[Marko Zegarac]

Hey Andrew,

I don’t want to spam the forum with my questions, so I will put both of my questions here.

1. Textual Inversions vs LoRA, Which one is better? And by ‘better’, I mean produce better results? Assuming that:

• Usage is on Stable Diffusion 1.5 (or its derivatives)
• Computational resources are not a problem for inference or training.
• File size does not matter, only the results.

With all that equal, which of the two is a superior technique and why?

1. Amazon Web Services. Which EC2 instance is the best for running Stable Diffusion remotely? Ideally, something that would be able to run Stable Diffusion 3 (once it comes out). Or, would you avoid EC2 entirely? If so, what service (from AWS ecosystem) would you recommend?

Thank you!

Marko

[Andrew]

Hi!

1. LoRA is better than TI, in the sense that it can change the model more.

2. Typically instances with T4 or V100 GPU. You can stick with 1 GPU for inference. AWS is not the most cost effective. You can consider Google Colab if you are ok with the interactive interface. They offer the same GPUs at a much lower rate.

[Author]

Posts