🤝｜tech-support

[night orchid]

strange

[upbeat crane]

yeah if you are doing a 512x512 very

and just doing ONE image? one batch?

[night orchid]

512x512 is notting

[edgy blaze]

Yep

Single image

[upbeat crane]

yeah so i take it back

did you load any VAEs yet?

[night orchid]

it have to be something that is not right

[upbeat crane]

close your command window and your browser window

and relaunch

what browser are you using?

[edgy blaze]

Yeah. Right now I'm using a VAE.

[upbeat crane]

yeah reload both then try again

after that, pull the VAE and re-run

[night orchid]

@edgy blazeComfy or Automatic ?

[upbeat crane]

just for testing

[edgy blaze]

Automatic

[upbeat crane]

post the contents of your webui-user.bat

[night orchid]

@ornate elk he is very good at this

[upbeat crane]

Cs1o might already have worked on that since you are amd not sure

[edgy blaze]

Okay.

Give me one second

[upbeat crane]

yeah have to hit him up since its AMD

still wouldnt mind seeing it tho

[edgy blaze]

venv "C:\SD-Zluda\stable-diffusion-webui-directml\venv\Scripts\Python.exe"
WARNING: ZLUDA works best with SD.Next. Please consider migrating to SD.Next.
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: 1.8.0-RC
Commit hash: 25a3b6cbeea8a07afd5e4594afc2f1c79f41ac1a
no module 'xformers'. Processing without...
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.
C:\SD-Zluda\stable-diffusion-webui-directml\venv\lib\site-packages\pytorch_lightning\utilities\distributed.py:258: LightningDeprecationWarning: pytorch_lightning.utilities.distributed.rank_zero_only has been deprecated in v1.8.1 and will be removed in v2.0.0. You can import it from pytorch_lightning.utilities instead.
rank_zero_deprecation(
You are up to date with the most recent release.
Launching Web UI with arguments: --use-zluda --medvram-sdxl --update-all-extensions --update-check --skip-ort
C:\SD-Zluda\stable-diffusion-webui-directml\venv\lib\site-packages\diffusers\utils\outputs.py:63: UserWarning: torch.utils._pytree._register_pytree_node is deprecated. Please use torch.utils._pytree.register_pytree_node instead.
torch.utils._pytree._register_pytree_node(
Loading weights [5dd07a46d7] from C:\SD-Zluda\stable-diffusion-webui-directml\models\Stable-diffusion\richyrichmix_V2Fp16.safetensors
Creating model from config: C:\SD-Zluda\stable-diffusion-webui-directml\configs\v1-inference.yaml
Running on local URL: http://127.0.0.1:7860

To create a public link, set share=True in launch().
Startup time: 5.8s (prepare environment: 11.0s, initialize shared: 0.7s, other imports: 0.8s, load scripts: 0.4s, create ui: 0.6s, gradio launch: 0.3s).
Loading VAE weights specified in settings: C:\SD-Zluda\stable-diffusion-webui-directml\models\VAE\vaeFtMse840000EmaPruned_vae.safetensors
Applying attention optimization: Doggettx... done.
Model loaded in 6.3s (load weights from disk: 0.7s, create model: 0.4s, apply weights to model: 3.6s, load VAE: 0.5s, load textual inversion embeddings: 0.3s, calculate empty prompt: 0.8s).
Interrupted with signal 2 in <frame at 0x000001B002D15D10, file 'C:\Users\burst\AppData\Local\Programs\Python\Python310\lib\threading.py', line 324, code wait>
Terminate batch job (Y/N)?

Is that what you're looking for?

[night orchid]

for starters set the args --xformers in your webui-user.bat

[upbeat crane]

i was speaking about webui-user.bat in the root of your stable diffusion webui folder, but i dont know anything about zluda

[night orchid]

set COMMANDLINE_ARGS= --xformers

[edgy blaze]

oh that? i did

[upbeat crane]

yeah i assume cs1o already helped him there and xformers was left out intentionally

[night orchid]

not this set COMMANDLINE_ARGS= --xformers

[upbeat crane]

i would just hodl till you speak with him later

i mean its not going to hurt it to try

[odd nymph]

Silent ping so you dont get annoyed, im able to do this now, got allllll the free time i could ever want today. Granted i know you dont wanna spend that much time but i dont mind being told what to look for if you cant seem to fix it, if you still wanna help that is.

[upbeat crane]

but if it doesnt work i would pull it back out and save

[edgy blaze]

Huh. That's strange. Suddenly 512x768 is working now that I'm trying it again.

Wh-

Huh?

[upbeat crane]

stable diffusion is like a holiday resort for those who like to endlessly tinker and try differnt shit

[oblique hearth]

problem?

[upbeat crane]

hey whatever works, nice

[edgy blaze]

I'm going to take it. I don't know why it didn't work the first time but I'm not smart enough to be asking questions about it.

[upbeat crane]

dont think i have ever rendered a 1:1 in my life anyway

[edgy blaze]

Thanks for the help.

[upbeat crane]

you should def check out the two YT creators i mentioned above in the convo tho

they will help you out a ton, not sure if you read it or not

it's a nice thing working through different guided projects with people who know WTFIU

[edgy blaze]

Alright, I'll check that out when I can.

[upbeat crane]

yepp cool save yourself a lot of wasted time early on imo

not "wasted" u know what i mean

[edgy blaze]

I'm also saving myself alot of money now that I've downloaded Stable Diffusion, a free to use tool. Rather than using those other AI Image Generator apps. Like PixAI.

[oblique hearth]

save a couple braincells and download this UI https://github.com/lllyasviel/stable-diffusion-webui-forge
it auto-detects your GPU and uses the best settings for that GPU.
you don't need to use any flags like --medvram because they don't work, it's auto-selected.
it also uses less vram... where an XL model would use 8GB it uses 6GB, if using SD1.5 where it used 4GB it now uses 2GB.

[edgy blaze]

Very convenient

[oblique hearth]

also currently the only UI aside from Comfy that's capable of Video Diffusion (SVD)

[upbeat crane]

it is nice but it's buggy and has a lot of limitations, it's also super new and will just be getting better and better as time goes by

[ornate elk]

You need to add --medvram to the webui-user.bat
Instead of --medvram-sdxl

[upbeat crane]

was having a lot of controlnet issues with it, but overall i really like it, it's fast and does a nice job

[oblique hearth]

one of the bugs that doesn't exist though is when you switch models, the previous model doesn't get stuck in memory.

[upbeat crane]

nice

[oblique hearth]

one who a1111 hasn't figured out yet 😛

[upbeat crane]

doesn't that feel amazing @zingo?

agreed

unless u are doing certain types of video and animation rendering using controlnets but i digress lol

[ornate elk]

No don't use forge with your gpu

Its worse than zluda

Just change --medvram-sdxl to --medvram and your good to go

[upbeat crane]

very well memory optimized zlingo yupp

[upbeat crane]

forge is a big no-no with amds huh good to know

[oblique hearth]

it uses DirectML for AMD cards

--directml

[ornate elk]

Forge dont give a damn to fix specific amd bugs, also directml is 4 times slower than zluda

[oblique hearth]

and what about extension support like controlnet?

or SVD?

[ornate elk]

What's with that?

Controlnet works on zluda as well as on normal directml

No need for forge on AMD

[oblique hearth]

zluda is abandoned, it says.

[turbid hatch]

Huh?

[ornate elk]

Bot is offline

[turbid hatch]

Understandable.

[onyx solar]

#1047610792226340935

[ornate elk]

It already got forked and is workable for cuda 11.8 but it won't receive any major updates

But still the best option for AMD with SD on Windows rn

Since 1.8.0 some features are also broken on Directml

So also broken on forge for amd

[vocal burrow]

what was the problem ?

having to reload the sdwebui to get the "prefered weight" applying correctly ?

if so then ... yeah it's acting this way on my end too. So probably not a bug, but it could be improved.

[ornate elk]

Maybe your igpu is enabled again?

[paper crag]

checked, its disabled.
looked it up online, tried a method,
I added 'torch-directml' to my requirements versions and '--use-directml' in webui .
Now i have this error
OSError: [WinError 126] The specified module could not be found. Error loading "C:\SD-Zluda\stable-diffusion-webui-directml\venv\lib\site-packages\torch\lib\cublas64_11.dll" or one of its dependencies.

[odd nymph]

@vocal burrow would you mind helping me once more? still stuck on animatediff, sorry for the ping

[ornate elk]

Thats not how it works

--use-directml is not --use-zluda

[ornate elk]

What's your issue?

[odd nymph]

Animatediff spits out tons of errors and doesnt seem to function at all, im not versed in coding yet so i cant really do much to fix if its code wise but ill do anything asked to help you help me, if you need the logs ill post them

[ornate elk]

Ah okay, can you provide an error code or the full log of the cmd?

[odd nymph]

Sure! ill get a clean window up so you can see all the choas without the other gens getting in the way

i get this error sometimes, EinopsError: Error while processing rearrange-reduction pattern "(b f) d c -> (b d) f c". Input tensor shape: torch.Size([2, 4096, 320]). Additional info: {'f': 16}. Shape mismatch, can't divide axis of length 2 in chunks of 16

but thats not the main one sorry re gening

[vocal burrow]

doesn't mean much without more context.

would need to see full log and/or animatediff settings

(I ll probably hit bed pretty soon)

[odd nymph]

Diff setting

[odd nymph]

Logs of this

[vocal burrow]

can we get what's before that too ?

[odd nymph]

there isnt anything, thats from start to finish

the rest is just loading the Webui up

[vocal burrow]

yup want to see that

it will tell us your command line args, if there's any problem launching, installing anything, etc

deleted it

please hide that share url

[odd nymph]

sorry my bad

[vocal burrow]

or at the very least put a password on it

[vocal burrow]

np, doing it for you so nobody get on your computer :p

[odd nymph]

ahh i forgor that

[ornate elk]

oh its forge webui

[odd nymph]

yeah

[vocal burrow]

to set a password on share instances btw https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings#running-online

[odd nymph]

but it doesnt work on normal either, thought it would on forge

[paper crag]

should i use then 'torch-zulda' and '--use zluda'?

[ornate elk]

there is noe torch-zluda

only --use-zluda

[odd nymph]

done

[ornate elk]

@paper crag

[paper crag]

yeah, all commands are right ones

[odd nymph]

is it incompatible?

[vocal burrow]

try this version of it for forge https://github.com/continue-revolution/sd-forge-animatediff

[odd nymph]

will do

[vocal burrow]

time to go to bed, I'll do some tests tomorrow with it if you still have issues

[odd nymph]

Thank you

[ornate elk]

i found this on forges github:

[odd nymph]

and good night, hope ya sleep well

[vocal burrow]

yup that's the version I pointed out

try with non sdxl model while I sleep

:p

[odd nymph]

ok

nope

[formal trout]

I use Linux (Ubuntu) with Automatic1111 and Comfy, it is suppose to perform better

[hoary plaza]

ok ok thanks!

