[TAForObvReasons]

Creators with decent-sized portfolios will just train their own models.

> He told me the training process took about 2.5 hours on a GPU at Vast.ai, and cost less than $2.

This is a SaaS waiting to happen.

https://waxy.org/2022/11/invasive-diffusion-how-one-unwillin... This was on the front page last year.

[mattnewton]

That’s for a fine-tune of an existing SD model that has already trained on a mountain of data. Training from scratch requires a mountain of image data and a lot more compute, so you would need a 100% clean base model as well, but then yes it’s totally doable.

[michaelmior]

There are a couple existing pretrained SD models that use all CC0/public domain data. I think at this point they're still significantly lower quality than other popular models, but I'm sure that will improve over time.