]

leave as default

[surreal sentinel]

assumed so, just wanted confirmation 👍

near end of venv reinstall I noticed this and:

    Traceback (most recent call last):
      File "C:\AI\stable-diffusion-webui\modules\scripts.py", line 527, in load_scripts
        script_module = script_loading.load_module(scriptfile.path)
      File "C:\AI\stable-diffusion-webui\modules\script_loading.py", line 10, in load_module
        module_spec.loader.exec_module(module)
      File "<frozen importlib._bootstrap_external>", line 883, in exec_module
      File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
      File "C:\AI\stable-diffusion-webui\extensions\sd-webui-EasyPhoto\scripts\easyphoto_ui.py", line 156, in <module>
        _scenes = os.listdir(os.path.join(models_path, "Lora"))
    FileNotFoundError: [WinError 3] The system cannot find the path specified: 'C:\\AI\\stable-diffusion-webui\\models\\Lora'```

says failed to find lora too, but it does see them

hmm, when i went to the main folders properties, it had a [-] icon by Read-only, I'm unchecking it so it isn't set to partial read only.

[woeful light]

When I try to start stable diffusion for the first time it keeps getting stuck on “collecting torch=2.1.2”

How do I fix that

[ornate elk]

Just wait and let it load

It can take a while

[woeful light]

''Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check''

[ornate elk]

What's your gpu?

If its AMD, checkout my AMD install guide in the Pinned Messages of this channel.
There are two guides. Depending on your gpu you can use zluda or directml

[surreal sentinel]

Seems some of my LORAs disappeared on their own? Some are showing, some aren't. There should be 23 loras.

[onyx solar]

Only compatible LoRAs will show up. Change to a 1.5 model and the others will show.

[surreal sentinel]

oof, that's a big ol' rip, was hoping to try Deliberate with the western dragons one

[wheat helm]

I launched webui and got this error (i think its an error?) and now my "deforum" tab isnt showing. Otherwise seems to be behaving normally, just missing the one tab. Anyone familiar?

better screenshot

[sand kiln]

I got zluda to work perfectly with automatic1111 on my AMD graphics card, which increased my speed by a ton. Anyone know if it is possible to use zluda for Fooocus also?

[ornate elk]

Should be possible when you install it normaly and then change the two DLLs in torch/lib like in auto1111

[ornate elk]

Make sure you updated your extensions

[sand kiln]

I tried that, but it didn't recognize the --use-zluda argument

[wheat helm]

thanks pretty sure thats what fixed it. just checked for updates, clicked apply and restart ui (which didnt immediatley fix it, so i closed completley out and relaunched). Thanks

i cant remember what i named my settings file surely theres a way to see them right 💀

[ornate elk]

Yea thats not how it works

Its not auto1111 directml fork and doesn't have zluda support.
You would need to replace some files and maybe adjust some scripts to make it work.
But I see no point in using Fooocus over auto1111

[ornate elk]

Np, you mean the Settings of deforum? Then idk

[wheat helm]

yeah lol like where it says save settings load settings etc

[burnt gazelle]

Hi, I am using automatic 1111 1.8.0 and trying to use the MeshGraphormer for fixing hands with adetailer. It is working, but sadly always using CPU for fixing hands with the below error message. Does anyone know why?

loading pretrained model /workspace/stable-diffusion-webui/extensions/sd-webui-controlnet/annotator/downloads/hand_refiner/hr16/ControlNet-HandRefiner-pruned/hrnetv2_w64_imagenet_pretrained.pth WARNING: All log messages before absl::InitializeLog() is called are written to STDERR E0000 00:00:1710236947.015089 5359 gl_context.cc:408] INTERNAL: ; RET_CHECK failure (mediapipe/gpu/gl_context_egl.cc:303) successeglMakeCurrent() returned error 0x3008; (entering GL context) I0000 00:00:1710236947.015213 5359 task_runner.cc:85] GPU suport is not available: INTERNAL: ; RET_CHECK failure (mediapipe/gpu/gl_context_egl.cc:77) display != EGL_NO_DISPLAYeglGetDisplay() returned error 0x300c E0000 00:00:1710236947.080494 5359 gl_context.cc:408] INTERNAL: ; RET_CHECK failure (mediapipe/gpu/gl_context_egl.cc:303) successeglMakeCurrent() returned error 0x3008; (entering GL context) E0000 00:00:1710236947.080561 5359 gl_context.cc:408] INTERNAL: ; RET_CHECK failure (mediapipe/gpu/gl_context_egl.cc:303) successeglMakeCurrent() returned error 0x3008; (entering GL context) E0000 00:00:1710236947.080588 5359 gl_context.cc:408] INTERNAL: ; RET_CHECK failure (mediapipe/gpu/gl_context_egl.cc:303) successeglMakeCurrent() returned error 0x3008; (entering GL context) E0000 00:00:1710236947.080612 5359 gl_context.cc:408] INTERNAL: ; RET_CHECK failure (mediapipe/gpu/gl_context_egl.cc:303) successeglMakeCurrent() returned error 0x3008; (entering GL context) E0000 00:00:1710236947.080646 5359 gl_context.cc:408] INTERNAL: ; RET_CHECK failure (mediapipe/gpu/gl_context_egl.cc:303) successeglMakeCurrent() returned error 0x3008; (entering GL context) INFO: Created TensorFlow Lite XNNPACK delegate for CPU.

[modern zodiac]

Hey guys! I want to let SD run creating some random prompts while I am working. So my idea is just to have a file with prompts I created earlier and load them with wildcards. BUT I would also like to use wildcards in those prompts. Is there a way to do that? So basically wildcards inside wildcards?