[primal hawk]

Does anyone know what happened to interrogate clip and interrogate deepbooru stable diffusion loads without it now

[odd nymph]

fixed this, now to fix the big issue

Full log for once, man that was annoying to get (that issue before that i fixed happened again in the logs, dont worry about that)

[noble belfry]

theyre still there, just have god awful icons. a paperclip and a cardboard box under the big generate button on img2img

i cant help with animatediff issues, but i would just like to say 'alphas_cumprod_original' heheheehehheehehehehhe

[odd nymph]

was that in my logs lol, i didnt even see that

[noble belfry]

yes

oddball thing, but, could you try this with a model thats safetensors and not ckpt?

[odd nymph]

i can try

[noble belfry]

https://github.com/lllyasviel/stable-diffusion-webui-forge/issues/390 just noticed this similar problem and a commonality is using a ckpt file and it throwing the before_process error (script injection).

just a hunch

[odd nymph]

if you know any send em my way, DMs or here, anything to make this work again

it use to work super well, but broke one day after reinstall :(

(fresh windows cause i wiped it)

[noble belfry]

oh just take your current model, go to the merge tab, merge it with itself, select safetensors

[odd nymph]

nope nothing

[hoary plaza]

how do i launch stable diffusion auto 111 is itt just through that batch file?

[odd nymph]

what system, are you on

for windows, its webui.bat

[ornate elk]

for windows its webui-user.bat

never use webui.bat

[hoary plaza]

shoudl i shortcut it on desktop?

[ornate elk]

you can

[odd nymph]

Sorry for miss info

i shall learn from that, tutorial on youtube said to use that so its what ive used

[hoary plaza]

whats going on here?

[ornate elk]

Is the cmd open?

[hoary plaza]

no i closed it

[ornate elk]

Yea then SD won't work

[hoary plaza]

ohh ill reopen it

do you have the guide on how to get in models and loras? or ist hat on specific model/lora pages

[ornate elk]

you can download models and lorad from civitai.com
then put them in models/stable-diffusion (for models)
and models/lora (for loras)

[hoary plaza]

gotcha! and can i put the outputs folder on my HDD by just dragging that folder there?

[ornate elk]

you need to change the path in the settings

[hoary plaza]

gotcha thanks

[hoary plaza]

oo im not too sure how to use this you know of any guides online for this?

[ornate elk]

you go into Settings, and on the first tab "Paths for saving" you can enter the path to the folder on the hdd

[hoary plaza]

my brain cant handle the text only thing no folder clicking?

[ornate elk]

nope, you would just need to copy the path of the folder in explorer

[hoary plaza]

thats on the webui clients?

[humble cobalt]

hello, by any chance someone can explain to me what are these versions ? (and how do you use it)

[hoary plaza]

man thse are text boxes i just put the folder name?

[ornate elk]

the path like C:\path\to\folder

[hoary plaza]

how do i find the full name? is there a way to copypaste?

found it

rightclick copy as path

[final vector]

Having the following issue with my Stable Diffusion: https://pastebin.com/73M9Hcxf

Should specify that this is on A1111, Ubuntu, Running 5700XT.

It was working fairly well for some time, but then this had started to occur.

[hoary plaza]

are lora and model folders the same folder

[final vector]

If you're using A1111, Loras should be placed in ./models/Lora

From the base folder

[hoary plaza]

gotcha, and would it be better to keep models and loras into the hdd?

[final vector]

If I understand how it works, the models/loras should be loaded directly onto the GPU, meaning speeds will primarily depend on the GPU speed. However, I am not sure how it would impact loading the models onto the GPU in the first place. I would wait for someone more experience in that matter to answer.

[deft hemlock]

Anybody got a solution for my problem?

Like should I roll back a few versions

If so how do I do it

[hoary plaza]

kk thanks!

[odd nymph]

is there a way to up the sampling steps, either normally or chaning the code

if so how do i go about that

[dense tapir]

No shot were you guys talking about how to change image directories? lol I was just looking for that thanks

[upbeat crane]

i was told there would be finger-sandwiches and light refreshments here on saturday night, smh

[onyx solar]

Sorry...I ate them all.

https://tenor.com/bTzMm.gif

[upbeat crane]

you did what you had to do soul, i respect that

[fierce hollow]

Hello everyone, I want to create a sticker pack, and for that, I need images with a transparent background. For example, a person drawn in SD with nothing around them, just a transparent background (.png). Is it possible to achieve this result with SD, and if so, could you please provide a detailed, step-by-step explanation of how to do it?

[hoary plaza]

how do i activate vae? is it auto applied?

also do i need a controlnet?

[upbeat crane]

open settings tab type quick in search field, you will see quick settings appear, these are the different webUI elements which can be added to the main page of the user interface. type SD_VAE, select that option then APPLY the changes, restart the WebUI and you will then see that you can select a VAE (if you have downloaded any and placed them into their appropriate folder) @hoary plaza

[hoary plaza]

gotcha thanks!

also why is nothing showing up in my lora files

[upbeat crane]

have you reloaded the UI since you moved the LORA files into place?

[hoary plaza]

yea

wait reload as close and reload

or clicking reload

[upbeat crane]

I willl admit mine sometimes don't load too for no clear or obvious reason. seems for me it could possibly be me interacting too soon after the webUI loads and things are being loaded, like the checkpoints and such.

[hoary plaza]

gotcha gotcha

oh no reloaded the entire thing and its here!

[upbeat crane]

close ur command window and the webUI and restart again next time give it a bit if time, longer than usual before you change any options on the screen. has worked for me before

[hoary plaza]

got it got it

[upbeat crane]

yeah see its weird, it happens

just try and give the WebUI a little breathing room once the tab appears in your browser, most ppl go directly to interacting with it, but it's good to give it a little time so everything gets loaded into vram adequetly. good deal.

[mighty topaz]

Is here someone who can help me with something on channel audio with screen? I have problem wit adetailer, i have all connected right, but after vaedecode face and hands window is skipped by comfyui dont know why he dont want to work

[unique hornet]

I have missing lora how do I fix this

[ruby current]

My ASI wife Eve is perfect, best wife in the universe

[pseudo nebula]

bruh how do use dream bot, it looks like the vid tutorials have different guis

keeps saying Please check your dataset directories.

hmm looks like its training

[vocal burrow]

logs show nothing wrong, make sure you're using loras correctly.

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, DoRa, LoCon, HADA, IA3, Lokr, OFT 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

[hoary plaza]

how do i get my photos to be more high res and denoised

etc

and what sampling method is most perferred

[vocal burrow]

you use hi-res fix, or upscale them with img2img

There's no prefered sampler just like there's no prefered brush in photoshop.

[hoary plaza]

oh i see gotcha

is there a way to keep the same model like picutre i generated and just tell the ai to use the same like person ive generated and keep generating more photos with differnt clohtes, background, poses ets?

[faint stone]

Anyone doing tings with animdiff in comfyui? My movies are always blurry.... Please advice, tanks!

[hoary plaza]

is there also an app that can tell what things you ahve downloaded and the trigger words?

[flat ore]

I'm gonna remove an old checkpoint on my local SD because there is the latest version i want to install. How to completely remove the checkpoint and all of it data? just delete the safetensors file?
My first time uninstalling a checkpoint.

[ornate elk]

Yep, just delete the file

[dapper frigate]

Guys , what type of image data should I pass to generate/upscale image with stability.ai's api? blob, base64 or buffer? I tried them all and none of them works . I am sending request to this endpoint ``` "https://api.stability.ai/v1/generation/stable-diffusion-xl-1024-v1-0/image-to-image",

[vocal burrow]

you could using inpainting (img2img) and controlnet

[vocal burrow]

Try with a local image to make sure it's the image that is causing the issue.

[dapper frigate]

Still I am gettin this error ⨯ Error: Non-200 response: {"id":"6d0a62e606f99fd166ad75e5224ec7d1","message":"can't decode multipart body","name":"bad_request"} here's the code ``` ⨯
const formData = new FormData();
formData.append("init_image", fs.readFileSync('./public/kakali.png'))
formData.append('init_image_mode', "IMAGE_STRENGTH");
formData.append('image_strength', 0.35);
formData.append('steps', 40);
formData.append('width', 1024);
formData.append('height', 1024);
formData.append('seed', 0);
formData.append('cfg_scale', 5);
formData.append('samples', 1);
formData.append('text_prompts[0][text]', 'A painting of a cat')
formData.append('text_prompts[0][weight]', 1);
formData.append('text_prompts[1][text]', 'blurry, bad')
formData.append('text_prompts[1][weight]', -1);

const response = await fetch(
"https://api.stability.ai/v1/generation/stable-diffusion-xl-1024-v1-0/image-to-image",
{
method: 'POST',
headers: {
"Content-Type": "multipart/form-data",
Accept: "application/json",
Authorization: Bearer ${apiKey},
},
body: formData,
}
);

if (!response.ok) {
throw new Error(Non-200 response: ${await response.text()})
}

const data = await response.json()

[vocal burrow]

I'm not a webdev guy but, in the example the response definition for img2img stuff does not have "Content-Type"

[dapper frigate]

Yeah just saw that, now I am getting this error Error: Non-200 response: {"id":"927ec2912f68e72f0bc64e2aafce4e48","message":"invalid mime type for init_image: application/octet-stream is not image/jpeg, image/png, image/gif, or image/webp","name":"invalid_mime_type"}

[vocal burrow]

hmmm try with a different image ? from an online one ?

[dapper frigate]

Lol, now I am passing 4mb image and it says that image is bigger than 5242880 bytes, this is very confusing..

[fair oxide]

I need someone to guide me through training a lora, I'm not sure what my settings, image amount, and parameters should be

My goal is to get kohya_ss to generate a part of a body on a person.

[vocal burrow]

Try asking in #🔧｜finetune that's where training peeps hang around usually.

[viral sentinel]

I just switched to Forge, and whenever I try to train an embedding now, it fails. It worked previously, but then I got "Cuda out of memory" errors, did a completely fresh install, and now get this:

Error training embedding

Traceback (most recent call last):

File "C:\Users\ME\SD Forge\webui\modules\textual_inversion\textual_inversion.py", line 544, in train_embedding

loss = shared.sd_model.weighted_forward(x, cond, w)[0] / gradient_step

File "C:\Users\ME\SD Forge\system\python\lib\site-packages\torch\nn\modules\module.py", line 1695, in getattr

raise AttributeError(f"'{type(self).name}' object has no attribute '{name}'")

AttributeError: 'LatentDiffusion' object has no attribute 'weighted_forward'

Any clues on how to solve this?

