Marlena is the latest of my models trained entirely on prompted images, no humans involved.
Trained on 76 prompted images, 15 vectors, learning rate 0.004, 15 steps pr image/embedding and 1650 steps in total. Responds well to all manner of checkpoints.