🤝｜tech-support

[worthy portal]

yes

[ornate elk]

If so change --no-half-vae to --no-half and it should work

[worthy portal]

ok i try

[blissful quarry]

Is there a controlnet that's good for capturing bodyfat? Or do I need to get a better mannequin with better shadowing to show body proportions, compared to the one I have? I'm trying various control nets (openpose, depth, normalBAE), but they kind of miss the body fat from this reference since it is only vaguely alluded to with shadow.

[worthy portal]

Now I see this
.
OutOfMemoryError: CUDA out of memory. Tried to allocate 40.00 MiB (GPU 0; 4.00 GiB total capacity; 3.35 GiB already allocated; 0 bytes free; 3.44 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
Time taken: 2.8 sec.

A: 3.39 GB, R: 3.44 GB, Sys: 4.0/4 GB (100.0%)

[stone hawk]

https://civitai.com/models/240700/perfect-thick-ass I can't get this Lora to work, it has 0 effect on images. Can anybody get it to work, and tell me how? (slight nsfw)

[ornate elk]

Okay what are your txt2img settings like resolution etc?

[worthy portal]

Generation
Textual Inversion
Hypernetworks
Checkpoints
Lora
Sampling method
DPM++ 2M Karras
Sampling steps
20

Hires. fix
▼
Refiner
▼
Width
512

Height
512

⇅
Batch count
1

Batch size
1

CFG Scale
7

Seed
-1
🎲️
♻️
Extra
Script

[ornate elk]

Seems normal

Have you tried The Animatrix model?

Also make sure your GPU driver is updated

[worthy portal]

Yes, but I still can only make images with a resolution of 400 and below

[worthy portal]

YES

[ornate elk]

Can you restart SD and screenshot the cmd window?

[worthy portal]

ok

[ornate elk]

From an other comment of that lora it seems like its broken and wont work. Falsely trained I guess as its only 4mb to

[worthy portal]

this is sceenshot
https://drive.google.com/file/d/1FGWFX4kLQLeXS1dWLm7g3XRaAcf4vW8S/view?usp=drive_link

[stone hawk]

I see, shame. Wonder where the ratings come from

[ornate elk]

Cant check that there, you can make a screenshot with Win+Shift+S
And then paste it here with Ctrl+v

[worthy portal]

i know
but i cant upload

[ornate elk]

You can upload here

[worthy portal]

upload failed
something is wrong
this shown

[worthy portal]

It's not a malicious link, it's just a link to my Google Drive

this is all in CMD
.
Already up to date.
venv "C:\AI\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.7.0
Commit hash: cf2772fab0af5573da775e7437e6acdca424f26e
Launching Web UI with arguments: --xformers --medvram --no-half
Style database not found: C:\AI\stable-diffusion-webui\styles.csv
Loading weights [5f6f1556d1] from C:\AI\stable-diffusion-webui\models\Stable-diffusion\animatrix_v20.safetensors
Running on local URL: http://127.0.0.1:7860

To create a public link, set share=True in launch().
Creating model from config: C:\AI\stable-diffusion-webui\configs\v1-inference.yaml
Startup time: 29.4s (prepare environment: 9.0s, import torch: 6.7s, import gradio: 2.8s, setup paths: 4.9s, initialize shared: 1.1s, other imports: 2.2s, setup codeformer: 0.2s, load scripts: 1.2s, create ui: 0.7s, gradio launch: 0.7s).
Applying attention optimization: xformers... done.
Model loaded in 13.7s (load weights from disk: 1.1s, create model: 0.7s, apply weights to model: 4.9s, apply float(): 0.7s, calculate empty prompt: 6.2s).

[ornate elk]

Okay looks all normal. Then you can try to change --medvram to --lowvram in the webui-user.bat

[worthy portal]

Is there a way to use both cpu and gpu to process data?

[ornate elk]

Nope not at the same time

[worthy portal]

It produces the image, but it gives this message in the last step, which is to show the image
Of course in 1080 px
OutOfMemoryError: CUDA out of memory. Tried to allocate 1.11 GiB (GPU 0; 4.00 GiB total capacity; 2.56 GiB already allocated; 0 bytes free; 2.60 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
Time taken: 3 min. 1.1 sec.

A: 2.85 GB, R: 3.38 GB, Sys: 4.0/4 GB (100.0%)

is there any --verylowvram ? 🙃

[worthy portal]

If you give me just a little bit of guidance, my problem will probably be completely solved
Thank you for your time

[woven smelt]

Hello im not sure if to ask this here or not but what should I use to control shots similar to this image, it seems random at the moment where the focus of the image is. Someone suggseted latent layer camera but I need it for XL thanks

[ornate elk]

oh, you cant to 1080p with your gpu. 1.5 models are also trained on a resolution of 512x512.
Better try 512x768
You can then later upscale the image in img2img with the sd upscale script

or you can also try use hires fix with these settings:
512x768
30 steps,
cfg 7
Hires Fix:
Hires steps: 10
Upscale by 1.5
denois: 0.5
Upscaler: Resrgan4x+

[ornate elk]

Maybe these will help:
close up shot
close up
close up portrait
portrait
cowboyshot
full body
wide view

[woven smelt]

ill give them a go

[worthy portal]

Thank you
you did me a favor
These settings are enough for me, because I can produce images with 1000 pixels, I do the rest with img2im or other software.

[vocal burrow]

you might want to use Tiled VAE then

[worthy portal]

what's the Tiled VAE ?

[ornate elk]

tiled diffusion with tiled vae is an extension that can help to not get the out of vram error when using hires fix or highres resolution

[worthy portal]

How can I install it?

[ornate elk]

go on the Extensions tab, click on Available, click on Load from:, then search for tiled diffusion, click install, and then restart the webui

then you have a tiled vae option you need to enable, you dont need to activate tiled diffusion

then you can try use Upscale by: 2

[worthy portal]

Thank you very much, I am installing

[worthy portal]

how can i enable ?

[ornate elk]

In txt2img you should see the Tiled VAE at the bottom

Open it and click enable

[calm cairn]

I have a question that (probably) someone knows the answer to
I'm looking at a model on CivitAI and it's only got one version (which is v1.0)
but then when I click download, there are 2 options
What is the difference? Should I already know this?
thanks in advance

[ornate elk]

hey, the large one is first trained model, the smaller one is the "pruned" model.
its fp16 and works faster on RTX and AMD cards.
It doesnt affect quality which model you use. both will get slightly different resuslts. So if you want to save space go for the 1.99gb one

[calm cairn]

ah thank you, I didn't realize size difference for pruning bad images would be that large
Thanks!

[woven smelt]

I can upsacle using hirex.fix or in img 2 img. Is there any difference in doing on or the other? If I use hirex 2x then img2img 2x will that be the same as img2img 4x?

[ornate elk]

its nearly the same, hires fix works better so if you can use that for the first upscale.
sd upscale script in img2img uses less vram so that works better for low vram cards or when going for 4k resolution

[woven smelt]

Ok thanks im more thinking about my process. I am geneating images at low res then picking the best and wanting to upscale those. Its easy if I can do that in img2img script. Using txt2img id have to import them back in one at a time with pnginfo , no batch possiblity

[ornate elk]

yea you can easy do batch upscaliing in img2img

[woven smelt]

can I apply adetailer to them in batch processing too?

[ornate elk]

i didnt tried but upscaling will fix the face anyway

[woven smelt]

On anohter note ive been using 0.3 denoising fro upscaling, .9 when inpainting from noise and .7 when inpainting from original, do those sound reasonable?

[mental locust]

Any hero up for taking a look at this comfyui workflow? Here's the latest iteration that works, and then I just tried plumbing in 'RAVE' and after triple checking I have it matching the example layout am still only getting static - just bypass the unsampler and stuff before it and you'll get the proper results :/

[woven smelt]

Thats what a rave look like when I go to one

[mental locust]

Try lowering the dosage and you'll see the other attendees...

[ornate elk]

yea if they work for you its good, for faces i inpaint at 0.5, upscaling with 0.2

[ornate elk]

whats the vae and smodel?

[mental locust]

I've been working with bb95furrymix and vae-ft-mse-84000-ema-pruned - the issue is that with most other models I've tried, the workflow creates great video to video, but with bb95 there is a very significant difference between the quality with and without animatediff. Without animatediff the image quality is great, but obviously there is no consistency. With animatediff the animation is smooth, but the image quality drops drastically. I have been talking to a lot of people and had a lot of great advice: I have put in place or had help with: modelsamplingdiscrete and rescalecfg, freeu_v2, freeinit, and the whole workflow is based on one I downloaded from youtube and modified thanks to another that had pieces I wanted (the removebackground).