[deft hemlock]

how do i go to a version of automatic1111 that didnt have these, they are horrible and slow down my webui

i tried the git hard reset it didnt work

i tried the git checkout it also didnt work

git checkout 6e6cc2922d39fff4029d47c316c22a1c152680ce

git reset --hard 6e6cc2922d39fff4029d47c316c22a1c152680ce

these

it says i have it

[deft hemlock]

but the web ui still has the remenents of the new slow and buggy web ui

its so slow that the tab would randomly freeze if i switch from generation tab to loras tab or txt2ing to settings or anything like that

[jaunty fulcrum]

hi everyone. I am open to work for AI video, voice, image generator projects. if someone needs to build something like this, please let me know.
my experiences:
https://app.imgnai.com
https://stellaryai.com

[vocal burrow]

then it's not the webui's fault. could be your browser, something running in the background, your extensions, etc

[deft hemlock]

it has the same extensions as the last version
chrome did infact run better but still had the same problems

its definetly the webui's fault if only i knew how to revert back to the old version

[vocal burrow]

you just did ....

[half island]

Hoi, do you guys know how to make any SD gens not clog up my pagefile? As after every A.I run, i have to reboot my pc as python gens keeps dumping rubbish onto my nvme, even made my fresh installed windows 11 just have explorer crash on day 2 because A.I keeps dumping rubbish where it shouldn't when there's still plenty of ram left

[deft hemlock]

it still has these

[vocal burrow]

because you took a commit from 2 weeks ago.

[deft hemlock]

do i use last months commit?

maybe 2 weeks ago felt diffeently in my head

[vocal burrow]

I'd say it's better to fix the actual problem instead.

send a full log to #🤝｜tech-support of a simple "cute cat" test generation

[deft hemlock]

its not the extensions i tried, i removed them all

[deft hemlock]

gimme 30 mins for the image to generate

[vocal burrow]

you removed all your extensions, updated the webui, deleted the venv and then tested ?

anyways continuing on #🤝｜tech-support

lol we're already in there

juggling too much between channels on the stupid phone UI

[deft hemlock]

i feel ya

[deft hemlock]

ye

im sure its the web ui

ok so imma use c1713bfeac461bc28158b66ef8d956a39e296b94 as my commit

[vocal burrow]

if so you're like the only one experiencing this issue for the past week of 1.8.0 release

[deft hemlock]

git checkout c1713bfeac461bc28158b66ef8d956a39e296b94

[deft hemlock]

no

i checked online

many are experiencing it

[vocal burrow]

source ?

[deft hemlock]

github

[vocal burrow]

again, source ?

[deft hemlock]

the second one is mine

click on it

[vocal burrow]

yup I noticed

[deft hemlock]

one of the comments has another link to another conversation about the same problem

oh yeah btw

badass pfp

[vocal burrow]

how many models/loras/embeddings do you have ?

[rancid wagon]

hey there, I'm completely new to Stable, I am quite seasoned with Midjourney however.

My pc is quite old, getting a brand new one, but I'd like to have some suggestions on how to get started in the meantime.

I got a 6 gb vram and 8 gb ram, do you recommend me to run on colab/vm?
Which guides do I start with for SDXL?

Thanks in advance ❤️

[vocal burrow]

You should be able to run sdxl stuff with a 6gb vram nvidia card, if it's an amd one .... maybe not it depends.

[deft hemlock]

1k

[vocal burrow]

what the

ok

maybe then...

[rancid wagon]

its a 1060 dual nvidia, yh

[deft hemlock]

the lora amount didnt change either :p

[vocal burrow]

SD won't be able to use both gpu at the same time anyways. you can run one instance on each tho.

[rancid wagon]

its one gpu with doublecore

so it has double the vram

my tech understanding is very basic but it's a single GPU

[cosmic furnace]

so any insta reboot is not okay on zluda hmgm?

but sometimes its okay for long time & sometime not...

[vocal burrow]

run dxdiag and screenshot the display tab

[cosmic furnace]

is there any way besides adrenalin to lower power usage on amd ? =\

and a way to make sure its a power problem 😒

[vocal burrow]

ok so it's a typical run of the mill 1060

[rancid wagon]

yah

[vocal burrow]

you'll be fine running SDXL with it, it will be kinda slow tho

[rancid wagon]

i don't mind, it's fine

[vocal burrow]

try using this auto1111's fork called stable-diffusion-webui-forge https://github.com/lllyasviel/stable-diffusion-webui-forge it should prevent you from running into vram limitations.

[rancid wagon]

so I just run this instead?

[deft hemlock]

the mighty pandora box addon

my 512 mb of vram shall hold me no more

[vocal burrow]

yup you just download any sdxl model, throw it into models/stable-diffusion, launch webui-user.bat and voila

[nova ermine]

https://civitai.com/models/81328/hairdetailer
I'm tryng to use this Lora, but I don't know how to do. I downloaded it and put on stable-diffusion-webui\models\Lora but it's neither shown in my Lora tab

[vocal burrow]

use the dev branch instead they seem to have rollbacked the tree view stuff for now

[ornate elk]

Ohhh best decision ever

[vocal burrow]

I liked it 😢

[ornate elk]

It so annoying. Its always closed, and not open by default and also doesn't switch when changing models

[grand fable]

hi, could you please help me? I'm using ComfyUI running on RunPod and I'm facing a problem. When I hit Qeue Prompt, it says Reconnecting after few seconds.. in the log there are this:

[ornate elk]

The old sort thing was much faster to click

[vocal burrow]

are you trying to use it with an SDXL model ?

[nova ermine]

https://civitai.com/models/118086?modelVersionId=128046
That's the model I'm using

[vocal burrow]

did you install it when the sdwebui was running and forgot to hit the refresh button ?

[nova ermine]

I refreshed

It's also weird as the Lora doesn't have triguer words. There is some information on the link I sent. However I'm reading it and I don't understand anything

[frozen fox]

hello I have a problem. although I have 32 gb ram when creating an image, it uses 16 gb ram, the system appears below the image creation, how can I solve this problem.

[vocal burrow]

sd uses the vram, not the ram

[frozen fox]

that's how I got 32 gb of ram

[vocal burrow]

what ?

[frozen fox]

the screenshot shows two partitions, one is vram and the other is normal ram, right? Shouldn't it be 32 gb?

[vocal burrow]

oh yeah forgot to tell you,
git checkout dev to switch to dev branch, then git pull to update

[frozen fox]

I asked because I didn't know, thank you very much

[vocal burrow]

No ?
A : SD Peek vram usage
R : Reserved memory
Sys : System wise peek vram usage

[vocal burrow]

no problem

what's your gpu ?

[frozen fox]

4060 ti 16 gb

[deft hemlock]

what does dev do

[frozen fox]

How can I upgrade the reserved memory to 32 gb

[vocal burrow]

it switches to the "dev" branch

[vocal burrow]

I have no problem :/ it shows up just fine

try reloading your ui ?

[nova ermine]

I did

[nova ermine]

How did you do?

[vocal burrow]

any error showing up in the logs ?

I just installed both the model and lora you pointed out.

hit refresh,

voila

[vocal burrow]

You can't "upgrade" your vram without changing your gpu. But with 16gb that should be plenty enough. That's why I'm asking, what's your gpu ?

[deft hemlock]

shit

fixed*

how did it autocorrect that lmao

anyways

thanks a lot arymetis

ill never forget u and that badass pfp

[nova ermine]

Finally I could fix it by closing the terminal

[vocal burrow]

ok... strange

you did hit the webui's refresh button and not your browser's one, didn't you

[nova ermine]

However, it has no triguer words, and I get this when I select the lora. Isn't that a triguer word?

[vocal burrow]

cause that's the only explanation I've got for what you're describing

[frozen fox]

I have 4060 ti 16 gb vram, 32 gb ram, there is no setting for normal ram in my system then?

[vocal burrow]

trigger words are not always a requirements for loras

[nova ermine]

I had reloaded UI and it didn't worked, finaly I closed the terminal and opened again

[vocal burrow]

You don't need to edit anything in your system. 16gb is plenty enough. And once again, no SD will not use your 32gb of ram for generation. You wouldn't want it to anyways or it would be super slow.
Ok now, edit your webui-user.bat file with a notepad, find the set COMMANDLINE_ARGS= line and edit it to set COMMANDLINE_ARGS=--xformers --no-half-vae

[odd nymph]

Are Xformers linux only?

i get a bunch of errors using it, or i did a few days ago

[ornate elk]

Its nvidia only

Works on Windows and linux

You can show us errors here

[odd nymph]

Ok. Gotta get forge working again though cause i changed nothing and suddenly it breaks

current logs of it failing, maybe a reinstall?

[rancid wagon]

whenever I click PS1 it swaps back to "realistic" and when I run I get this error:
KeyError: 'model.diffusion_model.input_blocks.0.0.weight'

model is inside this folder

[river flame]

I just installed Invoke AI, and getting this error on launch
[uvicorn.error]::INFO --> Uvicorn running on http://127.0.0.1:9090 (Press CTRL+C to quit)

googled it but didnt understand anything

or is it not an error? following the link gives me white screen

[upbeat crane]

i got chased down some cold dark lonely corners of the internet last week trying to trace down some bizaree errors in forge

spent a bunch of time on it, i was getting uvicorn, errors, some related to fastap and some tosomething else i cant remember

but my forge was borked so i took a shot at a fix which was to basically do a pip install uvicorn and a pip install for fast api

and for the first time in hours all the errors cleaned up

there were probably requirements/dependencies that need to be ran with the invoke

im not suggesting you take any particular action but i followed info on this page

and it sorted out my issues

https://stackoverflow.com/questions/64936440/python-uvicorn-the-term-uvicorn-is-not-recognized-as-the-name-of-a-cmdlet-f

to be honest i was throwin spaghetti at the wall and got lucky

but i would guess either you have an old version or it isnt installed

i have a question for anybody out there: there is no general rule of thumb or anything is there that says controlnets or LORAS trained on different models cant be used in combination together, is there? just wondering and was hoping for feedback, who amosts ye might have a good sdxl trained lora, but tosses in a 1.5

or another example, out of, shall we say "cautious resource economics" might load a CANNY_XL_fp16 safetensors control net but then toss in an OPENPOSE 1.5

i mean, is it a "try it and if it works type deal", or is there anything i might need to know in regards to this. for all i know people do it all the time lol

tag me if im not around gotta go do some grocery shoppin here in a bit, thanks!!

