Stable Diffusion how many ways to train a model

Stable Diffusion is an exciting and rapidly evolving field. I'm currently working on several projects involving Stable Diffusion, and one of the main ones is figuring out the best method to train a model to represent a person's face or likeness.

While there are currently five methods available (that I know off), these two are the most effective:

1. Dreambooth
   • Best option if you want to train a few models with similar steps. However, if you add various models and images, the results start to deteriorate, and you may have to compromise.
2. Textual Inversion Embeds
   • Although not as visually stunning as Dreambooth, I've achieved fantastic results with this method. The benefit is that I can use the same embed in multiple models. Keep in mind that you will need to fine-tune the settings for each model so that it resembles the person you are trying to create. However, once you find the settings, they can be reused across multiple embeds.

While the other three methods - Hypernetworks, LoRA, and Aesthetic Gradients - are viable options, they aren't as effective as the top two methods for training Stable Diffusion models to represent a person's face or likeness.

I hope this post has saved you time in your Stable Diffusion model training endeavors. Stay tuned for a more elaborate blog post on the top two methods mentioned above.