Train Stable Diffusion (textual inversion) with your own images
This repository contains tutorials to train your own Stable Diffusion .ckpt
model using Google Cloud Platform (GCP) and Amazon Web Services (AWS).
For platform specific instructions, see the following:
It's very cheap to train a Stable Diffusion model on GCP or AWS. Prepare to spend $5-10 of your own money to fully set up the training environment and to train a model.
As a comparison, my total budget at GCP is now at $14, although I've been playing with it a lot (including figuring out how to deploy it in the first place).
The spot price of A100 instance at GCP is under $2 per hour, and it takes under half an hour to train 500 steps. Smallest P3 instance is at ~$3/hour, and you should be able to train 500 steps in an hour.
Note that both AWS and GCP have limits on the number of GPU instances you can run if you have a new account. You'll need to follow their instructions to request a limit increase.
On my own testing it took 2 minutes for GCP to increase my limit from 0 to 1, while it took 7 days to do the same for AWS
This project is licensed under MIT license.