sd_dreambooth_extension VS dreambooth-training-guide

I am trying to use StableDiffusion via AUTOMATIC1111 with the Dreambooth extension

When I first attempted SD training, I was very frustrated. It wasn't until I found this obscure forum thread on Github that I actually started producing great results with Dreambooth. Because I have such satisfactory results, I'm very reluctant to beat my brains against LoRa and its related training techniques. I gave up trying to train TI embeddings a long time ago. And I never figured out how to train or how to use hypernetworks. I've only been able to get good results with Dreambooth directly because of that thread I linked above. I make LoRas by extracting them from Dreambooth-trained checkpoints. And I have no idea if I'm doing the extractions the right way or not.

Getting the same with automatic1111 and sd_dreambooth extension. Check out more here in the issues log: https://github.com/d8ahazard/sd_dreambooth_extension/issues/1266

It's several moths old and many things have changed. But the spreadsheet available through this thread on Github has been indispensable for me when I train Dreambooth models. I'm astounded no one talks about it. I bring it up all the time. The research presented there should be continued. I'd love to see similar research done for SD v2.1.

Training rate is paramount. Read this Github thread.

https://github.com/d8ahazard/sd_dreambooth_extension/discussions/547/ (in depth training principles understanding)

sd_dreambooth_extension https://github.com/d8ahazard/sd_dreambooth_extension.git main 926ae204 Fri Mar 31 15:12:45 2023 unknown

also the Dreambooth extension supports API (https://github.com/d8ahazard/sd_dreambooth_extension/blob/main/scripts/api.py) so i'm not sure where do you get those news :/

I dont have the time right now, but the rule of thumb for me was 80 unet learning steps for 1 image. Atleast 40 regularization images. Read more about regularization images here: https://github.com/nitrosocke/dreambooth-training-guide

This tutorial for dreambooth training has advice with regard to backgrounds which is probably also applicable to LORA. It recommends including images with solid, non-transparent backgrounds but not using them exclusively. Images that focus on the torso and face are probably most important unless your subject has very distinctive legs and feet. Removing other subjects is a must if you're training for a specific subject.

I found this, I'm going to go through this guide. Seems like I am using far too many images. https://github.com/nitrosocke/dreambooth-training-guide

Nitrosocke's guide already tells how much and what kind of images to use.

Dreambooth lets you do it in less than 100 images. https://github.com/nitrosocke/dreambooth-training-guide These folks say it's 5-15 to train on a person but I've not tested myself. https://www.reddit.com/r/StableDiffusion/comments/10tqy88/were\_launching\_a\_lightningfast\_dreambooth\_service/

See eg this tutorial for styles: https://github.com/nitrosocke/dreambooth-training-guide

The first time I tried to Dreambooth a style it went poorly. Then I found Nitrosocke's Dreambooth Training Guide and realized my problems were caused by a poorly redacted dataset.