🤝｜tech-support

[deep canyon]

i know now im readin em

at first thought it was its own thing

[ornate elk]

I think it processes cuda stuff in the background

[sweet basin]

I tried to generate in 2 browsers and canceled in one

dont know if I fcked up xd

it is like this now

[ornate elk]

yea restart it

and only one time

[vocal burrow]

probably not.try again and let it finish this time.

[sweet basin]

0%| | 0/20 [00:00<?, ?it/s]

started

still 0

https://tenor.com/view/lets-go-gif-drake-gif-26307972

done

--ckpt-dir "Path/to/models/stable-diffusion" --lora-dir "Path/to/models/lora" --controlnet-dir "..." --embeddings-dir doesnt work

[ornate elk]

You need to edit it with your paths lol

[sweet basin]

[ornate elk]

"Path/to/models" is a placeholder

[sweet basin]

E:\StableAI\stable-diffusion-webui-directml\models\Stable-diffusion

--ckpt-dir "E:\StableAI\stable-diffusion-webui-directml\models\Stable-diffusion"

so like this?

[deep canyon]

ok I am confused

am I just downloading the pix2pix model directly, just doing the stuff from the hugging face link, or both

these are the auto1111 instructions for pix2pix but they are vague to me

[ornate elk]

yes

[sweet basin]

yeah did it

[hazy cargo]

can someone help im braindead, im trying to launch stable cascade off of pinokio, everything appears to install right but when it opens up the proxy in a tab it just appears blank with the tab name saying loading....

[sweet basin]

this speed lmao

lol it doesnt even use full vram

[vocal burrow]

download the model from hf

[sweet basin]

so i can delete directml

and move folders from there to zluda

[deep canyon]

these dont appear to be windows instructions :P

assuming you need to do the github stuff first

[ornate elk]

I wouldn't delete it

Zluda with this webui doesn't support some extensions or sdxl models

[deep canyon]

also 0 amd instructions :(

@ornate elk youre pretty smart, any ideas as to how im supposed to follow these instructions

theyre breaking my brain

quite confused

[ornate elk]

your on colab?

[deep canyon]

?

idk what that is

im using Auto1111

[ornate elk]

okay good

[deep canyon]

maybe i was overthinking it

[ornate elk]

@deep canyon you only need this model:
https://huggingface.co/timbrooks/instruct-pix2pix/tree/main

and then load it into img2img

then you can use commands like "make her hair green" or "what would it look like when its snowing"

[deep canyon]

Ive never used this website before, what exactly am I grabbing?

[ornate elk]

the .safetensor file

[deep canyon]

oh okay i think I got that from the direct download link on the site

in the guide i mean

nvm i just got the checkpoint im dumb

no im double dumb

i got the safetensors and put it in the Stable-Diffusion folder

I assume I should get the checkpoint as well?

[ornate elk]

no

only the safetensor file

[deep canyon]

okie

[ornate elk]

and it goes into the models/stable-diffusion folder

[deep canyon]

right yeah i got that part

i think i was just expecting something more confusing lol

and ended up confusing myself

initial setup was a bit ass

[cunning tiger]

I use WebUI Forge, but when using the controlnet that comes with it, it gives an error
TypeError:'NoneType' object is not iterable
I know it's very generic, I'm asking to see if it's a common error.

[deep canyon]

no worky :(

i can make slight changes but i cant have it make an image another art style

its also just straight up dogshit

"make the wheels red" thanks g

[wraith bronze]

@ornate elk do you know if there is any extension or script available that will use the same seed twice in a batch generation and flip the resolution?
So for example 512x256 change into 256x512 and reuse the seed before continuing? I could just flip it around after the first batch is completed, but I am lazy and sometimes I set up a batch while I am not at home so it would be nice to have it flip automatically ^^

[onyx solar]

It's going to take a prompt like it's your slave. You're not giving it commands of what to do. This stuff works by describing what you want to see. But if you want the exact same car with just wheels that are red, you should be using inpainting, masking the wheels, and then describing "red wheels". (Though it'll work even better if you describe them in more depth and on what specific car.)

[deep canyon]

i tried like 40 different prompts that was just one example lol

for all sorts of different things

I managed to get it to look like an oil painting and a mosaic but thats the only two that worked

[onyx solar]

And I'm sure that if you prompted better and had your settings correct, you'd get better results.

Well...I lied a bit, apparently...you're trying to do the instruct pix 2 pix, which actually should take human language instruction. That said, denoising should probably be set to 1.0.

Also, you should bump the CFG and decreast the image CFG a bit if the effect isn't right. But I would focus on rewording the prompt, too.

[royal rapids]

hmm i might be blind but, whats the extension so that i can have pictures o nthe LoRas i use?

nvm i think i see it

[noble belfry]

theres a civit download helper extension that should load in extra info like that

[royal rapids]

ayee thanks

[cursive flame]

gm, looking to pick up a new laptop to use solely for stable diffusion video work. any recommendations? or if there is a better way to go about it, like renting access to a really good rig or something

i can also post in a different channel

[lavish jacinth]

So i tried doing a command to do an update check everytime i launch A1111 but for some reason it just returns this error

venv "D:\Stable-Diffusion\stable-diffusion-webui\venv\Scripts\Python.exe"
fatal: not a git repository (or any of the parent directories): .git
fatal: not a git repository (or any of the parent directories): .git
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: 1.7.0
Commit hash: <none>

[fierce nimbus]

Hey guys, hello and good evening. For many months, automatic1111 does not show my Lora tab in the extra networks section for some reason. I've tried reinstalling and I'm even using the release candidate but it just won't load. Models are in the /webui/models/LoRA folder. Anyone know how to fix this issue?

[vocal burrow]

Are you trying to use sd15 loras with an sdxl model (or vice versa) ?

[unique bear]

Is there a text model better than harrlogos?

[nocturne topaz]

Not for sdxl

[nocturne topaz]

they moved the location of it

[vocal burrow]

What do you mean ? it's still the same folder. And even in the 1.8RC's UI it's still at the same place on screen.

[nocturne topaz]

Compared to like 1.4 it moved

really old version had it on the right

[vocal burrow]

indeed

They talked about a "lora tab" so I assumed they weren't using that old of a version.

Guess we'll never know as they don't answer :p

[west harness]

Is there an easy way to convert from a diffusers pipeline to a single checkpoint/safetensors?

or maybe im misunderstanding, i have a unet/vae/text_encoder

[nocturne topaz]

Theres a tool on huggingface to do it, or just find the 6 lines of python to do it (i.e. import the right modules, load model, safetensors.save_model(), ??? profit)

[west harness]

it is surprisingly difficult to find considering it should be a fairly simple process

[river flame]

how do I install CUDA 11.8 into Forge? there s no venv folder

[nocturne topaz]

you run the system and it will download the correct torch + cuda for you

[river flame]

thanks

[mortal blaze]

Have you downloaded any ControlNet model ? You can't use the extension without themhttps://huggingface.co/lllyasviel/ControlNet-v1-1/tree/main

[nocturne topaz]

and the xl equivalents if using a sdxl model

[noble belfry]

can pt and pth files be converted to safetensors?

[west harness]

yes

safetensors supports dumping from state dictionary, and pt/pth can be converted to state dict

[noble belfry]

is there any drawback in load/unload to doing that? say for controlnet files

[west harness]

not sure what you mean, conversions are typically non-destructive

[noble belfry]

say i have something like forge, with a pile of ckpt files, pths etc everywhere. would there be any issues in running as normal?

if i batch converted anything safetensor-able to safetensors format

[river flame]

Is there any way to fix pixelation from FaceSwapLab? Hires.fix and refiner do not help

[west harness]

up to your hardware i suppose

[noble belfry]

its just that you sometimes get extensions say that will quietly download a bin or a pth from huggingface to use their features. it'd be nice if something instantly forced safetensor standardising, just on the offchance of malicious pth or more likely a hijacked download

I'm sure there's some reason it isnt done that already

anyway, thanks for the answer lizardy 🙂

[cosmic furnace]

@ornate elk Sooo. Did u tried whats better on zluda now? next or just directml?)

[vocal burrow]

Faceswaplab is pretty old and abandonned at that point. There are newer alternatives for that kind of things; reactor, ip-adapter, etc (I don't use them tho)

[vocal burrow]

I don't have AMD so I can't answer for them. But directml is definitely the slower option of the three. (assuming you're listing, directml, directml-zluda, sdnext-zluda)

[solemn atlas]

hi

iwanted to prompt and i got this

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument index in method wrapper_CUDA__index_select)

what does that mean

[river flame]

I tried it all. IP-adapter is not good enough, Reactor and FaceSwapLab are pixelated, and are not silimar to the original too. I use FaceSwapLab + IP-adapter with a different photo. The result is great but pixelated.

[vocal burrow]

Can you screenshot your webpage so we see what settings you used for your prompt ?

[solemn atlas]

sure

[vocal burrow]

what does the console says ?

[solemn atlas]

where did you read that

[vocal burrow]

the black window thing with white letter in it

stuff like that

Because in the UI it only shows the last line of the error log. So most of the time it's useless.

Also what's your hardware ?

[solemn atlas]

its the same as i send you. the problem is. when i close down SD and start it again it works for some generations and then stops. now it works but it will stop eventually.
you mean cpu? 4070S

when i start SD i also get this

[vocal burrow]

• "the same as i send you" ? You didn't send any console log, did you.
• cpu? 4070S. That would be your gpu. Open dxdiag, screenshot the system and display tabs

[vocal burrow]

That's because of TensorRT extension. Nvidia. If you're not using it, uninstall it and the error should go away. (Most likely you're not using it if it's bugged like that)

[cosmic furnace]

nah i meant zluda versions) . next is diffusers >_> . And seem to have problem with img2img. At least for me =\

[solemn atlas]

now it works bcuz i re-started it but after some while it does not work

[solemn atlas]

okay

[vocal burrow]

hmm :/ maybe you can try to whitelist the webpage in your adblockers. img2img had problems in the past with adblockers.

