🤝｜tech-support

[boreal bison]

done

ok now seems like it's using my 3090 yey

trying 2048x2048 see if it handles it

aaaand it does

Thank you very much @ornate elk

Now seems like I have to bother someone else for additional modules haha

[ornate elk]

1.5 models are trained on 512x512 and sdxl models are trained on 1024x0124 resolution, so going straight for 2048 will result in deformation

[boreal bison]

2048

[ornate elk]

for community made models, checkout Civitai.com
its the largest model database

[boreal bison]

not to my liking, I guess I have to dig into youtube for some tutorials now

[boreal bison]

Thanks!

I need to be able to train faces and do videos, can I do that?

[ornate elk]

one thing you should do:
go into Settings -> User Interface -> Quicksettings, there add sd_vae then hit apply and reload ui

[ornate elk]

yes training with 24gb vram is perfect and very fast

[boreal bison]

where is that settings?

[ornate elk]

also video stuff with AnimatedDiff or Deforum extension will work great

[boreal bison]

I need those! and control net and training modules

[ornate elk]

yes

[boreal bison]

I have no idea yet how these all work, I just heard stuff nothing else

[ornate elk]

for training you should install the khoya_ss tool, its a seperate tool

[ornate elk]

at the top, in the tabs

in the webui in browser

[boreal bison]

oh browser settings ok

[ornate elk]

not browser settings

[boreal bison]

I'm using brave browser

[ornate elk]

in this tab

on the right side

[boreal bison]

on snap

guess I'm blind

lol

first?

[ornate elk]

the third

sd_vae

[boreal bison]

done

what does it do?

[ornate elk]

then hit apply

[boreal bison]

did

[ornate elk]

it does give you the second dropdown in my screenshot

[boreal bison]

reloaded the page, still 1 drop down

[ornate elk]

you need to click reload ui

not reload the page

in the settings

[boreal bison]

done!

works

[ornate elk]

vae dropdown is needed for the future when you want to change vae files, needed for color/vibrance of images

[boreal bison]

ok

anything here that matches the quality of leonardo ai's vision XL?

[ornate elk]

these are samplers, not models

[boreal bison]

hmmm

[dreamy garnet]

also 2048 x 2048 is out of scope for most models, maybe sdxl based one can be pushed that far but I'd go lower and then upscale, hiresfix, tile, etc...

sdxl based models are trained at 1024 x 1024

[boreal bison]

it's fine I don't need 2048 by 2048, I just wanted to test

[dreamy garnet]

you can use other aspect ratios but they have to maintain close those numbers

16:9 is 1344 x 768

[boreal bison]

does automatic1111 have any built in models?

where are those? Anything good in there?

[dreamy garnet]

no you have to DL them

as @ornate elk said, civit.ai

[astral comet]

so my SD tells me to ''use the --no-half commandline argument'', how do I do that?

[boreal bison]

and what about samplers? What are those for?

[ornate elk]

yes your using the 1.5 model rn

[boreal bison]

I'm here

Juggernaut XL looks good

Are these free?

[ornate elk]

thats an sdxl model (1024x1024 resolution)

[ornate elk]

yes

[boreal bison]

holy moly

downloading juggernaut XL now

6.62 GB

[ornate elk]

to install models:
download the file
and if its larger than 1.98gb put it into the models/stable-diffusion folder

[boreal bison]

oh that's it? Than reload UI?

[ornate elk]

then click on the reload button near the first dropdown

there you can select it

[boreal bison]

got ittt

[dreamy garnet]

for info about samplers/schedulers and so on, best bet is reddit Stable Diffusion, at the tutorials tab you have stuff like this> https://www.reddit.com/r/StableDiffusion/comments/xmwcrx/a_comparison_between_8_samplers_for_5_different/

[boreal bison]

and this is all local and private?

[ornate elk]

completely

[boreal bison]

awesome! thanks

[ornate elk]

uncensored too xD

[boreal bison]

great haha

[boreal bison]

ahaha nice nice

[dreamy garnet]

[pearl musk]

is xformers working for Maxwell cards?

[viscid mist]

i wonder if its a bad idea to store my sd models on a nas

[ornate elk]

you can try it

it could work

if it doesnt work it will tell you in cmd

[pearl musk]

Thanks!

[ornate elk]

depends on the time it takes to load a model. but it doesnt change/affect the image generation speed

[boreal bison]

I got juggernaut installed, what sampler should I use with it for best results?

[ornate elk]

for sdxl models try a dpm++ sampler like dpm++sde or dpm++ 2m sde karras

[boreal bison]

ok

Can I generate a few images side by side at once? 😄

[little echo]

if a model supported it, could you use tags that changed the orientation your character/model is facing--"facing left" "from behind" etc
then also if you just alternated those, but kept the rest of prompt and used the same seed--theoretically would the model keep the same pose/shape in each batch?

[dreamy garnet]

dpm 2m / karras works really great

also ddim sched.

[boreal bison]

guys if the model type says Lora, does it go into a Lora folder in the models?

[little echo]

yes but its typically used differently

[boreal bison]

can find it here

[little echo]

with it you can add the tags it uses to make adjustments to a base SF model

[boreal bison]

oh ok

so it works with main model like jaggernault

[little echo]

but i think u have to enable loras in settings then go to the tab to add the lora tag to prompt

[ornate elk]

yes, with batch count (one image after the next one) and batch size (multiple at the same time)

[boreal bison]

awesome thanks

[ornate elk]

loras are also based on either 1.5 or sdxl

they only work with models based on the same

so 1.5 loras work with 1.5 models, but not with sdxl etc

[little echo]

if i use the same seed-but different poses-will the model still draw with slight different proportions i guess?
arms legs etc

[boreal bison]

got it

[little echo]

just tested--outfits etc change
i guess there is no way to make the figure look the same in all poses then

[ornate elk]

for that you need controlnet IP-Adapter

[little echo]

ok i will check it out thanks!

do you think it's possible to make it close enough to mimic a 360 panorama of IRL subject

[ornate elk]

no clue, but maybe

[little echo]

i'm looking to test in photogrammatry app for 3d model output

it would probably be a nuisance but it might work who knows

[boreal bison]

how do we add weights in automatic1111?

[ornate elk]

(word) or ((word)) or (word:1.2)

[boreal bison]

thanks

is there a weight limit?

[ornate elk]

nope, but going over :1.4 is like forcing something and going beyond :1.6 can do weird stuff to the image outputs

[boreal bison]

ok so 1.2 is good

1 is default?

[ornate elk]

yea i mostly do :1.3

[boreal bison]

got it!

[vale moat]

Hi, i am trying to install stable diffusion locally, but i keep getting error code "ERROR: Could not find a version that satisfies the requirement torch==2.0.1 (from versions: none)
ERROR: No matching distribution found for torch==2.0.1" anyone got any idea how to fix this?

[boreal bison]

@ornate elk do you know how I install those bone like things with which I can control the poses of the characters?

[lost crescent]

It never used to

[copper flame]

sorry if this is a dumb question but i was getting confused. I installed Automatic 1111

and i downlaoded this ckpt

https://huggingface.co/runwayml/stable-diffusion-v1-5/tree/main

and its working fine

but how would i then use this pokemon one?

https://huggingface.co/lambdalabs/sd-pokemon-diffusers/tree/main

i dont see a ckpt file 😅

so i think im missing something

actually i think this video is what i need

https://www.youtube.com/watch?v=-6CA18MS0pY&ab_channel=SECourses

[unique glacier]

i suppose you could try to reinstall torch. in the webui-user file. try to input the following command : --reinstall-torch

[dreamy garnet]

It’s a controlnet called openpose, depending on what model you’re using (1.5, 2.X, SDXL), they’re completely different https://huggingface.co/lllyasviel/ControlNet/tree/main/models

Each controlnet does different things, you can read about it in the modelcard.

[boreal bison]

This is getting too confusing too fast 😫

[ornate elk]

Hey, pls follow my install guide in the Pinned Messages of this Channel.
You have the wrong Python Version

[dreamy garnet]

Step by step, yes there are many things, for sdxl I’d suggest stability’s controlnets, also a user named Thibaud Zamora has a great controlnet repo.

[ornate elk]

Yes thats done with Controlnet

[boreal bison]

I’m watching tutorials on youtube, but so many random bad ones.. getting the ideas though

I’m guessing with comfy ui node based interface you can do better video to video. Or am I wrong?

[dreamy garnet]

You coming from vfx/3d world perhaps will find this useful, you build the openposes in blender, even animations and then bring those to the controlnets https://toyxyz.gumroad.com/l/ciojz

[ornate elk]

I think he needs it simpler xD

Not more advanced

[boreal bison]

I also really want to train on real faces and use them in my generated images. That’s the most important thing right now

[boreal bison]

I need the basics, understanding what is what. Advanced will follow 😊

[ornate elk]

There are many videos out there. Dont watch old tutorials as they get outdated fast.

[boreal bison]

give me couple good videos with real results please

[daring quest]

#🤝

[ornate elk]

@boreal bison do you already installed the controlnet extension?

