Software Engineering Courses (SECourses)•3y ago

kohya or dreambooth?

Pplayereckksdee kohya or dreambooth?

Furkan Gözükara SECourses•6/6/23, 4:50 PM

dreambooth of web ui

Furkan Gözükara SECourses•6/6/23, 4:50 PM

hopefully will make a video

Furkan Gözükara SECourses•6/6/23, 4:50 PM

i am processing school lecture

FFurkan Gözükara SECourses this realism amazing

felixsanz•6/6/23, 4:52 PM

what about flexibility?

felixsanz•6/6/23, 4:53 PM

can you share here the first 5 results of "ohwx man wearing a police costume, looking at camera" ?

felixsanz•6/6/23, 4:53 PM

not cherry-picking

Ffelixsanz what about flexibility?

Furkan Gözükara SECourses•6/6/23, 4:54 PM

flexbility bad

felixsanz•6/6/23, 4:54 PM

haha

Furkan Gözükara SECourses•6/6/23, 4:55 PM

this is for realism

Furkan Gözükara SECourses•6/6/23, 4:55 PM

i think it is bad

Furkan Gözükara SECourses•6/6/23, 4:55 PM

because of the resolution

Furkan Gözükara SECourses•6/6/23, 4:55 PM

because underlying model dont have info

Furkan Gözükara SECourses•6/6/23, 4:55 PM

in that resoluion

felixsanz•6/6/23, 4:55 PM

flexibility is important

felixsanz•6/6/23, 4:55 PM

you can't prompt at all if no flexibility

Furkan Gözükara SECourses•6/6/23, 4:55 PM

diferent purposes

felixsanz•6/6/23, 4:55 PM

yup

Furkan Gözükara SECourses•6/6/23, 4:55 PM

if you need shot for linked in

Furkan Gözükara SECourses•6/6/23, 4:55 PM

then this is it

felixsanz•6/6/23, 4:56 PM

depends

felixsanz•6/6/23, 4:56 PM

if you want a front facing picture looking at camera

felixsanz•6/6/23, 4:56 PM

how do you generate it?

felixsanz•6/6/23, 4:56 PM

haha

Furkan Gözükara SECourses•6/6/23, 4:57 PM

Furkan Gözükara SECourses•6/6/23, 4:57 PM

making some shots right now

Furkan Gözükara SECourses•6/6/23, 4:58 PM

will sort with AI

felixsanz•6/6/23, 4:58 PM

some pictures look like your face is photoshopped inside the suit

felixsanz•6/6/23, 4:58 PM

but yeah, good quality

Le•6/6/23, 5:02 PM

I noticed you got full body results from just using your face to train the LoRA. Do I need full body pictures to have a LoRA that respects prompt?

Original message was deleted

ashley•6/6/23, 5:02 PM

Don't you have to adjust the width and height from 512x512 for this?

Ffelixsanz some pictures look like your face is photoshopped inside the suit

Furkan Gözükara SECourses•6/6/23, 5:02 PM

i will use AI to sort

Aashley Don't you have to adjust the width and height from 512x512 for this?

Le•6/6/23, 5:03 PM

it is 512x513 for a missclick. But resolution doesn't impact anything. As you can see, I have full body pictures using the same resolution. If I use any textual inversion, I also get good results.

Le•6/6/23, 5:03 PM

No results on my LoRA

Original message was deleted

Furkan Gözükara SECourses•6/6/23, 5:03 PM

for distant shots you need either higher resolution

Le•6/6/23, 5:03 PM

I got TONS of face pictures using (((full body:1.3))) on the prompt

Furkan Gözükara SECourses•6/6/23, 5:03 PM

or fixing with inpaint

Furkan Gözükara SECourses•6/6/23, 5:03 PM

did you have distant shots in your training data

FFurkan Gözükara SECourses or fixing with inpaint

Le•6/6/23, 5:03 PM

I tried that too. I get weird results.

Furkan Gözükara SECourses•6/6/23, 5:05 PM

tried inpaint?

Le•6/6/23, 5:05 PM

this was inpaint

Le•6/6/23, 5:05 PM

none was blonde, none looked like me

felixsanz•6/6/23, 5:05 PM

but inpaint using "only masked"

Furkan Gözükara SECourses•6/6/23, 5:05 PM

latest version of kohya didnt learned

Furkan Gözükara SECourses•6/6/23, 5:06 PM

maybe you used that version

FFurkan Gözükara SECourses latest version of kohya didnt learned

Le•6/6/23, 5:07 PM

what should I do then? there is no way to use Kohya with a downgraded version, right?

LLe what should I do then? there is no way to use Kohya with a downgraded version, r...

Furkan Gözükara SECourses•6/6/23, 5:08 PM

there is a way certainly

Furkan Gözükara SECourses•6/6/23, 5:08 PM

https://github.com/FurkanGozukara/Stable-Diffusion/blob/main/Tutorials/Generate-Studio-Quality-Realistic-Photos-By-Kohya-LoRA-Stable-Diffusion-Training-Full-Tutorial.md

GitHub

Stable-Diffusion/Generate-Studio-Quality-Realistic-Photos-By-Kohya-...

Best Stable Diffusion and AI Tutorials, Guides, News, Tips and Tricks - Stable-Diffusion/Generate-Studio-Quality-Realistic-Photos-By-Kohya-LoRA-Stable-Diffusion-Training-Full-Tutorial.md at main · ...

Furkan Gözükara SECourses•6/6/23, 5:08 PM

do a new git clone

Furkan Gözükara SECourses•6/6/23, 5:08 PM

then do

Furkan Gözükara SECourses•6/6/23, 5:08 PM

git checkout 63657088f4c35a376dd8a936f53e9b9a3b4b1168

kohya or dreambooth?

Similar Threads