Software Engineering Courses (SECourses)•3y ago

tried a new one with ohwx woman as instance prompt and photo of woman as class and giving me a black

tried a new one with ohwx woman as instance prompt and photo of woman as class and giving me a black woman and this girl

sample_120-07140becac8bfdd3313027c71bcf05444ea4bb4e1.png

Zono50OP•3/23/23, 11:24 PM

i'm not using prior preservation with this, no class images, just my subjects images

Ffelixsanz why you don't use "photo of ohwx" as instance prompt?

Furkan Gözükara SECourses•3/23/23, 11:24 PM

you need to have class in prompt instance

ZZono50 tried a new one with ohwx woman as instance prompt and photo of woman as class a...

Furkan Gözükara SECourses•3/23/23, 11:24 PM

i cant tell much maybe your lora config incorrect

Furkan Gözükara SECourses•3/23/23, 11:25 PM

you can train on google colab

Furkan Gözükara SECourses•3/23/23, 11:25 PM

for free dreambooth

Zono50OP•3/23/23, 11:25 PM

Zono50OP•3/23/23, 11:25 PM

does that look right?

FFurkan Gözükara SECourses you need to have class in prompt instance

felixsanz•3/23/23, 11:25 PM

what do you mean?

felixsanz•3/23/23, 11:26 PM

I mean... you use "photo of" only in class prompt: "photo of man"

felixsanz•3/23/23, 11:26 PM

why not add "photo of" also to the instance prompt? why not "photo of ohwx" ? why only "ohwx"

Zono50OP•3/23/23, 11:28 PM

starting to come around at 26 epochs

sample_300-08803a85f117fe0367058e88c36231c90feb0899f.png

ZZono50 Click to see attachment

Furkan Gözükara SECourses•3/23/23, 11:28 PM

you dont provide classification folder you checked it generates those?

Ffelixsanz why not add "photo of" also to the instance prompt? why not "photo of ohwx" ? wh...

Furkan Gözükara SECourses•3/23/23, 11:28 PM

you can add. all those tokens will also get trained with your images

Furkan Gözükara SECourses•3/23/23, 11:28 PM

it is like using image captions

Zono50OP•3/23/23, 11:28 PM

i was using prior preservation but it never seemed to train the model, i went in the dreambooth chat and one of the mods said i didn't need to use class images to train

ZZono50 i was using prior preservation but it never seemed to train the model, i went in...

Furkan Gözükara SECourses•3/23/23, 11:29 PM

i made that test in this video

Furkan Gözükara SECourses•3/23/23, 11:29 PM

https://youtu.be/Tb4IYIYm4os

YouTubeSECourses - Software Engineering Courses

Automatic1111 Stable Diffusion DreamBooth Guide: Optimal Classifica...

Sign up RunPod: https://bit.ly/RunPodIO. Our Discord : https://discord.gg/HbqgGaZVmr. New best training settings for DreamBooth training in Automatic1111 Web UI. If I have been of assistance to you and you would like to show your support for my work, please consider becoming a patron on

https://www.patreon.com/SECourses

Playlist of #StableDif...

Automatic1111 Stable Diffusion DreamBooth Guide: Optimal Classifica...

Zono50OP•3/23/23, 11:29 PM

were you using lora or dreambooth tho? does it make a difference?

Furkan Gözükara SECourses•3/23/23, 11:29 PM

dreambooth

Furkan Gözükara SECourses•3/23/23, 11:29 PM

make difference

Zono50OP•3/23/23, 11:30 PM

i was using images generated by class, but it never seemeed to train the model correctly, their faces were either blurry, or looked nothing liek the subject

Zono50OP•3/23/23, 11:30 PM

i was using lora extended with lion tho, i think it's buggy

FFurkan Gözükara SECourses it is like using image captions

felixsanz•3/23/23, 11:30 PM

i'm using image captions and the face fidelity is perfect. photos looks like real ones, but not much styling

Zono50OP•3/23/23, 11:31 PM

captions like a woman standing in the park with green trees and mountains, like that?

felixsanz•3/23/23, 11:31 PM

yes my captions are like that

felixsanz•3/23/23, 11:31 PM

"photo of xzy wearing...... standing in....."

felixsanz•3/23/23, 11:31 PM

i don't include "woman" in the captions

Ffelixsanz i'm using image captions and the face fidelity is perfect. photos looks like rea...

Furkan Gözükara SECourses•3/23/23, 11:31 PM

the purpose matters

Furkan Gözükara SECourses•3/23/23, 11:32 PM

8 epoch done i am gonna test model on my pc

Furkan Gözükara SECourses•3/23/23, 11:32 PM

while training goes on

Zono50OP•3/23/23, 11:32 PM

do you use BLIP or just do them yourself?

felixsanz•3/23/23, 11:32 PM

btw i hope you can include a "conclussions" at the end of your videos

Furkan Gözükara SECourses•3/23/23, 11:32 PM

i tested captions. it was hard to get subject for me

Zono50OP•3/23/23, 11:32 PM

wish we could just use an alpha mask and get rid of backgrounds completely

felixsanz•3/23/23, 11:32 PM

after 1 hour of data i can't even remember and compare anymore XDD plus, i can't judge if you look similar or not

Zono50OP•3/23/23, 11:33 PM

why even need captions?

felixsanz•3/23/23, 11:33 PM

so a conclussions like "EMA is good; XZY is probably bad" would be awesome

felixsanz•3/23/23, 11:33 PM

but holy s., your videos are pure gold

Zono50OP•3/23/23, 11:33 PM

my worst fears have come, it turned my subject asian. That's why i had asian as negative prompt, her eyes are very asianish

Zono50OP•3/23/23, 11:34 PM

sample_480-1bcfc73f843f9401189d6ab24f1f81658f901728e.png

Zono50OP•3/23/23, 11:34 PM

well not my worst fears, but i suspected it would happen

Furkan Gözükara SECourses•3/23/23, 11:34 PM

in this video i used best settings so you can see them quickly @felixsanz https://youtu.be/m-UVVY_syP0

YouTubeSECourses - Software Engineering Courses

Training Midjourney Level Style And Yourself Into The SD 1.5 Model ...

RunPod: https://bit.ly/RunPodIO. Discord : https://bit.ly/SECoursesDiscord. Get your own #Midjourney level model for free via Style Teaching. If I have been of assistance to you and you would like to show your support for my work, please consider becoming a patron on

https://www.patreon.com/SECourses

Playlist of #StableDiffusion Tutorials, Au...