(Ideally the Dynamic Prompt extension would resolve wildcards all the way down recursively until it can't find any more)

[warm holly]

Create a python script to make you prompts.. load all those prompts in script section

then batch them!

But dynamic prompts extension can do it better

[modern zodiac]

So there is a way with DP do achieve this recursive behavior? Otherwise the script solution sounds like the way to go

[warm holly]

Simplest solution
5 files contain different lines of prompts
make a script to pick one line from each file and combine it with other lines from other files

Now you can have so many lines in each files and then combine
lets say u want to do around architecture

file 1 can have : Building type
file 2 can have : building color
file 3 can have : building background

just examples here

then create a longgg ass file with all perma and combo
use it in script.. batch it!

sleep

Guys I have an issue, I swapped 3070 with a 4090 and img2img is messed up

It was working al right.. .. Some one please help!!

the output images are corrupted of sorts... I tried all models.. Its the same..

[modern zodiac]

have you tried with an old prompt done with your 3070 and the same settings to make sure it's because of the new card?

[warm holly]

yes all that is done

I have a default prompt written in ui-config that opens when I start it..

[modern zodiac]

btw. I tried recursive wildcards and it is working out of the box with DP. I think this is new. All good. And it is amazing

[warm holly]

Solved it, one of my LORAS was corrupt!!.. Just hopping into server solved it Lol!!

[vast crest]

Hi, how did you solve the problem? I have a lora which a few days ago was called:lora:royta v1. Now it's called: Add Details-Style Lora. He changed his name on his own

[high bolt]

to have the same picture between graphic cards I think that you have to choose GPU seed / SDE GPU schedulers, or it will use a seed from the CPU, something like that

[ornate elk]

you would need to set it to CPU

[grim iron]

my sd just randomly stops doesn't give me an error or something it just stops in the middle of genarating and doesn;t do anything

[ornate elk]

whats the error you get?

also whats your GPU and whats inside your webui-user.bat?

[grim iron]

I don;t get any error

it just spops

[grim iron]

I have a 3080

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --api --medvram --no-half-vae
git pull

call webui.bat

[ornate elk]

change --medvram to --medvram-sdxl

also what settings do you use?

[ornate elk]

looks normal, did you used an sdxl or 1.5 based model?

[grim iron]

sdxl (juggernaught v9)

[ornate elk]

okay then change the --medvram to --medvram-sdxl and try again

[cosmic furnace]

@ornate elk Sooo how much stable is zluda now? espec about insta reboot & stuff?)

[grim iron]

still happening

[ornate elk]

totally stable for me

[cosmic furnace]

is there any way besides adrenalin to lower power usage on amd ? =
and a way to make sure its a power problem

[ornate elk]

can you show a screenshot when it hangs?

[cosmic furnace]

-10% power usage doesnt seem like a lot ... downvolting shud also lower overall power usage y ?

[ornate elk]

that doesnt make sense with your settings you showed?

[grim iron]

i'm using plot xyz atm

[ornate elk]

and the settings in there?

make sure you dont compare sdxl and 1.5 models with it

[grim iron]

just promt s/r

[ornate elk]

did you set any batch count or batch size?

or how many stuff do you have in the xyz prompt sr thing

[ornate elk]

yes downvolting would help maybe

[grim iron]

40 in total, I have done with more then 500 with no problem

nut I will try without xyz

[ornate elk]

xyz prompt sr is not intended for queing prompts.
therefore please use this extension:
https://github.com/ArtVentureX/sd-webui-agent-scheduler

[cosmic furnace]

Isnt it weird that sometimes it runs stable for an hour or two... and then after blackout reboot starts to often reboot? =\

[grim iron]

I use it on both x and y axis for mxing

yes ik that I can use dynamic promts for that

but that doesn't give me a legend

[ornate elk]

true, but i think its still the cause of the freezing, sdxl needs a lot of vram, have you tried again without xyz?

[cosmic furnace]

And btw is this deepcache thing worth it? And running with diffusers backend?)

[ornate elk]

i dont use sdnext, as its buggy

[grim iron]

its happening again without xyz

[sudden sleet]

how does one change the sampler in code?

using sdxl

[ornate elk]

i wish you would be clearer what your settings are, it says 20/80, so your using additional settings after the initial steps

my guess now is that your using maybe hires fix or refiner

@cosmic furnace do you get freezes on directml with zluda too? or just in sdnext?

[grim iron]

both are turened off

[ornate elk]

but something was or is active

if you press generate it should say 20/20

not 20/80

[grim iron]

20/80 is becayse the batch is 4

[ornate elk]

oh now i see it

[grim iron]

u can see that in the last ss

[ornate elk]

when it hangs, what does the taskmanager shows?

how much RAM do you have?

[grim iron]

32

10gb vram

[ornate elk]

that should be enough for sdxl

[ornate elk]

make sure you dont have Wallpaper Engine or other heavy programms running in the background

also do you use any sdxl vae ? the vae decode can hang sometimes when using sdxl
try this one:
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main

[grim iron]

I have it installed and the setting on auto

but juggernaught has it baked in

[ornate elk]

yea but you can still try it out, the baked in wont get used then

[cosmic furnace]

havent tried yet directml one 😵‍💫

yeah was sticking to next too much... hoping i can "make it work" like flawlessly )

the main thing i like about it is implemented styles & graphics change =\ . But pretty sure i can do it via promt like my friend does =\

[ornate elk]

yea you can do that with the styles sheet in auto1111 too

[noble belfry]

I use styles so much that if i forget to add my general style in ive found myself pulling and hauling at the generate settings wondering why my images wont pop.