If not, then yeah maybe it's a zluda quirk "for now".

cs1o will probably know when they get back

[cosmic furnace]

didnt tried that) hmmm ok

anyway next.sd seems to get updated to work better with zluda

no need to copy zluda files in last guide & etc

[vocal burrow]

As long as you get that error at startup I wouldn't trust tensorrt at all.

For now please, restart your sdwebui. trigger the error you initially came for. Copy paste the whole console log in a .txt and drop said file in there so I can take a look at it.

[solemn atlas]

I removed Tensorrt but i can't delate it. How does that work? I just removed it and re-started it.

i can't trigger the error but thats how it is now

[vocal burrow]

To remove an extension you delete its folder in stable-diffusion-webui\extensions so for tensorRT it's stable-diffusion-webui\extensions\Stable-Diffusion-WebUI-TensorRT.
It's also good practice to delete the stable-diffusion-webui\venv folder if the extension did some heavy modifications. Don't worry the venv will rebuild itself at launch, but it's gonna take some time.

In your case considering you've got many errors at startup that you ignore.... I'd recommend deleting venv indeed

so yup do that and try again

[solemn atlas]

Ok I delated both folders

[vocal burrow]

Then relaunch webui-user.bat, try to reproduce the error and send console log again if that happens.

[solemn atlas]

Okay I did it. It takes some time till it downladed everything. I must admit, since I started SD the first time there accured a lot of errors. I hope this time it works

[vocal burrow]

It should be faster than the first install as it will reuse some of what was previously downloaded.

[solemn atlas]

yes its faster than the first ime

time*

[sweet basin]

so my zluda sd was working fine yesterday but today getting "RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check
Press any key to continue . . ."

[vocal burrow]

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

[sweet basin]

for auto udpate?

[vocal burrow]

yes

[sweet basin]

doubt since I installed yesterday

[vocal burrow]

an update got released a couple hours ago, that's why I'm asking

Can you check please ?

[sweet basin]

error: Your local changes to the following files would be overwritten by merge:
modules/shared_init.py
Please commit your changes or stash them before you merge.
Aborting
Updating 77c0e5f2..828f558c
venv "E:\SD-ZLUDA\stable-diffusion-webui-directml\venv\Scripts\Python.exe"
fatal: No names found, cannot describe anything.
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: 1.7.0
Commit hash: 77c0e5f2e7f6e1f81f3a7477484a03f521465f7a
Traceback (most recent call last):
File "E:\SD-ZLUDA\stable-diffusion-webui-directml\launch.py", line 48, in <module>
main()
File "E:\SD-ZLUDA\stable-diffusion-webui-directml\launch.py", line 39, in main
prepare_environment()
File "E:\SD-ZLUDA\stable-diffusion-webui-directml\modules\launch_utils.py", line 596, in prepare_environment
raise RuntimeError(
RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check
Press any key to continue . . .

[solemn atlas]

another question why is SD so slow when genrating?

[vocal burrow]

considering it complains about merging files at the very first line .... I'm guessing that you do have git pull in your webui-user.bat

[sweet basin]

yeah I just added

[vocal burrow]

dghifjtgyy

[sweet basin]

something updated and same error

[vocal burrow]

no

I asked if you had it specifically because it can introduce errors if you update with some weird setup like zluda

[vocal burrow]

Might be because you're asking too much for it, because of the errors you had, etc

@sweet basin anyways.... what's in your webui-user.bat ?

[sweet basin]

iwill do a magic trick

and delete venv folder

[solemn atlas]

so it opend

[sweet basin]

Edit the webui-user.bat (right click), At the line COMMANDLINE_ARGS= You add: --use-zluda --medvram-sdxl --update-all-extensions --update-check --skip-ort

[solemn atlas]

what should i do now? Just generating?

[vocal burrow]

yes try to reproduce your error

[solemn atlas]

so the error when I started SD is gone away and generation also works. I need to generate a bunch of stuff to see if the error still accurs

[sweet basin]

venv deletion didnt worked

https://tenor.com/view/pai-gif-9478928798526726618

[sweet basin]

@echo off
git pull
set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --use-zluda --medvram-sdxl --update-all-extensions --update-check --skip-ort --ckpt-dir "E:\StableAI\stable-diffusion-webui-directml\models\Stable-diffusion" --lora-dir "E:\StableAI\stable-diffusion-webui-directml\models\Lora" --embeddings-dir "E:\StableAI\stable-diffusion-webui-directml\embeddings"

call webui.bat

[vocal burrow]

what does the console log says then ?

[vocal burrow]

should be good, let's wait for @ornate elk

[solemn atlas]

so here is the error again

[ornate elk]

did you replaced the files after deleting the venv ?

[vocal burrow]

Your sd-webui-roop extension is still bugged, so is your deforum and comfyui.

I would remove them too (and reinstall later on if needed) and rebuild venv

[solemn atlas]

these are my extensions which one should i delate? deforum, comfyui and the folder webui-roop?

also delating venv folder again

[vocal burrow]

Also for your actual "two cuda devices" error, try forcing sdwebui to use only your gpu with --device-id xxx xxx being your gpu device id.
You can find your gpu device id in task manager.

I have to go for now

[solemn atlas]

this is what i have

[solemn atlas]

but its already at 0 as i see?

[vocal burrow]

use --device-id 0 command line args in your webui-user.bat

[solemn atlas]

sorry where should i put it?

ahh i understand, open the command line and then putting in there

[vocal burrow]

Edit your webui-user.bat with notepad.
Find the COMMANDLINE_ARGS= line
Replace it with COMMANDLINE_ARGS=--device-id 0 --xformers

[solemn atlas]

did it

then save it and restart webui?

[vocal burrow]

yes

[solemn atlas]

okay so i restarted SD

tried to prompt 3times

this is what i get

[vocal burrow]

console log ?

(I'll check later, I have to run some errands)

[solemn atlas]

okay

[mighty needle]

Anyone knows what happened here? 281 settings changed

[sweet basin]

yeah

[ornate elk]

oh i see, you need to delete the shared_init.py its in modules/

then relaunch

[paper crag]

okay, i have issues with controlnet, specifically it doesn't appear in my sd page.
I've updated it, deleted and reinstalled it, file is in right place, it appears in extensions, but it still doesn't appear in generation pages.

[ornate elk]

which webui are you using?

[paper crag]

automatic1111

[sweet basin]

got an update but in the end same error torch not able to use gpu

[ornate elk]

yes, delete the file

then relaunch

[sweet basin]

deleted

update started and got same error

[ornate elk]

then drag and drop the two zluda files again into torch/lib

[sweet basin]

also did

nothing

update fcked up everything

added skip torch cuda

still somekind of error

guess another SD update will be soon

[ornate elk]

I updated to and everything works

Make sure you have --use-zluda in the webui-user.bat

[sweet basin]

dunno a lot of times when SD get an update it stopped working for me. but within a day another update incomes and fixes

happened like 4 times already

[ornate elk]

Yes thats the case for directml

But check your webui-user.bat because my guide updated too

[sweet basin]

well it is same as yesterday

[ornate elk]

You have --use-zluda in it?

[sweet basin]

ofc

time it worked

this update for sure

[cosmic furnace]

@ornate elk so have u tried next sd recently?)

[solemn atlas]

i would be happy if you could help me

[noble belfry]

could you copy and paste the errors here? they're normally in the window with the black text and white writing

sorry, just scrolled up, you already did. sorry if i sounded patronising 😄

at the bottom of the stable diffusion browser window it should have in tiny text something like version: f0.0.16v1.8.0rc-latest-268-gb59deaa3  •  python: 3.10.6  •  torch: 2.1.2+cu121  •  xformers: N/A  •  gradio: 3.41.2  •  checkpoint: f53945ec37 could you paste that in here?

[solemn atlas]

sure :9

🙂

[vocal burrow]

well give me the console log of your error :p

[vocal burrow]

the whole thing, not just a screenshot please

[solemn atlas]

sorry

[noble belfry]

oooh amd and cuda?

[vocal burrow]

they claimed to be using a 4070s

[noble belfry]

your 2-shot script looks ancient

april 2023, could you update it or check github to see if its deprecated?

[vocal burrow]

Might be but it's """just a warning""" so I doubt that's the problem. Plus they're not using it there

[noble belfry]

no you're great, you're 90% better than most people at giving info 😄

[vocal burrow]

what settings were you using for this one ?

sampler, prompt, etc screenshot whole webpage if possible.

[noble belfry]

yeah i read the gradiodeprecation thing and appearantly they suggest just suppressing it so its not it, you're right aryetis

[vocal burrow]

Yes @solemn atlas I'm asking the same stuff again and again but we're in a way better position than we were a couple of hours ago with all your errors and bugged extensions :p

[sweet basin]

and zluda working again wtf

[solemn atlas]

no worries 🙂

[sweet basin]

why your SD is white wtf

[noble belfry]

that used to be default. some people use discord in white too. heathens. 😛

[vocal burrow]

it adapts to the system theme

[sweet basin]

what the fck

[solemn atlas]

idk thats how i download it

[vocal burrow]

can you try without the lora Sarada ?

[sweet basin]

it hurt my eyes just looking at your screenshot

[noble belfry]

stupid things i'd do: disable every non default extension

there are some that cause issues even when not being used

[vocal burrow]

wait, "Composable Lora"

that might be the culprit

it's extra old and it used to hijack the built-in lora system

remove any lora from your prompt and try again

[noble belfry]

uncheck every one of these.

or, listen to aryetis, i think im helping over them and hindering. sorry! 🙂

[solemn atlas]

oh really? I wanted to make two characters in one image and it was said that I need that

[noble belfry]

interestingly you can do that in webui forge using masking now 🙂 (ill link after you get fixed)

[vocal burrow]

I'm not the sole possessor of the one truth. As long as we're not hundreds talking at the same time and hindering each other please feel free to help others.

(Some said, I might even be wrong sometimes)

[solemn atlas]

i'll uncheck them

[vocal burrow]

That's probably some outdated tutorial you watched then.

[vocal burrow]

to save some time, try my thing first.