TLDR: When animatediff with bb95, image quality goes bad. Latest workflow JSON without the rave components is attached here:

[ornate elk]

And its only with that model?

[somber halo]

good morning guys, does anyone know how to insert a specific image in a specific point through inpaint (even in a different way)?

I need to import a logo where I marked

[crimson leaf]

i have 5gb vram, should i use --med or --low?

auto1111

[ornate elk]

go with --medvram

and --xformers

[crimson leaf]

thanks

[ornate elk]

np

[azure sable]

I’m having a strange issue with inpaint. It keeps cutting off parts of the inpainted image, such as giving the person half of a head, missing limbs, weird proportions etc. I have two computers, and it’s only happening on one of them, and I’m using the same exact settings and model to inpaint on both of them. Is there a reason for that?

It also inpaints faces really big onto the face, completely ignoring the proportions of the image. I’ve already tried doing a complete reinstall of the SD A1111, and it didn’t fix the issue.

[ornate elk]

do you have examples of the broken image + the img2img settingd you used?

[azure sable]

Yes, give me a few minutes and I’ll get on my computer and get an example for you

[azure sable]

Here is an example I made just now. Steps: 35, Sampler: DPM++ 2M Karras, CFG scale: 7 Denoising strength 0.7, Mask Blur 4. Before inpainting a person I make a quick generation using Photoshop first just to put a person there so that inpaint has something to work with, Or I do a quick sketch if I'm trying to inpaint an object. You can see the head is partially cut off and so is one leg and both arms.

if I create the same image on my other computer, this doesn't happen, except maybe every once and a while but I can regenerate and it's fixed. With this computer, no amount of regenerations changes anything, it always gets cut off in one way or another.

[ornate elk]

thats strange, what you should do is to do exactly the same on both images with both PCs, then inpaint the same stuff in inpaint. then compare both Meta data of the output images to check for a difference.
Does both PCs use a nvidia gpu?

[lofty mauve]

Cassie what is your setup? Which models do you have running on your machines?

[azure sable]

Yes, both are Nvidia. One is a laptop with a 3060 and the other is a PC with a 4080. The PC with the 4080 is the one having the issue.

[ornate elk]

And if you go for a higher Denois? Still cut off?

[azure sable]

I’m using Realistic Vision 5.1 to generate these images but if I change the model it doesn’t seem to make any difference. I’ve tried most of the photorealistic models on civit.ai.

[azure sable]

I will try.

[azure sable]

bumping up the denoise doesn't seem to change anything, in fact, this is what happened. amusing but not any better lol

[ornate elk]

whats the prompt?

[azure sable]

I tried pressing generate a few more times and even when it does create full body pics, they're still cut off at the edges. I usually use pretty simple prompts, this is the generic one I used: "a man posing for a photo in a park, blonde hair, t-shirt, jeans, full body, "

[ornate elk]

also is the mask exact on the person or does it go over the surroundings of him too?

[azure sable]

I go slightly over the edge of the person so it learns the background a bit. should I be trying to be very exact?

[ornate elk]

nope, it should have the chance to blend over

[ornate elk]

do you use inpaint area "only masked"
and then set the resolution to 768x768?

[azure sable]

I do use only masked mode. I will have to create a new image from scratch as I don’t usually make square pics. I can try making one 768 by 768 now.

[ornate elk]

no no

you dont need to make square pic, but when you inpaint you need to set the resolution sliders to a square resolution

like 512x512 or 768x768 for larger images

[azure sable]

Oh ok I understand what you’re saying.

I will try that.

[ornate elk]

inpainting is like generating an image into an image

so you should go for the base trained model resolution

[azure sable]

I just tested that and it fixed it. No cut off parts. Weirdly I never used square resolutions on my old pc and never had that issue! Thank you so much.

[ornate elk]

no problem 🙂 yes you can be lucky if the inpaint resolution tiles equaly in the input image resolution, but to never get this the sqaure is the solution

[kindred tinsel]

Oh well I am running into problems with tiled diffusion here :c

:)

[ornate elk]

your gpu needs --no-half

[kindred tinsel]

so it did affect somehow I knew it lmao, let me add it some time later :c I closed the server and rn Im running comfy

Im downloading two clip models for comfy once those are done Ill get back to you, mind I ping you if I get any more errors?

[sturdy ginkgo]

ugh..updated ComfyUI and all custom nodes and it is now generating at like 60% the regular speed >_<

[kindred tinsel]

there's a comfy update?

[ornate elk]

sure, then you should try a 512x768 image, 30 steps, no sdxl model,
enable hires fix with following settings:
Upscale by: 1.5
Hires steps: 10
Denois: 0.5
Upscaler: Resrgan4x+

Then enable Tiled Vae and set
Encoder: 1024
Decoder: 128

Make sure you use a 2gb model, and your prompt token count needs to be below 225.
And then try to generate.

[kindred tinsel]

Okay? 2gb model right, I think the one I was using works for that, so tiled vae and hires fix do not go at the same time?

Or is that just for the test?

[ornate elk]

no, they need to be used at the same time, thats why i said Try to generate, as the last step

[kindred tinsel]

oke Ill take a look at it, webui is launching rn./

Also do you know why the entire webui is slow as hell, like feels sluggish and heavy to move around it, changing windows takes a ton of time stuff like that ?

[ornate elk]

no, but what browser do you use?

[kindred tinsel]

ms edge

[ornate elk]

hmm

if you can try in firefox ^^

[haughty pewter]

does anyone know how to adjust something like this so there is less change between images? more of a morph that abrupt shift?

[kindred tinsel]

really? the browser can be the problem? I even disabled gpu acceleration :c maybe that?

[ornate elk]

a well that can cause the slugishness too

but only if the cpu is stressed

[kindred tinsel]

is thios stressed? idk

Also I need another minute I forgot to put the no half in the launch commands

[sturdy ginkgo]

more frames?

[kindred tinsel]

Look, once the webui opened, like the new window form the auto open command this happened :=

[haughty pewter]

that seems to just increase the speed, Im trying to figure out how to reduce the shift between each individual image

[kindred tinsel]

@ornate elk Here is the output of the model with clearvae and it died in the upscale

[hallow latch]

How do I run this?

[kindred tinsel]

So it did this, and then failed, im not even sure what just happened :c

[kindred tinsel]

You don't really have to worry about pip unless you use python for work and the like.|

But if you want to update you should head to the folder that it is indicating and xecute that command

python.exe -m pip install --upgrade pip

[hallow latch]

I only use it for Stable Diffusion 😄

[kindred tinsel]

then its not really an issue, I get that message too when I install something with pip but not something too big to worry about AFAIK

[sturdy ginkgo]

can just ignore it then 😄

[kindred tinsel]

Im trying again but this time lowered tile size, lets see if that works out

[ornate elk]

it failed with the out of memory error?

[kindred tinsel]

ye[

I lowered the tiles size

it failed again !

Got to here and failed

[ornate elk]

if it failed ones you always have to restart SD

if not then it will fail always

[kindred tinsel]

you serious?

[hallow latch]

How can I install another UI?

Sorry Theme*Ä

[ornate elk]

thats how it is, not joking

[kindred tinsel]

so I have to sit here for 5 more minutes to open it up again?

Im going to check it in a minute, let the it im running rn run

or not, yeah Imma just go back to literal comfy, this is too much bs, comfy never fails on upsclae like this, or even if it does it allows me to pick up

[ornate elk]

wait

[sturdy ginkgo]

https://github.com/lobehub/sd-webui-lobe-theme

[ornate elk]

@kindred tinsel you said you use ultimate upscaler on comfy

thats also available in auto1111

[kindred tinsel]

yeah, there it works wonders, tiled vae and everything

[ornate elk]

yea that will work in auto1111 for you too

as it doesnt get out of vram

its in img2img tab

[kindred tinsel]

how do I install it?

[hallow latch]

How do run the git clone?

[ornate elk]

you either can use the sd upscale script or you install the Ultimate sd upscale extension

[hallow latch]

From the extensions tab ?

or from a bat file?