[tender axle]

gm

[karmic crown]

beacuase of those red letters

you do not have all necesary files installed

[tired sedge]

that doesn't make sense tho

[gloomy barn]

Imagem de uma noite estralada de um casa aconchegante e iluminada por dentro

[ornate elk]

What's your python version?

[smoky ravine]

hello, where can i find an up to date tutorial to train a lora (and also how to make a dataset)

[tired sedge]

3.10.6

okay it's got more weird

i ran it again, and it downloaded, but then like this happened

[ornate elk]

It seems like it didn't downloaded the default checkpoint (model) correctly

Check the models/Stable-diffusion folder

[tired sedge]

this?

[ornate elk]

Yes its broken

[tired sedge]

okay, what do?

[ornate elk]

You need to delete it and download an model by yourself like Dreamshaper v8 from here:
https://civitai.com/models/4384/dreamshaper

Put that file into the models/Stable-diffusion folder

Then relaunch the webui-user.bat

[tired sedge]

@ornate elk ty

[shadow kettle]

i know this sounds annoying, but can someone please tell me how to install it on an AMD gpu. Whenever i try i get error messages

[karmic crown]

check the pinned messages in this channel, you have two sets of instructions for AMD there

zluda is little more complicated, but it works faster on new AMD cards

[karmic crown]

btw. are you on windows, linux or mac?

[shadow kettle]

windoes 11

[quick forge]

does anybody know why I'm getting this error after reinstalling Ubuntu, and overwriting my f'ed up install of A1111.

MIOpen(HIP): Error [Compile] 'hiprtcCompileProgram(prog.get(), c_options.size(), c_options.data())' naive_conv.cpp: HIPRTC_ERROR_COMPILATION (6)
MIOpen(HIP): Error [BuildHip] HIPRTC status = HIPRTC_ERROR_COMPILATION (6), source file: naive_conv.cpp
MIOpen(HIP): Warning [BuildHip] /tmp/comgr-4a6d13/input/naive_conv.cpp:39:10: fatal error: 'limits' file not found
#include <limits> // std::numeric_limits
         ^~~~~~~~
1 error generated when compiling for gfx1030.
terminate called after throwing an instance of 'miopen::Exception'
  what():  /long_pathname_so_that_rpms_can_package_the_debug_info/data/driver/MLOpen/src/hipoc/hipoc_program.cpp:304: Code object build failed. Source: naive_conv.cpp
webui.sh: line 292:  9530 Aborted                 (core dumped) "${python_cmd}" -u "${LAUNCH_SCRIPT}" "$@"

[karmic crown]

than you can use those guides

[shadow kettle]

which guides

[karmic crown]

in pinned messages in this discord channel, anyway... here they are

#🤝｜tech-support message <- ZLUDA

#🤝｜tech-support message <- directml

that little pin in top right corner

[fossil dew]

I dont understand all the tecnical lingo

But Ive been using stable for months with no issue but lately my CPUs been crashing, tabs will crash randomly, wheni m using stable

I tried to use it this monring and i got this message

OutOfMemoryError: CUDA out of memory. Tried to allocate 576.00 MiB (GPU 0; 8.00 GiB total capacity; 2.67 GiB already allocated; 2.87 GiB free; 3.22 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Im not literate in this stuff 🙃

I had Chat GPT explain it to me in simpler terms 🙃

but I still dont understand how to fix it

Im using the same model ive always used

Batch size is always 1

aren't there settings in Stable Diffusion (i think im on 111)- that sort out use of memory and whatnot

Check Data Loading: Make sure you're only bringing in the data that's absolutely necessary.

Hopefully someone gets back to me soon- I can't be here long to fix 🥲

Need to use my PC for work lmfao

cant when shit keeps crashing- but it all started with stable so i figured somone here might be able to put it in simple terms for me

is..... this related

literally have 0 idea what im doin

[karmic crown]

https://github.com/viking1304/a1111-setup/releases/tag/0.0.8

[fossil dew]

I dont understand what that is

[karmic crown]

that is for mac users

script that install webui in one click, and now it can isntall forge as well

[fossil dew]

oh huh 👍

[karmic crown]

and it can also fix existing install automatically

[solar sundial]

anyone get stableswarmui running on wsl? just like a generic debian vm

trying to run on a radeon gpu, but its giving me the headache of no cuda compatible gpu even though i selected amd on the install

[frosty pawn]

How to solve a problem: NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check?

[karmic crown]

friendly advice, do not do that

[solar sundial]

yeah its not working anyway

i cant get the low vram option to actually work on my rtx 2060, only 6GB ram on that

[karmic crown]

in oder to get better results with AMD you need real Linux, not emulation in Windows

use Zluda or DirectML in windows, or make a dualboot insall

[solar sundial]

dang. i didnt think i'd need to go that far into it

changing gears then, where exactly do you go to change to the low vram option in SD webui?
I keep getting the CUDA out of memory message. And this only happens with the SDXL model loaded up

this is on a real linux machine

[noble belfry]

whats the formula syntax for this box. say for 1-10 but increments of 0.5? early on i used syntax here but havent in ages and ive forgotten. weirdly hard to google or find in docs

[karmic crown]

what is your GPU?

