These are models trained on film stills from some classic movies.

I used a sligthly different approach, but Oldskools great article inspired this.

I used a mix of LLava 13B natural language tagging and Wd1.4 tags for the datasets.

The models are not trained to reproduce individual likenesses but the style of particular movies, however stable diffusion base knows a lot about some of the actors already, so you can just prompt their names.

Charade:

Recommended Aspect Ratio: 1.85:1

Influence 1 or lower

Modern Times:

Recommended Aspect Ratio: 4:3

Influence 1 or lower

Eyes can be pretty messed up with this one, owing to the dataset

Caution: Low res Dataset

Mon Oncle:

Recomended Apect Ratio: 4:3

Influence 1 or lower

Caution: Low res Dataset