After several experiments with the model:

- negative weights, e.g., -0.5: hyper-realistic image, or soft illustration

- positive weights from 0.7 up to 1.0: illustration, pronounced outlines, bold lines