Additional details on LoRa and Embedding (SD)

A LoRA (Low-Rank Adaptation) is a 144MB file. 

• They are quick and easy to train, flexible, and produce good results, which has made them very popular. 
• They tend to memorize content (like tattoos and poses) rather than generalizing content. 

A TI (Textual Inversion) embedding is a 10-40KB+ 

TI can be applied to any model that uses the same base model, which is typically the base stable diffusion model. 

It cannot learn new content, but rather works on keywords trigger behind the scenes that push the model into creating certain output.