[boreal bison]

this one seems fine

[boreal bison]

yeah, but I’m not in my office now

watching videos on my tv now 😳

[ornate elk]

Thats okay, but its for comfyui

[boreal bison]

yeah

[dreamy garnet]

Yeah olivio is a great educator dwpose is also a controlnet preprocessor

[boreal bison]

things like that “preprocessor” I don’t understand yet. Haha

[dreamy garnet]

And for your specific question, there’s a user named Matteo, the developer behind Ip adapters, he just did a complete tutorial specific on FaceID and using faces, I’m not on the computer right now can link it later

[boreal bison]

please dm me when you are

I’ll check in the morning

[ornate elk]

Feel free to link it to me too

I already know how to use them but maybe he has some tricks

What I dont link is the additional need of a lora

Makes switching and testing slower

[boreal bison]

Guys do you prefer comfyuo or automatic1111? What do you use most? @ornate elk @dreamy garnet

[dreamy garnet]

@ornate elk @boreal bison > https://www.youtube.com/watch?v=oBKcjY-JO3Y

[dreamy garnet]

you should probably start with Auto1111, it's better

[boreal bison]

Can I do everything in it that comfyui does?

[dreamy garnet]

but if you are fluent on node-based and procedural workflows, comfyUI is where is at

[ornate elk]

Im using Automatic1111 mostly. I tried comfyui but I'm not a fan of setting up every step in the pipeline to get an image. Needing nodes for every extension and understand each setting again.
For me auto1111 is faster and easier to use. I know how to fix nearly any bug in auto too.

Comfyui is for the more advanced people who want to control every step in the image generating process

[dreamy garnet]

I personally haven't touched AUto since started with ComfyUI

[boreal bison]

I see

seems like I have to have to have both 😊

[ornate elk]

Yea that won't hurt xD

[boreal bison]

Who’s helping me install comfyui tomorrow?

[rough zenith]

I like AI)

[dreamy garnet]

it's not so complicated, just remember the first custom node would be the comfyUI manager, then you install eveything else with one click

[ornate elk]

Its easy. Check the link in the pinned message of this Channel to get to the comfyui site.
Then its just downloading a zip file you can unzip in a separate folder on your Z dtive.

[boreal bison]

do I need visual basics etc ? I really had a bad experience installing Risco Diffusion locally like 3 years ago, messed up everything

[ornate elk]

Yea comfyui ui manager is essential

[boreal bison]

that’s why I’m asking so many questions 😂

[rough zenith]

.

.

.

[boreal bison]

great guys! Will bombard you tomorrow with questions 😎

[rough zenith]

How can I run stable?

[ornate elk]

Localy?

If you have a GPU with 4gb vram or more then follow my install guide in the Pinned Messages of this Channel

[boreal bison]

This os exaxtly what I need

Does this work with automatic1111?

[dreamy garnet]

there should be extensions that do that, but i really don't know them, 100% comfyUI at this point

[boreal bison]

I’m def installing comfy ui!!

[ornate elk]

Yes

The FaceID IP-Adapter can be used in controlnet

[boreal bison]

Great, thank you

[sharp creek]

Hello, I entered this channel after 2 weeks today, and I noticed that I don't have access to the channels to generate images. Can an admin/moderator help me understand the reason or lift the restriction? Please.?

[dreamy garnet]

https://discord.com/channels/1002292111942635562/1100170153829871686

[covert jolt]

/drame

[fair oxide]

btw @ornate elk this is with the latest CN Update adding insta ID https://youtu.be/jjRFa1JugA0

you'd have to update CN and you'll see it listed in CN style box

hmm, not a very good video to show as an example, i'll have to look for specific video

https://github.com/InstantID/InstantID

[covert jolt]

Why can't I paint now

👀👀👀👀👀👀

[ornate elk]

nice, impressive how quick it got implemented:
also here is everything you need to know to get it run:
https://github.com/Mikubill/sd-webui-controlnet/pull/2580

[fair oxide]

the models are listed here https://huggingface.co/InstantX/InstantID/tree/main but i'm yet to try them out

[granite oar]

What is ID in this context?

[ornate elk]

the FaceID for face recognition - face swapping

[granite oar]

Ah

[vale moat]

hi thank you for your help, but after i installed 3.10.11, it would error out "No Python at '"C:\Users\kevin\AppData\Local\Programs\Python\Python312\python.exe'"

but when i looked under that folder the file is there

[ornate elk]

hey, you have to delete the venv folder and then relaunch the webui-user.bat

[vale moat]

it worked! thank you so much

what does venv folder do?

[ornate elk]

it stores the needed stuff for SD to work. and gets rebuild when deleting it.
there was a reference to your old python install in there, and now it uses the new path

[vale moat]

I see, thank you so much for your help

[fair oxide]

@ornate elk https://github.com/Mikubill/sd-webui-controlnet/pull/2580 describes the installation of instant_ID for a1111

[ornate elk]

yea i already added the link to my answer from before ^^

[fair oxide]

i keep getting bunch of errors when trying to generate a portrait

should i be changing the bin extension to something else ? ip-adapter_instant_id_sdxl.bin

[umbral mural]

[jovial apex]

hello i create and account

and pay the billing

but im not able to generate any image

[fair oxide]

@ornate elk not sure what i'm missing but it's not working for me, the error could be related to .bin extension that my system is not recognizing

also tried changing that to .pth but that obviously didn't work, think we need a properly compiled pth file?

[lament fog]

rocm linux users - beware upgrading rocm, have a way back. just spent 2 days thinking my GPU was going bad with image artifacts, but it was either the newer rocm or the newest torch. on a 6800 XT with 5.4.2 and very stable again

[lament fog]

I've been using opt-sdp-attention, is that the best option for AMD? I care about quality over speed

[valid gorge]

My SD (A1111) is acting up all weird again, I have two pictures created earlier, in both I have a person, one sitting and one standing, both also missing lower parts of the legs. So I wanted to do some outpainting (inpainting), expanded the canvases in Affinity, importer back in img2img inpaint and started working. With the first image all went well and worked as would expect, used fill (not original), mask only, cfg around 7 and denoise at aroun 0.64~0.7 and random seed. After about half a dozen renders, The standing person had full legs and a nice room around it. With second picture, person sitting, I tried the same approarch, at first all good and everything looked promising but then happened something and SD started to generate stuff like it did not see the image at all. Twekeked values and setting but nothing helped. Started SD twice from ground hoping things would get fixed, even rebooted computer. I’m stuck. Ideas?

[lament fog]

The fill option it's supposed to ignore the picture, unless I'm missing something

Original takes the content into account

Then there's whole picture vs masked only, masked only will apply the prompt to just that area

Another thing you could be doing is using the option "apply to non masked" I forget the exact wording

[valid gorge]

Yep, fill ignores what’s under mask but shouldn’t it understand the context anyway?

As i don’t have anything under the masked area, just empty canvas, it should work fine? This the workflow I used earlier anyway and worked as I expected

[delicate jolt]

pls help ,

The fill option it's supposed to ignore the picture, unless I'm missing something

[valid gorge]

Like here, in this workflow https://youtu.be/7IjJCEk-2mM?si=DMjW-Np2ti-DBZw0

[valid gorge]

Yep, as far I’ve understood, it shouldn’t ignore the picture

I might need to reinstall SD from ground up, perhaps have some sort of glitch from an update, idk.

Ok, I think it was some kind of bug. After changing model checkpoint to another one, and then back to what I was using, things seem to be working again

Nope nope, might have been positive too early. fck

[cobalt quiver]

I tried a model and it was working fine, but now I cannot use it

I got size mismatch for model

[dreamy garnet]

are you using something beside the model? lora? controlnets?

[cobalt quiver]

nope

I do have them installed, but not enabled rn

[dreamy garnet]

the mismatches are usually related to that, using 1.5 controlnets over sdxl models and viceversa, also loras and such

[cobalt quiver]

alright

but I entirely restarted and I still cannot use the model with nothing turned on?

[ornate elk]

Can you show the txt2img settings you use + model?

[cobalt quiver]

then I got tiled diffusion, tiled vae, additional networks(Lora), and Controlnet, all are not enabled

and I cannot even switch to the model

[ornate elk]

How big is the model size?

[cobalt quiver]

6.4GB only

[ornate elk]

Okay so its SDXL

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

[cobalt quiver]

it worked yesterday and this morning

however

I got a 3060

and

[balmy hare]

永远都用不了了？

[fair oxide]

its been stuck here for a min is that alright?

[ornate elk]

Yes let it run

Thats the webui.bat not the webui-user.bat

[cobalt quiver]

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--theme dark --no-half-vae --medvram --xformers --always-batch-cond-uncond --upcast-sampling
set PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.9,max_split_size_mb:512
set SAFETENSORS_FAST_GPU=1

call webui.bat```

[ornate elk]

A bigger screenshot could help too

To identify if everything is ok

[fair oxide]

it seems like its moved on i think

[ornate elk]

Do you use a 12gb 3060 or a 6gb one?

[ornate elk]

Looks good

[cobalt quiver]

6gb

[ornate elk]

Ahh okay

And you restarted the webui and now you can not load the model?

[cobalt quiver]

correct

[ornate elk]

What does the cmd shows when you switch to it?