[ornate elk]

you can search i nthe extenion list for it

[sturdy ginkgo]

from the extensions tab works fine

[kindred tinsel]

some come from the internet external githubs, latent couple runs on a separate extension haah

alright let me try once more to get this thing to work

[hallow latch]

I ran it in the default stabble diffusion folder.

Wait

I think I got it to work

Yeah

[kindred tinsel]

this thing?

[ornate elk]

Thats how you install extensions:

[ornate elk]

yes, if you use this you need to set the denois below 0.3 and the resolution to 512x512.
for the Upscaler try Resrgan4x+ ANime6b

[kindred tinsel]

where do I set that?

you got me confused

Should I generate first and then send it here right?

[ornate elk]

yes

first generate an image in txt2img without hires fix

then click Send to img2img

then set the resolution to 512x512

denois below 0.3

activate sd upscale script and set an upscaler

[hallow latch]

Yo, how do I disable this?
AssertionError: extension access disabled because of command line flags I try to run https://github.com/Mikubill/sd-webui-controlnet.git to install Controlnet

[kindred tinsel]

wait why so small for the resolution? Should I put like upscale by two or smth

[hallow latch]

I do 512x512 to find something I like

[kindred tinsel]

--enable-insecure-extension-access or smth like that in the bat file

[hallow latch]

Then I run the Hires. Fix

[ornate elk]

thats the Tiled resolution not the output res

[ornate elk]

you need to remove --listen or --share

[hallow latch]

Ohh :/

[ornate elk]

dont ever use enable-unsecure-extensions

[ornate elk]

you can add it back after the install 🙂

[hallow latch]

For sure

[kindred tinsel]

@ornate elk

I would like to report that this script is generating images on its own, with different poses, and the like

and Im assumingf that it will mix them up like an idiot

I do not like what this script did

[ornate elk]

then you didnt have set the denois below 0.3

[kindred tinsel]

the webui sent the prompt too, shoiuld I remove it?

[ornate elk]

you can, but its only needed if the denois is to high

if you go for 0.2 you can leave it

[kindred tinsel]

im testing 0.2, imma see if it comes out at least acceptable

Disregard the weird ass generation, this model tends to want to generate two girls even when prompted for one, and it can't fit them

but it looks better man thank you :)

[ornate elk]

no problem 🙂

[sturdy ginkgo]

got the same issue with non square aspect ratio's on every single generation with any model 😄 .. always a extra character

[ornate elk]

with any model? then the resolution was to high

[sturdy ginkgo]

768x512 for instance?

[ornate elk]

that would be okay for 1.5 models

[desert garnet]

what is generally better, anything V3 or aingdiffusion?

[stiff cobalt]

if anyone is using automatic webui im having an issue just now where i searched for 1 lora using the search bar but now it only displays like 7 of my loras. i have 50 installed

[sturdy ginkgo]

what does the command prompt say?

maybe an extension bugged

[ornate elk]

select an 1.5 model and refreh the loras

[stiff cobalt]

ty, lol

[mental locust]

Examples.

[untold marlin]

Okay sorry if this is a dumb question or has been asked 100 times but I am at a loss here. I have been watching videos on using Loras and have followed said videos to the letter but where they show the Loras showing Up in the LORA tab mine is still blank. Anyone know how i fix this?

[mental locust]

Did you put the lora in the lora folder? Did you turn stablediffusion off and on again, including closing the command prompt?

[untold marlin]

@mental locust yes and yes

@mental locust this is the folder i placed them in D:\StableDiff\stable-diffusion-webui\models\Lora

[sturdy ginkgo]

can you link the lora's you have tried

[untold marlin]

fixed it i feel so dumb

was not working with the checkpoint so it hide them lol

sorry i am new to this stuff

[sturdy ginkgo]

ah .. you were using a SDXL with a SD1.5 lora 😄

or vice versa

[untold marlin]

yea lol

[mental locust]

Do you have any tricks for managing this more easily? I have an absolute ton of loras and models and like... is there an idiot proof organisational technique?

[sturdy ginkgo]

i use comfyUI.. and i can put them in subfolders .. so in the nodes they say SD15/nameoflora.safetensor

not sure if A1111 can do that

[mental locust]

EPIC so obvious so easy. Thank you

[untold marlin]

ComfyUI that is the node based one right?

[sturdy ginkgo]

yes

[mental locust]

Hey can you tell me the filepath for CLIP models in ComfyUI, and where to download them? I'm coming up blank.

https://huggingface.co/openai/clip-vit-large-patch14/tree/main is all I can find

[sturdy ginkgo]

ComfyUI\models\clip

if for ClipVision it will be going in Clip_Vision

[mental locust]

again, obvious, thank you. I was looking in custom nodes and then in a1111 (as i link to models there)

ugh what have I done wrong this time. any help appreciated. thank in advance.

[sturdy ginkgo]

the clipvision model is not meant to be for the IPAdapter model used

https://github.com/cubiq/ComfyUI_IPAdapter_plus you can find the correct ones here

[mental locust]

Thank you

[open raven]

how do i instal gfpgan the faces are fucked

[celest fern]

How do I get the window on the right side of comfyUI to become locked, unmovable, and take up the entire column? I somehow made it like that while failing to type into a text box, but no amount of button mashing has been able to enable that again. Here's the image that shows what the windows looks like now:

[onyx solar]

Zoom your browser up to 120% or so.

[celest fern]

That makes the words way too big

[onyx solar]

The only other way I know of is to shrink the window down until it does the same thing, but then you have less room to work.

[celest fern]

Ahh, alrighty. Thank you!

[celest fern]

I don't know if there's a hotkey for it or not, I might've just accidentally increased the browser magnification back then, but regardless, I managed to do it with some CSS and a browser extension called stylus that enables custom css on websites.
https://www.reddit.com/r/comfyui/comments/15yd0vw/theme_the_comfyui_menu_and_fix_some_annoying/
For reference, here's the post I used that brought me to the fix, and here's the code for the userstyle: https://codefile.io/f/seArZMvMyP
Now all that's left is to modify it until I think it looks cool lol

[onyx solar]

Well yeah, you could adjust the css on the page, sure.

[crystal nacelle]

hey people. ive had multuiple problems while installing but managed to resolve alot of them and get stable diffusion running

Im using an AMD RX7600 8GB and i get this error when generating an image:
-RuntimeError: Could not allocate tensor with 9831040 bytes. There is not enough GPU video memory available!
Am using --use-directml --reinstall-torch --disable-nan-check --precision full --no-half --medvram

[lavish meadow]

is a1111 checkpoint merger good or nah

[sturdy ginkgo]

decent, there are better extensions for it with more control.

[lavish meadow]

i see

does it blend the two model styles?

im merging cetusmix and counterfeit

[mental locust]

ClipVision for video, or only for image?

[jaunty sphinx]

yo guys

how do i bypass this error bruvs

File "D:\Stable Diffusion Open Vino Version\stable-diffusion-webui\venv\lib\site-packages\diffusers\pipelines\stable_diffusion\convert_from_ckpt.py", line 1670, in download_from_original_stable_diffusion_ckpt
pipe = pipeline_class(
TypeError: StableDiffusionXLImg2ImgPipeline.init() got an unexpected keyword argument 'safety_checker'

i can see it in the python file

[crystal nacelle]

what gpu you using?

[jaunty sphinx]

intel xe graphics

with open vino

i just need the safety checker off

i see it in my python text file

but should i edit it?

to be true or false

safetychecker =

on line 1670

or do i edit safetychecker python file

thanks bruv

if you answer and i don't reply

[honest storm]

i've used chainner for upscaling so this might sound dumb
when using hires fix
do i have to use it when generating every image or can i apply it after generation

[ornate elk]

Hey, remove --reinstall-torch, --no-half --precision full --disable-nan-check

And add:
--opt-sub-quad-attention --opt-split-attention-v1 --upcast-sampling --no-half-vae

[ornate elk]

You dont need to use hires fix the whole time. If you like an image, load it into PNG-Info tab, then click on send to txt2img, then it will load everything up and you only need to set the hires fix settings.
Then you can recreate it

[cursive hazel]

Heya! I have what I think is probably a simple question, if anyone can help me out.

How do I stop Automatic1111 from opening itself up in a browser, when I launch the run.bat that starts it all up? It used to not do this, but a little while ago I did a fresh install, and it started to do this. I just don't like it automagically opening my browser when I start it up, as I actually use a different browser for playing with AI.