[noble belfry]

(outdated in SD land means 1 month old sometimes!)

yes.

[solemn atlas]

https://www.youtube.com/watch?v=kfoA0xWv-0Y it was this tut

4 months old

okay i unchecked them all

the extentions

[noble belfry]

try a simple img without a highres

[vocal burrow]

Only started to watch but I'd go with whatever "Regional Prompter" based solution he has.

also to work with "composable lora" iirc you needed to use --always-batch-cond-uncond command line arg

(which you didn't have set in your webui-user.bat)

[noble belfry]

i looked on their git, last touched 9 months ago

[solemn atlas]

so one generation worked but the next one didn't worked. I didn't used any hires

[vocal burrow]

edit your webui-user.bat and add the line set CUDA_VISIBLE_DEVICES=0 below the command line arg one.

[noble belfry]

jfc how many tokens is that

[river flame]

I m trying to install an upscaler SUPIR. I m so bad at this. Where do I write this?

[high knot]

Does the Reactor extension still work in A1111 v 1.8 for anyone else?

I'm getting this error despite having removed and re-git cloned it from source:

RuntimeError: unexpected EOF, expected 413788 more bytes. The file might be corrupted.

[solemn atlas]

like this

[noble belfry]

that sounds like a cutoff download

[vocal burrow]

that's probably a network error and not an A1111 one. delete venv folder, and run pip cache purge to recreate a proper venv from scratch

[river flame]

I dont know what s my version, intalled A1111 yesterday, Reactor works

[high knot]

Thanks, trying now 🙏

[vocal burrow]

wherever you cloned the repository, but that's not A1111, nor comfyui related, etc... Plus it's running on python 3.8 for some reasons ?!

[river flame]

ohhh(( okay thank you

[noble belfry]

"⚠ Due to the large RAM (60G) and VRAM (30G x2) costs of SUPIR, we are working on the online demo releasing." hahahhah, oh god

[river flame]

yes that s cute

[vocal burrow]

yeaaaah there's that too XD

demos are impressive tho

[noble belfry]

its not implemented yet at least in forge

[vocal burrow]

probably never will unless there's some breakthrough with those requirements

[river flame]

the new version works with 12GB 3060

[noble belfry]

well too rich for my blood "
With V8, NOW WORKS on 12 GB GPUs as well with Juggernaut-XL-v9 base model. "

still tho

[vocal burrow]

hmmm nice

[noble belfry]

i mean it does look like you could feed it a grainy photo of bigfoot and count the hairs on its elbow though.

[vocal burrow]

(still too much for my 2070s :p)

[noble belfry]

@solemn atlas any joy?

[vocal burrow]

yes like that

[noble belfry]

i really want to just cheat and tell you to go download webui forge and it will sort out all your problems for you (which it will) but i also want to know what this issue is.

[urban arrow]

hello! anyone knows what I'm doing wrong?
I'm chaining OpenPoseXL2, a depth map controlnet (tried multiple), and a canny map controlnet (tried multiple too) in Comfy UI

Here's my workflow and here's my result, it's all noise

[noble belfry]

im starting to sound like one of those guys who hears a windows issue and tells people "just install linux", but with SD 😄

[solemn atlas]

so now I tired to generate and it works without problems. THANK YOU !! ♥️ I hope it says like this 😅 but how do i create two characters if i delated the other stuff?

[noble belfry]

well. webforge has this feature using masking

will require you to setup a fork of a1111, but its so much better. it has lots of tweaks and its made by the guy who made the controlnet stuff

[vocal burrow]

you can try to reinstall your extensions and use the methods described in the tutorials.... Theorically it should still work. Make sure to use the """"""""updated"""""""" composable-lora extension if you decide to go with that route tho https://github.com/a2569875/stable-diffusion-webui-composable-lora

Otherwise you might have problems with LoCon and LyCoris versions of LoRa

But to do multiple character I'd use regional prompting instead. I find it much easier.

cf https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111?tab=readme-ov-file#regional-prompt-control

(boobies/cleavage warning, I don't choose the illustrations :p)

"masking" is also another variant of regional prompting

[solemn atlas]

thank you so much, I will try it. its really difficult to do 2 characters

[vocal burrow]

It's difficult at first but once you get the gist of the syntax, it gets really easy. Simple example (don't be scared by the amount of extensions I have on my page)

This might help you understanding the syntax (same syntax different extension) https://github.com/hako-mikan/sd-webui-regional-prompter?tab=readme-ov-file#2d-region-assignment

tldr if we we take my simple "divide ratio" formula,
1,1 => there is no ; => there is no row seperation => there is only one row and it's split 1,1
1+1 = 2 => 1 is 50% of 2 => there are two colum each taking 50% of the space

if we take something like
2,1,1 => no ; => still only one row
2+1+1 = 4 => 2 is 50% of 4 and 1 is 25% of 4 => there will be three columns, the first one will take 50% of the screen space while the remaining two will take 25%

It gets "more complex" when you have multiple row
2,2,1,1;4 ; 3,1,1 => there is only one ; => there is two rows => let's check the first number of each subformula to see how the rows are split
2 and 3 => 2*100/5 = 40 and 3*100/5 = 60 => the first row will take 40% of the vertical screen space and the second row will take 60% of the vertical screen space
And afterwards you apply the same method as seen previously to seperate the horizontal screen space for each row
2,1,1 => 50% ; 25% ; 25% for each column of the first row
1,1 => 50% ; 50% for each column of the second row

[urban arrow]

same with 1.5, very weird

[vocal burrow]

@solemn atlas

[solemn atlas]

thank you so much, I'll just try it out

[winged kayak]

hello when i add zelda lora it render a weird yellow image with grain thing , how to solve it ?

[vocal burrow]

its weight is probably too high.

lower it to 0.6

[winged kayak]

its worse now

[vocal burrow]

you didn't lower it tho ? also easynegative should be used in the negative prompt

[solemn atlas]

you have in presets "lastrun" i dont have that. does that matter or is it not important?

[vocal burrow]

not important

leave it empty

[solemn atlas]

okay thank you.

so i can prompt after this

[vocal burrow]

sure, just use ADDCOL as mentioned in the template section to seperate your two prompts

as shown in my example where I prompted for

cute dog  ADDCOL 
cute cat

[solemn atlas]

oh okay thank you

[vocal burrow]

no, if you use that you'll switch to another way of seperating the space.

[wanton stirrup]

howdy, having an issue when using upscale: AttributeError: 'DecodedSamples' object has no attribute 'to'
comfyui works fine, is just stable diffusion vanilla that gives me this error

[solemn atlas]

okay so i tried it

but there is a mistake i'm doing but i dont know where the mistake is

[winged kayak]

yes i lower at 0.6 like u said

[vocal burrow]

you haven't activated it

[vocal burrow]

that's not what your screenshot is showing

[winged kayak]

oh the 1 represent the weigth

i wasnt underqtand get it

thx

[solemn atlas]

🤭 sorry,

so i did it now but i get a boy with both hair colors

[vocal burrow]

what did you change to 0.6 then ?

[vocal burrow]

that's probably gonna set the default weight for the next time you use it

[noble belfry]

also putting easynegative in positive prompts tell it to make the worst image it can. 😄

[noble belfry]

yeah thats for clicking it into an image

[winged kayak]

oh

[noble belfry]

i mean its a negative prompt, its going to TRY to make it bad hahah

[vocal burrow]

something is weird here. Why doesn't your divide ratio does not match the preview ?
Set 1,1 and click the Visual and make template button

[noble belfry]

if you put it in negative it will look at the bad images its trained on and tell it NOT to do that 🙂

[solemn atlas]

i did but it makes it automatic 1,1,1

[vocal burrow]

(didn't know there was a lora for easynegative tbh, I thought there was only the embedding)

[noble belfry]

early days i tried getting a language AI to make me better prompts from my input. it misunderstood positive and negative. so when i asked for a cat flying through space its negative prompts were hilarious: stupid cat, no, cats do not fly etc.

[vocal burrow]

why is it still 1,1,1 ?

[noble belfry]

nor i im wondering which would be better to use, memory wise

[solemn atlas]

now its 1,1 but still one character

[vocal burrow]

base ratio is empty ? 🤔

that's weird, set it to 0.2

[solemn atlas]

so now i also did that

[vocal burrow]

weird, screenshot the whole webpage + console log

[noble belfry]

dont use ckpts

convert them to safetensor files

[vocal burrow]

that's a good advice to follow overall, it's better security wise (even though that's probably not what's causing the issue)

[solemn atlas]

what is ckpts? sorry

[noble belfry]

you can do that in a111 using model merge and putting in the same model twice but selecting safetensors

[vocal burrow]

there's a bunch of regional prompting error in there (probably due to the time were you ran it without any base ratio) try to restart sdwebui

And also try without lora for a first try.

[noble belfry]

its not necessarily the current issue but ckpt files are model files that can possibly (rarely) execute malicious code. so we convert them to safetensors which takes that ability from them

a ckpt isnt malicious, it just could be. its still a model file though

[solemn atlas]

okay I'll just do it

[solemn atlas]

good to know thank you

ah now it worked when i used presets

[noble belfry]

the short of it is, checkpoint files have like suspended or "pickled" code in them that they run. safetensor format is same model but doesnt let that happen the same way. 😄

oooh progress!!!

[solemn atlas]

I'll try with lora too

[solemn atlas]

thats why we can choose which one we want to download on civitai. I always wonderd what that is.
Btw, is it possible to do 2 checkpoints on SD? I saw someone on civitai using 2 loras at the same time but i dont know which page he/she used to generate that

[solemn atlas]

https://civitai.com/images/5118097

with lora i get still one character

[white stirrup]

if i have multiple instances of the webUI running and change the output folder for one of the instances it wont change it for all instances correct?

[noble belfry]

if its the same program then yes

it will

[vocal burrow]

use multiple checkpoint ? no

[white stirrup]

same bat file im runing just multiple instances

[vocal burrow]

settings will be saved to the same file soooo I wouldn't count on it.

it depends of the setting and how it's refreshed

[white stirrup]

well one way to find out

[noble belfry]

yes it still will change for all, that variable is stored in a file. one window will ask it to change the file and the other instances will listen

[vocal burrow]

the last part I'm not sure of

the instances might just keep running with the settings they read at startup.

