It's basically a compromise between and Doggy ;) touching on Spooning and .

Main points for the training data was: Capture different angles (pov, from side) and include different styles to be as style agnostic as possible and not flatten the underlying checkpoint.

Trained in 3:4, 1:1, 4:3.

Starting point:

xxsideways, 1girl, 1boy, lying, on side, solo focus, {, |}