[cobalt quiver]

Traceback (most recent call last):
  File "D:\Stable diffusion\stable-diffusion-webui\modules\shared.py", line 516, in set
    self.data_labels[key].onchange()
  File "D:\Stable diffusion\stable-diffusion-webui\modules\call_queue.py", line 15, in f
    res = func(*args, **kwargs)
  File "D:\Stable diffusion\stable-diffusion-webui\webui.py", line 199, in <lambda>
    shared.opts.onchange("sd_model_checkpoint", wrap_queued_call(lambda: modules.sd_models.reload_model_weights()), call=False)
  File "D:\Stable diffusion\stable-diffusion-webui\modules\sd_models.py", line 546, in reload_model_weights
    load_model_weights(sd_model, checkpoint_info, state_dict, timer)
  File "D:\Stable diffusion\stable-diffusion-webui\modules\sd_models.py", line 288, in load_model_weights
    model.load_state_dict(state_dict, strict=False)
  File "D:\Stable diffusion\stable-diffusion-webui\venv\lib\site-packages\torch\nn\modules\module.py", line 2041, in load_state_dict
    raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for LatentDiffusion:
        size mismatch for model.diffusion_model.input_blocks.4.1.proj_in.weight: copying a param with shape torch.Size([640, 640]) from checkpoint, the shape in current model is torch.Size([640, 640, 1, 1]).
        size mismatch for model.diffusion_model.input_blocks.4.1.transformer_blocks.0.attn2.to_k.weight: copying a param with shape torch.Size([640, 2048]) from checkpoint, the shape in current model is torch.Size([640, 768]).```

as to avoid spamming the channel, I get many many size mismatches after this

[ornate elk]

Okay then try to remove
--always-batch-cond-uncond
And the whole set Pytorch–Cuda-Alloc stuff

Thats normaly not needed

[cobalt quiver]

ngl normally I just run webui.bat

[fair oxide]

what have i done wrong

[cobalt quiver]

I dont use webui-user

[ornate elk]

Why that? That prevents every optimisation for your gpu xD

[cobalt quiver]

I forget

[ornate elk]

Ok. Always and only use webui-user.bat

[cobalt quiver]

fair enough

[fair oxide]

@ornate elk do you know how to fix this ?

nevermind i think its cause i forgot to install git

[cobalt quiver]

yeah, if it's not that make sure that you have git installed in the python that is your path

[cobalt quiver]

same error occurs

[fair oxide]

do i need a specific version of git or can i just use latest?

[cobalt quiver]

as to be expected as it occured when just runign webui.bat

[cobalt quiver]

I believe there is a range, you can try latest

[ornate elk]

You didn't followed my install guide

Downloading the webui as zip is the first wrong step

Check the Pinned Messages of this channel for a detailed install guide

[ornate elk]

And when you run the webui-user.bat ?

[cobalt quiver]

gives the same error

I tested also restarting my device, no change

[ornate elk]

Did you edited the webui-user.bat as I suggested?

[cobalt quiver]

yes

[ornate elk]

okay can you check your webui versions, in the webui scroll down to the bottom

[cobalt quiver]

requirements_versions?

[ornate elk]

no no in the browser

when the webui is opened

[cobalt quiver]

version: v1.2.1  •  python: 3.10.6  •  torch: 2.0.1+cu118  •  xformers: 0.0.17  •  gradio: 3.29.0  •  checkpoint: 98a1428d4c

[ornate elk]

ohhh wow, thats outdated

[vocal burrow]

that's vastly outdated

[ornate elk]

with that webui version sdxl wont work

[cobalt quiver]

that's odd, cause it worked yesterday and this morning

maybe my version got messed up at some point

@ornate elk should I install just the latest version of A1111? or?

[ornate elk]

you can update your existing webui, or you can install auto1111 in a fresh folder

[cobalt quiver]

fresh folder it is

[ornate elk]

okay, then checkout my Install guide in the Pinned Messages of this channel

[cobalt quiver]

simple enough pulling it with git

[valid gorge]

With the controlnet outpaint / inpaint method, can I control which way the picture is being extended?

[ornate elk]

yea and make the changes from Step 5

with that you have the best performance and compability

[ornate elk]

yea getting this error with InstantID:

2024-01-28 20:10:03,144 - ControlNet - INFO - unit_separate = False, style_align = False | 7/30 [00:07<00:18, 1.22it/s] 2024-01-28 20:10:03,144 - ControlNet - INFO - Loading model from cache: control_instant_id_sdxl [c5c25a50] 2024-01-28 20:10:03,145 - ControlNet - INFO - Using preprocessor: instant_id_face_embedding 2024-01-28 20:10:03,145 - ControlNet - INFO - preprocessor resolution = 512 2024-01-28 20:10:03,146 - ControlNet - INFO - Loading model from cache: control_instant_id_sdxl [c5c25a50] 2024-01-28 20:10:03,147 - ControlNet - INFO - Using preprocessor: instant_id_face_keypoints 2024-01-28 20:10:03,147 - ControlNet - INFO - preprocessor resolution = 512 *** Error running process: D:\Programme\AI-Zeug\stable-diffusion-webui-directml\extensions\sd-webui-controlnet\scripts\controlnet.py Traceback (most recent call last): File "D:\Programme\AI-Zeug\stable-diffusion-webui-directml\modules\scripts.py", line 718, in process script.process(p, *script_args) File "D:\Programme\AI-Zeug\stable-diffusion-webui-directml\extensions\sd-webui-controlnet\scripts\controlnet.py", line 1148, in process self.controlnet_hack(p) File "D:\Programme\AI-Zeug\stable-diffusion-webui-directml\extensions\sd-webui-controlnet\scripts\controlnet.py", line 1133, in controlnet_hack self.controlnet_main_entry(p) File "D:\Programme\AI-Zeug\stable-diffusion-webui-directml\extensions\sd-webui-controlnet\scripts\controlnet.py", line 1107, in controlnet_main_entry assert i > 0, "InstantID control model should follow ipadapter model." AssertionError: InstantID control model should follow ipadapter model.

[fair oxide]

@ornate elk

so i tried following the guide 😭

[ornate elk]

are you based in asia?

[fair oxide]

nah NA

[ornate elk]

hmm can you try again?

[fair oxide]

trying rn

[ornate elk]

did you installed git before?

[fair oxide]

i have git and python installed

tried it again and it failed

ill prob just download github desktop and clone it from there

[ornate elk]

you can also try restart your PC, or disabling third party antivirus systems

or changing the DNS server

@fair oxide i got InstandID to work

[fair oxide]

@ornate elk i cant just install this damn thing 😭

[valid gorge]

Still puzzling should I get 4060 TI 16GB instead of 3060 12GB, price difference has reduced a little, around 150 euros now comparing the two

I guess I found the inpainting / outpaint issue, looks like it's the resolution I'm using. 512x gives weird results, 1024x seems to be fairly consistend, above that the weirdeties come back.

So 1024 it is

[cobalt quiver]

@ornate elk it works on latest

[ornate elk]

something is strange with your git

or connection to github

close the cmd
then open up a cmd and type
pip cache purge
then delete the venv folder and relaunch the webui-user.bat

[fair oxide]

I’ll try this when I get home

[mossy glade]

Hello, I've been using SD for a while now and only recently noticed it keeps "loading 1 new model" for every generation I do. My question is, is it supposed to "load 1 new model" every generation? I would assume the model would have to be loaded only once and ready to go so long as i dont change it. I am on comfyui if that makes a difference

[plain wagon]

i have amd gpu and installed every packages needed from stable diffusion but some extensions still using cpu with onnxruntime! why?

like reactor for example, i dont have anything related to cuda installed?

[boreal bison]

@ornate elk Starting the comfy UI downloading process 🙂

@ornate elk how do I change the path here?

so I use my already downloaded models

[boreal bison]

After installing comfyUI manager, I can't run it, getting these errors :/

anyone???

[dreamy garnet]

I can help you with ComfyUI tomorrow, it’s really late for me already, off to bed : )

[boreal bison]

seems like I got it working for now!

thanks

I'll bug you with other questions tomorrow 😄

[fair oxide]

Does anyone know if it's possible to iterate through different LORAs' using xyz prompt?

Further, is there an easier/more efficient way to manage >100 LORAs and their trigger words?

[dim cave]

my brother getting this error when try to join

[lament fog]

Firstly you can organize them into folders, makes browsing a bit easier. You can also set a thumbnail image. Webui let's you pull the metadata, so you can usually see the relevant tags, you can also save the tags you want to come up when you load the Lora there.

[ornate elk]

There you input the path to the Stable-diffusion-webui like Z:/Ai/Stable-diffusion-webui

[lost crescent]

OpenPose is sooooo slow - like 1 hour for an 896x1152!!! Normal SDXL takes only 40 seconds for the same size. RTX2070 8Gb VRAM, 64Gb RAM, Win 10

[ornate elk]

What's in your webui-user.bat?

Also what openpose model do you use?

[lost crescent]

I am using

[ornate elk]

Why force fp32 lol

[lost crescent]

OK, but it works

[ornate elk]

That will slow down everything

[lost crescent]

So force 16?

[ornate elk]

Wouldn't force anything

[lost crescent]

OK, then how should my bat look?

[ornate elk]

I'm not comfyui user so idk what else you should add there

[lost crescent]

Alright 🙂 Thnx

[ornate elk]

In auto1111 it would be --xformers --medvram-sdxl --no-half-vae

[lost crescent]

I agree, and that is what I have in A1111

[ornate elk]

But what you can try to is to use a smaller openpose model

[lost crescent]

The 128 version? Alright, I will give it a go

[ornate elk]

That or the t2i-adapter-openpose-sdxl lora

[ornate elk]

But even the 256 lora version should be not taking an hour

But yea try without fp32 forcing maybe that will speed it up a lot

[lost crescent]

👍🏻

[lost crescent]

I say! Starting without --force-fp32 has made the world of difference!!! Thank you 🙂

66 seconds 😄

[ornate elk]

Nice! Np^^

[autumn void]

Hi, does SD UI work well on Windows 11?

