Kind 68001. Currently building beyond the not yet merged nip.i use j tag "text-to-image" and i tag "prompt", "text"

There will be easy ways to do this in clients etc or specific apps, but basically you need to send a 68001 event with specific tags, wait for a DVM to answer your request, pay a ln invoice of one DVM you choose and wait a couple of seconds.

It's a NIP currently being developed. It's a way for apps to use AI tasks without actually implementing or running them. You literally send to nostr what you want, and data vending machines might answer to do the task for you if they have the capabilities or capacities for it. They then respond with an event containing the result. Here's an example how this is employed with Highlighter for Speech-to-text transcriptions. I'm currently exploring further use cases like text-to-image or image-to-image.

https://www.youtube.com/watch?v=OJx6ExVTS7c