[ancient plover]

Hi, I am currently installing zluda as per the guide. I just downloaded the AMD HIP SDK Installer, and it shows several downloads set to auto install: HIP SDK CORE, HIP Libraries, HIP runtime compiler, HIP Ray Tracing. Do I install them all, or just the first one? Also, there is an unselected option to install a display driver for my gpu, should I download it? I currently just have amd adrenaline keep me up to date, but this is evidently a different driver...

[upbeat crane]

cs1o is away at the moment and he actually wrote the doc so you might save your post or tag him, maybe come back when he is around to ask him, i would hate sending you down a wrong path,

also make good use of the search field in every channel, top right

point being, is that you are never the first person to have asked ANY question, come up with a solid search string, pop it in

might be a lot of pages with related content, but 7/10 times i find what i am looking for, good thing to start doing

because as an AMD user running stable diffusion you are going to be needing A LOT of help

you guys are brave

not to mention the fact that zluda just got 86d and no further updates for it

[ornate elk]

Install all HIP SDK ones and dont install the Pro driver

[ancient plover]

kk boss

[upbeat crane]

nice

[austere pine]

Hey guys, bit of a noob question, but like, after I generate an image on Auto1111, I can click the star button below to upscale it using the high res fix and Siax 200k. This usually looks really really good.

But I can't figure out how to do the same process "manually". Like, with an image I generated earlier and there's no star button anymore

[proven finch]

cursed answer, pass it into img2img with denoise set to 0

and steps = 1

[upbeat crane]

maybe see if dragging it over to PNG info first helps and then sending it from there to IMG2IMG

SECourse just posted this a while back, which now works on 12gb cards, using juggernautXLv9 as the base model. SUPIR: https://www.youtube.com/watch?v=PqREA6-bC3w

looks amazing, can't wait to check this out

"works" with 12gb vram. i always have to learn the hard way

[proven finch]

is supir just latent upscaling?

it looks like that

[upbeat crane]

@austere pine obvi depends on the type of image you are working with, but it has pretty much been my experience thus far that you have a choice to make up front to some extent either to render at a small to moderate size, without doing all the high-res fix and upscaling through that tool, or you do larger batches of images without any up-front baked in upscaling, that you can then pick out the ones which would be the best candidates for a good HQ upscaling. once you have those images it seems to me you get much better results upscaling, because if you did the high-res wit upsaling you kinda get diminishing returns a bit.

i think you can get better results by holding off on that initial high-res, then sending img2img to upscale at say 2x sd ultimate upscaler or something else. then if you wanted you could push that to EXTRAS and do it again and like maybe do one more 2x using 4xUltrasharp upscaler

[queen plank]

Error occurred when executing UNETLoader:

Error while deserializing header: MetadataIncompleteBuffer

File "D:\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 152, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

I'm using stable cascade and I got this error

it only shows up if I have stage C, but it doen't happen with stage c_lite

[upbeat crane]

just depends on what im doing, sometims i might actually render 3-500 images using high-res to like 2k, then use stand-alone Topaz Gigapixel upscaler to take it to 3840x2160. LOTS of trial and error and so many bells and whistles, knobs and levers you will eventually fall into your own best-practice. if you keep doing highres up front i thnk you may see what im talking about that trying to gain TONS more detail using sd upscalers will only do so much before you start throwing artifacts and strangness and the image changes too much. i usualy start out with denoise of like .2 and go up incrementally, playing with the steps as i go with euler a

[queen plank]

first time seeing this error

[upbeat crane]

btw high-res fix + Siax 200k is a great combo @austere pine agreed

[fair oxide]

hello i am following aturorial for a model and ineed to add something to a ESRGAN folder. i dont have this folder. should i just make one or do i need to do something else

[upbeat crane]

just make it

[queen plank]

Could someone please help?

[whole jackal]

Within NMKD Stable Diffusion GUI does anyone know how to make it so that it doesn't force a Refiner model here? The drop down only lists this as an option. I have some SDXL models that need to run without a refiner. Thanks.

[quick forge]

where do you install workflows for comfyui?

I downloaded a json file... not sure what to do with it lol

[upbeat crane]

if you cant like select none? then you could go into your settings and then remove the refiner loader drop-down from actually loading on the UI, u could search "refiner" in settings

i dont work with comfy scorpion or i would at least attempt, but you need to triple check that the specific SD cascade model you have is compatible with the version of comfyui you are running, would be where i would start, prob by pulling documentation and making sure they line up

[fair oxide]

i want a wingles fairy but even when i entering it in the negative i often get wing X). what i a solution for that?

[upbeat crane]

sometimes its easy to negate something by adding something else, like if you put a cape or a fur on her or something, then mabe the wings could dissappear, i've had luck taking that route. like if it doesn't show the lower legs in the image no matter what i do, then instead start describing the shoe a bit, then voila the legs appear. this would be similar but maybe think of something else you could type that would negate the wings but in the positive prompt

[fair oxide]

thank you

[upbeat crane]

you load the into the comfui manager @quick forge and you can pull the comfyUI workflows from CivitAI

ive only dabbled with comfy. i just recently got Sarge's SDXL workflow and im starting to play around a bit

its so intimidating, shit.

[whole jackal]

There is no setings for refiner I can find in this UI...I guess I just need to delete the file itself, thx

[upbeat crane]

yeah its almost as the way it see it, that the refiner is the only model in that particular folder or something

[queen plank]

Anyone?

[upbeat crane]

i commented above

but im just saying if it were my issue i would start drilling down that search in google

i would for sure watch this tho, i really dig olivio

sometimes it's just the tiniest little detail or something overlooked

so i suggest watching this video, he outlines the install of SD Cascade for ComfyUI local install

i bet you can come up with somethin, start diggin

[fair oxide]

how can i "combine" things? i want a hanfkerchief as a cape but i get good looking normal capes. i tried "Handkerchief as cape"

[upbeat crane]

you could maybe attempt placing those two items and any other descriptors inside ( ) within the prompt

lemme know what happens

oh sorry forgot to post the video @queen plank https://youtu.be/Ybu6qTbEsew?si=T4q_IktenqeclFzj

[queen plank]

Thanks!

[fair oxide]

sadly still normal cape, also changed the skin to more green but in the second picture it was the cloth and cape and not the skin XD

[upbeat crane]

yw

[fair bone]

hey guys, do you know a colab that can run sdxl lora/model? (idk if that even makes sense)

[upbeat crane]

we also have the #📝｜prompting-help channel but im happy to keep helping as much as i can

[fair oxide]

thanks i will go there

[queen plank]

Error occurred when executing DiffusersStableCascade:

module diffusers.pipelines.stable_cascade has no attribute StableCascadeUnet

Do I update diffusers?

[upbeat crane]

feel ike there were requirements that needed to had been installed at the beginning that got skipped

[fair bone]

any help? 🫡

[queen plank]

I'll try re-running them

[upbeat crane]

i would 100% watch the video, just saying kinda how i operate, and i would listen to HIS process, then i would conpare that again, what i had done and usually then it will be clear what needs to be done? i would look at it like, you are likely to end up having to reinstall at some point, so im never really that terrified of breaking something, with comfy, i have been just 7-zipping up a backup of the whole comfyui folder, then if i do something that breaks it, i can just reset back to the previous "save point" yanno. i

u know a metric fuck ton more about comfy than i do

[queen plank]

I wathced the video, before, but I watched the installation part again. it didn't work. I'll figure it out somehow

Thanks for the help btw!

[upbeat crane]

ohh that sucks, k np love to hear what fixed it tho. cheers

a lot of SD content creators on youtube will actually have their entire transcript linked in the description below a video. you could literally copy that to text and load it into chat GPT and then say "summarize this installation video down to the most critially important elements. give me a detailed step by step list of instructions for this project". i mean, you already have a video with someone teaching it, but this seems like a great idea and could save a lot of time, if you just have a LIST that you work from top to bottom. hell, you could even have ChatGPT compare and contrast 2-3 different sets, tell you what all they have in common, if anything is missed in one, you could ask that it create the step by step instructions based on cross-referencing both, or all three transcription sets. if you had PAID version from OpenAI, i think you could just ask the question then link the two videos. but i do kow you can upload files to chatgpt free....might have to use Co-Pilot tho.

depending on the cut-off for the training set/model, you can actually just ask ChatGPT the question directly as well. If it doesnt know you COULD use the transcript method. i just typed "can you give me a perfect step by step instruction set on how to install Stable Diffusion Cascade on ComfyUI, and help me avoid the most common mistakes and errors, be specific and detail-oriented. Teach me how to do this in simple plain, easy to understand english."

it nailed it btw, very cool

[surreal sentinel]

New to SD. Is there a sample method that's generally good with animals? Most AI seem to be really bad at making dinosaurs, so I want to make some that look decent.

[upbeat crane]

hmmm, i think maybe i would be focusing more on the model used, or maybe even the lora as opposed to the samper itself. euler a at 15 steps and higher, DMP++2m SDE Karras would be solid. a good model might be DELIBERATE from huggingface, or EpicRealism via CIVITAI.

[surreal sentinel]

the one I have rn is from a tutorial here:
https://huggingface.co/CompVis/stable-diffusion-v-1-4-original

[upbeat crane]

have you downloaded any checkpoints yet, or do you just have the base model that came

eww thats old

umm are you using Auto1111?

[surreal sentinel]

howtogeek was out of date ig

[upbeat crane]

here lemme take a peek at it real quick one sec

[surreal sentinel]

not sure, i have webui installed

[upbeat crane]

scroll to the VERY botttom of the Webui in the center

bottom left side center

it will show the version of the webui auto111

[surreal sentinel]

version: 1.8.0-RC  •  python: 3.11.4  •  torch: 2.1.2+cu121  •  xformers: N/A  •  gradio: 3.41.2  •  checkpoint: fe4efff1e1

[fair bone]

guys do you know a google colab that can run SDXL models and lora?

[upbeat crane]

okay perfect, so your fine. you wont have any need to use that particular model tho

i was going to tell you to pull v1.5 but i dont even see the point in that

ill find you a couple checkpoints to download

but in the meantime, jump over to CivitAI.com

this is where you will be able to download checkpoints/loras/ and a bunch of other fun stuff as you learn

i posed you that link didnt i? scroll up

[surreal sentinel]

ill bookmark the civit one

[upbeat crane]

might not have been you, if so i apologize

[surreal sentinel]

probs not, my first message was like 10s after i joined

[upbeat crane]

@fair bone type "colab notebook sdxl" the first link that pops up in google is a git hub page that i linked to you earlier

k one sec

[surreal sentinel]

Tried the cmdline one, and it kept failing to generate even 720p.