[ornate elk]

Yes

And hey

[autumn void]

Fantastic, that'll make for an ever better evening^^

[ornate elk]

Checkout my install guide in the Pinned Messages of this Channel 🙂
For an easy start

[autumn void]

Is it different than for w10?

[ornate elk]

Nope

[autumn void]

I will if things go south but I've got it all laid out in xmind, metalinks and all ^^

nice and neat

One more thing, an nvme doesn't do squat for the performance right?

[ornate elk]

Most guides out there are outdated or missing essentials install steps. So I recommend if you want to use Automatic1111 to go with my guide first as it covers everything needed.

[autumn void]

will do then^^

[ornate elk]

It won't make the image generation faster. But its good for loading/switching models

[autumn void]

That's good to know, thanks again

[ornate elk]

No problem, and if you face any Problem or having questions while installing. Feel free to ask here

[drifting stirrup]

hihi !! i'm sorry if this is the wrong channel but i am having trouble with SD and no online resource has helped. basically, i have downloaded everything i'm supposed to and when i open the UI, it says that it doesn't detect python even though i have it. can someone link me some resources or give me a hand? ^^

[ornate elk]

Hey, what's your python version?

[drifting stirrup]

python 3.10 64 bit !

[ornate elk]

3.10.? SD supports 3.10.6 up to 3.10.11

[drifting stirrup]

sorry i was wrong it is indeed 3.10.6 xP

[ornate elk]

And you checked "add python to path" while installing it?

[drifting stirrup]

indeed

[ornate elk]

Okay then delete the venv folder and relaunch the webui-user.bat

[ornate elk]

If there isnt a venv folder thats okay too.
Then you have to open up a cmd and type
Where python
Copy the path.
Then right click the webui-user.bat and at the line Python=
Paste the path. Then save.
Then launch the webui-user.bat

[drifting stirrup]

there was indeed a venv folder! i think i can figure it out from here. i'll return with updates or if i need more help, thank you <3

[ornate elk]

No problem. Also checkout my install guide in the Pinned Messages of this Channel.
It covers the needed setting for the best performance depending on your GPU vram

[drifting stirrup]

i have returned! SD is up and running... kind of! i believe this isnt supposed to look like this

WAIT im sorry i was wrong

[ornate elk]

Make sure the webui is whitelisted in any browser adblocker

[drifting stirrup]

hi i am back ;w; so i am being met with the NansExcpetion error. I did the first thing I was prompted to do but i believe i may be misunderstanding the second set of instructions. i'm supposed to run the given cmd prompts in a seperate cmd window by themselves, correct..? i am unfamiliar with these kinds of things ^^;

[ornate elk]

Hey, what's your GPU?

[drifting stirrup]

NVIDIA GeForce GTX 1660 SUPER

[ornate elk]

Have you checked and made the changes to the webui-user.bat like seen in my install guide?

[drifting stirrup]

that completely slipped my mind, i was still waking up and just had my second cup of coffee xD i'll check that out now my apologies

[ornate elk]

NP, your GPU needs --xformers --medvram --no-half
In the webui-user.bat

At the line COMMANDLINE_ARGS=

Then after relaunching the webui-user.bat the error shouldn't appear again

[drifting stirrup]

unfortunately, the error did appear again. do you reccomend reinstalling everything and doing said instructions again?

[ornate elk]

If you installed the webui via git clone and the folder is called
stable-diffusion-webui then you dont need to reinstall.

[ornate elk]

Check the file size of the model you try to use.
It needs to be larger than 1.98gb

[drifting stirrup]

ok just tried to open the UI through said folder and was met with this

[drifting stirrup]

tried checking how big it is, was met with KB in the millions >_<

[ornate elk]

Why do you have two stable-diffusion-webui folders?

[drifting stirrup]

i just realized that 😭

[ornate elk]

Hopefully you didn't installed it twice

[drifting stirrup]

i deleted the other one

i might've misunderestood something, my brain is smooth but the second folder is gone now

[ornate elk]

Okay

[ornate elk]

If the model is smaller than 1.98gb its not a model(checkpoint)
So it doesn't go in the model/Stable-diffusion folder

Smaller stuff can be a lora file.
These go into models/lora

[drifting stirrup]

ok i see, i can try moving it into that folder then

ok so i dont think im understanding. the checkpoint i plan to use were given to me as checkpoints and work for others.. so if i were to move them into models/lora i wouldnt be able to use them as checkpoints ?

they are over the number you said they should be by the way, i just figured out how to check ^^;

[ornate elk]

Models are always larger than 1.98gb. These files go into models/stable-diffusion
Smaller files go in other folders depending what they are.

To use a lora you still need a base model

[drifting stirrup]

this is how big the models i plan to use are

[ornate elk]

As you had the webui twice you need to check the webui-user.bat again. Maybe you didn't made the changes there

[drifting stirrup]

idk why it wont just give it to me in gb

ok checking that now

[ornate elk]

That will work (is a model)

[drifting stirrup]

the necessary changes have been made to my webui-user.bat

[ornate elk]

Okay, then launch it again and show me the cmd output

[drifting stirrup]

the beginning or the whole prompt?

[ornate elk]

The whole thing

[drifting stirrup]

ok

[ornate elk]

Looks good

And now try to generate an image

With the 2gb model

[drifting stirrup]

currently attempting to generate an image

;-;

[ornate elk]

Can you show all the txt2img settings?

Prompt, steps, resolution etc

Oh wait

[ornate elk]

I know what you did wrong xD

[drifting stirrup]

oh shit

[ornate elk]

You added --no-half-vae instead of --no-half to the webui-user.bat

[drifting stirrup]

oh shi ok lets see

[ornate elk]

Changing that and it will work

My guide also covers that argument needed for gtx16xx gpus

[drifting stirrup]

ye i thought i did it right

[ornate elk]

change --no-half-vae to --no-half

[drifting stirrup]

ye i did, currently attempting to generate an image, just see the interrupt and skip button rn

[ornate elk]

Yea thats the beginning of the edit part, but a few steps below it covers the GTX 16 part

[drifting stirrup]

ok word my apologies! but, an image is currently visibly generating!! i think we did it :D thank you again for your kindness and patience, hopfully i wont have to bother you again anytime soon ^^;

[ornate elk]

No problem at all ^^ you shouldn't get any problems now with any files if they are in the right folders

Have fun 😁

[fair oxide]

Hello! Any kind soul have tips how to solve this? 🙂 "WARNING:tensorflow:From G:\a1111\stable-diffusion-webui\venv\Lib\site-packages\keras\src\losses.py:2976: The name tf.losses.sparse_softmax_cross_entropy is deprecated. Please use tf.compat.v1.losses.sparse_softmax_cross_entropy instead."

[hot spindle]

@fair oxide what Torch are you using? Screenshot the bottom of A1111,like this:

but thats only a warning, should work the same

[civic sentinel]

why I don’t have the pression to enter the room?is there anybody know？

[vocal burrow]

python 3.11, nope not gonna happen

compatible pythons are 3.10.6 -> 3.10.13

[worldly rain]

hey guys I have a problem, im on 3060 with 12gb of vram. until a few days ago i was generating my pics really quickly, i was gone a for a few days and today when I wanna generate a picture its getting stuck at %97 and my whole pc is starting to freeze so i have to restart it. I even tried to use --medvram but its not making a difference, why is this happening all of a sudden for no reason?

[dry loom]

I keep getting this, how do I fix it

@vocal burrow

@ornate elk

[ornate elk]

What's in your webui-user.bat?
Also make sure to have 15-20gb free space on C

[spark hatch]

Hey everyone!

Got a little problem. Sampling methods are missing for some reason (screenshot 1). They aren't hidden (screenshot 2) but at the same time they are downloaded (screenshot 3, 4, 5, 6).

Any ideas?

[spark hatch]

I'm interested in dpm++ 2m karras sampler, but as you can see it's missing

[dry loom]

Can someone help me?
Im trying to launch this but it cant seem to install torch and it says no module named pip

[ornate elk]

Delete the venv folder and relaunch The webui-user.bat

[dry loom]

Just did

[ornate elk]

K Diffusion isnt a sampler. Its used for all the samplers that have the Karras in name.

[ornate elk]

Run the python installer again. Choose Modify, next, then check "add python to system variables"
Then delete the venv folder again

[ornate elk]

Also scroll up and down in the sampler list to see all samplers (first screenshot)

[ornate elk]

Okay, then uninstall python. Restart your PC, and then follow my install guide in the Pinned Messages of this Channel

In AFK for now

[wind sun]

Good day, can someone help me with faceswaping videos with SD

i mean i have swaped a video in output folder already and i have around 2K photos face swaped in that folder but when i try to build it again in video, the video is getting 17minutes and the original video is smth like 30sec

can't fix that since yesterday

[spark hatch]

This didn't help (because it's actually all visible samplers)

I found the actual file with all samplers and it's totally fine, so I don't know what's the problem. Hope you can help somehow when you come back!