[ornate elk]

Hey, there should be a setting for this in auto1111 settings

[cursive hazel]

Ah! Thank you. I should have looked there
It's under Settings -> System, and it's like a 3-check box thing: Automatically open webui in browser on startup, with the options being Disable, Local, Remote
I have no idea what Remote does, but Disable sounds like exactly what I'm looking for!

[wintry tide]

hey there! I've tried to prompt images a few times, but sometimes it will prompt out videos instead and I'm not sure why it will suddenly do that.

[hollow schooner]

Is there anyway to calculate what resolutions my graphics card is able to make

Cause i just found out my graphics card is actually capable of making something higher than 512/512 without hires steps

[celest fern]

Here's the end result. Never actually fixed my issue with figuring out how to make it go to the bottom and stay where it's at by keeping everything at its defaults, so I spent a few hours on this instead.
https://github.com/SampleTexting/ComfyUI-Userstyle/tree/main

[celest fern]

Best way I've found is to just keep upping it until it fails, then go down to the last one that worked

[hollow schooner]

The hires steps or the width and height

[celest fern]

Both, try one until it fails, then try the other until it fails, then afterwards try em together until they both fail.
I've got 8gb of VRAM and I can pump out 1920x1080p images no problem using both, or either one of them alone, if that helps you any. (haven't really went any higer than that since it takes a solid minute or so)

[hollow schooner]

Huh that’s so weird cause i have 8gb of vram but I’m using amd

[ornate elk]

whats your gpu?

[celest fern]

It's likely because you're using AMD then if I had to guess. NVIDIA has cuda cores which speeds the process of AI applications up a lot thanks to the way it handles processing

[hollow schooner]

Amd Radeon RX 6600

[ornate elk]

and you use --medvram --opt-sub-quad-attention --opt-split-attention-v1 --upcast-sampling --no-half-vae?

[hollow schooner]

Yes

[ornate elk]

okay, then you can try to change --medvram to --lowvram

[hollow schooner]

Hm what does that do?

[ornate elk]

that splits heavy vram processes into smaller ones, meaning you can generate at higher resolution, but a bit slower

[hollow schooner]

Ohhhh

Okay done

Don’t mind the wait usually put it in auto mode

Okay thanks will try this out

[mental locust]

@ornate elk heya, I did more work on my workflow and got ipadapter going with it as well as freeinit, updated from bb95furryv11 to v13, - am I pursuing an impossible task? is this model simply not suitable for use with animatediff? I ran a bunch of models through animatediff to see which it liked or didn't like...

[ornate elk]

hey, it seems so, not every model is compatible with everything, some merged models can behave strange with certain samplers, upscalers etc, so i guess its the same here with animatediff

[mental locust]

Guess I will... learn how to make a model. 🤣

[junior snow]

------------------------------ --------- 2.0/2.6 GB 11.7 MB/s eta 0:00:56ERROR: Could not install packages due to an OSError: [Errno 28] No space left on device

 ------------------------------ --------- 2.0/2.6 GB 11.7 MB/s eta 0:00:56

Traceback (most recent call last):
File "C:\Users\erenu\OneDrive\Masaüstü\Yeni klasör (4)\webui\launch.py", line 48, in <module>
main()
File "C:\Users\erenu\OneDrive\Masaüstü\Yeni klasör (4)\webui\launch.py", line 39, in main
prepare_environment()
File "C:\Users\erenu\OneDrive\Masaüstü\Yeni klasör (4)\webui\modules\launch_utils.py", line 378, in prepare_environment
run(f'"{python}" -m {torch_command}', "Installing torch and torchvision", "Couldn't install torch", live=True)
File "C:\Users\erenu\OneDrive\Masaüstü\Yeni klasör (4)\webui\modules\launch_utils.py", line 116, in run
raise RuntimeError("\n".join(error_bits))
RuntimeError: Couldn't install torch.
Command: "C:\Users\erenu\OneDrive\Masaüstü\Yeni klasör (4)\system\python\python.exe" -m pip install torch==2.0.1 torchvision==0.15.2 --extra-index-url https://download.pytorch.org/whl/cu118
Error code: 1
Press any key to continue . . .

Got that error during download

what should i do

[ornate elk]

it says you have no space left on C

ake sure to have 15-20gb on C, even if you install it on an other Drive

also pls dont install it inside a Onedrive folder

[junior snow]

i was downloading to D.Why this error happened

[ornate elk]

because the files need for the install will be temporary downloaded to C

[vestal viper]

it seems that some of the models i downloaded are incomatible with AMD. some says "vram not big enough" or something, others straight up saying "no cuda no bueno" so is there anything i can do, or at least choose which models are ok for AMD users?

[ornate elk]

all models work for AMD rn, getting and out of vram error isnt due incompatibility

what was your gpu again?

[vestal viper]

rx 6600

[ornate elk]

did you tried to use an SDXL based model (6gb) ?

[vestal viper]

and yes i forgot to ss it but one of the modes did actually say "cuda not detected" and aborted the generation

[ornate elk]

okay, which model is that?

[vestal viper]

i have no idea, i just followed your guide and links

[ornate elk]

you can check the model sizes in the models/stable-diffusion folder

with 8gb vram you should stay with the 2-4gb models

[vestal viper]

ah so 4.99gb ones are no good?

[ornate elk]

they can work too, but then try a resolution of 512x768 max

if you can always go for the lower size

some models have multiple versions in different sizes

[vestal viper]

what about pruned models then, sometimes they put 2 files one full model which are usually 5gb and one pruned that's within 2-4gb

[ornate elk]

yea go for pruned ones

[vestal viper]

alrighty. but are there any differences?

[ornate elk]

they got unneeded stuff removed

the quality is unnaffectet

[vestal viper]

nice. thanks as always sir 🫡

[ornate elk]

np, the pruned will get slightly different outputs then the "full" versions, but not quality wise

at the beginning of SD all we had was 7-9gb models xD smallest one were 5gb

2gb is the smallest model size we can get now

[vestal viper]

oh but how big can i generate the pics with those pruned models? i wanted to make a wallpaper for my desktop but realized ain't no way now i'm gonna do that with 8gb vram

[ornate elk]

therefore you need to use the sd upscale script in img2img

it wont run out of vram and you can get 4k images, if you then load them into Extras tab you can upscale to 8k too

[vestal viper]

ah, ok. so i generate the original, then upscale it. thanks

[ornate elk]

yes, if you need any tip how to use the sd upscale scipt or the ultimate upscale extension, feel free to ask

[steady mountain]

Anyone knows how I can get my XTX running on my 22.04.3. Automatic1111 just ran on CPU after an install. I tried to reinstall the amdgpu-install_5.5.50503-1_all.deb but it did not work (probably because I mistakenly installed the driver for 20.04.6 HWE/ LTS) but it is weird it treats the 20.04.6 HWE(LTS) as newer than the driver for 22.04.3 (they have the same release date but 22.04 is of course a more resent distro but I am not a techwiz so what do I know) and thus refuse to reinstall it usual way

[junior snow]

i have 5gb space , its not enough?

[ornate elk]

nope

15-20gb needed

[fallen zephyr]

Which extension is the SD VAE pull down?

[ornate elk]

none,
Go to Settings - User Interface - Quicksettings, there add sd_vae, then hit apply and reload

[fallen zephyr]

Thanks!

[lavish meadow]

Are there any extensions that would make my life easier in a1111?

[ornate elk]

boorutag autocomplete for prompting, controlnet for everything, adetailer for face fix, reactor for face swap

[lavish meadow]

got the auto complete thang, idk what the rest does

[ornate elk]

the rest is for generating

[lavish meadow]

would they stop "weirdness" from happening

[ornate elk]

whats the wierdness exactly?

[lavish meadow]

a foot from a knee

two belly buttons stuff like that

[ornate elk]

that also depends on the model version and resolution

[lavish meadow]

someones glasses on their mouth once

yeah my denoise is at 0.5 fixed most but not all

i want a specific pose and stuff but idk how to do that

[ornate elk]

for that you need controlnet

with that you can control the pose

but can you show me an example of the issues you get rn?

[lavish meadow]

lemme see if i still got the images i probably deleted them

my pc lagging rn cause im doing a high res image ill try n find them after it calms down

[ornate elk]

whats your gpu?

[lavish meadow]

3060

[ornate elk]

do you use xformers?