[solar sundial]

rtx 2060 ko

[noble belfry]

(ignore that its steps, i just mean for x values)

[solar sundial]

6GB flavor

[karmic crown]

find webui-user.sh and open that in edit

[solar sundial]

there

[karmic crown]

find this part

[solar sundial]

k

[noble belfry]

have you done this already? works wonders on smaller gpus that get that cuda out of memory ... on windows 🙂

[karmic crown]

and try this
export COMMANDLINE_ARGS="--opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling --lowvram"

[solar sundial]

this is linux @noble belfry not windows

[karmic crown]

do not forget to remove # at the beggining

[noble belfry]

oh ffs, ignore me

[karmic crown]

or just add this below that line

and UNLIKE windows, start webui.sh

DO NOT run webui-user.sh

it is oposite than on windows 🙂

[solar sundial]

sorry what? unline windows?

[karmic crown]

sorry

typo

on windows, you edit webui-user.bat and run that one

on linux and mac, you edit webui-user.sh, but run webui.sh instead

thats what i meant

[solar sundial]

yeah that part I knew. been messing with it for awhile. it was the location of the configurations I wasnt entirely clear on

thats perfect. thanks @karmic crown

[karmic crown]

my linux install is in var/www/stable-diffusion-webui

and i run it as www-data user 🤣

and that is CPU only install on VPS

[solar sundial]

What’s VPS?

[karmic crown]

Virtual private server

A virtual private server is a virtual machine sold as a service by an Internet hosting service

[solar sundial]

I was just realizing that haha

I gave up on acronyms lol

[karmic crown]

10 CPU and 60 GB on SSD, but still slow as hell

[solar sundial]

Really??

[ornate elk]

What settings did you used?

[karmic crown]

i just did that for test

i am on mac

[noble belfry]

just to answer my own question should it come up again i found it. its 1-10 (+0.5) where 1 is start, 10 is end and (+0.5) is positive increments of 0.5

[solar sundial]

I’m running mine on proxmox with gpu pass through. Works great

best use for spare hardware you don’t want to put away

[solar sundial]

What’s this helping with?

[solar sundial]

Look at the help I received from Viking. Not sure how to direct you straight to it from my phone at the moment

[noble belfry]

a seperate question i asked earlier and found the solution to myself. just giving closure.

[proper pike]

Does anyone know how to fix this?

[solar sundial]

isn't red text usually indictive of a missing dependacy or package?

[balmy shore]

hey using SD Forge.. noticed some new samplers.. 'euler sgmuniform' and a dmp++2m variant.. are they okay to use with 1.5 models ? i think ive tried with fine results but just figured id ask if anyone has any more indepth info

[fossil dew]

i dont understand your question- Im not sure what settings your asking about ect

I figured out #1 my drivers needed updated

I cleaned up space in the C:// drive, I updated my drivers.

I even found that theres a ton of stuff conflicting on my PC competing to use the same ports; idk if that effects it much, but yet another issue to be fixed-

So I appreciate the help ect- I've gotten it to clean up a bit ❤️

Updating my Drivers helped the MOST

GPU driver haddent been updated since june 2023 🤡

[proper pike]

Does anyone know how to fix this?

[nova ermine]

I'm tryng to AI generate a character but eyes aren't nice at all. Is there a way to improve eyes? (Maybe a LoRA or something)

[fossil dew]

another thing that fixed my issue:

clearing all of the data in my browser

I use duck duck go- theres a little 🔥 thing that roasts all your data and clears it

did that now my GPU loves me again 🥺 🥲

so two things that helped:
(incase anyone has the same issue I had)

• Clearing my internet browser data
• Updating my GPU drivers

[noble belfry]

yeah you dont realise how greedy browsers are (especially chrome who i think is the worst offender) until you try to do something with them open

[steep belfry]

Good evening, my Stable Diffusion is having a strange error, before it could take images in less than 15 seconds, but for some reason, the image stops at 95% and there is a small delay. alkgeum know what this is?
like, will I have to reinstall Stable?

[ornate elk]

Its the VAE process

[steep belfry]

So just swap, right?

[ornate elk]

if your using sdxl, you should use the sdxl-fp16-vae

[soft fiber]

i had no idea we could insert formulas into the xyz script, thanks for this

[ornate elk]

@noble belfry @soft fiber checkout this:
there is a good amount of tipps and tricks (also the xyz syntax)
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features

[noble belfry]

aah! that's where its hiding. in plain sight!

confession. Despite generally using a lot of features, i didnt know what BREAK was in other people's prompts. i actually thought it was some popular anime or something!

[golden arch]

HI

I have a problem, just one week trying to install SD but always have problem with Python.
exit code: 9009 or exit code 3

stderr:
"C:\Users\Jos├®" is not recognized as an internal or external command, operable program or batch file.

Launch unsuccessful. Exiting.
Press any key to continue . . .

I Added Python to PATH
I instaled version 3.12.2 and 3.11.8

I'm desperate, please help me.

[ornate elk]

There is actualy a really good Anime Movie called BLAME!

[ornate elk]

Hey, What's your PC username? (It shouldn't contain any special characters, also no . or # etc)

And also uninstall python 3.12

Its not supported

After uninstalling python 3.12
Delete the venv folder and relaunch

Then send a screenshot in here if you get the error again

[golden arch]

Has a special character...😕
Uninstall, done. Which version I have to use?

[ornate elk]

3.10.11

Special character can break many things in Windows that require paths

You need to edit your Username

[golden arch]

Ok

[golden arch]

Do I add Python to PATH?