[dry loom]

It happened again

[worldly rain]

i have 12.9gb empty space on my C drive

but a few days ago i wasnt getting any lags and I still had the same amount of empty space

[royal meadow]

Can someone help me out with this animation API problem — When using xl-1024-v0-9 for generating keyframes, all works well (despite being utterly slow), but when I switch over to xl-1024-v1-0, the first frames until diffusion_cadence_curve = "0:(5)" → in this case the 5th, are rendered correctly but then quickly start fading to gray, what looks like 100% noise.

[ornate elk]

Is your webui updated?

[ornate elk]

Okay logs good

But 12gb doesn't need --medvram

[ornate elk]

Make sure to not have wallpaper engine or epic games launcher running in the background

[prime lion]

/ dream rabbit to eat radishes in the forest and then step on a big lizard under his feet in a realistic style

[ornate elk]

Do you use any antivirus software?

[dry loom]

Yeah

@ornate elk Could it have something to do with the fact that Python installs in an area I cant see even though I can see its path when I go to the path section on my control panel?

[ornate elk]

Its mostly because of that. It blocks stuff when installing python

Windows defender is fine. But everything else could block

[dry loom]

How do I manually turn off each anti virus software for a sec?

[ornate elk]

Which one do you use?

There should be a switch to toggle it off

[dry loom]

Honestly I dont know

[wraith bronze]

If you open CMD and type python --version do you get anything? Otherwise your PATH might not be set properly.

[ornate elk]

Any programm named Avira, avast, AVG, Norton, Bitdefender, Eset, 360security, kaspersky, panda

[ornate elk]

AVG

[ornate elk]

Open the app and disable it.
Then uninstall python.
Then install it again with "add python to path" checked.
Then delete the venv folder

[dry loom]

Trying to find the disable thing on it, cant find it

[wraith bronze]

Right-click the AVG icon in the System Tray. This is located in the lower-right corner of the desktop. If you do not see the icon, click the arrow button at the edge of the System Tray to expand the Tray and display all the icons. Click Temporarily disable AVG protection.

You will be asked for how long, and then just click OK

[dry loom]

DOWNLOADING OMG

THANK YOU ALL

Hopefully this is all I need

[wraith bronze]

I believe I use the same thing as you, and it was very straight forward. There may be some tweaking needed depending on your hardware.

[dry loom]

Ok so what do I do now?

[wraith bronze]

wait 😛 it should continue on its own

It might take a little while

[wraith bronze]

You will notice when its done ^^ so you can just leave it alone

[dry loom]

So when can I close the thing?

[worldly rain]

yeah I just wanted to add it to see if its gonna fix it

[wraith bronze]

If I remember correctly after this is done, it will open the webUI, if it do you need to keep the terminal open as long as you run the webUI

[worldly rain]

also I only have them both working on the background, i'll turn it off and try it again and get back to you:)

[dry loom]

@wraith bronze "RuntimeError: "LayerNormKernelImpl" not implemented for 'Half'"

@ornate elk

[wraith bronze]

You may need to add some parameters in your "webui-user.bat" filed in the webui. Try change set COMMANDLINE_ARGS= to set COMMANDLINE_ARGS= --no-half

[ornate elk]

whats your GPU?

[dry loom]

Dont know...

[wraith bronze]

press the windows key and write "dxdiag" and look under "display"

[ornate elk]

open up the taskmanager, go to Performance tab, check the GPU 1 and GPU 0

[fair oxide]

"save the tags you want to come up when you load the Lora there" can you further explain this?

[dry loom]

@wraith bronze @ornate elk Is this it?

[wraith bronze]

Under "display 1"

[ornate elk]

your in the wrong channel/server for complaining discord updates.

[wraith bronze]

You need to be in the "Display 1" tab if you opened DXDiag you have a set of tabs at the top of the window

[dry loom]

I dont even know what that means

[wraith bronze]

Did you open DxDiag or the task manager?

[dry loom]

Im about to check task manager

[wraith bronze]

In DxDiag I assume you opened first the tabs are at the top, you click them

[wraith bronze]

In task manager you have to go to "Preformance" and select the GPU

and it will say the name in the top right corner

[lament fog]

In webui, there is a Lora tab to access the loras you have. When you navigate there and select one,it populates in your prompt. If you look close around those buttons, there's little widget that gives more info about the Lora when you hover over it

[dry loom]

@wraith bronze @ornate elk

[wraith bronze]

That doesn't show the GPU name/model

The red boxes is the information we need

[ornate elk]

make the same screenshot as Kimchii

[wraith bronze]

Do you have a GPU 1 as well?

[ornate elk]

okay

[wraith bronze]

Oh, OK, I don't know if SD can run on a integrated gpu, but if it can you probably need to add the flag for low vram

[ornate elk]

@dry loom you have to edit the webui-user.bat
at the line set COMMANDLINE_ARGS=
you add: --skip-torch-cuda-test --use-cpu all --no-half --opt-sub-quad-attention

then save tand relaunch the webui-user.bat

[ornate elk]

only on AMD integrated like yours

and maybe irisXe

[dry loom]

@ornate elk @wraith bronze

Like this?

[wraith bronze]

Tht looks correct

[dry loom]

Now what am I missing....

[worldly rain]

I closed epic launcher and wallpaper engine, updated my gpu and restarted my pc, still got stuck at %97 and my pc froze...

[ornate elk]

you may need to add --precision full

[dry loom]

Same thing happened

[ornate elk]

did you relaunched the whole cmd?

the webui-user.bat

can you show the cmd output as screenshot after you launched it?

[dry loom]

I think it works now

[worldly rain]

any recommendations?

[ornate elk]

does it stuck with sdxl or 1.5 models?

or both?

[worldly rain]

sdxl

i havent tried 1.5

i dont use it

[wraith bronze]

Check that you have read/write permission in the folder ^^

[ornate elk]

then you should use --no-half-vae
in the webui-user.bat

[worldly rain]

alrightt

[pine finch]

hi 👋

im new here... maybe silly question... can I use sd1.5 trained Loras on an img to txt with base prompt of SDXL?

guessing not, but I thought I remembered success with it earlier on

[wraith bronze]

I have no clue, but if you know what result you should get, try it? I haven't had any problem when I randomly download loras and just try them ^^

[worldly rain]

Same thing, its still freezing. I generated one without controlnet and it worked even tho it slightly lagged at %97. I generated another one while using controlnet and it got stuck again at %97. All extensions are updated

[pine finch]

Yeah, I'm not getting the expected results. not even close. That's why I'm asking here, I thought you could but I may have remembered wrong, or had some other details wrong when I had used the lora earlier.

[wraith bronze]

Check the repo for the controlnet you are using, if it is a compatibility issue, others probably asked about it as well.

[ornate elk]

so it only stcusk when using controlnet?

[worldly rain]

Repo?

[wraith bronze]

repository, they probably have a page on github

[worldly rain]

It slightly lags at %97 without it. It crashes at %97 with control net so yes.

[worldly rain]

Okay ill check now

[wraith bronze]

Just google the control nets name and add +github and if they have one it should pop up near the top

[ornate elk]

it crashes? you only said frezing before lol

[worldly rain]

Alright

[ornate elk]

thats a big difference, as kimchi said check that your using controlnet models made for SDXL

you can find them here:
https://huggingface.co/lllyasviel/sd_control_collection/tree/main

[worldly rain]

Yeah thats what i meant my bad

Okay will check now

[hollow drift]

Does anybody know what causes this? Running automatic1111 (lshqqytiger fork, DirectML) and recently highres-fix is just completely broken half the time. I'm running on the latest commit (d500e58a6) with the following arguments:

--lowvram --opt-sub-quad-attention --upcast-sampling --opt-split-attention-v1 --data-dir 'D:\SD' --skip-torch-cuda-test --use-directml

The only thing that recently changed that I can think of is that I switched from lowvram back to medvram and then back, plus I recently deleted a ton of stuff from my temp folders.
This happens with every checkpoint I load. Sometimes it will do the first few images just fine and then it just breaks down again. Rebooting or restarting SD doesn't seem to help.
My specs:

• AMD RX 5700 XT
• AMD Ryzen 7 3700X
• 48GB RAM
• Running everything off a Samsung 850 EVO SSD, if that matters

The initial image is always perfect and doesn't have strange artifacts, it's the highres-fix that seems to implode

[ornate elk]

remove --skip-torch-cuda-test annd add --no-half-vae
also whats the resolution and sampler, etc

[hollow drift]

512x512, dpm++ 2M Karras

I've already tried changing resolutions and switching samplers but it seems to happen on everything

I'll try removing those two arguments, hold on

[leaden socket]

I'm having the same issue. I'm trying to Uninstall XL and replace it with SD1.5 so I can use these loras

[ornate elk]

you need to add --no-half-vae
and remove the --skip-torch-cuda-test

[hollow drift]

oop

Yeah got it on the 2nd read lol, thanks

[ornate elk]

no that wont work

[ornate elk]

you dont need to "uninstall" sdxl, just switch to an 1.5 model

[hollow drift]

Alright, trying again with the same seeds and stuff

I use the latent upscaler by the way on 1.5, dunno if that matters

[ornate elk]

also whats the model and upscaler?