[lavish meadow]

yup

someone told me to add it so i did

[ornate elk]

good

[lavish meadow]

idk what it does

[ornate elk]

but your pc is lagging?

[lavish meadow]

yeah generating

[ornate elk]

xformers enhances the performance and vram management

[lavish meadow]

its tuck at 2 seconds

stuck*

[ornate elk]

do you have any other heavy programm, 20 browser tabs or wallpaper engine open?

[lavish meadow]

wallpaper engine yes

20 tabs no

[ornate elk]

close and exit wallpaper engine while using Stable-diffusion

then nothing should lag

[lavish meadow]

ohhhh

that makes sense

i got a dog background where they munching on food

absolutely adorable mate

[ornate elk]

haha but uses gpu too

[lavish meadow]

[ornate elk]

you can even get crashes when both is running at the same time

[lavish meadow]

that probably explains the lag

[ornate elk]

yes i had it too

now i exit it before using SD and its fine

[lavish meadow]

ill keep that in mind cause i got on auto start up

[ornate elk]

yea same xD

[lavish meadow]

LOL

gotta ask is the model merge good or no

[ornate elk]

model merge is a cool feature, but i would say checkout the models on civitiai

there is mostly a model for everything

[lavish meadow]

i like counterfeits style

but cetus has better prompt resonence if u get it

[ornate elk]

you could merge them

and test

[lavish meadow]

akr have

tho its buggy

missing limbs

foot out of knee

a hand a foot

[lavish meadow]

@ornate elk it lags really badly before it ends

about 3s or 2s before its done but screws up

idk what is causing it

[ornate elk]

what are your txt2img settings?

resolutions, steps, hires fix etc

[lavish meadow]

lemme get a ss

[ornate elk]

and your using a 3060 with 12 or with 6gb vram?

[lavish meadow]

12

[ornate elk]

hmm okay and what error do you get in cmd?

[lavish meadow]

thats the thing

no error

it says 100% done in cmd

but page says otherwise and the image dosent go to anywhere

not the txt2img folder

just pops outta existence

[ornate elk]

whats your browser?

[azure magnet]

help, how to install this https://github.com/Coyote-A/ultimate-upscale-for-automatic1111 ?

[lavish meadow]

opera

gx

[ornate elk]

operagx can be an issue

[lavish meadow]

what should i try?

[ornate elk]

try edge for testing

[lavish meadow]

will do

[ornate elk]

better would be firefox, but edge is preinstalled on every windows pc

[lavish meadow]

yeah

was thinking of switching to firefox but cant be assed

[ornate elk]

using firefox my whole life, private and for work, best browser so far ^^

[lavish meadow]

truee

i like customization

but could the merged model be also contribuating to it

?

[ornate elk]

i dont think so, you would get an error when using it

[lavish meadow]

weird

[ornate elk]

not when the image is already done

[lavish meadow]

now my pc is lagging again

[ornate elk]

did you closed wallpaper engine? xD

in the tray menu

[lavish meadow]

yup

idk what is doing it

[ornate elk]

check the taskmanager

and is ypur gpu driver updated?

[lavish meadow]

i believe so

its maxing my gpu

no way

rpic games launcher was doing that

[ornate elk]

xD

[lavish meadow]

wth??

now its nice and smooth wtf

[ornate elk]

yea epic games launcher has some issues xD

[lavish meadow]

that caused the issue...

it immeidantly worked after

[ornate elk]

good to know

[lavish meadow]

yeah

now if anyone got issues

just ask

"is your epic games launcher open?"

[ornate elk]

xD

yea goes on my list to the wallpaper engine

[lavish meadow]

never would of thought it was the launcher

like how the hell do they make a launcher that bad

[ornate elk]

better not have epic in autostart

[lavish meadow]

i do

I like free games

[ornate elk]

yea but no need for autostart

[lavish meadow]

yeah just turned it off

had it on cause i couldnt be assed to turn it off

gotta be honest with that one

[ornate elk]

😄

[lavish meadow]

now it loads smooth

[ornate elk]

now SD should work stable

[lavish meadow]

aye it also solved some problems i had with generation

[ornate elk]

does the image gets now saved in the outputs folder?

[ornate elk]

ah alright

[lavish meadow]

still some problems

but not too bad

btw how do i add control net to a1111?

[ornate elk]

via the extensions tab

[lavish meadow]

got the link to it?

[ornate elk]

click on availavle, then on load from

then search for sd-webui-controlnet

and click install

[lavish meadow]

d-webui-controlnet

https://github.com/Mikubill/sd-webui-controlnet

this?

[ornate elk]

dont install it via the url

just do the steps i mentioned

[lavish meadow]

gotcha gotcha

"ValueError: unknown url type: 'sd-webui-controlnet'"

[ornate elk]

Click on extensions, click on Available, then on Load from:, then search sd-webui-controlnet

[lavish meadow]

dats wut i did

[ornate elk]

youve done it wrong

you removed the url

you should have just clicked the load from button

[lavish meadow]

what url should be in there

[ornate elk]

Load from: https://raw.githubusercontent.com/AUTOMATIC1111/stable-diffusion-webui-extensions/master/index.json

[lavish meadow]

i see what ya mean now

[ornate elk]

the bottom one is it

[lavish meadow]

ye

which one

top or bottom

i just saw what u said im blind dawg

so how does it work, is it easy to do?

[ornate elk]

you now need some extra stuff to get it work

also you need to click install and after its done you need to relaunch the webui

You need to download the Controlnet models from here:
https://civitai.com/models/38784/controlnet-11-models
Make sure to download each model + its config file (check the screenshot).

Then put everything into the models/ControlNet folder

[lavish meadow]

all of them dear god

why all of them?

[ornate elk]

no need for all of them, openpose, canny, depth, normal, tile, should be enough

if you have the space and want all stuff available then you need all

for making a pose you would only need openpose for example

for transferring a drawing into an image you would need canny, or sketch or anime lineart

[lavish meadow]

openpose would be what im looking for

so when i download it what do i do them?

then*

[ornate elk]

read my messages carefully xD

[lavish meadow]

dawg

Im slow okay

[ornate elk]

yea np, but i dont type everything doubled if not needed ^^

after you have moved them, you can relaunch the webui

[lavish meadow]

yes sir

relaunching it rn

it lauched again

what now

[ornate elk]

now you have a new tyb in txt2img at the bottom

ControlNet

there you select OpenPose

and click enable

then you have to drop in an image with a person in a pose

[lavish meadow]

gotcha gotcha

[ornate elk]

here is an example

you can use different types of the preprocessor

[lavish meadow]

what settings should i use for it?

[ornate elk]

openpose-full will use Skeleton+Face+Hands

it depends on what you want to do

but full is the best if the input image contains face and hand

[lavish meadow]

I see

so thats it?

pretty simple

[ornate elk]

yes

thats it

[lavish meadow]

thank you

your very helpful

gotta ask whats ur favourite model?

cause im tryna use counterfeit but have more well rounded prompts

[ornate elk]

here for example with the controlnet Depth model it makes a nearly identical image

[ornate elk]

i have a lot of models, counterfeit was good until he baked in the vae in v3.0,
i like AOM3, AnythingV5, CetusMix, Divineanimemix, ToonYou

[lavish meadow]

is there anything like counterfeit?

cause love the style but it is restrictive

[lavish meadow]

dark sushi i like

is it good with humans

[ornate elk]

yea its quite good

there is also the 2.5d darksushi mix mix

[lavish meadow]

Ooo ill check it out

thank you

also what does baked in vae mean

[ornate elk]

that the model already includes a custom vae

but counterfeit 3.0 baked in vae is oversaturated

[lavish meadow]

oh shit thats why some gens have been meh

i see i see

gonna try some dark sushi

also could i do a doodle

and control net would copy said doodle

like a stick figure

[ornate elk]

for that you need the controlnet model Sketch

[lavish meadow]

I see thank you

also what should i use here

[ornate elk]

first go for balanced

if the person doesnt follow the pose, you can say controlnet is more important

[lavish meadow]

I see

[ornate elk]

also here ive done some comparrison a while ago, open that in browser for better view

[lavish meadow]

I like more grit

if that makes sense

[ornate elk]

grid view?

[lavish meadow]

no more dark

[ornate elk]

ahh

[lavish meadow]

like dark sushi and counterfeit style

[ornate elk]

you can enhance dark images with the epiNoiseOffset lora

[lavish meadow]

realistic?