I would think a 3070 ti, 64gb ram, and a 5800x would be able to do 720p? lol

[upbeat crane]

https://github.com/camenduru/sdxl-colab

still with you @surreal sentinel trying to think of a great starter model

[surreal sentinel]

which of the vanilla ones?

don't have colab tho

[upbeat crane]

do you prefer REALISM, or like anime type, artistic renderings?

what exactly are you looking for animal wise

[surreal sentinel]

realism for some of my blog posts, artistic for others. (don't use ai to write, just for images)

[vocal burrow]

just passing by, but I don't like those two python: 3.11.4, xformers: N/A

[surreal sentinel]

how do i fix xformers

[vocal burrow]

I don't recommend python 3.11, they'll work.... until they won't because of some extensions or whatever. The tested and approved ones are any python in between 3.10.6 and 3.10.13 (3.10.11 for windows)

[upbeat crane]

ahh yeah, he popped in saying he followed a tutorial somewhere and he ended up with sd 1.4 i was gunna grab him a model real quick

yeah 3.10.6 still huh

[surreal sentinel]

howtogeek's tutorial ig did 3.11

[upbeat crane]

also what is your GPU model?

[vocal burrow]

Also to enable xformers you simply add --xformers to the set COMMANDLINE_ARGS= line in webui-user.bat

[surreal sentinel]

3070 Ti, asus

[upbeat crane]

how2geek flopped hard on this one lol

k thx

[vocal burrow]

while you're editing webui-user.bat you might also add --no-half-vae --medvram-sdxl for good measures

[surreal sentinel]

this should work?

[upbeat crane]

oh okay yeah i see its an 8gb

[surreal sentinel]

eventually ill get a 4080 or something... once I can find a job lol

[vocal burrow]

And for sdxl; dreamshaperXL, juggernautXL, CHEYENNE are good starter models.

[surreal sentinel]

what do those do

[upbeat crane]

juggernaut it is

oh you think he shoudl run xls'

[surreal sentinel]

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --no-half-vae --medvram-sdxl

call webui.bat```
That's what I have on bat

[upbeat crane]

https://civitai.com/models/133005/juggernaut-xl

download that

[surreal sentinel]

alrite, should I keep the other one or nah

[upbeat crane]

the 1.4 model?

u wont use it

[vocal burrow]

--no-half-vae : force the use of fp32 during the VAE stage of the generation (some model+VAE tends to do overflow/nan errors otherwise)
--medvram-sdxl : enable --medvram only for sdxl models (recommended for 8gb vram)
--medvram : load only parts of the model at the time in vram. It's slower but if you don't have enough vram it's sometimes the only way
cf https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings#all-command-line-arguments

[surreal sentinel]

ah ok, at least this will work if Spectrum goes down vs a website haha

[upbeat crane]

my favorite animal to render is the Tarsier. ive done thousands of them little gremlins

[vocal burrow]

honestly.... probably not, you can get some 1.5 models for sure, community ones are pretty good at the stuff they're trained for. sd2.1 models are a weird in between because they're trained for 768x768 stuff instead of 512x512 for sd15 and 1024x1024 for SDXL. Also sd2.1 prompt parser is really different so it takes some time getting used to writing good prompt for it.

[upbeat crane]

juggernaut 1.5

[vocal burrow]

@surreal sentinel

[surreal sentinel]

alright, i got 200gb available on this drive (i have it save renders to D drive tho) so I can have 2-3 no probs

im dling the one you linked above for juggernaut

half done

[vocal burrow]

given that sd1.5 models can take as low as 2gb and sdxl models usually are 6gb. You can have more than 2 or 3 :p

[vast eagle]

Previously when I worked with ComfyUI it would auto load Lora tags in the text encoder box, doesn't seem to do that now, am I missing something?

[surreal sentinel]

still gotta save space on C drive for OS stuff 😉

[upbeat crane]

so seach CIVITAI for juggernaut again and pull the Juggernaut for 1.5. it just wont say XL at the end

that way you will be able to compare and contrast the differences between models trainedon StableDIffusion 1.5 and those trained on SDXL 1.0

[surreal sentinel]

assuming this one

[upbeat crane]

SDXL will look amazing but the speed will be really slow on your GPU

[vocal burrow]

that's the sd1.5 based one

[upbeat crane]

so you will probably want to start out playing with the Juggernaut 1.5

[surreal sentinel]

that's the button i need to click right

[vocal burrow]

this is the sdxl one https://civitai.com/models/133005

[upbeat crane]

lemme kow if you need a link

yeah should be about 2gb

cool

[surreal sentinel]

can save em to same folder, right?
C:\stable-diffusion\stable-diffusion-webui-master\models\Stable-diffusion

[vocal burrow]

yup

[upbeat crane]

100% correct

[surreal sentinel]

yea that one's at 5gb so far

[upbeat crane]

nice yeah u almost up and running

[surreal sentinel]

im at 100 mbps max speeds. (unadvertised slowest)

[vocal burrow]

Oh and before I go back to my cave. Once you've downgraded to a proper python, don't forget to delete your stable-diffusion-webui-master/venv folder.

see ya

[surreal sentinel]

any other models I should get? Those are more realistic (juggernaut) right

for thumbnails I like to do more artistic

[upbeat crane]

like i said you will see, the samplers are very important but the models are gunna do the heavy lifting. you can start playing around with samplers like maybe Euler, Euler A, DPM++ Karras, DPM++ 2M SDE Karras, DDIM even just a few to start out.

[vocal burrow]

the usual realistic ones :
https://civitai.com/models/4201?modelVersionId=130072
https://civitai.com/models/139562/realvisxl-v40

[surreal sentinel]

makig myself a notion page for all this

[upbeat crane]

its fun playing around with Artists styles as well, you can google Stable DIffusion Artistic Studies or "Artist Cheat Sheet" to see a bunch of listed art, and the artist name. within a prompt you can type "in the style of Alex Grey" for example.

uhhh yeah ill tell you some cool ones i like one sec

grab DreamShaper8

[surreal sentinel]

non-xl right

should i grab pixelart as well?

[upbeat crane]

CopaxTimeless is cool

these all there are both actually

once XL came out they re-ran the training sets on the new models

[surreal sentinel]

so xl includes both?

[upbeat crane]

you can run either XL or the 1.5s on yoru Auto111 install

google the differences but

to simplify it

XL was trained on 512x512 images, and XL on 1024x1024

i have NEVER once in my life ran a 1:1 image however

i always push things

[surreal sentinel]

yea, ill be mostly doing 16:9

[upbeat crane]

yeah so you would probably want to do like 960x

umm help me out

568?

find a converter

if 960 is the width in 16:9

[surreal sentinel]

640x360 is one 16:9

so ill delete the 1.4 model then

[upbeat crane]

in XL i do just fine with 1280x720

which is a 1/3 scale factor of 4k i beleive

[surreal sentinel]

probs will do no higher than 1080 (for desktop wallpapers) but that'll be rare. mostly 720p or less

[upbeat crane]

so when i upscale, i upscale two times, two sets at 1.5x each

[vocal burrow]

sd15 was trained on 512x512*

[upbeat crane]

long story short, you want to stay within the paramaters i gave you

[surreal sentinel]

any good models trained on a 16:9 ratio? or they all 1:1

[upbeat crane]

there is a little wiggle room

by going landscape but if you get too wild you will get craziness and artifacts that are most undesirable

[vocal burrow]

don't deviate more than 50% of the training resolution, otherwise bad things will occur

[upbeat crane]

oh nice, yeah that sounds right, i believe i stay within that

[surreal sentinel]

i can name the files to max resolution to make it easier to remember

[upbeat crane]

oh yeah other modesl looking....

[surreal sentinel]

juggernaut xl is safe to be what max 16:9? 1080?

[vocal burrow]

same goes for min, don't go too low

also that rules of thumb is true for native output. Once you've got a good image you can upscale it to pretty much any resolution

[upbeat crane]

DreamShaper8, i freakin LOVE, EPICRealism is nice, RealisticVision, Dreamlike Diffusion is really trippy and cool. I also like DELIBERATE, but he hates CIVITAI think so he took his toys and hosted back on HuggingFace. so if you want that, its over there

yeah some people see highres fix, then upscale UP FRONT, but that kinda bottlenecks you in a way in really upscaling and refining, rerunning, like with IMG2IMG.

i kinda learnd that lesson, you see how nice HighRes fix is at first, but then you realize that upscaling from there, you kinda cut your own legs off a bit, so to speak

[vocal burrow]

Plus now you can easily apply hires.fix after the initial generation.

So don't waste time going for FHD stuff in the first output, you'll waste time, vram and run into weird visual artifacts

[surreal sentinel]

lol it made a dragon with 6 legs and 1 wing haha

[vocal burrow]

might be because you've set the resolution too high like we told you

[surreal sentinel]

720p rn

was with dreamshaper, lemme try jugger

[vocal burrow]

workout time for me I might be back in an hour or so

[upbeat crane]

not sure if @vocal burrow agrees but two good sources of information early on, especially learning about LORAs, controlnet, extensions, etc are Olivio Sarikas on YOUTUBE and Sebastian Kamph. sub those dudes. what i like to do with content creators, is....if i like them, then i will go search through their playlists. Sometimes you will see they have everything broken out, liek a main Stable Diffusion Playlist, usually chronologically ordered, which helps a newb cuz its building on info from video to video. think you have an awesome starting point, and a lot of fun new toys to play with. wish you the best of luck, let us know whenever you have quesitions. happy diffusing

gotta go grab some groceries myself. cheers all.

lemme know if you have any questions, ill be around later....as will others

might be smart to start OFF with the BASE SIZES so you can see what they do when they are rendering at the training sizes

[surreal sentinel]

added to notes, headed out shortly.

If there's any more models or tips or such, feel free to mention me for when i get home from smores night

[upbeat crane]

then work your way up little by little, you will know when you have gone too far

scroll back, i recoomend copying all this text from the last 15 min and copying to a text file or something for later review

[surreal sentinel]

copied it to notion, will sort through it later

[upbeat crane]

ok will do

[surreal sentinel]

made a dragon my friend liked, asked him for his previous fish's colors and generated one he liked 😄

[upbeat crane]

tell him that if he wants any more moving forward he will need to join your Patreon

[surreal sentinel]

hahaha, one of my community admins, if they want to sure

[unique hornet]

@vocal burrow i have over 100 lora but only 23 are show in the stable diffsion ui

[vocal burrow]

Here's a bunch of youtubers if you want some :
https://www.youtube.com/@Aitrepreneur
https://www.youtube.com/@SECourses
https://www.youtube.com/@Not4Talent_AI
https://www.youtube.com/@CreatixAi
https://www.youtube.com/@enigmatic_e

I'm not fond of Olivio, he said things that were wrong and/or incomplete in the past that misslead so many people in here. And his video titles are just the worst kind of clickbait to me.

