This contains some brainstorming about the project.
Version 0: Dump all of the dreams into a prompt, and ask a model to generate a new dream.
Version 1: Fine-tune a model on the dreams.
How to set this up? Would the prompt for each dream just be "generate a dream", or should it include a summary?
Version 2: Include the generated images as part of the models output.
The simplest way to do this is to have the prompt as part of the dream output. This limits me to images that were generated purely from a prompt.
Version X: Train using brainwave data taken when I'm sleeping.
Difficulties
It's been very difficult to recall dream details and write them down in a way that adequately describes them. I've read from people working on dream journals that this is a skill that takes time to develop.
The images that I've created don't match what's in my dreams very well. To improve them, I'll need some drawing and digital art skills. I've ordered a cheap drawing tablet to get started on this.