I haven't had time to open up the laptop to get my prompt but here is the rest.
The model is three models joined together in Stable Diffusion: JuggernautXL_v8, animagineXLV3_v30, and digitalLife_v12
I googled for photos of traditional pirate ships and used that in image to image so Stable Diffusion would have a base to work from. It helps reduce getting random weirdness when you're trying to do something it's not trained for. In this case when I wanted a pirate ship in space it would give me a boat in water or a spaceship, but image to image helped it get closer to what I wanted but as you can see all but one had some sort of land and they all still had water.
In the negative prompt I typed water 20+ times which got rid of the water but still had land. In the top two images I then used inpainting to paint out the land and used different prompts and iterations until I got something I liked. The title image worked out the best but I think I went through 24 version, 8 at a time (the max versions in SD).
To do the inpainting instead of a mouse I use a Kamvas 16 Pro. I spent $450 on it and I also draw and paint with it.
Does that help?


