Software Engineering Courses (SECourses)•3y ago

Trained with 20 images and no filewords on Liberty model. very good results but not very flexible. g

Trained with 20 images and no filewords on Liberty model. very good results but not very flexible. going to work with filewords moving forward.

TTrimac okay then i have to watch 30mins video lol. thank you

P1pSl1pOP•3/8/23, 2:31 AM

use the settings from the first part of the video where he sets up test_0. it doesn't take much time..

PP1pSl1p Trained with 20 images and no filewords on Liberty model. very good results but ...

Trimac•3/8/23, 2:36 AM

looks consistent results, have tried some stylish prompts? maybe futuristic and let see how effective it is

Trimac•3/8/23, 2:37 AM

The question is what based ckpt will be use if we want stylish result

Trimac•3/8/23, 2:39 AM

Seacourses only used the 1.5 original and that model is old for me and not so impressive in creativity

TTrimac looks consistent results, have tried some stylish prompts? maybe futuristic and ...

P1pSl1pOP•3/8/23, 2:44 AM

The model isn't flexible at all. Most likely due to the lack of filewords

TTrimac Seacourses only used the 1.5 original and that model is old for me and not so im...

P1pSl1pOP•3/8/23, 2:46 AM

yeah, I originaly trained on 1.5 model. it was okay but flexible. but then I started experimenting with other models and got better results but they weren't as flexible.

P1pSl1pOP•3/8/23, 2:48 AM

2k photos should give solid results I imaging. lol

P1pSl1pOP•3/8/23, 2:48 AM

did you use filewords?

Steel•3/8/23, 2:48 AM

well if you watch videos they all talk about using few photos like 20 good once

Steel•3/8/23, 2:48 AM

yes. I fileword the clothing

Steel•3/8/23, 2:48 AM

and haircolor

Steel•3/8/23, 2:49 AM

this is also ai

P1pSl1pOP•3/8/23, 2:49 AM

that's my next step. up the dataset count and filewords meticulously

Steel•3/8/23, 2:50 AM

and this is a real photo of her

P1pSl1pOP•3/8/23, 2:50 AM

nice results

Steel•3/8/23, 2:51 AM

very.. I tell you if you use dreambooth and make normal models you are wasting time

Steel•3/8/23, 2:51 AM

kohya lora, only way to go

P1pSl1pOP•3/8/23, 2:51 AM

guess I have a mission for today

P1pSl1pOP•3/8/23, 2:51 AM

got any good resources to get started?

Steel•3/8/23, 2:51 AM

what do you mean by resources?

P1pSl1pOP•3/8/23, 2:52 AM

links to tutorials... etc.

Steel•3/8/23, 2:52 AM

oh and train with realistic vision

Steel•3/8/23, 2:52 AM

no, cause all tutorials do it wrong

P1pSl1pOP•3/8/23, 2:52 AM

lol

Steel•3/8/23, 2:52 AM

I tried folloing tutorials which gave me crap

Steel•3/8/23, 2:53 AM

dont listen, they only focus on either stylized photos of their own faces

Steel•3/8/23, 2:54 AM

512,512 is NOT a a good idea

Steel•3/8/23, 2:56 AM

just do one thing.. when training put the unet at 0.0001 or 0.00008

Steel•3/8/23, 2:56 AM

somewhere in between

Steel•3/8/23, 2:57 AM

Steel•3/8/23, 2:57 AM

my settings

Steel•3/8/23, 2:58 AM

also dont use Regularisation photos, they messed up my faces badly

SSteel Click to see attachment

Trimac•3/8/23, 3:02 AM

Lora in khoya seemed not good in training with faces a all i tried multiple times yesterday it seems dreambooth is working well from the video, im testing it now ill be back with my results

Trimac•3/8/23, 3:02 AM

i always have this message: Please configure some concepts

Steel•3/8/23, 3:03 AM

can you share results

Steel•3/8/23, 3:03 AM

comparisson

Steel•3/8/23, 3:04 AM

but key is to have high res photos. if you crop them to 512 you wont have good results

Trimac•3/8/23, 3:09 AM

Anyone solved this problem? Please configure some concepts

P1pSl1pOP•3/8/23, 3:13 AM

I don't even know where to install khoya. lol

Steel•3/8/23, 3:17 AM

https://www.youtube.com/watch?v=9MT1n97ITaE

P1pSl1pOP•3/8/23, 3:17 AM

cheers

dave5555•3/8/23, 3:22 AM

anyone know what "lora extended" option means and does?

SSteel https://www.youtube.com/watch?v=9MT1n97ITaE

Trimac•3/8/23, 3:35 AM

i think i wasted my time in this video

Steel•3/8/23, 3:36 AM

what do you mean

TTrimac Anyone solved this problem? Please configure some concepts

dave5555•3/8/23, 3:44 AM

it does that, just press train again and it will go

Melz•3/8/23, 4:03 AM

im new to making lora thing, and when i create lora on kohya, from 20 img and 100 img source, my file size only 9mb, so what happend with that? i see ppl make lora above 100mb

Melz•3/8/23, 4:06 AM

i mean i try using 20 img as source first and 100 img latter, but size still 9mb

Virus•3/8/23, 4:07 AM

Do you guys know any information on how to utilize ControlNet to indicate the location of eyes and mouth in a sketch when working with Stable Diffusion?
We have the openpose, but as far I know, it works only in humanoid images.

MMelz im new to making lora thing, and when i create lora on kohya, from 20 img and 10...

Steel•3/8/23, 4:27 AM

network rank decides the size

Trained with 20 images and no filewords on Liberty model. very good results but not very flexible. g

Similar Threads

Similar Threads

Similar Threads