[ornate elk]

Yes

When installing python check "add python to path"

If you dont see that option. Run the python installer again. Click modify, next, and then check "add python to system variables"

[golden arch]

Done

[golden arch]

Done

[ornate elk]

Then delete the venv folder

And relaunch the webui-user.bat

Depending on your GPU vram you need some performance settings.
You find them in my guide in the pinned message of this channel.

[balmy shore]

yes many ways

theres a ti called pretty eyes, but describing eyes in prmpt too

[golden arch]

Done.

[nova ermine]

Is that a LoRA?

[golden arch]

Done, same result. 9009.
I think the username editing didn't work.
I'm going to restart.

[round edge]

I'm trying to train a model in dreamtime but my images come out like this in the preview this is at 30% is this normal I'm guessing not

[solar sundial]

super random, but I'm just noticing that my SD webui is informing me that xformers is "N/A".... any suggestions here?

[fossil dew]

Can I dm you a question I have; I dont wanna post the SS here & you've been helpful to me before~

I'm just cleaning up program folders on my PC and wondering if you'd be able to sort something out for me ❤️

[drowsy creek]

switching to linux soon, is there a way to run GPU (faster then a minute per image) stable diffusion on an AMD RX 5600 XT?

(imma be using arch linux if that matters since most stuff I saw was for ubuntu, but then again I'm a linux noob so idk, all I know is arch power user go brr)

[solid night]

I installed SD Forge. while looking at the log I see this "Device: cuda:0 NVIDIA GeForce RTX 3070 Laptop GPU : native
Hint: your device supports --pin-shared-memory for potential speed improvements.
Hint: your device supports --cuda-malloc for potential speed improvements.
Hint: your device supports --cuda-stream for potential speed improvements.
VAE dtype: torch.bfloat16
CUDA Stream Activated: False" does this mean my CUDA processors are not currently being used for SD?

[torn tide]

Hi all,
I am using this API from stable diffusion: "https://modelslab.com/api/v3/text2img" to generate a unique avatars for my application.
This is giving me URLs of images generated by AI but instead of URL can i just get a buffer data of that image as I want to store this image in my private storage

If anyone help me out on this, It would be a great help.

Thank you.

[ornate elk]

Sure I can take a look

[ornate elk]

Rocm is your way to go

On Linux its the fastest you get

[noble belfry]

what do you get if you set base64 to true?

havent used the api personally but one would assume that would send the image as base64 data string

[noble belfry]

https://docs.modelslab.com/realtime-stable-diffusion/text2img/#response also, theres a little chatbot you can ask questions to about the api 🙂 (it seems to hallucinate a little but still)

[dreamy carbon]

Hello my friends. I was attempting to make some AI art of Buddhist ink style painting. During my journey to find out how to install it, I ran into a couple of problems. Everytime I change the user.bat to install xformers. No matter what I do when I start the webui.bat it would always says “no module “xformers” like 3 times and skips it. I am trying to do the 1111 and I just can’t get xformers to install. Does anyone know a solution? Please and thank you

[drowsy creek]

(help the person above first ^^^)
btw quick question, are there any good CLI terminal interfaces for SD on linux? ik it's impractical compared to a gui but it's mainly just for fun

[dreamy carbon]

I am extremely new to these things so please have some patient with me

[ornate elk]

You have to add --xformers to the webui-user.bat in the line COMMANDLINE_ARGS=

And always start SD with the webui-user.bat and dont use the webui.bat

@solar sundial

[dreamy carbon]

What does SD stand for

[dreamy carbon]

Also I use —xformers —reinstall-something can’t remember

So just xformers?

Tyvm I’ll try it in the morning

[lunar kite]

NotImplementedError Traceback (most recent call last)
<ipython-input-7-953c5d278381> in <cell line: 155>()
156 render_animation(root, anim_args, args, cond, uncond)
157 elif anim_args.animation_mode == 'Video Input':
--> 158 render_input_video(root, anim_args, args, cond, uncond)
159 elif anim_args.animation_mode == 'Interpolation':
160 render_interpolation(root, anim_args, args, cond, uncond)

16 frames
/usr/local/lib/python3.10/dist-packages/xformers/ops/fmha/dispatch.py in _run_priority_list(name, priority_list, inp)
61 for op, not_supported in zip(priority_list, not_supported_reasons):
62 msg += "\n" + _format_not_supported_reasons(op, not_supported)
---> 63 raise NotImplementedError(msg)
64
65

NotImplementedError: No operator found for memory_efficient_attention_forward with inputs:
query : shape=(1, 14080, 1, 512) (torch.float16)
key : shape=(1, 14080, 1, 512) (torch.float16)
value : shape=(1, 14080, 1, 512) (torch.float16)
attn_bias : <class 'NoneType'>
p : 0.0
decoderF is not supported because:
max(query.shape[-1] != value.shape[-1]) > 128
xFormers wasn't build with CUDA support
attn_bias type is <class 'NoneType'>
operator wasn't built - see python -m xformers.info for more info
flshattF@0.0.0 is not supported because:
max(query.shape[-1] != value.shape[-1]) > 256
xFormers wasn't build with CUDA support
requires device with capability > (8, 0) but your GPU has capability (7, 5) (too old)
operator wasn't built - see python -m xformers.info for more info

tritonflashattF is not supported because:
max(query.shape[-1] != value.shape[-1]) > 128
xFormers wasn't build with CUDA support
requires device with capability > (8, 0) but your GPU has capability (7, 5) (too old)
operator wasn't built - see python -m xformers.info for more info
triton is not available
requires GPU with sm80 minimum compute capacity, e.g., A100/H100/L4
Only work on pre-MLIR triton for now
cutlassF is not supported because:
xFormers wasn't build with CUDA support
operator wasn't built - see python -m xformers.info for more info
smallkF is not supported because:
max(query.shape[-1] != value.shape[-1]) > 32
xFormers wasn't build with CUDA support
dtype=torch.float16 (supported: {torch.float32})
operator wasn't built - see python -m xformers.info for more info
unsupported embed per head: 512

Hi! Has there been an update to SD Deforum? Since yesterday I'm getting these message saying that it cannot start because of CUDA issue

Something to do with the xformers?

