If I'm reading the terms correctly they are CC0 1.0 Universal Public Domain Dedication by default:

https://stability.ai/stablediffusion-terms-of-service#:~:tex...

Obviously this could change in the future.

Side question (let's set aside for a moment the issues about the underlying data set and photoshop editing), is the generation deterministic?

I mean, given a set of prompts (text) to a specific engine, will the same set of prompts on the same engine after - say - one week produce a same image?

Or in the meantime changes on the data set or whatever will generate a different image?

Yes it is deterministic, and it's beautiful, now works of art will be forever recoverable by prompt/seed/inputs.

Even if ipfs and cloudflare loose your jpg, as long as civ is humming, and SD is runnable somewhere, your art will live. Forever

Using the same seed value and sampling method with the same every other value... Yes. It's the same. Time changes nothing.

So - in theory - one could copyright the exact prompt (and seed)?

It wouldn't be much different from copyrighting a short poem or the lyrics of a song.

Wouldn't this case apply here too, making it ineligible for copyright protection? https://www.smithsonianmag.com/smart-news/us-copyright-offic...

Seems very weird to me. Where's the difference between me using a Photoshop plugin that generates some parts of the image and me using Stable Diffusion that generates some parts of the image? How do you know how much of the image was generated and how much was created by my own hand in Photoshop? What if I have Stable Diffusion as Photoshop plugin?

Note: I know there are hot debates about copyright of underlying data training data set. I’m asking about what Stable Diffusion distributors allow/don’t allow.

Google "transformative use."

I mean what do the licenses of this software allow.