[hollow drift]

this model is beautifulRealistic_v60 but I've had it happen on practically every other model

[ornate elk]

try with Resrgan4x+ upscaler if that works

[hollow drift]

epicrealism, stock SD, dreamshaper, etc.

first picture came out correct, you might've already fixed it for me lol

I'll let the batch of 4 finish, see if it still acts weird

that's weird though, I don't remember adding or removing "skip torch cuda test"

[ornate elk]

--skip-torch-cuda-test makes it run on the cpu

only needed when someone dont have a gpu

as the directml webui updated to 1.7.0 everyone got the error to add that.
but instead you had to ad --use-directml

so thats probably why it was in there

[hollow drift]

ooohhhh

odd how it degraded with every image too, and only when upscaling

this seems to have fixed it, they're all coming out correct

thanks!

[ornate elk]

nice no problem 🙂

[spark hatch]

Yeah, I even have git pull in bat file

[ornate elk]

okay, whats in your webui-user.bat?

[valid gorge]

Testing out ComfyUI, are there similar boot argumets like when using A1111?

[frigid dome]

Can anyone tell me the maximum value for clip skip? As far as I can tell sd15 is 12, sd14 is possibly something less (?), but I have no clue what sd21 and SDXL might be.

[vocal burrow]

here they are : https://github.com/comfyanonymous/ComfyUI/blob/ef5a28b5971761a9a71bc5b0af5bac6778e5f57a/comfy/cli_args.py#L98

[lament fog]

no idea, but curious why you'd be asking. you generally want to set it to whatever the checkpoint was trained with, which is almost always (ie 99.9999%) 1 or 2

[frigid dome]

Adding functionality to a bot

Doesn't matter anyways, I looked at the API and it only accepts 1-12

[queen plank]

Help

I can't run comfy ui

it was working fine, till now

I tried uninstalling then reinstalling pip, it sayas 2.1.2, but then in the gpu.bat it sayas 2.2.0

I treis searching online for hours

it's currently 3:38 AM

it happend at about 11:30P PM

I need help, urgently!

if I have to re-intall comfy ui, how do I keep all my checkpoits and custom nodes?

[sonic fable]

Hi, I have downloaded the model sdxlNijiSpecial but while trying to load, I am getting below error message. Any inputs would be great. Thanks.

[hollow trench]

when using thelastben's collab doc.

TypeError: check_deprecated_parameters() missing 1 required keyword-only argument: 'kwargs'

this is on step 6 "test the trained model"

[craggy jewel]

im trying to load a new .safetensors model but im getting these errors does anyone know the fix?

model is monsterDiffusionSDXL_v10SDXL

[ornate elk]

Ah okay, your using the onnx version. Thats not compatible with all the samplers.
You can switch to normal directml by removing the --onnx
Then delete the venv folder.
Then add: --opt-sub-quad-attention --opt-split-attention-v1 --upcast-sampling --no-half-vae
To the --medvram --use-directml
Then launch the webui-user.bat

[ornate elk]

Is your webui updated? And the Extensions too?
The error is from controlnet and doesn't recognise sdxl it seems.

You can check the webui version when scrolling down in the browser

[ornate elk]

Is your webui updated? Check the version at the bottom in browser

[unique glacier]

You can just go to the original folder and copied it to the new installation.

[craggy jewel]

is this the latest version? if not where do i download the newest one

[ornate elk]

No that's a very old version

[craggy jewel]

oh dang lmao i guess thats why it wasnt working

[ornate elk]

To update it, edit the webui-user.bat.
There add
Git pull
Into an empty line before webui.bat
Then delete the venv folder.
Then open up a cmd and type
Pip cache purge
Then launch the webui-user.bat

[craggy jewel]

yooo bet, ty i thought i would have had to download completely new one

[ornate elk]

Only if your webui folder has the name:
Stable-diffusion-webui-master

Or any name other than stable-diffusion-webui

[craggy jewel]

aww man

[ornate elk]

Thats fine

You should be able to update

[craggy jewel]

this is good for the bat file right

[ornate elk]

Also what's your GPU?

[craggy jewel]

lemme check

GTX 1650

[ornate elk]

Okay, then add --medvram --no-half behind the --xformers

[craggy jewel]

ok

aight i did everything i am opening the webui.bat

[ornate elk]

No

Not the webui.bat

Only ever launch the webui-user.bat

[craggy jewel]

yee thats what i meant

living in the middle of nowhere means L internet but its aight

[ornate elk]

Ah damn 🥲 can take a while.

But then your updated

[craggy jewel]

yoooo nice

[ornate elk]

Sdxl models are large so they are slower than 1.5 models with GPUs with 6gb or less

[craggy jewel]

it says 12Gb of gpu memory here so i should be good

[ornate elk]

Thats shared GPU memory.
Dedicated is the memory (vram) of your GPU

In that case 4

[craggy jewel]

oh lmfao

im guessing that is pretty yikes in my case

[unique glacier]

Tough to be an ai enjoyer with old gpu.

[ornate elk]

Yea its not much. So I would recommend not using models that are larger than 6gb

[craggy jewel]

hopefully they still work tho im aight with them being a bit slower

[unique glacier]

Luckily for you. there are tons of SD 1.5 base model to fit that need. You gonna be fine.

[ornate elk]

They still should work

[craggy jewel]

i might check the specs on my laptop it might actually be better, it has an RTX 4050 im pretty sure

[ornate elk]

That one would have 6gb vram

[craggy jewel]

ah ok so it would be a bit better

[ornate elk]

Yea it also has more cuda cores and better timings so it would be a lot better

[craggy jewel]

aight bet im prob gonna use laptop for generating AI

it is laptop gpu vs pc gpu so idk if that makes a big difference

[ornate elk]

Yea Laptop GPUs are always less powerful, smaller, and have less vram than their counterpart in a PC. But this GPU would be still faster than your gtx1650

Also the laptop GPUs get hotter

[craggy jewel]

hmm ok, hopefully it doesnt end up overheating

[ornate elk]

I think it doesn't. But you can check where it has the vents for air intake. Then place it so they aren't covered

[craggy jewel]

aight bet sounds good, i'll copy my entire stable diffusion folder over to laptop sometime tomorrow once its done updating

[ornate elk]

That won't be a good idea. As the venv folder needs to be recreated anyway.
Moving the folder will cause the webui to not update anymore.

[craggy jewel]

hmm ok

even if i use an external drive?

[ornate elk]

Yes

[ornate elk]

Best would be to follow my install guide. You find it in the Pinned Messages of this Channel.

[craggy jewel]

aight sounds good, or could i just follow this again once its on laptop

[ornate elk]

Thats what I mean. Follow that on the laptop

To safe some download time you can copy your models folder over to the new models folder.
Before launching the webui-user.bat for the first time.

[craggy jewel]

yee i have about 7 models so imma do that

[vestal turtle]

Nothing works in the sandbox

Just shows Generating, but nothing works

I mean when using image to image generation prompts

[vocal burrow]

That's not very helpful. You don't go to the mechanic and say "car does not work"

What software are you using ? What's your hardware ? What's the exact error / show logs ? @vestal turtle

[valid gorge]

Thank you!

[fair oxide]

Has anyone made use of the official script of SDXL and created something without ComfyUI and Automatic1111 on thier system or Collab with T4? If yes. Please dm. Need assistance in training on my dataset.

[grand fable]

Hi, do you have any idea, why ComfyUI can't see the IPadapter model?

[grand fable]

Solved, RunPod restart helped.

[grim iron]

0it [00:00, ?it/s]
make buckets
min_bucket_reso and max_bucket_reso are ignored if bucket_no_upscale is set, because bucket reso is defined by image size automatically / bucket_no_upscaleが指定された場合は、bucketの解像度は画像サイズから自動計算されるため、min_bucket_resoとmax_bucket_resoは無視されま す
number of images (including repeats) / 各bucketの画像枚数（繰り返し回数を含む）
D:\programmeerstuff\stable\kohya_ss-master\venv\lib\site-packages\numpy\core\fromnumeric.py:3504: RuntimeWarning: Mean of empty slice.
  return _methods._mean(a, axis=axis, dtype=dtype,
D:\programmeerstuff\stable\kohya_ss-master\venv\lib\site-packages\numpy\core\_methods.py:129: RuntimeWarning: invalid value encountered in scalar divide
  ret = ret.dtype.type(ret / rcount)
mean ar error (without repeats): nan
No data found. Please verify arguments (train_data_dir must be the parent of folders with images) / 画像がありません。引数指定を確認してください（train_data_dirには画像があるフォルダではなく、画像があるフォルダの親フォルダを指定する必要があります）

I get this error when trying to train a lora using kohya even though there are images in the input folder can anyone help?

[valid gorge]

Could someone link it again, how do I use SDXL model together with SD 1.5 Lora in Comfy, pls? I think there was some guide to this earlier.

[valid gorge]

I think it was this one, have to test it today and see how it works, https://civitai.com/models/129283/snoodler-or-use-sd-loras-with-sdxl-models-or-workflow

[wraith bronze]

Did you put train_data_dir as the parent folder? rather than train_data_dir/YourFolderWithTheImages?

[buoyant laurel]

Hello everyone. I have a personal archive of texture scan data. Each "scan" consists of an RGB image of the surface as well as a greyscale image that represents the height/depth of that same surface. To summarize, i have thousands of scans and I'd like to train a model to generate scans. I assume I could encode the height into the alpha channel of the RGB and use that as training data, but I'm not sure if the stable diffusion models will utilize it.

Is it possible to train a lora/safetensor/checkpoint on a 4-channel RGBA dataset and get inference to create a 4-channel texture?