[noble belfry]

maybe, but the renderer on the server wont, it should listen to the last thing you told it

[vocal burrow]

it depends on how it's programmed and that I don't know

[white stirrup]

im testing it now

[noble belfry]

each instance wont tell it to change and post to THIS folder each time, they'll say put in the folder you use. only one changed the name

[white stirrup]

one instance is the default folder the other is the one i put

[solemn atlas]

ok it wont work whatever i try, sadly

[noble belfry]

well then, colour me suprised

[white stirrup]

indeed youre correct they will both go to the same folder

[vocal burrow]

I'm busy atm but for lora you probably want to use the "Latent" mode instead of "Attention"

[noble belfry]

still its fun to test out things like that and get the feel of software. that should hold true for anything that runs gradio, or most likely anything that runs as a server on your computer you access through your browser

[white stirrup]

i just got into AI stuff a few days ago

still trying to feel things out and get better at promt generation, loras etc

i feel as if im getting better

[solemn atlas]

tried it and still doesnt work sadly

[noble belfry]

yeah mine was a terrible malformed ball of god knows what when i first started.

nobody learned SD stuff from SD university, it doesnt exist 😄

that said, $60 to sign up for wit's SD university degree. Courses start same place as that willy wonka chocolate experience place 😛 jk

[noble belfry]

try adding "volumetric lighting" to your prompt for that.

it looks a little washed it might be a vae thing too. theres lots of things can make images pop and yeah, experiment and ask questions like crazy. 99% of what i know is from trial and error and other peoples trial and error.

god, i sound like one of those animes. together with friendship and working hard we can achieve our dreams if we just believe! 😛 or just lurk and nab other people's prompts 😉

[white stirrup]

yeah i go through Civitai from time to time to see how people make prompts

but tbh i want to learn more than copy

because i wont be able to find everything i want online

[white stirrup]

looks kinda blurry

but tbh thats because i switch to a different model

[noble belfry]

which model is it?

the dim wash looks like it could do with a better vae. is it sd1.5?

[white stirrup]

it was think diffusion XL

but i mostly use dreamshaperXL it yields the best results imo

but TDXL seems to have a more accurate dataset

[white stirrup]

what is a vae?

[noble belfry]

hard to explain in one sentence but it kind of tells the model rendering where to put colours and really makes an image pop. your sky would be more vibrant, the trees less dull. etc.

how many steps are you using for it?

[white stirrup]

75

[noble belfry]

something isnt quite right then. how long does that take to render?

[white stirrup]

about a minute and a half

im using an rtx 3080

[noble belfry]

you should be getting light years better than that

could you post a screenshot of your screen like this?

[noble belfry]

the bit above it too that shows the model

[white stirrup]

trying the DPM++ 3M SDE Karras sampler

nah it made it worse

[vocal burrow]

it's probably because both prompt describe a person, so it get confusing for sd. Try to use a "common prompt" as per my example here, it helps sd get an idea of what the overall composition should be like.
(still busy and probably will be for the rest of the night, covid is a bi***)

[noble belfry]

@vocal burrow not to multitask you but what would you say is wrong with AC3s images not popping?

[vocal burrow]

I didn't follow everything but ... wrong vae ?

[noble belfry]

whats a good vae for thinkdiffusionxl?

[vocal burrow]

hmmm I don't think I have any custom vae for sdxl apart from the fp16 fix one

[noble belfry]

link it up?

i really only use lightning models which are fine without vae honestly

[vocal burrow]

https://civitai.com/models/140686?modelVersionId=155933

[noble belfry]

put that in your models\vae folder AC3

[white stirrup]

I will

Thanks

[vocal burrow]

if all you want is for your output to "pop" (colour wise) you can also try to correct them afterwards with https://github.com/hako-mikan/sd-webui-cd-tuner

or https://github.com/KohakuBlueleaf/a1111-sd-webui-haku-img/

[noble belfry]

thanks! 😄

but sometimes it is nice to just push button after your thoughts are typed in and receive a lovely image

the other thing is that dreamshaper is still at the end of the day finetuned on faces and portraits and action shots

holy LOL. try this: replace tram with train.

i think its just tram-ist since it isnt well trained on them

and maybe only has olde photos or san francisco to go on

[vocal burrow]

Yup I get you. tbh fever is getting the better of me so I can hardly focus and read their 5 feet tall prompt. I'm sure there is some room for improvement in the prompt phrasing itself or maybe it's just that the model itself tends to do grey stuff

[solemn atlas]

Get well soon!!

Okay I'll try it. Thank you so much for your help so far. Its also late here so I try it tomorrow

[noble belfry]

no problems

none at all, i like a good mystery 😄

[vocal burrow]

for further questions please read https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111?tab=readme-ov-file#regional-prompt-control and https://github.com/hako-mikan/sd-webui-regional-prompter. Those are long readme for sure but it's because regional prompt control has a lot of features. There's also https://stable-diffusion-art.com/regional-prompter/ (not using the same extension but the overall idea should be the same)

[high knot]

Still same error message:

RuntimeError: unexpected EOF, expected 413788 more bytes. The file might be corrupted.

Created new venv, cleared venv cache. Also tried reinstalling the extension, dependencies, even python.

Any other ideas? 😬

[vocal burrow]

what's the complete log ?

[solemn atlas]

thank you so much!!

[vocal burrow]

urgh >< reactor again

why is it always reactor

Can you please put the full log in a .txt and drop it in here

[noble belfry]

as a rule i dont help with faceswap things. i know you have to help, but in that case i tap out as you dont know who you're helping do what. 🙂

plenty of legit uses of it, but its iffy and ive come into a few cases of people asking weird stuff with dubious premises. not throwing shade at you tblake, its just if support falls silent over faceswapping thats generally why

[vocal burrow]

I also tend to stay away from faceswap thing... It's not my thing and I never use them.

[high knot]

Understandable. I'm just trying to put my wife in various historical eras for funzies (with her blessing btw!) 🙂

[vocal burrow]

And luckily for me cause loop and reactor seems to be really bad at installing their dependencies XD

[noble belfry]

its one of those "how can i dispose of an animal roughly the size of my 70 year old grandmother" questions 😄 😄

[solemn atlas]

okay, i could do exactly one time two characters in one picture

but it only happend once. So i guess SD does what it wants

😅 .

[noble belfry]

but you couldnt give her pants. damn 😛

[solemn atlas]

the prompts do what they want haha

I prompt "shorts" and got this haha but getting 2 characters at once only happend once sadly

[noble belfry]

they do! and thats another thing. putting two people together will often just get bleh couples poses. one thing you CAN do for interesting poses is to partially make an image with anime models which love all kinds of striking poses, then ask a photorealistic one to do the heavy lifting after

gives the photo model enough to build over

[river flame]

I m trying to install StableSR upscaler, the guide says to put it into the 'models' folder in the sd-webui-stablesr folder, but there s no 'models' in here, where so I put it?

[noble belfry]

also it could be the 19:1 or whatever ratio that is. try 4:3

[vocal burrow]

Try rephrasing the common prompt maybe, also you could set a a "base prompt" https://github.com/hako-mikan/sd-webui-regional-prompter?tab=readme-ov-file#use-base-prompt the more you constraint sd the more likely you are to get what you want. Following that spirit you can tweak the cfg value. In theory the higher it is, the more it should follow the prompt (but passed a certain value it will be counterproductive, usually if you go higher than 10 you're probably doing somehing wrong)

[solemn atlas]

that sounds interessting. So i should use an already anime fanart and use that as reference or how exactly can i do it?

[noble belfry]

you got it!

also, such a small thing but this will improve images so much:

[vocal burrow]

what's your hardware ?

[noble belfry]

SD is trained on images. people who make couples images dont make wallpaper resolutions. so if you make them standard photo resolutions it has more of an idea

[high knot]

4090 / 7950x3d

[noble belfry]

4:3 or 3:4 aspect ratio is a very good one

[solemn atlas]

that ratio is def. better

[noble belfry]

well, its half tech half art, half jiggling things till they work and the other half fractional math

[solemn atlas]

So i read that part but I'm a little bit confused

a woman and one man holding hand ADDCOMM
1boy, Sasuke, a man, handsome, with purple skirt, wearing SSK_outfit, posing with karin for a picture, left shoulder exposed, abs, tattoo, lora:sasuke:0.8 ,
ADDCOL
1girl, karinuzumaki, red hair, red eyes glasses, purple jacket, long sleeves, shorts, thighhighs cowboy shot, best quality, highly detailed, ertestyle, lora:naruto_karinuzumaki-11:.8

a woman and one man holding hand ADDCOMM
1boy, Sasuke, a man, handsome, with purple skirt, wearing SSK_outfit, posing with karin for a picture, left shoulder exposed, abs, tattoo, lora:sasuke:0.8 ,
ADDCOL
BREAK
1girl, karinuzumaki, red hair, red eyes glasses, purple jacket, long sleeves, shorts, thighhighs cowboy shot, best quality, highly detailed, ertestyle, lora:naruto_karinuzumaki-11:.8

like this?

[solemn atlas]

what can i do with that or do i just activate it?

[noble belfry]

yeah https://github.com/thomasasfk/sd-webui-aspect-ratio-helper install it

it adds little boxes to pick aspect ratio

and a little padlock icon to keep the aspect ratio and scale up/down

as i say a lot of finetunes are trained on standard photo sizes so it makes it a bit smarter.

[vocal burrow]

1/ you probably want to use --opt-sdp-attention or --xformers command line arguments
2/ you also want to use the SDXL fp16 vae I linked above because right now you have nans error in your VAE stage
3/ you're using CUDA 12.1 , you might want to downgrade to 11.8 https://github.com/Gourieff/sd-webui-reactor/issues/310

[noble belfry]

sdxl is 1024x1024 but you still will get better images if they're shaped like actual photographs

[vocal burrow]

assuming you're not using a base prompt here, it looks good to me

[high knot]

Really appreciate your time, thanks so much! I'll give all of this a go 🤞

[solemn atlas]

thank you so much but how is it if i put BASAE prompt? Do i do it at the beginning and also mention it on the template from regional prompter too?

