Trained on portrait images, merging two different faces vertically.

You can use use a prompt like:

left half: describe the person and scene on the left half

right half: describe the person and scene on the right half

Should work with both natural language and tags.