[lavish meadow]

ill try that out

[ornate elk]

there i used it

[lavish meadow]

That looks good

ill defo pick that up

[ornate elk]

also the add-detail lora ist a must

im afk now, have fun!

[lavish meadow]

thanks for your helo

help*

[ornate elk]

no problem 🙂

[simple tartan]

ok @ornate elk , I need your help. I attempted to go the ubuntu route last night to try out the linux version but got a wall of errors. before I try and fix that, can i, or more, should i roll into Arch? I'm planning SD, Ooba, basically all of my AI work on that drive

[signal cargo]

any idea how come on last step of image generation the image becomes screwed up?
xD
like some filter that goes over and makes everything layered with random color and overly sharp

[simple tartan]

from my experience, thats either a VAE issue or the size is wrong

[signal cargo]

VAE?

[simple tartan]

the part of your SD that does the color and quality stuff

A variational autoencoder (VAE) is a technique used to improve the quality of AI generated images you create with the text-to-image model Stable Diffusion. VAE encodes the image into a latent space and then that latent space is decoded into a new, higher quality image.

[main axle]

Hi, I am trying to create sdxl lora for product images

I have 6 product images with white background

But it struggles in color

After I finish training it sometimes generate different color products

does anyone can help me?

[rotund lava]

try to ask in #✨｜sdxl chat,theres a few ppl there that train loras and could help u

[signal cargo]

i forgot I added VAE for eyes

xD

ill try w/o it

yep

that fixed it

thanks

xD

[main axle]

Got it.
Thanks

[fair oxide]

how do i fix this issue

im installing automatic1111

[untold marlin]

I have a question. Once I generate a character i like for my visual novel how to i make it to where SD will use that same character for other prompts? The body shape and face will need to remain the same. Clothing might be different at times but kinda need the rest to stay looking the same.

[ornate elk]

That means you installed the webui the wrong way, by downloading it as zip file

Checkout the Pinned Messages in this Channel and follow my install guide

[ornate elk]

You can use controlnet IP-Adapter for this

[lofty mauve]

Is there a programmatic way to install hugging face models locally for Automatic1111 WebUI? I have been downloading them and manually moving them into the correct folder. Also, how about the controlnet models? I am very new to this and any sort of written guide will be much appreciated.

[simple tartan]

i think only civitai has an extension for webui, but i coudl be wrong

[ornate elk]

Here is a full guide for ubuntu and amd for Auto1111:
https://www.youtube.com/watch?v=NKR_1TUO6go

I would say try that. Arch needs some other commands but I've found some helpful guides for that too if needed

[ornate elk]

You can download into folder directly with the wget command.
Controlnet models go into models/controlnet

[lofty mauve]

I have a 4070 M gpu and it appears to be working as normal when I simply click the button. is that not a normal thing?

What is the difference beween a checkpoint and a model?

[simple tartan]

Thanks! it's not really a question of needing Arch, its just that is my preferred distro, but if its easier ot get support on a ubuntu build, why not save some headaches. If it's the same either way, i'd go arch

[lean laurel]

hey hi, its me again ❤️
sorry i m back again to bother, how should i install this model?? and how do i use it to transform a photo into digital art?

[ornate elk]

What button?
Checkpoint = model

[lofty mauve]

In the video he had 2 drop downs, one had a model label and the other checkpoint. Good to know that they're the same thing

My version only has "checkpoint" drop down

I am very new to this so I am learning as much as possible. I am a programmer though so it shouldn't be too difficult!

[ornate elk]

You also need some adjustsments for the best performance. (For a 8gb GPU)
Edit the webui-user.bat and at the line COMMANDLINE_ARGS=
you have to add: --xformers --medvram-sdxl --no-half-vae
Then save and relaunch

[lofty mauve]

why is that? What do those values do?

any chance you have a link to some docs explaining them?

[ornate elk]

Sure, give me a sec

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

[lofty mauve]

so how about an introduction to what I am even doing with this app, lol.

I am planning to make a meme photo of myself doing multiple roles with different hair styles and outfits so I am doing inpainting for this.

[lean laurel]

thats sound fun ❤️

[lofty mauve]

The meme is going to be a picture of me as a full stack developer, then another one as Product Manager, and another one as software architect, and another one as marketer, and another one as salesmen.

And I'll post it to my Twitter account for engagement and on my website.

I do custom SaaS consulting as a solo developer so it is accurate and funny.

[lean laurel]

i like the variety of the pics u gonna make

sorry i m still learning how install models

i wanna download this one

https://civitai.com/models/159094/final-fantasy-style
buts there 2 options to download

which one should i click??

[lofty mauve]

same here, I just downloaded one from civita.i. but I was trying to download them from hugging face

[lofty mauve]

click the down arrow

do the top one

[lean laurel]

oh we are at the same moment learning lol

[lean laurel]

the model safetensor??

[lofty mauve]

I had the same problem yesterday

I thought I downlaoded a model but I downloaded pictures or something

I couldn't figure it out and that's how I ended up here

https://civitai.com/models/76937?modelVersionId=103767

[lean laurel]

oh i see, i actually ended here for install all, i couldnt install SD without the help of this channel ❤️

[lofty mauve]

for example, the tutorial I followed on YouTube said to go here to download hair styles

so I went there and downloaded one to use and I still have yet to get it working like the dude did in the video. He skipped so many steps with his video editing

[lean laurel]

i dont like when the tutorials skips steps

[lofty mauve]

https://www.youtube.com/watch?v=kHZMzpw0Y2I

that's the video I tried following

[lean laurel]

ok after i download what should i do next??

[lofty mauve]

You have to put it in the right folder

what operating system are you on?

[lean laurel]

at least have a nice video cover lol

[lean laurel]

windows

[lofty mauve]

find the folder "stable-diffusion" in your programs installation folder and drag and drop it there

for example, mine has this

so I have 2 models

I'm on Ubuntu though, lol.

[lean laurel]

stable diffusion > models

into de models folders?

[lofty mauve]

yup

so find the stable diffusion ui folder then keep digging in there til u find models and then within the models folder there should be another one called Stable-diffusion

[lean laurel]

oh the stable difusion folders inside the models folders

[lofty mauve]

yup

[lean laurel]

found it!

[floral wolf]

Hi there,
Is the limit "This API is rate-limited to 150 requests every 10 seconds." connected to 1 account or 1 API key on the account? In other words, As I can generate 10 API keys on my account, can I have 1500 requests via 10 api keys from 1 account?"

[lean laurel]

hi ❤️

ok i moved it, now should i open the webui-user file??

[ornate elk]

Check the model size. It needs to be larger than 1.98gb

Before you move it into models/stable-diffusion

[lean laurel]

it must be heavir than tat?

the one i downloaded and moved to my folders size 36 mb!

thats bad??

[ornate elk]

Did you made the performance improvement?

[ornate elk]

Then thats not a model. Its a so called Lora

Goes into models/lora

[lean laurel]

ohhh ok ok

but i opened the page and it show the options, so i close the page and the cmd and delete it?

[ornate elk]

What do you want to do?

[lean laurel]

i wanna transform a photo into the videogame style i saw in the model page

[ornate elk]

Just move the small file into models/lora
And everything larger than 1.98gb goes into models/stable-diffusion
Then everything will work

[lean laurel]

ohh ok

[lean laurel]

but i ll be able to do that or should i download more models?? a model larger than 1.98?

[ornate elk]

You can do that in img2img

[lean laurel]

can i do that only with the lora i downloaded?

[ornate elk]

But you need at least a model

[ornate elk]

Loras can only be used together with a model

Not alone

[lean laurel]

ohh ok ok, thanks, so i gonna close the page and the cmd and move the small file

[lean laurel]

how should i do this?

[ornate elk]

You dont need to close it

[lean laurel]

my bad sorry, i gonna move it and open it again

[ornate elk]

You would need to click the reload button near the model dropdown.
But a restart of the webui+cmd works too

[lean laurel]

ohh but if i closed and open it again, will it works??

[ornate elk]

The browser tab or the cmd?

[lean laurel]

both... sorry

[ornate elk]

Yes if you close and open both again it will work

[lean laurel]

ok i found this

https://civitai.com/models/7240/meinamix

[ornate elk]

Thats a model

[lean laurel]

this one actually is a model, right?

ohh great!!

so download it, and move it into the model>stable-difussion ?