[ornate elk]

Stable Diffusion

[ornate elk]

Remove --reinstall stuff

What's your gpu?

[ornate elk]

Hey, are you on Windows? What's your gpu and what's inside your webui-user.bat?

[lunar kite]

I should remove the model from my google drive?

I’m on a Mac

[lunar kite]

Not sure about webui-user is that just for windows?

Just using the google collab

[ornate elk]

Okay

Can't help with google collab

Does your Mac has an M chip?

[lunar kite]

Is that the latest chip?

[ornate elk]

M3 is the latest

You can run stable-diffusion on Mac locally if it has an M1-M3 chip

Deforum should work on webui version 1.8.0

[lunar kite]

I don’t have ~ have to use google collab

Issue just started yesterday

[ornate elk]

Okay do you have a file you launch for the collab?

[lunar kite]

It accesses a model that’s saved in my google drive ~ could that be it?

Protogen_V2.2.ckpt

[ornate elk]

I dont think its because of the model. Do you get that error only when using deforum?

[lunar kite]

Yeah I only use collab and stable with Deforum

[ornate elk]

Did you activated FILM maybe in deforum?

Thats broken on MAC rn

[lunar kite]

What is that?

[ornate elk]

A checkbox in deforum for interpolation

[lunar kite]

Here’s what I have for interpolation?

[ornate elk]

Okay, and when you set the 20 to 0 ?

[lunar kite]

And check the box as well?

[ornate elk]

No

What's strange is the xformers error in your log

[karmic crown]

wait

[karmic crown]

it is broken ONLY on MAC

it works perfectly fine on other platforms

[ornate elk]

Okay corrected it

[karmic crown]

@lunar kite what mac do you have and do you use local version or colab?

if you are on colab, your computer is not important, it can be anything

but if you are running local version on mac, than it is isportnat what mac it is

and how it is set up

[ornate elk]

He's on collab

[lunar kite]

Running on collab

[karmic crown]

please show me what you have there

btw, better go to https://discord.com/channels/1011314943380959313/1013994958346256454 for deforum issues on colab

[lunar kite]

Is this for me? If so where do I find this?

[karmic crown]

when you run a1111 on colab, you should have same interface as local webui, right?

i have never used webui on colab, i only used local versions on windows, mac and linux

[lunar kite]

Oh I see ~ it’s only webui

Looks different

[karmic crown]

you should have extensions tab there

where you can check which versions od extensions you have

i want to see which version od deforum you are using

and to show me exact version of a1111 that you are using

[lunar kite]

Not sure it’s the same ~ version “Deforum Stable Diffusion” (v0.7.1)

[karmic crown]

thats why i advised to go to deforum server, they know exactly how those things looks there

[lunar kite]

This one? https://discord.com/channels/1011314943380959313/1013994958346256454

[ornate elk]

Its not the webui interface, most collabs are scripts that have a minimal ui

[karmic crown]

yes that is for deforum on colab

[lunar kite]

Cool posted the issue there 🙏

[karmic crown]

ok, looks like some other guys had the same issue and found the solution there
xformers==0.0.24 <- looks like xformers update broke the whole thing

i posting this here in case someone else have the same issue later

[viral fox]

morning all. Quick question. Can i have multiple UI's on SDXL. I.e. I am currently running ComfyUI, but could i Also install say Automatic111 and InvokeAI. Im looking to compare and contrast so I can get a balanced view of them

Basi caly can they co-exist

[noble belfry]

yes 🙂

you just need to point the other ones to your existing models folder so you dont have to copy them all 🙂

[karmic crown]

Laurence just confirmed on colab-notebook, that this solution works

[karmic crown]

it is probably the best to move all models to a1111, and setup comfy to use a1111 models

it is easier

actually, all models, loras...

@viral fox check this out #🤝｜tech-support message

but since you alredy have comfy, move your models to a1111
yes, you can do the oposite, but i think this is more convinent than to set path for every single thing in a1111

[noble belfry]

fair enough. point is, yes, they can coexist and dont interfere with each other :>

[karmic crown]

100% agree, i just want to make it as easy as possible for him, instead of maintain 241324 parameters in a1111 for every single thing 🙈

[noble belfry]

im still pleased that forge deals with it in one single line now 😄

[karmic crown]

i like that one too 🙂

i need go get back to work, talk to you guys later

[noble belfry]

forge is just the hold my beer edition of a1111 and i think people should move over to it

even if you turn half the optimizations off that slow some things down, its still just a bit more polished.

[fair oxide]

Which ai tool is the best to create videos?

[fossil dew]

I've sorted it out lol I appreciate it. Whole issue was a mess but all's well now. Thanks anyways 😊🍒

[torn tide]

I have set base64 to true but still couldn't get base64 string

[smoky ravine]

hey, anyone got a good upscale comfyui workflow? basically something that just does what you can do in a1111 img2img upscale but on comfy, without generating the image first

[noble belfry]

what does a response look like then? could you show me one of the responses you get from the api?

