Software Engineering Courses (SECourses)•3y ago

maybe interesting for you @mikemenders : https://github.com/bghira/SimpleTuner/releases/tag/v0.7.0

GitHub

Release v0.7.0 - tune SDXL in just 12G VRAM! · bghira/SimpleTuner

DeepSpeed lives!
Now correctly integrated, DeepSpeed allows for training SDXL's full U-net at 8 seconds per iteration on just ~12G of VRAM. See the documentation for more information!
Features
...

sharkdodo•10/22/23, 1:06 PM

a question on lora with various checkpoints: in the previous sd1.5 era, i could train a lora on sd1.5 or anylora checkpoint and use that lora on almost any checkpoints without fail. However, loras trained on & tested ok with sdxl model doesnt seem to work at all with other checkpoints (such as realVisXL) - it generates a lot of artifacts as if overtrained. Does anyone have idea why?

Sandstormleader•10/22/23, 1:31 PM

anyone have a link to the automatic1111 sdxl auto installer for windows?

Abolfazl•10/22/23, 3:57 PM

لطفا به فارسی چت کن

Abolfazl•10/22/23, 3:57 PM

سلام

Abolfazl•10/22/23, 3:58 PM

Hello

GGeorgeClintonGeorgeBush What's the current best SDXL checkpoint for realistic photos of people?

Furkan Gözükara SECourses•10/22/23, 4:35 PM

sdxl 1.0 really good

Furkan Gözükara SECourses•10/22/23, 4:35 PM

havent compared others yet

Pphillip Is there a youtube video or something how to use this kind of caption for Lora? ...

Furkan Gözükara SECourses•10/22/23, 4:35 PM

i dont have yet but you can test each one do a train and compare

Furkan Gözükara SECourses•10/22/23, 4:35 PM

best approach

MMarcelo Sdxl quality is insane, thank you for everything @Dr. Furkan Gözükara . Its 2 pm...

Furkan Gözükara SECourses•10/22/23, 4:36 PM

awesome thank you

DR.Siva•10/22/23, 4:46 PM

ok it seems lobe theme needs internet to properly load

BBenjamin is it maybe possible to save more checkpoints if we put them in a temporary fold...

Furkan Gözükara SECourses•10/22/23, 4:47 PM

yes but kaggle has a problem right now with temp folder

Furkan Gözükara SECourses•10/22/23, 4:47 PM

i got reply from kaggle employee

Furkan Gözükara SECourses•10/22/23, 4:47 PM

it works super slow

LLEGACIES @Dr. Furkan Gözükara is it more accurate to crop my dataset with your auto crop ...

Furkan Gözükara SECourses•10/22/23, 4:47 PM

i think if you crop same way would be same

Furkan Gözükara SECourses•10/22/23, 4:47 PM

but you dont need to waste time

Ssharkdodo a question on lora with various checkpoints: in the previous sd1.5 *era*, i coul...

DR.Siva•10/22/23, 4:48 PM

yes i too felt the same. Lora`s dont play well with models imo. Also i found , setting them to a 0.9 or less than that results in much controlled output.

Mmikemenders @Dr. Furkan Gözükara check this PR 😉 https://github.com/kohya-ss/sd-scripts/pul...

Furkan Gözükara SECourses•10/22/23, 4:48 PM

amazing news

Furkan Gözükara SECourses•10/22/23, 4:48 PM

after this sd 1.5 can be researched throughly

SSandstormleader @Dr. Furkan Gözükara do we have an up to date version of dreambooth and automati...

Furkan Gözükara SECourses•10/22/23, 4:48 PM

older working only

Furkan Gözükara SECourses•10/22/23, 4:48 PM

but after EMA of kohya we won't need it

DR.Siva•10/22/23, 4:49 PM

qwen-flux-wan-such Will try this soon..

BBenjamin maybe interesting for you @mikemenders : https://github.com/bghira/SimpleTuner/r...

Furkan Gözükara SECourses•10/22/23, 4:49 PM

i think instead using kaggle better

Furkan Gözükara SECourses•10/22/23, 4:50 PM

you can train with 2 hours 160 epochs

FFurkan Gözükara SECourses i think if you crop same way would be same

LEGACIES•10/22/23, 4:50 PM

how can i use bucket in dreambooth extension for sd 1.5?

Ssharkdodo a question on lora with various checkpoints: in the previous sd1.5 *era*, i coul...

Furkan Gözükara SECourses•10/22/23, 4:50 PM

because sd 1.5 models are vastly merge

LLEGACIES how can i use bucket in dreambooth extension for sd 1.5?

