The model I think is fairly self explanatory.

There's a dataset bias towards princess peach. Other generations of other characters may be a bit erratic in applying the style at a strength of 1, seems more reliable at a strength of 1.3. Previews made with autismmix confetti.