[white stirrup]

@noble belfry how exactly do you use a VAE?

[noble belfry]

one second ill help you put the menu on your main screen

[white stirrup]

thanks

[vocal burrow]

if you tick the base / common prompt and then click the "Visualize and make template" it will show you exactly what to do in your prompt.

[vocal burrow]

VAE quick guide :
1/ What is a VAE.
It's a part of the stable-diffusion pipeline that encode/decode information from tokens to latent space and from latent spaces to pixels. Aka it transforms math statistics in pictures
2/ Where do I put my VAE ?

• VAE with .vae.pt, .vae.ckpt, .vae.safetensors extensions go into the models\Stable-diffusion folder
• VAE with .pt, .ckpt, .safetensors go into models\VAE
  3/ How do I use my VAE ? Three possibilities :
• Either you name it similar to another one of your model (eg : Anything-V3.0.safetensors + Anything-V3.0.vae.pt), by doing that it should automatically load the VAE when you load the associated model.
• You manually load your VAE by going to Settings -> Stable-Diffusion -> sd_vae and selecting your VAE
• You add an easily accessible VAE dropdown at the top of your page to quickly switch back VAE by adding sd_vae to your Settings -> User Interface -> Quicksettings list

[noble belfry]

that saves my boot time 😄

also do the last step, its useful now and then

[solemn atlas]

ok now I'm here again. 😢 i closed it and started webui again but yeah, thats what i get the whole time

[vocal burrow]

I have a long copypasta.txt file with lots of tuts ready to be thrown in here.

[noble belfry]

oh ffs

[white stirrup]

i just needed the settings part thanks

[noble belfry]

i'm sure ive a few i could add to it 😄

[white stirrup]

oh god

[noble belfry]

its....bright! 😄

[white stirrup]

wait i didnt switch to euler a

[noble belfry]

try another prompt that isnt a mechanical landscape

[white stirrup]

nah

[noble belfry]

describe the bits of the tram

[noble belfry]

itll make the model focus more on it

[vocal burrow]

cfg value is probably incorrect, or you're using lora with some incorrect strength settings

[white stirrup]

no lora

[vocal burrow]

then what's your cfg ?

try lowering it a bit

[white stirrup]

its at 3

[vocal burrow]

oh

XD

btw that might be more a topic for #📝｜prompting-help

I'll be off for now, head is spinning, sleepy time it is.

[noble belfry]

yeah prompting help might be better

[covert anvil]

Hi, is it possible to run Stable Cascade in ComfyUI using Cuda 11.4?

[solemn atlas]

Good night, get well soon and thank you for your help 👍

i dont understand this 😭

[magic hollow]

hello im not sure if my graphics card is enough for stable diffsuion, im trying comfy with directml but it crashes my pc when i try to create an image my card is an amd rx 480 and im not sure if there is a version of stable diffsuion i can run or not.

[ornate elk]

Auto1111 can run it with some tweaks

You find my directml guide in the pinned messages of this channel (you need the Directml guide, not the Zluda guide)

[magic hollow]

well thank you for letting me know this.

[ornate elk]

NP, But I can't guarantee that it won't crash

But its worth a try

You would need --lowvram and use 2gb models

Then it should work

[magic hollow]

alright good to know.

do i still use lowvram if its an 8vram card?

[ornate elk]

Yes because directml isn't optimised for vram management

You can try --medvram first

But if it crashes use --lowvram

[magic hollow]

well alright and i think ill try lowvram first

[ornate elk]

Also you can't go much higher than 512x768

In resolution

But upscaling should be possible

In img2img

[magic hollow]

Well dang but i guess i can live with that

[ornate elk]

Yea its not an issue if you can upscale

[magic hollow]

thanks for letting me know all of this, and ill probably have more issues in the near future.

[ornate elk]

No problem, for any questions feel free to ask here

[ornate elk]

If you have set everything up like in my guide, and tested, let me know if it worked

Also fp8 support got added recently so that could help a lot too

[magic hollow]

give me moment im installing python

yeah so i go this error ERROR: Could not find a version that satisfies the requirement torch-directml (from versions: none)
ERROR: No matching distribution found for torch-directml

[ornate elk]

What's your python version?

[magic hollow]

i just install 3.10.6

but i also have a newer version installed aswell for coding.

[ornate elk]

It probably uses the wrong one then

[magic hollow]

how do i switch it?

[ornate elk]

Whats the other python?

[magic hollow]

i believe 3.11.8...

[ornate elk]

Ah okay

[ornate elk]

So then you need to open up a cmd.
Then type
Where python

Then copy the path it shows for the python 310
Copy that.
Edit the webui-user.bat and paste it at the line PYTHON=

Then delete the venv folder and relaunch the webui-user.bat

[flat olive]

i have rx580

im using directml and the --use-directml on commandline_args

[magic hollow]

now i have to wait for torch to install.

[magic hollow]

does it generate images well?

[ornate elk]

Wrong arguments, remove --skip-torch-cuda-test

[flat olive]

the rx580 yep, but you cant make videos

[flat olive]

ok

same thing

[ornate elk]

Delete the venv folder and relaunch

[flat olive]

ok

[magic hollow]

what about reapting textures or openpose?

[ornate elk]

Also please take a look at my AMD install guide from here. You need more COMMANDLINE_ARGS.
AMD Directml Guide:
#🤝｜tech-support message

[magic hollow]

i must say this guide is very helpful.

[wispy wasp]

Got an amd and nvidia gpu, I went with the directml version because the nvidia is faster for text generation. I'm currently trying to get it to run on the amd card and it just won't. It still uses the nvidia card.

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --use-directml --skip-ort --medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling --device-id 1 --reinstall-torch
set CUDA_VISIBLE_DEVICES = 0

call webui.bat

[ornate elk]

What's the AMD GPU?

[wispy wasp]

580

[ornate elk]

Okay then you need to do the following:

Open up a cmd and type
Pip cache purge

Then you need to remove --reinstall-torch --device-id 1
And remove set Cuda_visible_devices 0

Then delete the venv folder and relaunch the webui-user.bat

[magic hollow]

so i got it up and i put easynegative in the embedding folder before launch and this is what i get.

[wispy wasp]

still didn't work

the only thing i could think of is maybe the torch version is wrong?

==============================================================================
You are running torch 2.0.0+cpu.
The program is tested to work with torch 2.1.2.
To reinstall the desired version, run with commandline flag --reinstall-torch.
Beware that this will cause a lot of large files to be downloaded, as well as
there are reports of issues with training tab on the latest version.

Use --skip-version-check commandline argument to disable this check.

[magic hollow]

weird now it shows up

alright after 1 image so far it hasent crashed my pc, so thanks.

[noble belfry]

we take cats as thanks

😛

[magic hollow]

is that true?

[noble belfry]

yes, the devs gave me a tour and SD is run entirely on cats, its like the matrix, but cats. the horrible price we pay.

[ornate elk]

What's your webui called?

Stable-diffusion-webui or stable-diffusion-webui-directml

[ornate elk]

Nice!

[wispy wasp]

stable-diffusion-webui-directml

[ornate elk]

Okay good.

[wispy wasp]

i'm disabling my nvidia gpu and seeing what happens

[subtle quarry]

Okay so I can't seem to get stable diffusion to reinstall xformers despite using --reinstall-xformers

and --reinstall-torch throws me a bunch of warnings.

[wispy wasp]

it ******* worked

[subtle quarry]

this is all that it does when throwing --reinstall-xformers at the command-args

[ornate elk]

You can always delete the venv folder and it will reinstall everything.
If you want to make sure to get the latest stuff you should open up a cmd and run
Pip cache purge

Then you dont need --reinstall-torch and --reinstall-xformers

[subtle quarry]

Do I need to update any of this stuff?

[ornate elk]

You dont need to but you can

[wispy wasp]

Shot in the dark but does anybody know how to choose amd gpu over nvidia gpu on the directml fork?

[ornate elk]

Maybe with --device-id but without cuda visible devices

[wispy wasp]

why does it decide to "work" now?

It uses the amd gpu then errors out with modules.dml.pdh.errors.PDHError: Something went wrong.

restarting to see if it happens again

it did

[noble belfry]

if something says "could not locate" or "import error X is not a Y of Z" nearly every time its either your client is out of date or the extension is. someone is pointing at something in another house they have moved.

when in doubt, update 🙂

it might be one or the other so you can rarely eyeball which one is at fault but updating everything always helps.

cs1o what happens when they finally stop using python 10 bud. what do you do? flight to mexico?

i applaud it as i was one of those last ones who stuck with 2.7 over 3. 😄

[wispy wasp]

solved it, --lowvram was the culprit.

[warm tusk]

Do lightning models need their own loras or do xl ones work?

[noble belfry]

loras do not work well with lightning at all

a lightning model in its own way IS a kinda big lora in what it does, it pushes certain things towards you. you wouldnt have much luck with a lora on one. you can try taking a lightning render and using another model, say an sd1.5 to add loras after but its a mess

if you have say a 30XX or higher, dont really use lightning models unless you want to generate a lot of images quickly

[warm tusk]

Do other similar things like textual inversions and hypernetworks work?

[noble belfry]

not on lightning models, no not really. you can do great things with prompts though

embeds might work though i havent tried.

i dont see why they should though

there might be some changes eventually, but otherwise consider them tweaked already, and any other tweaks might muddy it up

you can try though

[dire hawk]

Thank you.

Newbie here. I tried to run my stable diffusion and looking for the answer from youtube and end up here. This solves the issue

[brittle oak]

Hello everyone, does anyone know how to solve this in A1111 with colab pro....?