[dense sandal]

how can I clear the nsfw content in CIVITAI? I can't find the filter setting

[wraith bronze]

Its under account settings

[dense sandal]

ty found it

[drifting aspen]

Hello,
I'm considering building equipment for Stable Diffusion. Many recommend RTX cards, but they are expensive. The machine will be a computer/server running 24/7, and I'll be executing commands remotely. I don't need very fast drawing creation, but for example, a resolution of 1920x1080 within 15 minutes for an image. What hardware should I choose for this? Which graphics card is suitable for this purpose?

[ornate elk]

hey, if your planning on making a Linux server, then go for AMD gpus with at least 12gb of vram. (best linux support)
If you want that to be your ai windows server then go for any Nvidia GPU with 8gb or more vram (best ai support)

[drifting aspen]

hey, the server will be Debian. Currently, I do it using easy diffusion, but one small image takes 30 minutes. I wanted to build something at a low cost, a nice AI base. Is Nvidia Tesla C2050 enough for starters?

[ornate elk]

that card has only 3gb of vram, not good

vram is the most important part for ai stuff

[drifting aspen]

NVIDIA GRID K1 is ok?

what graphics cards do you recommend? which ones have a good price-quality ratio?

[ornate elk]

should be newer than 2016

a rtx 3060 is a good price/vram gpu

[drifting aspen]

https://www.techpowerup.com/gpu-specs/grid-k2.c1700

Keppler 2013

Ist old

[ornate elk]

Yea too old for the new ai stuff

[drifting aspen]

Rtx 2060 is not ok?

[ornate elk]

its okay

[drifting aspen]

I will check the prices, if there is a small difference from the 3060, I will buy the 3060. Does this card really have much higher performance than the 2060?

[ornate elk]

that shows the fastest gpus for Stable-Diffusion:
source: https://www.tomshardware.com/pc-components/gpus/stable-diffusion-benchmarks

[dull echo]

i have trained a hypernetwork but i cant find where the trained pictures from hypernetwork are

[short tulip]

Hello, I am using the auto111 webui, I am testing the new controlnet face-id_V2 control(model?), I am using the Lora alongside it as shows in the documentation for it. My question is, I get much better results using SD 1.5 than I do with SDXL. Like SDXL I get a version of the face but more... cursed... Is this an issue where I need to wait for improvements or is there something I might be missing?

[ornate elk]

hey, do you have an example of the cursed face image?

[short tulip]

Sure, I grabbed an image of Angelina Jolie off google. (I was originally trying with my own face). The left is sdxl, the right is 1.5. I've tried multiple models. Both at CFG 5 and 40 steps with the faceid lora at around 0.6 weight

[ornate elk]

Was the sdxl used with a 1024x1024 resolution?

[short tulip]

I tried 1024x10x24 as well as a 768x1024. Regardless of resolution, the face usually has some odd details, usually.

[ornate elk]

Hmm okay, I will try it too in a few minutes

[ornate elk]

Did you used 0.6 as lora strength?

That is recommended

[short tulip]

I did, yes. I also tried it at 1, 0.5, etc

[ornate elk]

the latest controlnet update changed the way it handles the ip adapter, so i think thats the cause for this

[fair oxide]

oh great, im curious to know how you managed it, did you use the .bin file ?

[fathom vine]

guys can a lora-maker expert help me? i've trained a lora 3 times, modified the dataset several times, changed the number of steps, and i always get as a result someone who looks VAGUELY like the subject, it doesn't follow the likeness properly, and I'm just dumbfounded, idk what's going wrong.

[inner raptor]

Hello why i cant use rhe bots ?

[ornate elk]

bin and safetensor is needed, it works only with 2 controlnet units combined, here is how its going to work:
https://github.com/Mikubill/sd-webui-controlnet/discussions/2589
renaming the files is important

[ornate elk]

hey, its under maintenance: #1047610792226340935

[inner raptor]

Ooh thx

[fair oxide]

yes i followed that guide closely... here are two images of my settings

output happens to be this

which makes me think its not working

[spiral depot]

my A1111 keeps saying error connection errored out, i am using paperspace to run SD, does anyone know why this might be happening?

[fair oxide]

@ornate elk i wonder about this text clutter on console when i press the preview icon, something could be amiss but not sure, any suggestions?

[ornate elk]

yea that could be the cause it doesnt work

but havent seen this specific error

@fair oxide did you installed the C++ build tools or the standalone insightface wheel?

[fathom vine]

anyone who knows howt o do loras with kohya that can advice me? i ned help 😦

[fair oxide]

hmm, i didnt install any build tool, all i did was download the models rename that as specified and put them in model folders, only other thing that i did was to update the controlnet extension

[ornate elk]

yea then its missing the insightface model it seems, did you maybe used reactor extension or roop before?

[fair oxide]

nope, didnt use any of those

[ornate elk]

okay yea then your missing insightface

[fair oxide]

so i have to install insightface separately? or is that from extension tab?

[ornate elk]

nope, do this:
https://github.com/Gourieff/sd-webui-reactor?tab=readme-ov-file#viii-for-windows-users-if-you-still-cannot-build-insightface-for-some-reasons-or-just-dont-want-to-install-visual-studio-or-vs-c-build-tools---do-the-following

[fair oxide]

ok thanks, im on it

[ornate elk]

make sure to only follow the auto1111 steps so use cmd, not venv/activate stuff, also not the steps for portable a1111

[fair oxide]

these are the steps i followed through ...

[ornate elk]

yea no need for venv/activate

just copy the file in the root folder

then pip install it

[ornate elk]

if you installed it with pip, launch the webui-user,bat and test instantID again

[sour wharf]

.

[fair oxide]

yes, trying an image now, think its starting to work with what i see., give me few seconds

with the source image

with the source

i think i can get better resemblance by turning off adetailer

btw @ornate elk were you able to make instant ID work with sd 1.5 models?

[ornate elk]

nope its only for sdxl rn

[fair oxide]

ahh ok. its interesting how you can generate the facial resemblance and close variations of a source image

[ornate elk]

yea, but for me ip-adapter-face-full still works the best

[fair oxide]

here is another variation

[ornate elk]

nice one

did you changed some controlnet end steps or weights?

[fair oxide]

no, kept that to default but yes i could change some values to make more variations

@ornate elk are you making use of these https://huggingface.co/h94/IP-Adapter-FaceID/tree/main

[ornate elk]

i tested these a few days ago and a few hours ago again xD

[fair oxide]

i couldn't get them all to work, i get some errors when attempting to generate image with face id models

[ornate elk]

they are nice if the input image is good

[ornate elk]

you can show it to me

[fair oxide]

can you give an example with screenshot of your settings?

[ornate elk]

there is one new error without description, ive seen today, (but it basicly means, face not found)

[fair oxide]

ok let me try and render one with face id

@ornate elk this is what im going for

[ornate elk]

wrong model

[fair oxide]

oh?

i got bunch of errors too lol

[ornate elk]

you need to download the .bin files and put them into the controlnet/models folder.
the .safetensor (lora) files go into models/lora

ive made the same mistake the first time

[fair oxide]

oh wow... i have to redo those files then, i got everything mixed up

[ornate elk]

i also made a cnet folder inside loras to quickly find them

[fair oxide]

@ornate elk so for an example ... i'll download ip-adapter-faceid_sd15.bin and put them in models/controlnet folder
and ip-adapter-faceid_sd15_lora.safetensors into models/LoRA folder?

those from this link https://huggingface.co/h94/IP-Adapter-FaceID/tree/main

[ornate elk]

exactly

[fair oxide]

great. let me fix those and try one image

[molten coral]

I can't get any xl model to run they all freeze or crash when I load them and if they are loaded they crash auto1111 when generating.
Any idea how to get this fixed?

[ornate elk]

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

[molten coral]

GeForce RTX 3060 Lite

just the standard webui-user.bat that ships with auto1111

I have not made any changes

also I'm on linux so I use the .sh

[ornate elk]

How much vram does it have?

[molten coral]

12gb

[ornate elk]

Okay then you need to edit the webui-user.sh,
uncomment the line #set COMMANDLINE_ARGS
There add --xformers --no-half-vae

Then save and relaunch the webui.sh

[molten coral]

like this: export COMMANDLINE_ARGS="--xformers --no-half-vae"

[ornate elk]

Yes

[molten coral]

it worked once and now it crashes when generating

so I've tested it and it's basically the same as it was before the change

still slows down and then crashes while generating

[ornate elk]

What's the error you get?

What's the resolution too?

[molten coral]

no error 1024x1024

[ornate elk]

What does the cmd shows?

[molten coral]

it happens with any rez tho

[ornate elk]

Does the terminal crashes? Or any error in there?

[molten coral]

the only thing of notice is: WARNING:sgm.modules.attention:Attention mode 'softmax-xformers' is not available. Falling back to native attention. This is not a problem in Pytorch >= 2.0. FYI, you are running with PyTorch version 2.1.1+cu121.

but I read that I can just ignore this

[ornate elk]

Do you use the development branch of auto1111?

[molten coral]

it just shows```
Model loaded in 46.8s (load weights from disk: 8.0s, create model: 0.8s, apply weights to model: 36.8s, apply half(): 0.1s, move model to device: 0.1s, load textual inversion embeddings: 0.6s, calculate empty prompt: 0.3s).
0%| | 0/20 [00:00<?, ?it/s]Killed

just the standard version from: https://github.com/AUTOMATIC1111/stable-diffusion-webui

[ornate elk]

Hmm okay

[molten coral]

I've updated it many times I've had this issue ever since xl models came out

guess I'll just wait for the next model

[ornate elk]

But 1.5 works normaly?

