
Ever since I stumbled on Stable Diffusion, there is one particular use case that I have been really interested in: generating avatars of myself with fine-tuned models.
The usual method I've seen is for people to spin up Automatic1111 in a google colab notebook and fine-tune a model with an extension.
I find this to be quite cumbersome to be honest, I've had to go through hours of youtube videos to get something that is average. That is when I decided to code the tool I would like to use myself: selfdiffusion.
It's a CLI tool that allows you to easily assemble your own dataset and launch a fine-tuning job on a cloud GPU. The image is a sample I got with the tool (on the left, a training image and on the right, a generated sample).
I want to sell this tool. The way I would make money is:
Selling credits for people to be able to run their training job and run the inference pass.
Offer a platform where custom model creators could themselves make money for every inference that is made from their model.
I've started hacking the MVP for this but I would love to get some feedback on this version of my product idea. I cannot quite separate myself between what I would want and what other people would want.
Would you be enticed by the ability to create avatars of yourself using the CLI ?
It would also mean the world to me if you can "roast" this idea or offer criticism.
Thank you !