[torn tide]

yes sure

this is what response look like

[noble belfry]

it doesnt recognise "yes" set it to true (no quotes

[torn tide]

Actually I have set it true earlier but that wasn't working as well so i tried "yes" instead

[torn tide]

see here

[solar sundial]

Thank you

[hot spindle]

Hi, I know its probably a bit outside the scope of this but can anyone point me to why I'm failing to install torchmcubes (torch is supposedly installed):

[dapper finch]

Hi,I have a problem with the last part for automatic 1111 installation.Can someone help me with an advise?

[hot spindle]

@dapper finch just do what it sayss there: add --skip-torch-cuda-test

after the webui.sh

[ornate elk]

Hey, what's your GPU?

[ornate elk]

We dont know if the laptop has an dedicated GPU. So that won't help as first advice. Also its windows so adding anything to webui.sh won't do anything

[dapper finch]

If I try to add here,don’t work

[ornate elk]

Can you check the taskmanager under the performance tab ans then show the GPU part

You definitly installed the wrong webui version. As AMD needs a different one

[ornate elk]

@dapper finch
Thats not the task manager.
But in any case you would have to follow my AMD install guide from here:
#🤝｜tech-support message

Worth mentioning is that its pretty limited on that Laptop because of the 8gb RAM (and the integrated GPU)

[dapper finch]

Thank you so much

[willow parrot]

my friend needs help with running SDXL, he's getting 18s/it on a 3070ti on a 832x1216 image with these settings

he is running xformers, has the latest torch version

can't figure out what's wrong with his installation for the life of me

[vocal burrow]

what's his full log for startup + 1 generation ?

what's in their webui-user.bat ?

[willow parrot]

this is what he sent me earlier after i told him to reinstall venv while he was updating torch & stuff

[willow parrot]

just --xformers and --odt-sdp-attention

[vocal burrow]

makes no sense to have both, they're not compatible

[willow parrot]

that's not what's causing it tho

because i have the same on a 3060 and i get like 1.5it/s with those settings

[vocal burrow]

oof

[willow parrot]

yeah that counts the download of venv folder too

[vocal burrow]

still can you get one FULL log ?

[ornate elk]

3070 ti has 8gb vram and therefore needs --medvram-sdxl

[willow parrot]

he can't do it right now but yeah later

[willow parrot]

would not having that really slow him down this much?

it's just a 832x1216 image

[ornate elk]

Yes its slow without

[willow parrot]

i'll tell him to do that then

[ornate elk]

Its like 10 minutes to 1:30 on a GTX1080

[vocal burrow]

without it, it wil probably fall back to their system's ram when the vram is full

=> slow down

[willow parrot]

jesus

i just have --xformers on my 3060 12gb, should i be using anything else for a speed-up?

[ornate elk]

And he need to remove --opt-sdp-attention

[willow parrot]

that doesn't sacrifice quality

[willow parrot]

yeah i did

[ornate elk]

Your fine, but if sdxl takes long for you too, then you can add --medvram-sdxl

[willow parrot]

it's good enough for me for now

just noticed im running python 3.9 instead of 3.10 is that slowing me down at all?

[vocal burrow]

upgrade it to 3.10.11 and delete your venv folder

[ornate elk]

No but that will result in problems

[willow parrot]

alright

ty

does adding --medvram--sdxl sacrifice anything noticeable?

[karmic crown]

that is just for us on macs

[willow parrot]

quality wise

[vocal burrow]

speed, but if you have under 8gb you need it soooooo not really

[ornate elk]

Nope there is no command that results in worse quality

[willow parrot]

ah i see

[opal bluff]

Before i ask is there anyone running linux?

[opal bluff]

But i have 3 24gb GPUs. That should make all the images and make them amazing with no hand deformations.

Well anyhow, I will leave the following question here incase someone can help. I am running a linux mint install on a server with three GPUs.
Question 1: Is there a way to start up stable diffusions webui.sh in the background on startup? I can start it up via terminal but as soon as i close terminal it stops my session. I want to be able to start the server and let it run then remotely do an images.

Question 2: Is there an argument or variable i can change somewhere to spread the load of batches across multiple GPUs?

[ornate elk]

Launching the webui in terminal is the programm, it needs to be opened to run stable diffusion and generate images.
You cant spread the load of the batches, but you can run a seperate webui instance on each of the GPUs.
checkout stableswarm-ui, it can help whith using mutliple gpus

[desert grove]

What does the paint over tool thing usually used for inpainting, do for Canny in the Controlnet?

Does it do anything? I can't seem to find a difference in the output.

[opal bluff]

Okay Thank you. So there is no chance of the program launching on startup without it being visible? How about starting up locally on the server on startup with no other human intervention?

[desert grove]

What I'd like to be able to do is use canny on specific parts and have it generate regularly for the rest of the image. For example, I usually use a mannequin tool to set up characters before I draw them, but when I put them in and use Canny or Depth, the outputs copy the bald head of the doll. OpenPose also always fails to copy the poses properly.

[soft fiber]

you could probably write a script to do that on startup but this requires a little bit of knowledge

[opal bluff]

I can do a script. Prviously i was trying to use ubuntu server. Apparently it hates me but i am used to linux mint so if a script will work i can probably get it done. Thank you all.

[karmic crown]

1. should be easy - ./webui.sh & should put it in the background, but i am not sure if will it keep working after you close the terminal. it depends if there is something that explicitly requires the terminal

2. sorry but no, 1 GPU and that's it

[opal bluff]

yeah the ./webui.sh & does not seem to work like that. At least remotely via SSH

too bad on the multiple GPUs. May have wasted some cash.

