Software Engineering Courses (SECourses)•17mo ago

traning people/characters seem to be figured out mostly now, I'm curious, whats next? using the thou

traning people/characters seem to be figured out mostly now, I'm curious, whats next? using the thousands of reg images to further finetune Flux dev?

Ddaroth traning people/characters seem to be figured out mostly now, I'm curious, whats ...

Furkan Gözükara SECourses•8/29/24, 3:20 PM

i will do fine tuning with my reg images

Furkan Gözükara SECourses•8/29/24, 3:20 PM

also a style training

Furkan Gözükara SECourses•8/29/24, 3:20 PM

will share style dataset too

Furkan Gözükara SECourses•8/29/24, 3:20 PM

and also then there is single image training challange :d

Furkan Gözükara SECourses•8/29/24, 3:20 PM

but also there is fine tuning config / workflow

Furkan Gözükara SECourses•8/29/24, 3:20 PM

i bet it will be way better than lora

FFurkan Gözükara SECourses i bet it will be way better than lora

�

𝔸𝖨ΞӾ•8/29/24, 3:24 PM

https://tenor.com/view/hell-yeah-brother-hulk-hogan-nodding-hogan-gif-12315629892838477060

Tenor

FFurkan Gözükara SECourses i will do fine tuning with my reg images

dsienra•8/29/24, 3:24 PM

did you try training with regularization but using no class? all your trainings looked amazing but all man end up looking like you

dsienra•8/29/24, 3:28 PM

may be doing that you maintain resemblance linking you to the token ohwx only and not to the token man

dsienra•8/29/24, 3:36 PM

I will watch your video now!!!!

novsky•8/29/24, 3:45 PM

is Q8 gguf flux model worse than Q5? becuse I have speed boost on Q8 and don't understand why

Ddsienra I will watch your video now!!!!

Furkan Gözükara SECourses•8/29/24, 3:58 PM

awesome

Nnovsky is Q8 gguf flux model worse than Q5? becuse I have speed boost on Q8 and don't u...

Furkan Gözükara SECourses•8/29/24, 3:59 PM

here comparison :

Furkan Gözükara SECourses•8/29/24, 3:59 PM

this is made for LLMs but same logic

Ddsienra did you try training with regularization but using no class? all your trainings ...

Furkan Gözükara SECourses•8/29/24, 4:00 PM

so how am i gonna this ? so it will be ohwx token of me, and man token for reg images right?

Furkan Gözükara SECourses•8/29/24, 4:00 PM

good idea i can test

Furkan Gözükara SECourses•8/29/24, 4:00 PM

but the issue is flux itself has inner captioning

Furkan Gözükara SECourses•8/29/24, 4:00 PM

so it still caption me as man :d

FFurkan Gözükara SECourses here comparison :

novsky•8/29/24, 4:02 PM

if it's only size issue, and speed boosts almost twice (6,3s to 3,7s/it) I will just need bigger ssd)))

novsky•8/29/24, 4:03 PM

thx for explaining

Dreamer@TDFilmStudio•8/29/24, 6:06 PM

@Dr. Furkan Gözükara After seeing your tutorial I am confused. We dont need to caption our training images for flux lora, right? (To achieve better quality)

DDreamer@TDFilmStudio @Dr. Furkan Gözükara After seeing your tutorial I am confused. We dont need to c...

Sharinel•8/29/24, 6:23 PM

I've done a few character loras now and just had to put in the trigger name then ,woman. And I don't think you even needed 'woman'

SSharinel I've done a few character loras now and just had to put in the trigger name then...

Dreamer@TDFilmStudio•8/29/24, 6:24 PM

So no point in using something like joycaption to caption all images meant for training?

Sharinel•8/29/24, 6:25 PM

I did that with the first couple, then just did the trigger word with the next few. No appeciable difference imo

SSharinel I did that with the first couple, then just did the trigger word with the next f...

Dreamer@TDFilmStudio•8/29/24, 6:25 PM

Ok thanks for confirming

DDreamer@TDFilmStudio @Dr. Furkan Gözükara After seeing your tutorial I am confused. We dont need to c...

Furkan Gözükara SECourses•8/29/24, 7:04 PM

for character yes just ohwx man or woman

Furkan Gözükara SECourses•8/29/24, 7:04 PM

for style i will test

DDreamer@TDFilmStudio So no point in using something like joycaption to caption all images meant for t...

Furkan Gözükara SECourses•8/29/24, 7:04 PM

yes not at this moment

Furkan Gözükara SECourses•8/29/24, 7:04 PM

we will see in fine tuning

FFurkan Gözükara SECourses for character yes just ohwx man or woman

Dreamer@TDFilmStudio•8/29/24, 7:07 PM

The vae path, T5-XXL and Clip-L Paths options are not showing in my latest installed Kohya GUI

Dreamer@TDFilmStudio•8/29/24, 7:08 PM

The options are visible in your tutorial of v19

FLUX_LoRA_Training_Simplified__From_Zero_to_Hero_with_Kohya_SS_GUI_8GB_GPU_Windows_Tutorial_Guide.mp4_snapshot_00.32.34.504.png

DDreamer@TDFilmStudio The vae path, T5-XXL and Clip-L Paths options are not showing in my latest insta...

Furkan Gözükara SECourses•8/29/24, 7:08 PM

select model first

Furkan Gözükara SECourses•8/29/24, 7:08 PM

it will become available

Furkan Gözükara SECourses•8/29/24, 7:08 PM

select dev fp16 flux model

FFurkan Gözükara SECourses but the issue is flux itself has inner captioning

dsienra•8/29/24, 7:11 PM

Really???? ohh that's cool!! but a huge problem at the same time, must to be be a way to override this, its ok if the model recognize you as a man but not any man only "ohwx man"

Ddsienra Really???? ohh that's cool!! but a huge problem at the same time, must to be be...

Furkan Gözükara SECourses•8/29/24, 7:12 PM

i dont know any overwrite way atm

Furkan Gözükara SECourses•8/29/24, 7:12 PM

it is how it is designed

Furkan Gözükara SECourses•8/29/24, 7:12 PM

but fine tuning may overcome this issue

dsienra•8/29/24, 7:15 PM

I don't understand how it works , to auto caption in have to analyze the images first caption them and then train, but I don't think is doing that, may be is the way that organize its weights, and it auto detect that the patterns relates to the weights that overlaps with the concept of man, but really idk

Ddsienra I don't understand how it works , to auto caption in have to analyze the images ...

Furkan Gözükara SECourses•8/29/24, 7:17 PM

it uses attention of both image itself (this is where auto captioning happens in neural network) and the prompts we provide

dsienra•8/29/24, 7:19 PM

I edited my message to make it more clear, hehe

dsienra•8/29/24, 7:20 PM

ok I see your point

dsienra•8/29/24, 7:21 PM

this is documented somewhere or everybody is making assumptions?

Ddsienra this is documented somewhere or everybody is making assumptions?

Furkan Gözükara SECourses•8/29/24, 7:23 PM

well i asked this to top developers and 2 said similar stuff