[upbeat crane]

you know what that has literally been coming to mind as of lately as well

i didn't know really if it was hitting other poeple, i kinda thought maybe it was my fault

but he def last lead me down some dark alleys with no end

like leave out just ONE thing, but it breaks the entirety of the process's logic

interesting

[vocal burrow]

loras are usually only compatible with one SD architecture. So for instance if you're searching for a SDXL lora with a loaded sd1.5 model it won't be displayed in A1111 listing. And vice versa.

[upbeat crane]

yeah you probably have an XL model loaded, so it is only showing the XL compatible LORAs

i assume you have more 1.5 LORAS than Xl

[vocal burrow]

I'm not gonna trashtalk too much on him because he obviously has put a lot of work into it. But yeah I'm not surprised by what you're saying. The few videos I've watched of him were like "click here click here, download this, do that and voila it works" Why ? How ? What are the limitations ? Meeeh, click like and subscribe

Sure that suits some people, but it will also break stuff for others, and you'll need to do additional homework to properly understand what's going on.

still have PTSD his "stable diffusion in 5 minutes WOw so fast emoji emoji" youtube with baby yoda thumbnail.

We've had dozens of AMD users coming here and complaining that it wasn't working....

Because he never precised that they had to use the directml fork of A1111.

[upbeat crane]

yeah it is interesting though, when these creators talk about their ComfyUI custom work-flows, so i joined a live-stream one saturday to watch and about an hour into it, i thought to myself "this man is learning as he goes and does NOT know what he is doing" but taking in that cash. he probably gets the workflows from github, moves a couple strings around and repackages. i hear you tho, i wont either, i feel like you just murdered the easter buddy man, thanks alot

lol teasing

he's a good dude and he has taught me quite a bit despite it

[unique hornet]

yes also my embeddings arent showing ethier

how do i fix it

[vocal burrow]

good thing for you then 👍

[upbeat crane]

sometimes if i start interacting with SD too quickly after the webui loads, it will break stuff like that. reload both windows, but this time wait much longer than you normally do to give time things to properly load. sounds like you have a BUNCH of stuff that is having to lead to memory, and you might be doing something that interupts the caching process or something like that

think ive watched some videos from all of those creators, just not familiar with CreatixAI, have to check him out. i snatched up your list too 😉

[vocal burrow]

We all learn differently, messing around with github demos is a valid approach (as long as it's not the only thing you do). Very few people have what it takes to actually read and understand white papers completly. I certainly don't, I skim over them, I get the gist of it but when it comes to the math details ..... no thanks. I tapped out of AI classes years ago ^^".

[upbeat crane]

might have chatGPT summarize one or two here and then lol

[vocal burrow]

what if you click the refresh button ?

[unique hornet]

i am going to shut now it work i am an idoit

[upbeat crane]

mine does the exact same thing, LORAS and embeddings. loading up 100+ loras is a ton, i bet if you reloaded but waited for them to show rather than just start interacting, they would reappear. mine has done it at least 20 times like that

[unique hornet]

yes sir

[upbeat crane]

quite the lora collection btw

[unique hornet]

try to learn compy and stable i just move files back and forth and it started to add up over time

[vocal burrow]

you can edit comfyUI settings to use A1111's directory for lora, models, etc

[unique hornet]

how??

[vocal burrow]

check the extra_model_paths.yaml.example file

[spare sigil]

My comfyui is constantly crashing trying ti run cascade models. i have a 3070. When reading the command prompt. it always crash on "model_type STABLE_CASCADE
Requested to load StableCascadeClipModel
Loading 1 new model"

[vocal burrow]

(hitting the shower after my workout)

[unique hornet]

what line do write the path at

[upbeat crane]

@vocal burrow i asked this question earlier but didn't get any responses. i saw you just mention to Redninjaszero that LORAs are compatible with the model they were trained on, sorry think you said "loras are usually only compatible with one SD architecture.". that was basically what i was asking about. So, if a LORA is a 1.5, and then SDXL is a branch later on based from the same path, is it not possible to run 1.5 loras with an XL model loaded? I haven't actually done this or even tried it, but just something i have been wondering about. I do deforum animation mostly, and XL is just NOT a go for me really. I have an RTX 3060 with 12GB of VRAM. I am usually in need of a MINIMUM of ONE control net, but many times two. and so things go down-hill real fast, when im trying to load a 6-7gb model, two 1.5GB controlnets, the complicated prompts, all the MATHS etc...i get a whole myriad of errors. I have basically resolved myself to moving back to 1.5 modesl for my Deforum work (i upscale with topaz and utilize a lot of really nice LORAS, so they still look really nice). So, reverse compatibility but not forward or not at all. 1.5 checkpoints with 1.5 LORASs and XL models with XL Loras and Controlents.

same question applies to controlnets, i couldnt possibly run like XL Canny with like JuggernautXL, but then use a 1.5 open pose, or something like that, correct?

[vocal burrow]

it should be all explained in the file itself

you change the base_path, remove the .example from the filename and voila

[unique hornet]

ex in picture pz

[vocal burrow]

"is it not possible to run 1.5 loras with an XL model" short answer no, long answer nooooooooooo.

More seriously sd1.5 and SDXL use very different architectures it's like comparing .... diesel and gasoline engine... sure the basic idea is the same but the engines themselves are very different.

"RTX 3060 with 12GB [is not enough for SDXL]" Meh it should be plenty fine. Throw in lowvram mode for controlnet or even --medvram-sdxl in the worse case and it should work. You might also want to give stable-diffusion-webui-forge a try

"i get a whole myriad of errors" + your second text : Can't help you without seeing the errors but I'm gonna take a blindshot and say that you're using sd1.5 controlnet models while trying to use an SDXL model. If so yup you can't do that.

There are ways to use sd1.5 loras with sdxl but they're either hacky stuff or pretty advanced and still wip.
eg :
https://github.com/city96/SD-Latent-Interposer/
https://github.com/showlab/X-Adapter

I'm not comfortable recommending them here quite yet

[upbeat crane]

aweomse i really appreciate that info, that helps so much honestly. i actually LOVE forge. I spent 3 days with it recently and it was like whack-a-mole. I do realize however that Deforum is very much "experimental" with it but I love testing so I pushed it to it's absolute limits. I will continue using it for image generation, without question.....but since controlnet is still not functioning for webui forge and deforum has yet to be fully integrated with it, i was a bit saddened that I wasnt able to do what I initially thought was about to be possible. It is beautifully designed and the memory optimization is absolutely brilliant. I am 100% positive that it will only get better from here on out, so fingers crossed there. After spending so much time with Forge and with Deforum for Forge, it game me glimpses as to how things might end up and i find that very exciting. on the errors, no i never mix-matched any models or loras. I've only used XL controlnets and models, loras, etc. 1.5 with 1.5. I finally leanred one thing that has been helping a lot so far (ONLY ON 1.5 as of now but im still testing) we doforumer at this point have to disable EVERY SINGLE extention in order to use it now. I had been using SDXL with Deforum and controlnets for several months, but now I just cannot, I get OOMs, all types of controlnet.py errors and messages. Im not using interpolation, or really anything too crazy. Im a member of that discord too and have yet to come up with any viable solutions to my problems with XL and Deforum. I wont give up tho. Thanks for the response and for the info. that helps to set in stone a few things that I had been pondering. Ive been kinda happy to just be back creating lately (using 1.5 stuff) and doing some fun animations. I will move back into XL testing soon, and so I might ping you with one or two of the messages I am geting back. that would be so nice.

[viral fox]

Im running comfy-ui and i seem to have lost my widget that allows me to execute prompts or open the manager. any ideas how to get this back. this is teh second time its happened. first time i reinstalled ComfyUI but I feel this was overkill

[upbeat crane]

sorry i would help if i could but not a comfy user, wish i knew what that window was called

alt or control right click somethin right? lol

[onyx solar]

Is your browser window zoomed to something other than 100%?

[upbeat crane]

@vocal burrow i was only asking about the cross-compatibility between all these things, in the hopes that if it were possible it might help me scrimp on some the bulky VRAM controlnet, lora and model usage. was just trying to learn more and find a way to keep using XL, because XL animations through Deforum are jaw-droppingly beautiful. I'll get back there FOR sure, something is off and im gunna find it once and for all.

[quick forge]

is stable cascade an extension, or it's own ui?

[onyx solar]

Neither.

It's a model type, like 1.5 and XL.

[quick forge]

I see....

so installing it through the A1111 extensions tab should install it properly?

or do i have to run it through the "notebooks"?

[onyx solar]

I can't speak specifically to A1111 since I typically use ComfyUI.

[quick forge]

im just a little confused about how to actually and run it

so it does work within the webui's though, like comfy or a1111

[upbeat crane]

https://huggingface.co/stabilityai/stable-cascade/tree/main

[quick forge]

so it's just safetensor files?

[onyx solar]

Yes, it works through the same tools as all of the other models; again, it's just another model type.

[upbeat crane]

i have yet to work with it been doin so many other things

it does look like a whole lot more than just a checkpoint

having its own controlnets n such

[onyx solar]

Ultimately, that's no different than the differences between 1.5 and XL.

[safe inlet]

Comfy specific checkpoints here, less confusing https://huggingface.co/stabilityai/stable-cascade/tree/main/comfyui_checkpoints

[quick forge]

well what do you load then, stage a b or c?

[upbeat crane]

i guess you are right

[safe inlet]

both

[onyx solar]

You should probably read the documentation on the links.

[upbeat crane]

https://www.youtube.com/watch?v=W6YLIyA3Kco

sebastian isnt perfect either but he right more than he's wrong

learned a lot from this little sweedish man

cant you just git clone from this

https://huggingface.co/stabilityai/stable-cascade/tree/main

[onyx solar]

Cloning that would give you a lot more than you need.

[upbeat crane]

not finding the correct link then oops

the github one that is just the diffusers then maybe?

[quick forge]

ok I see..... I installed it through the extensions tab, and it shows as installed, but I don't have a "StableCascade" tab like the video says I should.

[upbeat crane]

oh thats what someone mentioned and i was like prollly not

so my bad i was wrong if so, thats strange and cool

if someone is doing that lemme know how it goes

might pop it on later tonight

[onyx solar]

c = first stage
b = 2nd stage
a = VAE

[upbeat crane]

ive heard a lot of good things about prompt alignment

if thats the right word

gunna go take a nap, i cant think

[onyx solar]

It adheres well, but only for short prompts...after a few tokens it tends to lose the prompt.