[karmic crown]

it is in background as you can see

let me try to close the terminal

it can be run in bacground, but it will be closed as soon as you close the terminal

someone would probably need to make a smiple daemon script

[karmic crown]

check out this
https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/9569

my Linux server is too slow (CPU only) for me to bother with that, sorry

[sand dawn]

I have an API question regarding A1111 / Forge - API.
I thought I could use the endpoint sdapi/v1/unload-checkpoint as a means to free up resources in my discord bot script, which controls both textgenwebui and A1111/Forge.

If I go to http://127.0.0.1:7860/docs while running Forge and click Test > Execute, nothing happens in the cmd window.
If I go to http://127.0.0.1:7860/docs while running A1111 and click Test > Execute, it seems to just reload the current model.

I feel like I'm missing something here? Or is this endpoint just bugged?

I get response code 200 in my bot script but nothing actually happens

[desert grove]

Anyone know whether or not the painting tool in Contolnet affects Canny or anything, or is it just that for certain Controlnets that it does affect?

[karmic crown]

https://civitai.com/models/119012/bluepencil-xl
what's going on here?

ok maintenance

i got 404 a second ago

[vocal burrow]

civitai is under maintenance

[karmic crown]

I was confused because I was getting 404 for bluepencil-xl, but when I posted the link, I saw "We're adjusting a few things, be back in a few minutes..."

[vocal burrow]

yeah civitai is probably down 5% of the time....

at least it feels like so :p

[karmic crown]

i was afraid that bluepencil was taken down

https://www.liblib.art/modelinfo/689ac4eb9d3549a4bfdde984735f6687

🙂

i am comparing the speed between a1111 and forge on my new mac
i finally managed to get almost the same speed with 1.5 model
i now want to check xl and to see which one works faster

[vocal burrow]

On my hardware, memory usage difference is much more noticeable than speed

[karmic crown]

i had around 2it/s on a1111 and 1.5it/s on forge and comfy
now it has 2it/s on both a1111 and forge with some crazy forge params
--skip-torch-cuda-test --attention-pytorch --all-in-fp16 --vae-in-fp16 --use-cpu interrogate

i hope xl will work faster with this on forge

download speed from liblib here is slow

and i am lazy to find my hdd with backup 😉

[vocal burrow]

what about fp8 ? I wonder how apple silicon handles it.

[karmic crown]

--all-in-fp32
--all-in-fp16
--unet-in-bf16
--unet-in-fp16
--unet-in-fp8-e4m3fn
--unet-in-fp8-e5m2
--vae-in-fp16
--vae-in-fp32
--vae-in-bf16
--clip-in-fp8-e4m3fn
--clip-in-fp8-e5m2
--clip-in-fp16
--clip-in-fp32

those are options on forge

you cant use fp8 for all

[vocal burrow]

yup

but still

[karmic crown]

ok, this makes more sense

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3745620006, Size: 1024x1024, Model hash: f4b9e61cc0, Model: bluePencilXL_v4.0.1, Version: v1.8.0

Time taken: 1 min. 33.4 sec.

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 294492588, Size: 1024x1024, Model hash: f4b9e61cc0, Model: bluePencilXL_v4.0.1, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7

Time taken: 1 min. 16.7 sec.

i will experiment further to find if some params can make it a little faster
but this is way too slow compared to MCurtos M2 Ultra (but he has 192GB of unified RAM, which is the rough equivalent of having GPU with 128GB) 🙂

[sick ginkgo]

anyone know why the "choose file to upload" tab is gone?
https://i.imgur.com/ZNTuS8b.png

[karmic crown]

what happen if you click here?

[solemn atlas]

hey guys, thats Wow whats that? A page to download loras?

Hey guys, a question. What is a "wildcard" and in what folder do i put it in?

[solemn atlas]

It's not possible to download it i guess?

[karmic crown]

it will be, just a little later

maintenance

it is back online

[chrome rampart]

evening all. question what open pose model/file do i need if i want to use with sdxl?

[karmic crown]

https://huggingface.co/lllyasviel/sd_control_collection/tree/main

those with xl in name

[chrome rampart]

thks viking

[karmic crown]

ok i can live with this

[chrome rampart]

sorry to ask viking but see several. For me I use Juggernaut and would like to use poses kohya_controllllite_xl_openpose_anime.safetensors
kohya_controllllite_xl_openpose_anime_v2.safetensors
t2i-adapter_diffusers_xl_openpose.safetensors
t2i-adapter_xl_openpose.safetensors
thibaud_xl_openpose.safetensors
thibaud_xl_openpose_256lora.safetensors

or do i need them all?

[karmic crown]

thibaud_xl_openpose.safetensors
thibaud_xl_openpose_256lora.safetensors

dowload the smaller one

t2i-adapter_xl_openpose.safetensors
thibaud_xl_openpose.safetensors

those are different implementation of same thing

i have a new laptop, so i cant check what i had at this moment

[chrome rampart]

thks. cant get her to pose lol will keep trying. Have fun with new laptop

[karmic crown]

this might help

[little echo]

can anyone give me tips how to use zero123 in forge?
does it need configuration during installation?

[karmic crown]

as far as i can tell it requires a model

[little echo]

i have the checkpoints in ..forge_ui/models/z123 though

tested 3 different and they won't show in drop down

using a runpod template and haven't installed it myself b4 so i don't know if requires options on install

should it work out of the box ?