[river eagle]

Howdy howdy, all! Hey -- first a question... So, I've been creating in SA with a sad old RTX 2070 Super (8gb). It can chug through OK, but yes, it does crash and burn quite a lot more than I'd like. Thinking about making the jump to Red Team and upgrading to a RX 7800 XT 16GB. Anyone have trouble with AMD cards running SA or alls well and rockin'? Oh, and anyone have the 7800 XT yet? And if so, are you loving it?

[brisk cedar]

How can i make controlnet units side-by-side instead of having to scroll down to the next unit? (forge)

[vocal burrow]

I would stick to nvidia if I were you .... AMD is barely starting to look good on windosw thanks to zluda.

And even then, it has a lot of little quirk that makes it incompatible with some extensions, it tends to break much easier than green team, use more vram for similar operation, etc

[ancient plover]

can someone please clarify the usable weight range for loras? I have read somewhere that all lora's are strictly weighted from 0 to 1, (or -1 to 1 in some cases) but i have seen a few pages on civitai that tell you to use weights over 1, or the regular posted images with prompts that include values over 1 for loras... So what is the truth? And if it is the case that loras can have different ranges, how can you easily determine what the range for some given lora if it isnt explicitly stated by the author on civiai?

[onyx solar]

Depends on the LoRA. Refer to each LoRA's documentation for the ranges they suggest you use.

If not stated, you have to toy with the settings to be sure.

[ancient plover]

ok, I thought so, thanks, Also, this is random, but is there any issue with renaming your lora files? Will a renamed lora file still work properly assuming you change the lora-call text to match?

[onyx solar]

Depends on what tool you use...it just may mean you need to change how you refer to it.

[ancient plover]

Im using webui auto 1111, so for example I renamed one of the detail adding loras, that came with a super long name to : "add-detail-xl" and changed the call phrase to "lora:add-detail-xl:1"

[solemn atlas]

good morning guys, it's me again,. Sadly I have the same problem as yesterday

so I delated the venv folder and now opening webui again

but it still doesnt work

[river flame]

I found this DarkLight Lora, fell in love with it, installed, reloaded, but it doesnt show in loras ._.

[paper crag]

total noob question, but if models aren't appearing in my lora page even though i have them in the folder, that means the said lora is not compatible with my current base model right? (i.e. pony or sdxl)
So that means i have to get that base model that is compatible with it.
So can i just update to that model if im using basic one?

[nocturne topaz]

you need to use a full model as the checkpoint, not a lora?

[solemn atlas]

what do you mean by that?

yesterday it worked with that

[shut eagle]

Why I can't use prompt?

[ancient plover]

yeah, loras and embeddings don't work with models between 1.5 and SDXL, so depending on whatever base model you currently have selected, you will only see loras and embeddings that are compatible show up. if you go on civitai, you can sort models by their base versions so you will only see things that work together,

[paper crag]

can i update to sdxl?

[ancient plover]

depends on what you are trying to do, SDXL can do some new stuff way better than SD 1.5, but it also has alot less models to choose from, so in some ways sticking to 1.5 with way more embeddings and loras, in quantity and usefulness may be better... Also, XL seems a bit slower and requires a beefier pc if that means anything.

[vocal burrow]

then you were very very very lucky. Loras are not SD models, they're patches for SD models

LoRa/Lycoris quick guide :
1/ Where do I put my LoRa : stable-diffusion-webui\models\Lora, Loras are NOT stable-diffusion model but rather patches for said models.
2/ How do I use my LoRa, you can either :

• Invoke it directly in your prompt by using the following syntax <lora:filename:weight> (or <lora:alias:weight> depending of the setting in setting->extra network). eg : <lora:animeTarotCardArtStyleLora_v20Offset:1> for using the Lora animeTarotCardArtStyleLora_v20Offset.safetensors
• Use the UI "Extra Networks" / card system by doing as described in my screenshot. Clicking an associated lora card should automatically insert the lora prompt liked mentioned above. https://cdn.discordapp.com/attachments/1002602742667280404/1191779742534348870/loraHowTo17.jpg
  3/ Be careful, LoRa are trained specifically for some SD version. Lora trained for SD1.5 (and derivatives) models will not load on SD2 models and vice versa (same goes for SDXL). Also even though Lora trained on SD1.5 models should work for every SD1.5 derivatives, they'll probably work best with the model that has been used for its training. Therefore, (by default) only Loras compatible with the model you currently have loaded will show up.
  4/ LoRa, LoHa, LoCon, what are these ? They're all commonly refered as "Lora" but differ slightly technically :
• LoHa : Lora with hadamard decomposition
• LoCon : Lora extended to redisual blocks.
• More explanations : https://stable-diffusion-art.com/lora/, https://www.sdcompendium.com/doku.php?id=lora_0014, https://www.sdcompendium.com/doku.php?id=locon_training_0071, etc

[vocal burrow]

If the questions is "can I update my sd1.5 loras to sdxl loras" or "can I somehow use sd1.5 loras with sdxl" .... Then the answer is "yes... kinda..." It's still very experimental, cf this repository https://github.com/showlab/X-Adapter?tab=readme-ov-file and as far as I know there is no implementation for auto1111

And if you don't understand any of the things written in there. Then take it as "no it's not possible yet"

[vocal burrow]

#1047610792226340935

[ornate elk]

@cosmic furnace hey sdxl works now with directml zluda

[solemn atlas]

Wait if I understand that correctly. Whatever I download from Civitai is not Lora? I downloaded what people call basemodel like "aniverse" and put them on the StableDiffusion Models folder

this is my Lora folder

and this my models-> SD

I always chose "base models" for example "aniverse" in the SD-checkpoint at the beginning before I start txt2img

I thought the Sakimi-chan is a base model so I put it in the SD folder for "checkpoint"

these are my LORAs but the don't have a picture on it sadly

Now all of a sudden I can generate again, I really don't understand this and later probably it won't work again

[jade ether]

Hi, im trying to get SD up and running for the first time.
Im using Win 10.
I ran all the commands and got stuck on:
source .pt2/bin/activate
Whats the Win equivalent?
I want to install as little extra software as possible.
Meaning, if possible, no Anaconda...

If you want to help, you can DM me to not clutter the chat if you like

[vocal burrow]

There's a bit of everything on civitai, SD models, Loras, Embeddings, etc...
It's specified everytime

On the cheyenne's page you can also notice that there are multiple versions of this model

[vocal burrow]

what are you trying to run exactly ?

[solemn atlas]

so I should put the Sakimichan lora into the lora folder

so everything which has the name checkpoint in it comes into the SD folder and every other which is called Lora into the Lora folder

Maybe that was my mistake

[jade ether]

https://github.com/Stability-AI/generative-models
Im trying to set up SD for the very first time. I followed the Install Steps in the readme and got stuck on that one:
source .pt2/bin/activate
IDK what it wants to do exactly, because it only works on UNIX based systems.

[vocal burrow]

• "so I should put the Sakimichan lora into the lora folder" => most likely yes. one easy way to determine if you're dealing with a lora or not is : read the description (duh) and if you can't read check the file's weight. SD models are at least 2gb. Therefore if it's less than that it's probably a lora or an embedding.
• "everything which has the name checkpoint" => probably but some names can be confusing so I wouldn't use only that to determine the type of file I'm dealing with.

[solemn atlas]

thank you so much, that really helps me lot. I will just do that

this is what i can do now. Thank you so much for all your help. Now i need to find out how to make them even better

[vocal burrow]

While that is indeed the official "stable diffusion" repository I would advise you to use a third party UI to run stable diffusion. As they come with multiple optimizations, extensions, an actual freaking UI, etc...
The most popular one is https://github.com/AUTOMATIC1111/stable-diffusion-webui/ => (mostly known as auto1111) fast, very large community, many many extensions, etc... But it supports only nvidia gpus on windows (let me know if you're an AMD user)
The second most popular https://github.com/comfyanonymous/ComfyUI/ => fastest (by a little), node based UI, more complex to learn but much more tweakable, good amount of extension
And for the third spot it would probably be either https://github.com/lllyasviel/Fooocus => fast, very easy to learn, not really tweakable, it's basically for beginners and people who just want a "press button to get pretty images" but don't want to learn how to properly use SD
Or https://github.com/Stability-AI/StableSwarmUI => fast, made by stability.ai team, it's kinda an hybrid of comfyUI and auto1111 as it's using those as backend for rendering, there is some extensions for it too, it's been maintained for quite a while so by now so it's also a serious solution.

sd.next is also worth a look https://github.com/vladmandic/automatic it started as a fork of auto1111 and it's kind of a "bleeding edge" version of it. If there's a new tech, someone is gonna jump on it and implement it for sd.next (wheter it makes sense or not :p).

There are tons of (local install) options out there. But I believe that those are the main ones

[jade ether]

Alright thank you very much. I start by giving webui a shot. I guess I come back if I run into another problem 😄 cheers

[solemn atlas]

Can someone have auto1111 and comfyUI at the same time?

[vocal burrow]

sure can, that's why they use "venv".

I have comfyUI, auto1111, forge and sdnext installed atm

tldr "venv" folder is kind of a "virtual machine" for python.

instead of using your system's python and libraries, auto1111/comfyUI/etc creates a python "venv" thus copy pasting your python stuff in there and installing all their own stuff in this very same folder.

[solemn atlas]

Oh I didnt know that. Wha is the diffrence between them?
so if i want to have comfyUI too, do i download it from extensions or from github via URL?

[vocal burrow]

That way instead of installing a bunch of stuff in your system it's all contained in the venv

[vocal burrow]

Get the standalone comfyUI. In my opinion unless you have a very specific workflow you should never use comfyui auto1111 extension.

It makes no sense to have both of them running at the same time and sharing same folder.

[solemn atlas]

i understand, so the standalone means downloading it?

