Works best on 4mp resolution. Captions are not best yet, but working on a v1.5. general style, look and feel can be done with this model, but tends a bit to give men data skin. The dataset was created from 1-3 screencaps per episode of the full series, so around 400 images