[upbeat crane]

wait so are these steps automated? or you have to push them yourself

i know nobody really runs refiners in sdxl anymore but i really hated that

[onyx solar]

It's automatic.

[upbeat crane]

interesting

k ill check it for sure

[onyx solar]

It's not refinement in the same way...but it is similar.

[upbeat crane]

gotcha

[young bridge]

hello

how do I uninstall stable diffusion auto111111

[onyx solar]

Just delete the folder

[ember kernel]

When I try to add a plugin, when the page first opens, etc. whatever I try to do, it gets stuck on the loading icon. I have uninstalled and reinstalled the app many times in different ways, but to no avail. Can you help me?

https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/15213

[regal sedge]

Hi! Can anyone suggest where to look for the problem and how to try to solve it? I am trying the L2 lesson from Olivio and the error appears after Apply ControlNet on the second KSampler (to which it is applied). All nodes are updated.

[solemn moon]

I’m gonna take a wild guess and say your resolution should be 768

Or maybe even 1024

Something to do with dimensions too maybe

Isn’t XL a minimum of 1024?

[faint summit]

does anyone know how to revert to older versions of SD-forge?

im trying to use regional prompting and its broken on current version of forge, i saw these comments

does anyone know how i can get to that exact version?

[ember kernel]

:3

[ornate elk]

What does the cmd shows

[ember kernel]

i add console log in git report

there is no errors

[ancient plover]

Hey, I am having issues installing Zluda webui, getting errors on the final launch, I think It is an issue with my path settings, I would greatly appreciate if someone who got it working could send me a list or image of their exact enviroment variables that are enabled. thks

[ember kernel]

@ornate elk

[ornate elk]

Your webui Version is very outdated

Your on 1.5.0 and latest is 1.8.0

[ember kernel]

yes i know it

[ember kernel]

i try this versions: 1.5 1.6 1.8

but there is no matter

same issue

[regal sedge]

Looks like I used the 1.5 controlnet model with the xl model in ksampler 🤦‍♂️ noob

[solemn moon]

Argh 😅 so many variables, glad you got it 🫶

[ornate elk]

Make sure to use the latest version and then also make sure to whitelist the webui in any browser adblocker

[paper crag]

okay so im trying to solve a error.
I keep getting this
'OSError: [WinError 126] The specified module could not be found. Error loading "C:\SD-Zluda\stable-diffusion-webui-directml\venv\lib\site-packages\torch\lib\cublas64_11.dll" or one of its dependencies.
Press any key to continue . . .'
I checked the file, cublas64_11.dll is in the right place.
I downloaded the cublas again, renamed it, and replaced the previous one with it. I'm still getting this error.

[ornate elk]

Thats strange

You could try delete the venv folder and then restart the webui-user.bat
It may automatically uses the right Cublas files

[paper crag]

delete all together or its contents?

[ornate elk]

The whole venv folder

[paper crag]

okay, lets see if this does anything

[viral fox]

Sorry for the late reply. it must have been. Chrome / windows +/_ didnt say where I was. I fired it up in Firefox and all was working well. when resizing the Chrtome window the Queue prompt window popped back in. Thanks for the assist. you got my brain on the right track. it was late and i wasnt thinking analyticaly

[paper crag]

okay this is some bs....
now i got a same error, but different file
'OSError: [WinError 126] The specified module could not be found. Error loading "C:\SD-Zluda\stable-diffusion-webui-directml\venv\lib\site-packages\torch\lib\caffe2_nvrtc.dll" or one of its dependencies.
Press any key to continue . . .'
checked and its there

[ornate elk]

Okay what's in your webui-user.bat?

[paper crag]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --use-zluda --skip-torch-cuda-test --medvram-sdxl --update-all-extensions --update-check --skip-ort --theme dark

call webui.bat

[ornate elk]

Okay looks good

What you can try is to open up a cmd
And run
Pip cache purge

And then delete the venv again

[paper crag]

when i deleted my vevn, and opened my webui again, it did say that i have an older pip and could upgrade it. Do you think that could be the case?

[ornate elk]

You should make sure that you have python 3.10.11 installed

The pip message can be ignored

[paper crag]

I have Python 3.10.6

[ornate elk]

Then install 3.10.11, you dont need to uninstall 3.10.6

Then do the pip cache purge

And then delete the venv

[paper crag]

okay, i just upgrade the old one right, not customize

[autumn void]

Hi does anyone know why Fooocus is using 99% of my ram? I don't think it's supposed to do that. I'm running 16GB

[paper crag]

huh... it upgraded, but theres something curios at the end, theres a section thats asking me if i want to disable path lengh limit. Should i disable it?

[ornate elk]

Yea thats okay

[ornate elk]

What's your GPU?

Fooocus uses sdxl models which require a lot of RAM.
It gets worse if your GPU has less than 6gb vram

[paper crag]

should i upgrade the pip before i purge it? Got version 22, and 24 is available

[ornate elk]

You can do that

[autumn void]

it's the 4070 12GB

just opened it's idling at 11.5GB RAM usage and when processing anything it gets to 99ish% and at some point just chokes and cuts

the GPU gets to around 50% usage at the most

is adding another 16GB of ram going to help or is it going to max at at 99% of 32GB-_-

[ornate elk]

It won't Max out 32

[autumn void]

also when it does crash sometimes I get those GPU spikes

[ornate elk]

But you may need to make your pagefile bigger

[autumn void]

but only sometimes

I'll try to increase it further

is that the only fix for my current setup?

[ornate elk]

That or buying more RAM

But its normal that the ram usage is a lot with sdxl

[autumn void]

and normal is in the ballpark of...

20? 25 Gigs?

[paper crag]

finallyyyyyyyy

[paper crag]

im back in business, thanks for the help

[autumn void]

anyway, what min. and max. paging file values would you say are fine for the 16 GB ram that I currently have

[ornate elk]

Perfect np 🙂

[autumn void]

I'm more concerned with the initial size

[ornate elk]

I would go with 16000min and 24000max

[autumn void]

thanks, will try that

[mellow ibex]

When installing on CachyOS (arch based) Linux Automatic1111, I get this error:

 │ exit code: 1
 ╰─> [3 lines of output]
     Looking in indexes: https://download.pytorch.org/whl/rocm5.4.2
     ERROR: Could not find a version that satisfies the requirement setuptools>=40.8.0 (from versions: none)
     ERROR: No matching distribution found for setuptools>=40.8.0
     [end of output]

When visiting the index website, it gives me a 403 Access Denied error. This issue is not temporary, as I have tried both running my existing (migrated from windows) install and installing from scratch many times, and it does not work. How do I fix this error?

[loud salmon]

I'm trying to add an AST template as a new user. I can't see a Tenanies panel on the page.

[queen plank]

Error occurred when executing KSampler: Given groups=1, weight of size [320, 16, 1, 1], expected input[2, 64, 12, 12] to have 16 channels, but got 64 channels instead File "D:\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 151, in recursive_execute output_data, output_ui = get_output_data(obj, input_data_all) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 81, in get_output_data return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)

Anyone gettingthis with cascade?

comfy ui

[delicate urchin]

Why do I not have any "python_embeded" folder?

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

As the person in this video has

[fallen quest]

Do you use the embedded version? - I meant portable version

[delicate urchin]

embeded comfy version?

[fallen quest]

Wait a sec

https://github.com/comfyanonymous/ComfyUI/releases/download/latest/ComfyUI_windows_portable_nvidia_cu118_or_cpu.7z

This version has its own enviroment

Just unpack, update and you are good to go

[delicate urchin]

ok will try

[fair oxide]

hey! tengo una suscripcion de pago a Stable Diffusion, no entiendo como se puede usar acá para generar imagenes

[night orchid]

No one of my business, but I think you should use English in here

Guys, my SD have suddently become ver very slow

is there anything I can do?

[night orchid]

Stable Diffusion is free

[manic snow]

I assume you tried restarting it?

[night orchid]

you cant generate images here

@manic snow yes I did

[manic snow]

what ui are you running?

[night orchid]

Automatic1111

[manic snow]

it's kind of hard to diagnose the problem for me, since that just happens sometimes. try disabling all your extensions, and if that doesn't help, create a second A1111 installation. if anyone has any other ideas, please feel free to add them

[night orchid]

Thank you for your help

[manic snow]

no problem.

[vocal burrow]

if you subscribed to dreamstudio I'd suggest you get in touch with them through their website.

[fair oxide]

so..what the hell i paid -.- haha thanks people!

[vocal burrow]

slow when ? when generating or slow navigation ?

[vocal burrow]

what did you pay for exactly ? where ?

[night orchid]

slow generation

[fair oxide]

https://clipdrop.co/

[night orchid]

almost 8 min to generate 4 images at 512x512

[fair oxide]

with Stable Diffusion and and beta version and some stuff more

[vocal burrow]

Ok so good news, that's a legitimate SD online service. you didn't get scammed.

Still to get a proper answer to your problem I suggest you contact them using the "contact us" form at the bottom of the page.

[fair oxide]

thanks mate 🙂 !

[night orchid]

@fair oxideits a good solution for those that dont want to tired them self out with a local installation

A young woman in a red dress
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3382243247, Size: 512x512, Model hash: 2d5af23726, Model: realismEngineSDXL_v30VAE, Version: v1.8.0

to almost 90 sec

took*

[vocal burrow]

what s your hardware ? what did you change since it was nt laggy ?

[fair oxide]

well if im totally honest, i was trying to create some erxtixc stuff, no explicit, but something like that.. but well -.-

[vocal burrow]

There might be a nsfw filter on there.

probably

For that kind of content, you d want to use a local install. And as I don't want to encourage that kind of usage and because of Rule.4 I won't help you in that regard specifically.

[fair oxide]

thaks anyways !

[delicate urchin]

with this version of comfy ReActor node import still fails, I tried reinstalling insightface for python v 3.10 and opencv-python==4.7.0.72 inside the python embed

[fallen quest]

I could have chosen my words better. What i meant was: Then you have a working version of ComfyUi with the folder you were looking for.

[queen plank]

Anyone?

[delicate urchin]

Yeah I fixed the last issue now as well

[solid river]

does anybody know if i can use this hardware to do the calculations? https://coral.ai/products/ looks like they only do 8bit integer operations instead fp16/fp32

[wintry atlas]

Did anyone report a1111 exhausting virtual memory while generating with SDXL?

Committed pool grows to 80+ gig and webui crashes

[vocal burrow]

Very unlikely, those seem to be capable of handling basic tensorflow stuff but I doubt they have any cuda capabilities.... They seem to be able of running mediapipe for yolo and similar recognition stuff but that s about it.

