Fine-Tuning Datasets

[EleFreelance]

Hi,

Thank you for the release, this is a great project, congrats!

1、How much training data is currently required for finetuning existing frameworks, especially for the relighting task?
2、In image generation tasks, SDXL can achieve facial identity transfer with just a few dozen images using LoRA training. Can Lumina-Accessory also achieve similar results with only a few dozen facial images?

Thank you very much :)