Cleaned up dataset, re-captioned all images and greatly decreased training steps.

Previous version was well overcooked but worked great for sdxl turbo

Version 1.1 is more flexible and works better with traditional sdxl models and configurations

Version 1.0 Turbo

Training Specifications:

Image Count: 150

Resolution: 1344,768

Model: SDXL base 1.0

Steps: 8000 (with learning rate decay)

Image generation:

Trigger word: "The Mandalorian"

Model: Turbo DreamShaperXL

Steps: 10

CFG: 1.5

Lora strength: 0.6

Also take note that most of the training images were of the chest and above so this model really excels at portrait shots and close ups of the helmet.

This is also my first Lora so any feedback would be great!