[ornate elk]

You can verify that by looking at the Info box of the model on the model page

[lean laurel]

this info box??

[ornate elk]

Checkpoint or checkpoint merge is a model

Base model is the version its based on

I'm AFK now

[lofty mauve]

what is a Lora?

[lean laurel]

ok, thanks you CS! and have a nice day ❤️

[simple tartan]

fracking aye, i installed ubuntu 23.10 and rocm doesnt support it, wtf

Some packages could not be installed. This may mean that you have
requested an impossible situation or if you are using the unstable
distribution that some required packages have not yet been created
or been moved out of Incoming.
The following information may help to resolve the situation:

The following packages have unmet dependencies:
 libwayland-amdgpu-client0 : Depends: libffi7 (>= 3.3~20180313) but it is not installable
 libwayland-amdgpu-server0 : Depends: libffi7 (>= 3.3~20180313) but it is not installable
 rocm-gdb : Depends: libtinfo5 but it is not installable
            Depends: libncurses5 but it is not installable
            Depends: libpython3.10 but it is not installable or
                     libpython3.8 but it is not installable
 xserver-xorg-amdgpu-video-amdgpu : Depends: xorg-video-abi-24 but it is not installable
E: Unable to correct problems, you have held broken packages.

[ornate elk]

Rip yes you need 22.04

Its LTS and supported until 2027

[simple tartan]

yea, its downloading....

[ornate elk]

23 only until July 24

[lean laurel]

@ornate elk sorry for ask, u always help people and know about everything, are u one of the authors behind the SD?

[ornate elk]

A lora is a small filed trained on a specific character or style to help a model create that.

[lean laurel]

oh and who i use it?

[ornate elk]

No ^^ I'm just a Tech enthusiast who works since the release of auto1111 with SD stuff as hobby

[lean laurel]

thats still pretty admirable ❤️

[ornate elk]

You need to select it in the lora option

In txt2img or img2img

AFK now cya later

[lean laurel]

how do i upload my photo into img2img?

[lean laurel]

cya later too @ornate elk !! thanks for everything!!

[simple tartan]

yea, kudos to @ornate elk he's been helping me get ramped up as well. I played with SD back in the early befor 1.0 days, and its so different now with A1111. I'm working as fast as i can to get to the same level of knowledge, i'm obsessed with learning and its way easier now to get full into all the LLM options

[lean laurel]

u sound like u know a lot too!! is great that people like u both are here so u can help others learn

do u know how i chosee the lore i want int a img2img??

[simple tartan]

depends on what you can "a lot". I know a little bit about almost everything, but not many things have I mastered yet, lol

there should be a "Lora" tab near the bottom of the ui

[lean laurel]

that mean the lora is already added??

[simple tartan]

if you havent installed any yet, I HIGHLY recommend the civitai browser+ extension, so much easier to grab new ones

yes

but weight of 1 is a bit heavy, start with 0.3 on most lora's

and add your personal sauce to the prompt along with the lora

[lean laurel]

thats great, SD looks like have a loooooot to learn, a lot of things

[simple tartan]

for sure

[lean laurel]

i just wanna to turn the photo i uploaded into the style of the model

do u recommend me anything to writte into the promp?

[ivory bramble]

Hi, I have web ui A111 installed, but I'm watching a tutorial on YouTube and this person suggests using a HiRes.fix file called "8x_NMKD-Superscale_150000_G.pth". So I downloaded it, I put the pth file in C:\a1111\stable-diffusion-webui\models\ESRGAN but it's not showing up as a sampling method:

[simple tartan]

depends on if your are doing people, animals etc. I always use phrases like masterpiece, photorealistic, extremely detailed face and eyes, etc

[ivory bramble]

Sorry wrong screenshot: https://i.imgur.com/SwmqkqP.png

[simple tartan]

negatives are really helpful too, things like bad hands, bad fingers, bad anatomy

[lean laurel]

ohhh thanks, gonna add that ones

always separed by a , right?

[simple tartan]

yes

and you can use () to strenthen the weight of words

[lean laurel]

ohh with () heavier the word?

now click generate right?

[simple tartan]

(blue eyes) for example will try harder to generate blue eyes, or (((blue eyes))) pretty much will force it

yep, generate and wait

[steady mountain]

Anyone know how long Preparing metadata (setup.py) ... - \ should take?

[simple tartan]

@lean laurel once you are rolling and have everything working, I'd head over to #📝｜prompting-help where you can find more details and help on good prompts

[ivory bramble]

This is what his window looks like https://i.imgur.com/QNtX6Id.png

Under Hires.fix, I don't have the 8x_NMKD-Superscale_150000_G option

This is so confusing

[simple tartan]

did you reload SD after you uploaded the sampler?

[ivory bramble]

Yeah, I refreshed the page and also closed down the whole web ui and started it up again

[simple tartan]

i'm rebooting to install another build of nix. make sure its in teh right folder, not sure if samplers go in the root or have a subfolder of their own off the top of my head

[ivory bramble]

I'm trying again

[simple tartan]

BBIAF

[ivory bramble]

No, it's still not there

[lean laurel]

hey anyone know why i m having a "runtime error" trying to make img2img??

anyone know what to do now?

[dim ermine]

Hello Guys, any idea why my 4x-UltraSharp.pth is not loading in the UI Menu? Did reload UI, and still not showing in the menu. Thank you🤘

[vocal burrow]

Look at the console log for more details

[lean laurel]

hi Aryetis!! i closed it and open it again and look like is working

[lean laurel]

is this normal?

[simple tartan]

are you team red?

[lavish meadow]

fellas if i did this

would this work with controlnet?

[lean laurel]

a masterpiece

can someone help me with the img2img please??

i upload this to the img2img

i added a checkpoint with anime models

and recibed this

lol

what should i do?

[lavish meadow]

too much denoise?

[simple tartan]

did you create a prompt?

[lavish meadow]

lower the denoise to abt 0.3

[lean laurel]

i didnt

[lean laurel]

oh where is it?

i found it

[lavish meadow]

bingo

[lean laurel]

i had 0.75

[lavish meadow]

if u have that too high it messes it up

lower it alot if u want it closer

if u want it slightly different

higher

remember to prompt what they look like

[lean laurel]

should i also mode the checkpoint??

[lavish meadow]

what features you would like

checkpoint should be fine

run it again

[lean laurel]

oh so in the prompt i should describe the photo i uploaded?

[lavish meadow]

yup

lemme get an example

[lean laurel]

runing it ❤️

ohh thanks

[lavish meadow]

low denoise

too high

see the differnce

maybe try toonyou

its what i used

[lean laurel]

oh toonyou is the model??

ohh yeah i totally see the different

i can actually see the diferent in the preview, sorry my pc is a bit slow

[lavish meadow]

its cool

[lean laurel]

where i download the toonyou model?

and thanks a lot @lavish meadow

[lavish meadow]

civitai

and np

i had that problem

[lean laurel]

oh a lot of us should go throught the same problem

[lavish meadow]

yup

denoise is a pain at first

[lean laurel]

i m watching it, a to low denoise change almost nothing, and a hight denoise look like make anything de IA want lol

any advice for negative prompt to avoid face deformation?

[lean laurel]

https://civitai.com/models/30240/toonyou

this one do u recommend me?

[lavish meadow]

yup thats it

[lean laurel]

thanks i m gonna download it! ❤️

no deformation is a good negative prompt?

[lavish meadow]

ive never tried that one

[lean laurel]

do u have an advice u can give me to get no deformed faces?

[lavish meadow]

not really, a hope and a dream

describe their face their eye colour stuff like that

might help

[lean laurel]

i see, and if i m having multiple characters in the photo?? how i describe them?

[simple tartan]

same as you do in RL. "man on left, woman on right", or "woman with black hair in front, woman with red hair in background"

[lean laurel]

ohh thanks thanks

[simple tartan]

i think of prompts liek i do GPT, just imagine your talking to the artist

or the analogy "explain it to a blind man"

[lean laurel]

thanks thanks ❤️

[ornate elk]

you need to use an similar resolution to

like 768x512

[lean laurel]

ohh how do i check the resolution of the original pic??

hi @ornate elk ❤️ how u doing?

[ornate elk]

hey fine, back at home

[lean laurel]

did u had a nice day?

[ornate elk]

in the image details

[lean laurel]

ohh i see

so if my image have

532 x 768

so i should create a pic with that resolution

one thing @ornate elk , i said i have 2gb vram, is ok is a make pics with that resolution??