[molten coral]

yea

[ornate elk]

What you can do is to use --opt-sdp-attention instead of --xformers

[molten coral]

yea no change

[ornate elk]

Okay, and your webui is updated ?

Version 1.7.0 ?

[molten coral]

how do I check which version it is?

[ornate elk]

In the browser at the bottom

[molten coral]

1.6.1

[ornate elk]

Okay not the latest but this version still supports sdxl too

[ornate elk]

You can try update the webui via git pull and then delete the venv folder before launching the webui.sh

That will make sure you have clean venv

[gray oyster]

I need help with Oobabooga in case anyone is using it here

[gray oyster]

So whenever I wanna load a model the webui just crashes silently:

Though it does seem to use bitsandbytes for cu118 which it shouldnt

it should use cu121

it also didn't ask me what cuda version to use

despite doing it everytime usually

its version 1.7.0 of the webui

im using a 4060 Ti with 16gb

it also had the same issue with cuda toolkit 11.8

except there it asked me if i wanted to use cuda 11.8 or 12.1

[molten coral]

@ornate elk I don't like to git pull to often tho as that pretty much always messed up auto1111 completely. I had to wait many month once until I got a working auto1111 version again

I always land in py package dependency hell when I git pull after I got it working

[gray oyster]

same, but i guess that is with most repos

[molten coral]

I never had issues with that in other languages tho only py is wired like that

[vapid lily]

。

[hollow trench]

AttributeError: partially initialized module 'charset_normalizer' has no attribute 'md__mypyc' (most likely due to a circular import)

[hollow trench]

using thelastben fast dreambooth collab doc

[hollow trench]

none of these collab docs i can find work out of the box, every one of them has some sort of error for me... i purchased the 10$ pro and im using the best hardware available to me as well

[grim iron]

not sure what u mean, I don't have a folder named train_data_dir, this is my folder structure, I followed a tutorial and this is how tthey did it

I fixed it by putting the images in the 7_ folder

but now I get this error

I treid reinstalling xformers but that didn;t fix it

[wet gale]

#✍🏼｜rules-and-tos

[radiant saffron]

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.

Im not sured whats going on i did everything it said and its still not working keeps doing that mid gen, i have a 4060ti so it shouldn't be vram

[thick igloo]

Hi there I'm trying to get Stable Diffusion working on my alienware computer with an RX 560 AMD GPU. Initially I was following the corridor crew tutorial but after a number of error messages and finally sorting them and being able to open the URL and not generate anything I realized that I would have to start again with AMD in mind.

I found this video from AMD (https://www.youtube.com/watch?v=kw0WT5sDBIY) and tried to do this via Miniconda, Github and an updated Adrenalin Driver but at the end of the day I still get the same error message as before which is: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this.

The issue is I did this in the past attempt from the corridor crew tutorial and doing it meant I couldn't generate anything. Was really hoping going this AMD route would let me get the GPU working. Secondary problem is I couldn't for the life of me figure out how to change the base file location in miniconda. It says it's the base where drive I installed it when I envs_dirs but the base it reads there defaults to my C: drive which doesn't have a ton of extra space. Tried everying in PATH environment variables and nothing worked.

Anyway I have spent days on this at this point so if anyone can help I'd really appreciate it. Thanks.

[ornate elk]

Hey, what's in your webui-user.bat?

[radiant saffron]

i just have --xformers now

[ornate elk]

Hey, AMD GPUs can work on Windows with the auto1111 directml fork.
Amd uses this too but in combination with onnx.

The best you should do is to remove everything youve done.
And then checkout my install guide for AMD.
Its in the Pinned Messages of this Channel.

[ornate elk]

Okay, then what's the file size of the model you try?

you can also add --no-half-vae

But first check the file size of the model

[radiant saffron]

Socamerge AniMix XL 6.62gb

[ornate elk]

Okay so its an sdxl model

Is your webui updated to 1.7.0 ?

[radiant saffron]

i just stopped using stability matrix and installed webui by itself

yes 1.70

python 3.10.6

Launching Web UI with arguments:
no module 'xformers'. Processing without...
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it. is that bad?

[ornate elk]

Okay, make sure you unticked the upcast float32 button in settings

[ornate elk]

Yes

[radiant saffron]

ah

unticked now

[ornate elk]

You always ever need to launch the webui-user.bat to start SD

Dont run the webui.bat

[radiant saffron]

oh

well i feel like a dumbass

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers

call webui.bat

should that have more in it

ah installing xformers now cool

[ornate elk]

Add --no-half-vae --update-check

[radiant saffron]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --no-half-vae --update-check

call webui.bat

yay working again

ty muchly

[ornate elk]

Nice no problem 🙂
Should be much faster now too

[radiant saffron]

On stab matrix it was slightly faster at the 50% mark but was getting endless loading checkpoints and broken lora loading

but i am happy to be generating again

ty

[ornate elk]

If you use hires fix make sure to always set hires steps to 10-15 and not using 0

That will make it much faster when hires fix is used

[radiant saffron]

ah am using 25 will decrease

[ornate elk]

Yea thats the issue then. You normaly dont need that much and thats why it takes so long

[radiant saffron]

can't type today keep making spooling mistooks and grameer eroors

[unreal valley]

Can somebody help me with this, I cloned the control lora folder from huggingface but the models are missing

[ornate elk]

You have to download then manually or use wget command for each file

[unreal valley]

No worries, I used the manager for Comfy and got all of them

[azure sable]

Does this error while training a lora mean anything? Training continues as normal after it pops up. it comes up after the first sample is generated.

[pine finch]

I don't think you need to uninstall anything... just use a 1.5 checkpoint model

[pine finch]

oh I see @ornate elk already said this... sorry

[pine finch]

have you tried updating that .py file?

[azure sable]

I have no idea how to do that or what I would update it to. I don't know anything about coding and I'm fairly new to this.

[pine finch]

I havnt seen that error before... but the error does say that the value in that file should be 0. You might try changing it... its not too hard, if you navigate to that files location it says... c:\Users\cassi\kohya_ss\venv\lib\site-packages\diffusers\piplines\stable_diffusion\pipeline_stable_diffusion.py and edit it with notepad...

just change that steps offset number from 0 to 1... save it, then reopen it and check?

I mean... reopen kohya_ss

[azure sable]

hmmm. well according to that file, it IS set to 1. So I'm not sure why it's telling me it's 0.

[pine finch]

that is wierd... sorry idk. If it was me I'd probably just delete kohya and reinstall it 🤷‍♂️

[azure sable]

thanks, I will be.

[deep canyon]

Does anyone know how to make this go above 1GB

ive tried setting the flags in Fooocus to always high vram which worked, but still only at 1gb

figured it out

[valid gorge]

I'm again struggling with InPainting + Lora (in A1111). Results are quite random, and with luck I get a seed seed that works but most of the time it does not. Literally pretty much hit and miss situtaion.

I've tried tens and tens of combinations of CFG, Denoising, Sampling Steps, Masked padding and having masked content in both fill and original . Feeling totally lost ugh, lol.

Not sure anymore if it's the settings, A1111 or if things actually are so randon. Now doing an XYZ plot (again) testing different values and combination WITH fixed seed.

Next stop giving up lol.

[valid gorge]

XYZ plot gave me something. Still need to keep on testing how much everything and anything is affecting to everything. Position of the planets need to be taken with the equation too.

[ornate elk]

How?

[deep canyon]

Go to \Fooocus\backend\headless\fcbh\model_management.py

change mem_total = 1024 * 1024 * 1024 to mem_total = X * 1024 * 1024

replace X with VRAM quantity in MB

should be just after line 96

[ornate elk]

Nice thanks!

[deep canyon]

np

[ornate elk]

if you want to use Auto1111 with directml for AMD, then checkout my install Guide in the Pinned Messages of this Channel,
it has much more features than foocus

[deep canyon]

what makes it different?

Fooocus is the first "AI" tool ive utilized

still no ROCm for my stupid XTX lol

[fair oxide]

hi @ornate elk i had a little net issue yesterday and couldn't stay online longer to ask few more quesitons about face id, but about the things you said about putting bin files in controlnet/models folder and lora files in models/lora folders, you'd then have to include loras like ip-adapter-faceid_sd15_lora into the positive prompt?

[ornate elk]

Everything, Auto1111 was the first webui and still has the best support for any models and community extensions that are not available on Foocus.
Comfyui is also nice but for the more experienced users.
Foocus only can use SDXL models. these are very large and slower on directml rn, and Foocus doesnt support community extensions afaik.
you have much more freedom in auto1111.
foocus is for people who search a "Midjourney like experience"

[ornate elk]

i have the 7900xtx too ^^, on linux it would be much faster, but for me it works good in Auto1111 to

[deep canyon]

when you say community extensions, do you mean something like LoRAs?

or checkpoints

[ornate elk]

no i mean plugins with more features, like upscaling, video animation stuff, UI enhancements, etc

[deep canyon]

you can do a decent amount of that

I know you can add your own upscalers

[ornate elk]

Loras work in auto1111 too

[deep canyon]

ill check it out

[ornate elk]

controlnet for example is an essential extension

[deep canyon]

what separates Comfyui from Auto1111?