LoRA vs Dreambooth vs Textual In

这几个东西的用法实在是比较像，让人搞不懂具体原因，今天查了这几个功能的原理，下面链接是讲的最简单清晰的。

https://www.油土比.com/watch?v=dVjMiJsuR5o

Textual Inversion - Captures a likeness, or a particular style. Multiple Textual Inversions can be called in your prompt, and they combine (if they're Styles), somewhat.

Hypernetworks - Affect the image as a whole - like overlaying a filter on top of the image. This can be to capture an artist's style, or to apply a cinematic look (noir, etc.). They can technically capture a likeness, but I think they're far more useful to capture a style.

LoRA - LoRA, or "Low-Rank Adaptation for Fast Text-to-Image Diffusion Fine Tuning" is a training method, much like Textual Inversion, but it creates a kind of mini-checkpoint. They're meant to be quicker and more forgiving to train than a TI. It can capture a subject, or a style, or both - and can be called in your prompt just like a Textual Inversion. You can also combine it with textual inversion, like so;

from:
https://civitai.com/questions/125/textual-inversions-hypernetworks-and-loras-oh-my