Software Engineering Courses (SECourses)•17mo ago

I'm trying to get SwarmUI working and it's not being very cooperative

FFluffyQuack I'm trying to get SwarmUI working and it's not being very cooperative

Furkan Gözükara SECourses•8/26/24, 1:14 AM

make cfg 1

Furkan Gözükara SECourses•8/26/24, 1:14 AM

for flux

Furkan Gözükara SECourses•8/26/24, 1:15 AM

i made joycaption multi threaded

Furkan Gözükara SECourses•8/26/24, 1:15 AM

it is slower than 1 thread but 8 gpu works great

FluffyQuackOP•8/26/24, 1:15 AM

Ty, now the result looks correct

Furkan Gözükara SECourses•8/26/24, 1:16 AM

FluffyQuackOP•8/26/24, 1:28 AM

Testing another LoRA

0327-realistic_film_still_of_a_live_action_mo-flux1-dev-7410248-1.png

dembel sow•8/26/24, 1:28 AM

Hello all, if train of 8 photos on two different model will the training fail? or will it come with an average of the two faces?
I'm asking because i trying to mix two models as to come as with a new likeness but the training refuse to finish

FluffyQuackOP•8/26/24, 1:28 AM

Was trained on images of this character

Ddembel sow Hello all, if train of 8 photos on two different model will the training fail? o...

Furkan Gözükara SECourses•8/26/24, 1:44 AM

it may get mixed bleed

dembel sow•8/26/24, 1:46 AM

yeah i'm looking for that, i want a new person that is neither

dembel sow•8/26/24, 1:46 AM

as long as each output is the new one

dembel sow•8/26/24, 1:46 AM

i'm happy

dembel sow•8/26/24, 1:47 AM

because i wanna create a virtual influencer that ressemble neither of them two

dembel sow•8/26/24, 1:47 AM

But instead each output give me new one

Furkan Gözükara SECourses•8/26/24, 1:49 AM

try it

dembel sow•8/26/24, 1:59 AM

Is the fune fill tuning course up for sale?

Ddembel sow Is the fune fill tuning course up for sale?

Furkan Gözükara SECourses•8/26/24, 1:59 AM

it is shared here

Furkan Gözükara SECourses•8/26/24, 1:59 AM

https://www.patreon.com/posts/kohya-flux-lora-110293257

Patreon

Kohya FLUX LoRA and Fine Tuning Training Full Tutorial For Local Wi...

Get more from SECourses: Tutorials, Guides, Resources, Training, FLUX, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözü on Patreon

Furkan Gözükara SECourses•8/26/24, 1:59 AM

it keeps gettinbg updated

Furkan Gözükara SECourses•8/26/24, 1:59 AM

only for lora

Furkan Gözükara SECourses•8/26/24, 1:59 AM

fine tuning next hopefully

Furkan Gözükara SECourses•8/26/24, 2:06 AM

yep it is super faithful but it overfit as well. i think fine tuning will be way better than lora

AIGambino•8/26/24, 2:24 AM

Dr how big are your safetensors from flux trains? Mine are 300-400mb instead of 6gb files lol

AAIGambino Dr how big are your safetensors from flux trains? Mine are 300-400mb instead of ...

Furkan Gözükara SECourses•8/26/24, 2:30 AM

you train lora?

AIGambino•8/26/24, 2:30 AM

yeppers

AAIGambino yeppers

Furkan Gözükara SECourses•8/26/24, 2:46 AM

what is lora rank

Furkan Gözükara SECourses•8/26/24, 2:46 AM

and save preicision

AIGambino•8/26/24, 2:49 AM

message.txt2.23KB

AAIGambino Click to see attachment

Furkan Gözükara SECourses•8/26/24, 2:53 AM

AIGambino•8/26/24, 2:54 AM

32 linear under lora?

AAIGambino 32 linear under lora?

Furkan Gözükara SECourses•8/26/24, 2:57 AM

yep

AIGambino•8/26/24, 2:58 AM

What are you reccomending 64?

AAIGambino What are you reccomending 64?

Furkan Gözükara SECourses•8/26/24, 3:03 AM

128

AIGambino•8/26/24, 3:04 AM

https://tenor.com/view/coroca-keanu-reeves-gif-19058801

Tenor

AIGambino•8/26/24, 3:04 AM

I'm training without regularization too, seems like theyre working ok so far... not sure how to add regularization in ai toolkit yet

AAIGambino I'm training without regularization too, seems like theyre working ok so far... ...

Furkan Gözükara SECourses•8/26/24, 3:10 AM

reg images doesnt help i tested

Furkan Gözükara SECourses•8/26/24, 3:10 AM

huge updates

Furkan Gözükara SECourses•8/26/24, 3:10 AM

https://www.patreon.com/posts/110613301

Patreon

JoyCaption : Amazing Image Captioner - Batch Processing Included - ...

Get more from SECourses: Tutorials, Guides, Resources, Training, FLUX, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözü on Patreon

AIGambino•8/26/24, 3:10 AM

My man! ty sir

Furkan Gözükara SECourses•8/26/24, 3:11 AM

GomezBro•8/26/24, 3:27 AM

@Dr. Furkan Gözükara question bro, making a lora for a person is one thing. what do you recommend for settings and image amounts if i want to make a style lora? For instance, there is an NFT I want to make a style lora of. How many images should I do? 100? 500? How many steps? To do a Lora with 32 images alone at 1024x1024 took 14 hrs and 10 mins on FLUX. I did a lora once on sdxl with 3000 images and that only took like 9hrs. 512x512 of course. So I just want to know why it takes so much longer on FLUX and the amount of images I should use?