[vocal burrow]

standalone means "don't get the comfyUI auto111 extension but the proper independant comfyUI installation from https://github.com/comfyanonymous/ComfyUI/"

[solemn atlas]

okay so I download it normal not with the Code

[vocal burrow]

never use the Code->"Download as a zip" button unless explicitely specified to do so

[solemn atlas]

okay

so I extract it

[jade ether]

So, Im currently trying to get webui up and running. Setup seems easy enough. Clone the repo, click the batch. Running the batch throws some errors:
Loading weights [fb503fecb4] from C:\Users\stable-diffusion-webui\models\Stable-diffusion\v1-5-pruned-emaonly.safetensors
loading stable diffusion model: SafetensorError
Traceback (most recent call last):
File "C:\Users\me\AppData\Local\Programs\Python\Python310\lib\threading.py", line 973, in _bootstrap
self._bootstrap_inner()
File "C:\Users\me\AppData\Local\Programs\Python\Python310\lib\threading.py", line 1016, in _bootstrap_inner
self.run()
File "C:\Users\me\AppData\Local\Programs\Python\Python310\lib\threading.py", line 953, in run
self._target(*self._args, **self._kwargs)
File "C:\pathToRepo\Repos\stable-diffusion-webui\modules\initialize.py", line 148, in load_model
shared.sd_model # noqa: B018
File "C:\pathToRepo\stable-diffusion-webui\modules\shared_items.py", line 133, in sd_model
return modules.sd_models.model_data.get_sd_model()
File "C:\pathToRepo\stable-diffusion-webui\modules\sd_models.py", line 621, in get_sd_model
load_model()
File "C:\pathToRepo\stable-diffusion-webui\modules\sd_models.py", line 706, in load_model
state_dict = get_checkpoint_state_dict(checkpoint_info, timer)
File "C:\pathToRepos\stable-diffusion-webui\modules\sd_models.py", line 331, in get_checkpoint_state_dict
res = read_state_dict(checkpoint_info.filename)
File "C:\pathToRepo\stable-diffusion-webui\modules\sd_models.py", line 305, in read_state_dict
pl_sd = safetensors.torch.load_file(checkpoint_file, device=device)
File "C:\pathToRepo\stable-diffusion-webui\venv\lib\site-packages\safetensors\torch.py", line 308, in load_file
with safe_open(filename, framework="pt", device=device) as f:
safetensors_rust.SafetensorError: Error while deserializing header: MetadataIncompleteBuffer

Stable diffusion model failed to load
Applying attention optimization: Doggettx... done.

[river flame]

I guess that s a tech question. Upgrading my 1080ti. 2080ti is 50% more it/s than 3060ti and 50% more expensive. But it s also 3gb vRAM more - is it a big deal, should I spend more?

[noble belfry]

I'd wait a little and see what SD3's releases are.

That said, 10xx is a bit old hat, i have one myself and 30xx is going to last you a good few years in non ai stuff. but at the minute a gpu is just a tax on your new machine. its really, really silly prices

[vocal burrow]

what's the full log ?

[jade ether]

venv "C:\pathToRepo\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: v1.8.0
Commit hash: bef51aed032c0aaa5cfd80445bc4cf0d85b408b5
Launching Web UI with arguments:
no module 'xformers'. Processing without...
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.
Loading weights [fb503fecb4] from C:\pathToRepo\stable-diffusion-webui\models\Stable-diffusion\v1-5-pruned-emaonly.safetensors
loading stable diffusion model: SafetensorError

Traceback (most recent call last):
File "C:\Users\me\AppData\Local\Programs\Python\Python310\lib\threading.py", line 973, in _bootstrap
self._bootstrap_inner()
File "C:\Users\me\AppData\Local\Programs\Python\Python310\lib\threading.py", line 1016, in _bootstrap_inner
self.run()
File "C:\Users\me\AppData\Local\Programs\Python\Python310\lib\threading.py", line 953, in run
self._target(*self._args, **self._kwargs)
File "C:\pathToRepo\Repos\stable-diffusion-webui\modules\initialize.py", line 148, in load_model
shared.sd_model # noqa: B018
File "C:\pathToRepo\stable-diffusion-webui\modules\shared_items.py", line 133, in sd_model
return modules.sd_models.model_data.get_sd_model()
File "C:\pathToRepo\stable-diffusion-webui\modules\sd_models.py", line 621, in get_sd_model
load_model()
File "C:\pathToRepo\stable-diffusion-webui\modules\sd_models.py", line 706, in load_model
state_dict = get_checkpoint_state_dict(checkpoint_info, timer)
File "C:\pathToRepos\stable-diffusion-webui\modules\sd_models.py", line 331, in get_checkpoint_state_dict
res = read_state_dict(checkpoint_info.filename)
File "C:\pathToRepo\stable-diffusion-webui\modules\sd_models.py", line 305, in read_state_dict
pl_sd = safetensors.torch.load_file(checkpoint_file, device=device)
File "C:\pathToRepo\stable-diffusion-webui\venv\lib\site-packages\safetensors\torch.py", line 308, in load_file
with safe_open(filename, framework="pt", device=device) as f:
safetensors_rust.SafetensorError: Error while deserializing header: MetadataIncompleteBuffer
Stable diffusion model failed to load
Applying attention optimization: Doggettx... done.

[vocal burrow]

most likely something went wrong during the installation

network error or something like that

try deleting the model in stable-diffusion-webui\models\Stable-diffusion and start again

[river flame]

I want to upgrade now. Does vRAM affect image quality or just possible resolution and generation speed?

[vocal burrow]

it doesn't impact image quality, only what's possible to do

[river flame]

but what you mean what s possible? sounds frighening to downgrade from 1080ti 11gb to 3060ti 8 gb

what will I lose? will I still be able to make 1024x1024 pics?

[vocal burrow]

you'll have to use --medvram-sdxl or use fp8 mode with forge to get sdxl model to fully load on your gpu without offloading to your system's ram.... tldr you'll get slower results than if you had 12gb

[river flame]

ohhh.. I almost bought it)))

[vocal burrow]

now will it be slower than your 1080 ti ? probably not

but it would be slower than a 3060 with 12gb that can fit the whole model into its vram

[river flame]

I was just looking at PudgetBench

[vocal burrow]

https://vladmandic.github.io/sd-extension-system-info/pages/benchmark.html

here's some benchmark for you to decide.

just know that the less vram you have the harder it is (but still possible) to get high resolution output

the harder it is to train lora

etc

[river flame]

okay I guess I have to go higher with 2080ti

[vocal burrow]

There are 3060 12gb version too

[noble belfry]

i assumed thats what we were talking about! misread

[lament wyvern]

can we still run auto1111 or warpfusion specifically online?

[modern zodiac]

Is there a way to save the metadata (parameters) created by A1111 when editing the image in a photo editor?

[vocal burrow]

not sure I understand the question but it sounds like it depends about your photo editor.

[feral sigil]

Hi, is there any way to run SD on Colab free? I get a runtime error before I even manage to start the application.

[modern zodiac]

Thx. Yeah I guess mine just does not save it. Bummer...

[shut cedar]

Is it possible to run SD on my AMD?

[karmic crown]

probably, wich one?

[ornate elk]

Yes you can checkout the two guides in the pinned messages of this channel for AMD

Directml works on any AMD GPU while zluda works only on 6000 series or newer (but zluda is much faster)

[shut cedar]

7800 XT

[shut cedar]

Sweet, thanks!

[karmic crown]

then you have two choices as CS1o wrote

[dense sandal]

I updated Auto1111 but it says it's still using torch 2.0.1 + cu118

Okay I put --reinstall-torch --reinstall-xformers I think

[dense sandal]

How can I revert to 1.7.0 now? xd

[karmic crown]

whats wrong?

[hoary lodge]

Any ComfyUI users here? if so, anyone having any CUDA Issues lately? I tried runing a simple SVD workflow, and I got this:
File "C:\ComfyUI_windows_portable\python_embeded\lib\site-packages\torch_ops.py", line 755, in call
return self._op(*args, **(kwargs or {}))
RuntimeError: CUDA error: invalid configuration argument
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

[dense sandal]

deforumation doesn't work on 1.8.

[karmic crown]

you need to update all plugins as well, not only a1111
have you do that?

[dense sandal]

yup

[karmic crown]

show me the log

[dense sandal]

the author of the plugin told me I had to revert

[karmic crown]

hmm

i have a new laptop

so i just installed everyting, but i havent checked yet

give me a second

ModuleNotFoundError: No module named 'basicsr'

this one?

[dense sandal]

if you need to know.. that's what I got.

I am reverting to 1.7.0 now.

also keep in mind I said deforumation, not deforum.

[hollow harbor]

How can I hide the PNG Info tab from the Automatic1111 webui entirely?

I use another extension for that and I don't need both to be there.

[noble belfry]

there should be a "hidden ui tabs" dropdown in the settings. im on forge so it might have moved things around

but you should be able to do this

[limber pier]

i wouldn't go lower than 12gb vram

if it's at all possible for you to afford a 12gb card

i was using a 12gb 3080 before and i was sooooo glad it wasn't the 10gb version cuz with sdxl you're constantly noticing you're at 10.8gb, 11.2gb, etc

if you get within 0.5gb of your limit, at least on windows, it'll slow to a crawl

[rancid harness]

Since update 1.8, deforum isn't working for me anymore. Is anyone else experiencing issues with it? Deforum isn't visible for me

[karmic crown]

i just experienced that myself

[hollow harbor]

Brilliant! I missed it rushing through the settings. Thanks.

[karmic crown]

you have two choices, to revert to 1.7.0 or to wait for a fix

easiest way to revert a1111 to 1.7.0 is this

git reset --hard cf2772f

when they fix deforum you can update a1111 with

git pull

[olive snow]

guys hello

can someone please explain to me why it says that

i am trying to do a SDXL model training

and just after it is reading the catching latens.. it says this

have tried it 5 times

[solemn atlas]

hey guys, now i have a question. how can i download Composable Lora, controlnet and Latent Couple without its bugged like yesterday

[rancid harness]

Thank you 🙂 I'll try it out right away

[karmic crown]

i tried to fix this, and i managed to make it appear in the menu, but clicking the button does not do anything 🤷🏻‍♂️

there is some library conflict, and i do not want to play with that, i can wait

strange thing is that everything looks fine, and there is no errors i log, but button does not do anything 🤷🏻‍♂️

[rancid harness]

it can be quite time-consuming.

