Software Engineering Courses (SECourses)•15mo ago

If this works I'll eat my hat

KallamamranOP•10/7/24, 1:42 PM

https://tenor.com/view/allenxandria-eating-hat-gif-22720770

Tenor

KKallamamran If this works I'll eat my hat

Furkan Gözükara SECourses•10/7/24, 1:45 PM

should work :d

KallamamranOP•10/7/24, 1:50 PM

Seem to

KallamamranOP•10/7/24, 1:50 PM

Will try yet another imposible training now

KallamamranOP•10/7/24, 1:51 PM

Geometrically advanced rear headlights

Furkan Gözükara SECourses•10/7/24, 1:52 PM

importance of RAM - Virtual RAM

KKallamamran Geometrically advanced rear headlights

Furkan Gözükara SECourses•10/7/24, 1:52 PM

FFurkan Gözükara SECourses ohwx + class

tazztone•10/7/24, 1:54 PM

i wonder if other class prompt than "man" would be better. my character has no hair, but most men have hair

so i thought about using "skinhead" or smth. wdyt?

Ttazztone i wonder if other class prompt than "man" would be better. my character has no h...

Furkan Gözükara SECourses•10/7/24, 1:59 PM

you can make

Furkan Gözükara SECourses•10/7/24, 1:59 PM

ohwx bald man

Furkan Gözükara SECourses•10/7/24, 1:59 PM

and prompt as bald

Furkan Gözükara SECourses•10/7/24, 1:59 PM

but i think either way should work for flux

KallamamranOP•10/7/24, 2:03 PM

Don't think I'll run this

Furkan Gözükara SECourses•10/7/24, 2:03 PM

haha 5k images

Furkan Gözükara SECourses•10/7/24, 2:03 PM

train 1 epoch and see results

Furkan Gözükara SECourses•10/7/24, 2:03 PM

you can continue with generated checkpoint later

KallamamranOP•10/7/24, 2:03 PM

Really isn't...

KallamamranOP•10/7/24, 2:04 PM

It's 243 images

Furkan Gözükara SECourses•10/7/24, 2:04 PM

nope you have 5040 images

KallamamranOP•10/7/24, 2:04 PM

Nope. I really dont!

Furkan Gözükara SECourses•10/7/24, 2:04 PM

then you have mistake somewhere

Furkan Gözükara SECourses•10/7/24, 2:05 PM

KallamamranOP•10/7/24, 2:05 PM

Including .txt files

Furkan Gözükara SECourses•10/7/24, 2:05 PM

your repeating 10?

Furkan Gözükara SECourses•10/7/24, 2:05 PM

or 20 ?

Furkan Gözükara SECourses•10/7/24, 2:05 PM

you have something wrong for sure

KallamamranOP•10/7/24, 2:07 PM

40 was default in the config file. I will reduce this

KallamamranOP•10/7/24, 2:07 PM

Still won't be a giant diffrence just by that

KallamamranOP•10/7/24, 2:13 PM

I'm not going to wait that long

KallamamranOP•10/7/24, 2:13 PM

I'll revert to using AI-Toolkit and LORA instead

KallamamranOP•10/7/24, 2:13 PM

That's soooo much easier

tazztone•10/7/24, 2:51 PM

just the kohya dataset prep is a bit weird to understand in their UI

Ttazztone just the kohya dataset prep is a bit weird to understand in their UI

Furkan Gözükara SECourses•10/7/24, 2:57 PM

actually it is easy once you understand logic

Furkan Gözükara SECourses•10/7/24, 2:57 PM

i explained in this video in details

Furkan Gözükara SECourses•10/7/24, 2:57 PM

https://youtu.be/nySGu12Y05k

YouTubeSECourses

FLUX LoRA Training Simplified: From Zero to Hero with Kohya SS GUI ...

Ultimate Kohya GUI FLUX LoRA training tutorial. This tutorial is product of non-stop 9 days research and training. I have trained over 73 FLUX LoRA models and analyzed all to prepare this tutorial video. The research still going on and hopefully the results will be significantly improved and latest configs and findings will be shared. Please wat...

Furkan Gözükara SECourses•10/7/24, 2:57 PM

its worst part is reg images usage

B0nd•10/7/24, 2:57 PM

@Dr. Furkan Gözükara have you tried training a Lora on flux1-dev2pro.safetensors instead of the regular Flux dev model? I've read you get better results

FFurkan Gözükara SECourses actually it is easy once you understand logic

tazztone•10/7/24, 2:58 PM

ye but even after i understood it, i prefer to do folder name by hand