[ornate elk]

it can be limiting but can work, you have to try it out. Also everytime you encounter an out of memory error, you have to restart the webui cmd

[lean laurel]

besides, idk why, but i m having a problem, i cant make 2 pics in a row, always when de second pic is gonna be ready, there a problem, i need to close the page, the cmd, open it again and i can make other pic, do u know if i can do anything?

[ornate elk]

try 512x768 in txt2img

[lean laurel]

i m having that problems!

how do i restart the webui cmd? just close and click again the webui-user?

[ornate elk]

what settings you used?

[ornate elk]

yes

[lean laurel]

what settings about what??

when i installed it??

[ornate elk]

when you get the error

the txt2img settings

[lean laurel]

actually i only manipulate one thing

let me check

this settings??

i only manipulate denoising strength, make it lower

my computer wont explote if i put a hight resolution right lol?

[ornate elk]

nope, it will be fine xD

[lean laurel]

should i change any other setting?

[ornate elk]

nope, should be fine

[lean laurel]

btw

what mean

the 5/75??

[ornate elk]

tokens

can be ignored but dont go over 225, that needs more vram,

[lean laurel]

ohh ok ok, thanks

[dim ermine]

Hello Guys, any idea why my 4x-UltraSharp.pth is not loading in the UI Menu? Did reload UI, and still not showing in the menu. Thank you🤘

[lean laurel]

@ornate elk do u know why give me this error?

[ornate elk]

what does the cmd shows?

[lean laurel]

wait i ll show

[stiff lava]

Hi all, I'm trying to build a machine to run SD locally, something entry to mid range. I see a lot of people mention the 3060 as a good starter GPU, and the 3090 comes up a lot, but no one seems to mention the 3070. Is there a reason for that?

[idle cairn]

Hello !
Is anyone else having this issue with ControlNet (Openpose) that turns the image or part of it yellowish ?

[ornate elk]

hey, nope, can you show us an example?

[ornate elk]

hmm strange error, does that happen in txt2img too after a restart?

[lean laurel]

yeah, sadly, may be the resolution?

[ornate elk]

try 512x512 for test

[lean laurel]

i m trying

if a image have a hight resolution, can i use a multiple number for the resolution?

[idle cairn]

It seems to be intermittent, now, since I've changed the the sampler...
My examples contain nudity, I can't post them, I assume (even blurred) ?

[ornate elk]

what do you mean?

[ornate elk]

yes, you can't post them here ^^, but good that it works now

[lean laurel]

is the pic i wanna transform with img2img is 1280x720, can i put the resolution in 640x360?

[ornate elk]

yes

or 768x512

[lean laurel]

i didnt get the error with low resolution

[lean laurel]

why that number?

[ornate elk]

because 1.5 models got trained on a resolution of 512x512

and 4:3 would be 768x512

[lean laurel]

and thats proportional to 1280x720?

[ornate elk]

nope but it will still be okay to use

[lean laurel]

ok ok thanks @ornate elk !!

[idle cairn]

It started again. Here is an example

[sturdy ginkgo]

3070 is just a waste of money as it has almost 0 performance increase compared to the cheaper 3060

[stiff lava]

ooh ok thanks! when the docs say that SD only runs on NVIDIA graphics cards, does that mean literally only nvidia branded, or would like a Gigabyte/MSI/Geforce branded one still work?

[sturdy ginkgo]

Nvidia Framework .. brand doesn't matter

[stiff lava]

awesome, thank you @sturdy ginkgo !

[ornate elk]

a 3070 has only 8gb vram, you dont want 8gb vram in 2024

12 should be minimum

[stiff lava]

maybe a 12gb 3080 then?

[ornate elk]

for SD vram is the most important

[ornate elk]

would be faster than a 3060 but has the same vram

[sturdy ginkgo]

My GTX 1070 has lasted long enough with 8 GB 😄

[ornate elk]

and you tried an other sampler?

[idle cairn]

Yeah, I thought it solved it, but it didn't.
Also tried multiple lighting prompts, but it really just disappears when I disable ControlNet

[sturdy ginkgo]

if using SD 1.5 models: try https://civitai.com/models/211032/nsfw-filter-slider-or-tool-lora

[ornate elk]

what are your controlnet settings?

[sturdy ginkgo]

hmm .. try using multiple passes .. unit 0 being prompt is more important, unit 1 controlnet more important with a depth or canny controlnet

[idle cairn]

I'm gonna be honest, I just started using ControlNet, I'm not really comfortable with it yet

[sturdy ginkgo]

i use comfyUI and mess around with literally every single value and order till i get it to work like i want 😄

[idle cairn]

I use A1111 so I'm a little bit more limited, but I do the same thing 😉

[sturdy ginkgo]

Love control 😄

[lofty mauve]

Can you guys help me with controlnet? I have the extension installed but I have no idea how to get the right model, I think? I don't think my controlnet settings do anything. Also, how do I upload an image specifically just for controlnet?

[ornate elk]

have you tried openpose with an 1.5 model?

[idle cairn]

Once, a few weeks ago, but couldn't figure out how to use it, so I left it alone

[ornate elk]

its the exact same what you did rn, you only need to select an other model thats not sdxl

and you need the 1.5 compatible controlnet models

[sturdy ginkgo]

for the images civitai has a searchfilter for that.
for models: https://huggingface.co/lllyasviel/sd_control_collection/tree/main
those with XL in the name are for SDXL
those with 15 are for SD1.5

[ornate elk]

would recommend using the pruned controlnet 1.5 models to safe space and faster controlnet:
https://civitai.com/models/38784/controlnet-11-models

[untold marlin]

Does comfy work with SD 1.5?

[ornate elk]

yes

[valid elbow]

@valid elbow

[untold marlin]

okay next question someone who is new to this which would you recommend?Comfy or a1111? I rather start on the one stick with to learn on then have to relearn cause need to switch UIs.

[ornate elk]

auto1111 for sure, its easier to switch some toggles and press some buttons than make a complete workflow or understanding each setting, auto1111 has also enough settings you need to understand ^^

comfyui is good if you want the absolute full control of everything. More suitable for the already experienced SD users

[untold marlin]

10-4 thanks

[ornate elk]

checkout the pinned messages in this channel for my installation guide of auto1111

[untold marlin]

Already have it installed played with it last 2 days. But before I dug any deeper wanted to see which was best to use

[obtuse hare]

PermissionError: [Errno 13] Permission denied: '/Users/user/.cache/huggingface/hub/models--openai--clip-vit-large-patch14'

[ornate elk]

ah okay, installing comfyui is easy too. and you can use the models in both webuis if you want to compare

[untold marlin]

Think i am doing okay so far

[ornate elk]

looks really good!

[untold marlin]

Next got learn to do the whole upscaling thing

[obtuse hare]

looks an awful lot like a content stealing "youtuber"

(i am joking)

[ornate elk]

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

[obtuse hare]

I can't run .bat files i'm on mac

[ornate elk]

ohhh

[obtuse hare]

should be similar tho

[ornate elk]

but you can still run pip cache purge

and delete the venv

[obtuse hare]

where is venv

[ornate elk]

inside the stable-diffusion-webui folder

[ornate elk]

yea upscaling is the key for good quality

feel free to ask here if you have any question about it

[untold marlin]

Yea why i asked about comfy cause most guides i found online was talking about upscaling workflows. But I think for now til I upgrade my pc I will stick to Auto1111.

16gb ram but only a gtx 1650

[obtuse hare]

started reinstalling stuff

[obtuse hare]

didn't work

it says

PermissionError: [Errno 13] Permission denied: '/Users/user/.cache/huggingface/hub/models--openai--clip-vit-large-patch14'

[ornate elk]

Ah okay, yea the optimisation in auto1111 is good even for slower cards

[ornate elk]

Which guide did you followed for the install?

[obtuse hare]

a youtube video, I am a visual learner

[obtuse hare]

https://www.youtube.com/watch?v=Jh-clc4jEvk

[untold marlin]

@obtuse hare I feel that. All this reading I have been doing and still have to watch a video to fully grasp the concept

[ornate elk]

Checkout the Pinned tabs in this channel, a Member here made a fully auto install script for mac that should work without problems

[obtuse hare]

the video is just a video version of the guide i saw

i used the auto install too i believ

[ornate elk]

That video is 11 months old