[karmic crown]

nope

[rancid harness]

Great, it worked. Thanks again 🙂

[karmic crown]

there is nothing in the log

[rancid harness]

So there's nothing we can do about it then?

[karmic crown]

hmm... i see a pull request (which is the same thing i did)
but that only helped with showing the ui, since some libraries have versions conflicts
https://github.com/deforum-art/sd-webui-deforum/pull/950/commits/a5fd48cde5458afa9fd26f6aa29e272f009c8292

[rancid harness]

I initially thought it might be related to ControlNet. I uninstalled all extensions and left only Deforum, but that didn't help either. Then I saw the version number 1.8 and knew it had to be related to the update 😄

[karmic crown]

hmm forge f0.0.17v1.8.0rc-latest-269-gef35383b with sd-forge-deforum works without any issues

now it works in 1.8.0

[rancid harness]

only with Forge WebUI?

[karmic crown]

it works on my computer in both now, after i changes some things

[noble belfry]

i think a lot of the work forge does is stop us putting unnecessary flags on thing

[olive snow]

@rancid harness

do you maybe know man?

would help a lot

[karmic crown]

it is nice, but works slower than a1111 on my laptop

crazy thing is that a1111 works even faster than comfy 🙈

i know what is your next question
Macbook Pro with M3 PRO 🙈

[noble belfry]

it depends on what you're doing with forge. it can be slower if you have a process that switched out models. its great if you are running batches without much changing around. its especially good on peasant machines like my 1060

[karmic crown]

basic 1.5 512x512 generation is slower

[olive snow]

sorry I just found the mistake. it was because i was out of cuda memory finally 😅

[karmic crown]

https://github.com/deforum-art/sd-webui-deforum/pull/950

🙈

any volunteers to test deforum fix on 1.8.0?

[karmic crown]

https://github.com/deforum-art/sd-webui-deforum/issues/949#issuecomment-1975267207

@rancid harness try it, please

[noble belfry]

in fairness almost nobody uses sd1.5 and at 512x512 resolution, its not really a good benchmark. but yes as i said its optimizations can work out slower in some cases, depending on when you start your timer 🙂

i guess its like comparing a car with a low top speed but fast acceleration to one with slow acceleration but a high top speed. neither will win every race

[inland seal]

hey guys, how can I fix the batch size to give me properr results - separate images, not one the the same several times

[noble belfry]

not sure which program youre using, but if theres and option for seed, set it to -1/random

[river flame]

yes, I just bought 2080 ti 12 Gb, thank you

but I m getting "Torch is not able to use GPU" in Forge now. (A1111 is ok). Reddit says I should delete VENV folder, but there s no such folder in Forge

what do I do

[vocal burrow]

what's the full log ?

[river flame]

Traceback (most recent call last):
File "D:\Forge\webui\launch.py", line 51, in <module>
main()
File "D:\Forge\webui\launch.py", line 39, in main
prepare_environment()
File "D:\Forge\webui\modules\launch_utils.py", line 431, in prepare_environment
raise RuntimeError(
RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check

[vocal burrow]

the full log

[river flame]

Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: f0.0.16v1.8.0rc-latest-268-gb59deaa3
Commit hash: b59deaa382bf5c968419eff4559f7d06fc0e76e7
Traceback (most recent call last):
File "D:\Forge\webui\launch.py", line 51, in <module>
main()
File "D:\Forge\webui\launch.py", line 39, in main
prepare_environment()
File "D:\Forge\webui\modules\launch_utils.py", line 431, in prepare_environment
raise RuntimeError(
RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check

that s all

[noble belfry]

oh god i heard that in my old school teachers voice!

[wind sun]

@karmic crown ay can you help my G here, i will appreciate it ❤️

[vocal burrow]

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

[karmic crown]

as some people already pointed, -1 for seed is needed to get different images

[vocal burrow]

ok that's weird it should work out of the box like that. Did you install it using the git branch method or the one click package ?

[river flame]

1-click

[noble belfry]

also batches tend to be samey. if you up your batch count to 20 and reduce your number of batches to 1, it should get more variance too. one think though, if youre using lightning/turbo/lcm models, they seem to have a lot less creativity

[karmic crown]

only there dark is actually added to params, everything else is just a comment

[vocal burrow]

what's in your webui_forge_cu121_torch21\webui folder ?

[wind sun]

Is that for us

[noble belfry]

yes 🙂

[karmic crown]

even i am not sure if @REM is same as REM on windows, i forgot

[river flame]

D:\Forge

[wind sun]

We are trying it now

[noble belfry]

rem is just used as a commenting out

[wind sun]

We are like 5 people in discord call trying to figure this out because it’s annoying

[river flame]

and now when I m trying to generate in A1111, I get OutOfMemoryError: CUDA out of memory(without any controlnets)

[vocal burrow]

@REM <=> :: it's just for commentary lines

[karmic crown]

@noble belfry i know about REM, but I am not sure what @REM do, I know about @echo off

[vocal burrow]

rem = remarks (I guess)

[karmic crown]

i use windows like 3 times a year lately 🙂

[vocal burrow]

I'm not asking were it is but what's in it.

That's where the venv should be

[karmic crown]

i am saying that i have no idea if REM with @ befor it is the same as REM without or not

but both should be comments

[vocal burrow]

OH, yeah same thing I believe

[vocal burrow]

ok and you ran the update.bat before running run.bat ?

[river flame]

should I just reinstall both Forge and A1111

[wind sun]

Okay so How can I generate different images in 1 batch?

[vocal burrow]

no, just run update.bat

[river flame]

yes

[wind sun]

Like i want different results but from my promps

[river flame]

I just did again, all the same

[wind sun]

Idk if you guys can understand me

[noble belfry]

do you have an option for "seed" or random seed?

[vocal burrow]

what does update.bat says ?

[wind sun]

Ye we have seed options but the thing is if we put -1 it basically generates randoms but we just need to generate all of the pictures with the same parameters in the inserted data

[karmic crown]

@vocal burrow REM This will be shown during execution @REM This will not!

that is the diffrence

[river flame]

When I ran it a minute ago, it updated like 5 things, I closed it, now it says Up to date

[karmic crown]

so, same as @echo off

[wind sun]

“The problem Is that when he generates Images In batch, all of the Images that come out separately come out the same”- this question also

[karmic crown]

i use windows only to compile some legacy apps i made a long time ago, and some testing so i forgot 🤷🏻‍♂️

[wind sun]

Viking i think we will need your help here idk

[noble belfry]

thats the nature of randomness. you cant have both consistent and a variety without using maybe a lora.

[karmic crown]

https://global.discourse-cdn.com/bluescape/original/2X/7/75bfe6dd9740361dd7bde13a4c600fa9ac98dbbc.jpeg

put -1 here

[noble belfry]

a controlnet could hold the pose and style of the image too, but its going to be different. you wont get exactly the same person with the same hair with the same clothes in different poses, for example.

[wind sun]

They come exactly like this but each image’s are completely the same, there is no change at all

[storm vector]

Alright so the Images that he generates comes out like yours but instead of the images being different one Image from the 6 will be exactly the same over and over.

[noble belfry]

i think this is a communication issue more than a technical problem. im finding it hard to articulate the answer

[vocal burrow]

lol by default they use set SKIP_VENV=1, I didn't even notice as I have set mine before even launching it for the first time and installed it using git.
So yeah, there's no venv for the one click package of forge. It's running straight from the system folder

[noble belfry]

i never noticed that

[karmic crown]

@inland seal please show the complete UI, where we can see all the settings, especicially SEED

[noble belfry]

TIL

[wind sun]

There is just no difference in the pictures that are coming out

[river flame]

so do I just reinstall then?

[vocal burrow]

reextract the system folder from the .zip and see if it helps

[noble belfry]

so forge has no venv? its just portable?

[vocal burrow]

it comes with its own """venv"""" by setting a bunch of system variables

environment.bat

@echo off

set DIR=%~dp0system

set PATH=%DIR%\git\bin;%DIR%\python;%DIR%\python\Scripts;%PATH%
set PY_LIBS=%DIR%\python\Scripts\Lib;%DIR%\python\Scripts\Lib\site-packages
set PY_PIP=%DIR%\python\Scripts
set SKIP_VENV=1
set PIP_INSTALLER_LOCATION=%DIR%\python\get-pip.py
set TRANSFORMERS_CACHE=%DIR%\transformers-cache

[noble belfry]

as if i didnt have enough python versions and venvs to worry about now theres this lol

flips table

[karmic crown]

actually no, it use its own python

see the relative paths

[vocal burrow]

Or you can do as I did and install it using the git method + having a seperate webui-user-forge.bat with set VENV_DIR=venv-forge in it.

yes it does

it has everything in it so don't worry

[karmic crown]

forge over git use venv

[vocal burrow]

it shouldn't mess your system

[vocal burrow]

it just doesn't create one at start because when you download the .zip there is already one in it.

(thus the set SKIP_VENV=1)

[karmic crown]

as i said put -1 for seed

[noble belfry]

gonna be honest, ive gathered much too much environment clutter from AI projects over the past 18 months, i should just clean it all up

[wind sun]

Doing it

[inland seal]

same shit

[wind sun]

Those two have to be separated

That’s the problem

They are coming out together

[karmic crown]

that is just a grid view

[noble belfry]

press the right arrow on your keyboard buddy

[wind sun]

Nope

Thats not the problem

[karmic crown]

this is fine

[noble belfry]

this is a bit like pictionary now

[karmic crown]

also this

this is a problem

it should not be like this

something is f--- up

[noble belfry]

https://tenor.com/view/dignity-oh-come-on-luann-you-know-what-this-is-kurt-van-houten-gif-20607258

[wind sun]

They are together, in one picture and we need them separated in different pictures

[karmic crown]

but generated images are diffrent

first "image" is not an image

[wind sun]

Ye thats the good point but the bad one is we need them in separated pictures not coming in just a one

[karmic crown]

it is a grid that shows all generated pictures