Furkan Gözükara SECourses•10/22/23, 4:50 PM

i think sd 1.5 dont have bucket option

Furkan Gözükara SECourses•10/22/23, 4:51 PM

or it is doing auto

Furkan Gözükara SECourses•10/22/23, 4:51 PM

cant remember

SSandstormleader anyone have a link to the automatic1111 sdxl auto installer for windows?

Furkan Gözükara SECourses•10/22/23, 4:51 PM

here https://www.patreon.com/posts/86307255

LEGACIES•10/22/23, 4:53 PM

i saw in your video that for auto cropping script that you have sent the dataset into the train tabl to resize it, do i need to do the same with my dataset after cropping it 512px x 768pxt?

LEGACIES•10/22/23, 4:54 PM

LLEGACIES i saw in your video that for auto cropping script that you have sent the dataset...

Furkan Gözükara SECourses•10/22/23, 4:57 PM

auto1111 cropping so bad

Furkan Gözükara SECourses•10/22/23, 4:57 PM

dont use

Furkan Gözükara SECourses•10/22/23, 4:57 PM

use my script

Sandstormleader•10/22/23, 5:35 PM

is kohya something you run locally?

SSandstormleader is kohya something you run locally?

Furkan Gözükara SECourses•10/22/23, 5:42 PM

yes

Furkan Gözükara SECourses•10/22/23, 5:42 PM

you can run anywhere

Furkan Gözükara SECourses•10/22/23, 5:42 PM

it is an external trainer script

Sandstormleader•10/22/23, 5:46 PM

everything i see is a notebook @Dr. Furkan Gözükara

Sandstormleader•10/22/23, 5:47 PM

also does it only do LORAs i've never made a lora

SSandstormleader everything i see is a notebook @Dr. Furkan Gözükara

Furkan Gözükara SECourses•10/22/23, 6:05 PM

watch this tutorial please : https://youtu.be/sBFGitIvD2A

YouTubeSECourses

Become A Master Of SDXL Training With Kohya SS LoRAs - Combine Powe...

In this tutorial, you will learn how to install Automatic1111 Web UI for SDXL. How to use LoRAs with Automatic1111 SD Web UI. How to install Kohya SS GUI scripts to do Stable Diffusion training. How to train LoRAs on SDXL model with least amount of VRAM using settings. All of the details, tips and tricks of Kohya trainings. How to do x/y/z plot ...

Furkan Gözükara SECourses•10/22/23, 6:05 PM

then this : https://youtu.be/EEV8RPohsbw

YouTubeSECourses

How To Do Stable Diffusion XL (SDXL) DreamBooth Training (Full Fine...

Patreon post for config files : https://www.patreon.com/posts/89213064

SSandstormleader also does it only do LORAs i've never made a lora

Furkan Gözükara SECourses•10/22/23, 6:05 PM

it can make lora and dreambooth as well

Sandstormleader•10/22/23, 7:00 PM

hmm i was unable to get the dreambooth tab to start working with out regulization images

SSandstormleader hmm i was unable to get the dreambooth tab to start working with out regulizatio...

Furkan Gözükara SECourses•10/22/23, 7:06 PM

it should work i did a lot of such training

Sandstormleader•10/22/23, 7:07 PM

i'll have to try again when i have more time

Kallamamran•10/22/23, 7:58 PM

@Dr. Furkan Gözükara Hey... I don't remember... What CLIP-model did Stable Diffusion use? Default in CLIP Interrogator is VIT-L-14 and that's not it right?!

EyeSpyBekoAI•10/22/23, 8:02 PM

Runpod users, question for you. Do you know if it's possible to clone a pod, exactly as it is, and basically recreate an instance exactly with all the files, setup, etc., in it? For example, let's say I setup pod "X", downloaded Kohya, A1111, all extensions, files I need, etc. And then, let's pretend I come back to that pod tomorrow, and it has unavailable GPU. Is it possible to clone/recreate the whole setup, so that I can have a pod that has GPU available with my setup immediately setup?
Thanks!

KKallamamran @Dr. Furkan Gözükara Hey... I don't remember... What CLIP-model did Stable Diffu...

Furkan Gözükara SECourses•10/22/23, 8:10 PM

laion/CLIP-ViT-bigG-14-laion2B

maybe interesting for you @mikemenders : https://github.com/bghira/SimpleTuner/releases/tag/v0.7.0

Similar Threads

Similar Threads

Similar Threads