Software Engineering Courses (SECourses)•3y ago

Thanks, It'll be helpful. I just read your comment on someone's query on the lora video, that it's n

Thanks, It'll be helpful. I just read your comment on someone's query on the lora video, that it's not possible on 4gb Vram, wish i had read earlier. haha, great scratching my head, learned a lot though.

Kkashy Thanks, It'll be helpful. I just read your comment on someone's query on the lor...

Furkan Gözükara SECourses•4/30/23, 10:24 PM

ye 4 gb not possible :/

Isaac Brock•4/30/23, 10:34 PM

what is the significance of the loss value during lora training?

Isaac Brock•4/30/23, 10:34 PM

is a lower value more desirable?

Isaac Brock•4/30/23, 10:34 PM

im finding that i need to redo training 3 or 4x before getting a good result

Isaac Brock•4/30/23, 10:35 PM

like restoring from the previous state, not fresh every time

Isaac Brock•4/30/23, 10:35 PM

so a total of like 15-20k steps

Furkan Gözükara SECourses•4/30/23, 10:45 PM

lower loss is desired but doesnt mean well training aone

Furkan Gözükara SECourses•4/30/23, 10:45 PM

it can memorize and have 0 loss

Furkan Gözükara SECourses•4/30/23, 11:00 PM

lora vs dreambooth pretty close quality

Furkan Gözükara SECourses•4/30/23, 11:00 PM

Furkan Gözükara SECourses•4/30/23, 11:01 PM

but dreambooth has better background flexibility

Furkan Gözükara SECourses•4/30/23, 11:01 PM

i used 256 rank for lora

Vlad•4/30/23, 11:01 PM

Hello
Is this Google colab still working with dreambooth?
https://colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast_stable_diffusion_AUTOMATIC1111.ipynb
After I've installed the dreambooth extension, when I try to restart all I receive this:
ModuleNotFoundError: No module named 'discord_webhook'

Google Colaboratory

VVlad Hello Is this Google colab still working with dreambooth? https://colab.researc...

Furkan Gözükara SECourses•4/30/23, 11:02 PM

i think google colab banned it

Vlad•4/30/23, 11:03 PM

so run pod is the solution?

VVlad so run pod is the solution?

Furkan Gözükara SECourses•4/30/23, 11:08 PM

yep

Furkan Gözükara SECourses•4/30/23, 11:08 PM

https://news.ycombinator.com/item?id=35653698

apetresc

Google Colab has started banning Stable Diffusion WebUI users

Hacker News

•

4/21/23, 1:48 PM

Furkan Gözükara SECourses•4/30/23, 11:09 PM

i am training a nice voice to do voice over and see how well it be on next video

Furkan Gözükara SECourses•4/30/23, 11:09 PM

it says that voice itself is not copyrightable :d

FFurkan Gözükara SECourses lora vs dreambooth pretty close quality

hopeyan•4/30/23, 11:23 PM

256 rank？what is that

Isaac Brock•4/30/23, 11:34 PM

what does it mean when your lora is only good at drawing the face up close? Further away, like full body shot you dont really see the face at all, it more matches the base face of the model

Isaac Brock•4/30/23, 11:34 PM

is it because you didnt have enough far away shots or does it just need more training?

Hhopeyan 256 rank？what is that

Isaac Brock•4/30/23, 11:35 PM

it affects how large the file ends up being, i think 256 would be like 300mb

Isaac Brock•4/30/23, 11:35 PM

if i want better results, should i be setting my rank to 256? i just do 128 like your last lora tutorial

Hhopeyan 256 rank？what is that

Furkan Gözükara SECourses•4/30/23, 11:38 PM

lora network size

IIsaac Brock if i want better results, should i be setting my rank to 256? i just do 128 like...

Furkan Gözükara SECourses•4/30/23, 11:38 PM

i think go 256

Isaac Brock•4/30/23, 11:49 PM

will that fix the face not working if its not close up?

Isaac Brock•4/30/23, 11:50 PM

if i have 12-15 pics, is 40 reps, 600 reg images 1 rep, 12 epochs for like 8000k steps enough?

Isaac Brock•4/30/23, 11:51 PM

i see people with much better loras on civitAI that clean to only used 2-3k steps, i am training at like 15k and not getting as good results

Isaac Brock•5/1/23, 12:19 AM

im wondering why my loras dont work very flexibly, is it because i am not labelling the training set?

Isaac Brock•5/1/23, 12:19 AM

i can prompt just for the character and get good close up results, but if i try to prompt a scene the lora basically appears to stop working

Isaac Brock•5/1/23, 12:23 AM

would doubling the learning rate help?

Isaac Brock•5/1/23, 1:27 AM

Do you lower the repeats the more images you train with and retain the same quality?

ZZono50 biggest thing when doing captions is to just use the instance prompt, if you use...

rafraf•5/1/23, 2:14 AM

Hey just one thing. Dr doesn't use captions in the tut, but here you mention having captions, but you basically just use the command to add your instance prompt to the caption right? So you see a benefit to having a caption text files for Subject Lora's? That's the only thing that didn't seem clear to me

Furkan Gözükara SECourses•5/1/23, 2:22 AM

i tested captions today and they didnt improve for me once again

IIsaac Brock i see people with much better loras on civitAI that clean to only used 2-3k step...

Furkan Gözükara SECourses•5/1/23, 2:22 AM

training certain subject much different than those models on civitai

Isaac Brock•5/1/23, 2:53 AM

wow i get much better result after only 3 epochs doubling the learning rates to 0.002 and increasing rank to 256

Isaac Brock•5/1/23, 2:57 AM

0.0002*

Isaac Brock•5/1/23, 3:07 AM

anyone else have a bug where the first time you start to train it wont see the regularization images ?

Isaac Brock•5/1/23, 3:19 AM

oh for some reason the copy folder data didnt do reg image folder

Isaac Brock•5/1/23, 3:36 AM

how did you say I could speed up training by increasing batch size or batches?

Isaac Brock•5/1/23, 3:37 AM

ive been sharing your tutorials in all the SD groups im in

Isaac Brock•5/1/23, 3:42 AM

can you just keep training a lora and it will get better? or can you overbake them like embeddings?

IIsaac Brock anyone else have a bug where the first time you start to train it wont see the r...

Khang•5/1/23, 4:03 AM

I put 1500 images in the regularization folder and SD still try to create more. Later I found out that some of my images are not 512x512 so I resized them and everything work fine. Also don't forget to put "/" before your directory path if you are using Runpod (such as: /workspace/stable-diffusion-webui/test)

Isaac Brock•5/1/23, 4:03 AM

he said you dont have to resize images to train a face lora

Isaac Brock•5/1/23, 4:03 AM

higher resolution better

IIsaac Brock he said you dont have to resize images to train a face lora

Khang•5/1/23, 4:05 AM

Oh is that so? Did he mean training images or regularization images?

IIsaac Brock can you just keep training a lora and it will get better? or can you overbake th...

Khang•5/1/23, 4:08 AM

All methods can be overcooked. And in my experience it usually is around 150-200 epochs

Isaac Brock•5/1/23, 4:09 AM

how can people have so many epochs? i do 4k steps in 3 epochs is this wrong?

Isaac Brock•5/1/23, 4:10 AM

i feel like it shouldnt be taking me this long to train loras for not very good results

Thanks, It'll be helpful. I just read your comment on someone's query on the lora video, that it's n

Similar Threads