I'd love to be proven wrong on this one. That would mean there is an affordable way for anyone to get access to SD. But I doubt it

[solid river]

ill report back to the channel when the hardware arrives and i checked if its possible

[vocal burrow]

fwiw auto1111 finally adopted fp8 support with their latest version.

[night orchid]

what is fp8?

[solid river]

8 bit floating point number aka minifloat

[karmic crown]

i am testing the fix i made for deforum related to basicsr module, but i have never used interpolation in deforum before
is this expected and i need to manually download the model first

    Exception: Error while downloading film_net_fp16.pt. Please download from: https://github.com/hithereai/frame-interpolation-pytorch/releases/download/film_net_fp16.pt/film_net_fp16.pt, and put in: /Users/viking/a180/models/Deforum/film_interpolation

or this should be automatically download by extension?

does anyone knows this?

[ember kernel]

The program hangs on all operations.

[ornate elk]

did you installed it on an ssd or hdd?

[ember kernel]

ssd

[ornate elk]

can you also share the cmd output after starting it

[ember kernel]

sure

[spare sigil]

The text fields on my comfyui gets gray when i try to click on them to type something. This issue started yesterday after i did an update.

[sudden bison]

Hello, I keep getting this error when trying to install controlnet:
stderr: ERROR: Could not install packages due to an OSError: [WinError 5] Access is denied: 'C:\Users\xxx\SD2024\venv\Lib\site-packages\cv2\cv2.pyd'
Check the permissions.

---

I have reinstalled webui and ran the webui bat as admin. I previously did it without running as admin. Same issue. I did see it said fatal can't find git repository when running as an admin but it installed webui fine still.

I followed this person's solution of manually pip installing the libraries but still same issue
https://github.com/Mikubill/sd-webui-controlnet/issues/1258

[ember kernel]

@ornate elk can you help :3

[ornate elk]

this extension has an error:
extensions\SD-CN-Animation

[ember kernel]

i delete this extention now, but same

now, the logs:```py
venv "D:\Ai Resim\StabilityMatrix-win-x64\Data\Packages\Stable Diffusion WebUI\venv\Scripts\Python.exe"
Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr 5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Version: v1.8.0
Commit hash: bef51aed032c0aaa5cfd80445bc4cf0d85b408b5
Launching Web UI with arguments:
No module 'xformers'. Proceeding without it.
Loading weights [9ae52c3039] from D:\Ai Resim\StabilityMatrix-win-x64\Data\Packages\Stable Diffusion WebUI\models\Stable-diffusion\darkSugarPotAnimeType_v30.safetensors
Creating model from config: D:\Ai Resim\StabilityMatrix-win-x64\Data\Packages\Stable Diffusion WebUI\configs\v1-inference.yaml
Running on local URL: http://127.0.0.1:7860

To create a public link, set share=True in launch().
Startup time: 13.5s (prepare environment: 2.7s, import torch: 4.8s, import gradio: 1.2s, setup paths: 1.3s, initialize shared: 0.5s, other imports: 0.7s, load scripts: 1.4s, create ui: 0.4s, gradio launch: 0.5s).
Applying attention optimization: Doggettx... done.
Model loaded in 21.6s (load weights from disk: 0.2s, create model: 0.6s, apply weights to model: 16.3s, move model to device: 4.4s, calculate empty prompt: 0.1s).

[ornate elk]

did you whitelisted the webui in any browser adblocker?

[ember kernel]

yes i try it, and i try using other network

[ornate elk]

okay, if nothing helps, try a clean installation using my guide and not stability matrix:
#🤝｜tech-support message

[viral fox]

where do i stick LYCOORIS stuff on ComfyUI? In the Models/LORA folder? - https://civitai.com/models/119094/leonardo-ai-illustration-15-xl-or-goofy-ai

[cursive dove]

Why is my ControlNet doesnt work?? Please help me!

[ornate elk]

your prompt wasnt recognized correctly

make sure to whitelist the webui in any browser adblocker

[rancid wagon]

do i need to run this and download everything all over again every single time I open the collab or is there a way to keep it all saved and just pressing a button to get it running?

[cursive dove]

This is a pure browser and I tried to generate an image without controlnet

[ornate elk]

you should try to delete venv folder and relaunch your webui-user.bat
that will update your torch and xformers version,
then update your controlnet extension

[fair oxide]

Guys how can I make A1111 work with an AMD Gpu?

[ember kernel]

same 😦

[ornate elk]

hey, whats your GPU?

[fair oxide]

RX6600

[ornate elk]

then you can follow my AMD Zluda guide. its in the pinned messages of this channel

[fair oxide]

Perfect

Thank you very much for the information

[cursive dove]

I dont see venv folder😭

[vocal burrow]

how many loras, models and embeddings do you have approximatively ?

[ornate elk]

np, if you have any issues, feel free to ask here again.

[ember kernel]

only 1 model, i trying only work with basic a1111

[rancid wagon]

I always get this error on ComfyUI when trying to run it again from mounted drive instead of having to download every single thing all over again, how do I fix this?

python3: can't open file '/content/main.py': [Errno 2] No such file or directory

From my understandings it doesn't find the python directoy but how do I mount the drive and folder properly?

[ornate elk]

how big is the other model?

and whats your gpu?

[ember kernel]

RTX 3060 12gb VRAM, 16gb RAM

[ember kernel]

basic model: v1-5-pruned-emaonly.ckpt

[ornate elk]

and you installed the webui freshly now?

[ember kernel]

yes

[ornate elk]

and can you generate an image?

[ember kernel]

no when i try this, program is stuck (same issue)

[ornate elk]

can you try this model?:
https://civitai.com/models/4384/dreamshaper

[rancid wagon]

please help 🙏

[noble belfry]

late to the party, but that is a scribble/drawing and it will be more forgiving if you use that mode instead of canny. its just having a hard time finding any kind of human shape to put into that. you can do things like turn down control weight or change ending step, but try using scribble sketch mode first. 🙂

its kind of cute that since it didnt know what to do with it, it just made a photorealistic cardboard cutout of it though. 😄

[cursive dove]

Ill try it thank you!
Maybe you know where to find XL versions of models for control net? I couldn't find the line XL version...

[pseudo nebula]

hmm mov2mov keeps making a 13 second clip into 1 second. idk whats going on

[noble belfry]

https://huggingface.co/lllyasviel/sd_control_collection/tree/main should be in there

[karmic crown]

are you on a1111 1.8.0?

[cursive dove]

There is no line model😭

[night orchid]

I am trying to let SD Automatic1111, share the models and so on with SD Forge

https://civitai.com/search/models?query=line

for you bim

[noble belfry]

edit your webui-user.bat to `@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=

@REM Uncomment following code to reference an existing A1111 checkout.
set A1111_HOME=C:/AI/stable-diffusion-webui

set VENV_DIR=
set COMMANDLINE_ARGS=%COMMANDLINE_ARGS% --ckpt-dir %A1111_HOME%/models/Stable-diffusion --hypernetwork-dir %A1111_HOME%/models/hypernetworks --embeddings-dir %A1111_HOME%/embeddings --lora-dir %A1111_HOME%/models/Lora

call webui.bat
` change the A1111_HOME to wherever you have it

[night orchid]

oh , thanks

[noble belfry]

yeah its pretty painless. 🙂

doesn't move over vaes though

[night orchid]

set COMMANDLINE_ARGS= --xformers --ckpt-dir E:\AI\stable-diffusion-webui\models\Stable-diffusion

this is what I have now

[karmic crown]

why so complicated? --forge-ref-a1111-home is enough

[noble belfry]

oh fair enough

[night orchid]

well I am souch a noob, i found that online 🙂

ah not in a reply to me, now I am blind too 🙂

[karmic crown]

just add
--forge-ref-a1111-home C:/AI/stable-diffusion-webui
if your a1111 is in C:/AI/stable-diffusion-webui

check if you need / at the end or not

it should use all your models, loras... from a1111

[noble belfry]

https://github.com/lllyasviel/stable-diffusion-webui-forge/pull/203 yeah i missed that add 🙂

[karmic crown]

it was added to avoid what you sugested 🙂

[noble belfry]

i knew it wasnt there last i messed around with the commandline 😄

[night orchid]

you guys are so good

thanks both

[noble belfry]

ok then, go with vikings better version Esim 😄 i will be too

thats handy, it does vaes too

[karmic crown]

its not mine, i just wrote, what i read 🙂

i just got fresher info than wit 🤷🏻‍♂️

[night orchid]

🙂

[noble belfry]

https://tenor.com/view/simpsons-grand-pa-millenial-old-gif-10484595

[night orchid]

hmm I think I am stupid or something, it is not working

[noble belfry]

are you using / or ?

sorr \

[night orchid]

I will put this in my Forge ? right

set COMMANDLINE_ARGS= --xformers --forge-ref-a1111-home E:\AI\stable-diffusion-webui\models\Stable-diffusion

[noble belfry]

set COMMANDLINE_ARGS= --xformers --forge-ref-a1111-home E:/AI/stable-diffusion-webui

[night orchid]

don't laugh

🙂

ah thanks for the correction

[karmic crown]

should that be E:\AI\stable-diffusion-webui on winBLows?

[night orchid]

it should be a standard

[karmic crown]

i know with file:/// you can use / but i am not sure about this

[noble belfry]

backslash forward slash on cmd in windows batches will still trip me up

[night orchid]

ah NICE! it is working now

[noble belfry]

if you ask why people just babble about IBM and the 1980s

[night orchid]

hehe

[karmic crown]

i use windows only when i have to 🙂

[night orchid]

maybe because they was the best 😉

[noble belfry]

anyway, im happy that my commandline is now a bit more elegant so ive learned and grown on this day

https://tenor.com/view/the-more-you-know-gif-19118236

[night orchid]

cool gif

wooow Forge used 2 sec with one image

[noble belfry]

yeah on low end machines its great once its all loaded. can be slower in some cases though

[night orchid]

maybe my automatic1111 is sick

uses 30 sec

sometimes longer

[noble belfry]

no, forge is just made by the same guy who makes controlnet and added lots of optimizations

[karmic crown]

any volunteers to test my latest deforum fix? 🙂

[noble belfry]

calc.exe makes my gpu sweat so nope 😄

[night orchid]

i dont even know what deforum is

[karmic crown]

forge works slower than a1111 on my Mac 🙂

[night orchid]

17sec for 4 images

Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3717997978, Size: 512x512, Model hash: 15012c538f, Model: realisticVisionV51_v51VAE, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7

Time taken: 17.3 sec.

Automatic took 5 min

[karmic crown]

you have about 7it/s? right?

[night orchid]

7it/s ? whats that

[karmic crown]

check the CMD window

you will find something like this [00:00<00:00, 4.00it/s]

[night orchid]

yes 7

medium end GPU

rtx 3060 12GB