The author has created a dataset called TFHQ that consists of 186 thousand high resolution face images extracted from movie trailers, aiming to address the lack of diversity in expressions in existing datasets like FFHQ. The process involved filtering out low-quality images, deduplicating similar frames, and associating different images of the same person. The dataset is available for download under the Creative Commons BY-SA license, and the author encourages users to share any cool projects they use it for. The unique approach of using movie trailers for a diverse range of emotional faces sets this dataset apart from others.
https://www.justinpinkney.com/blog/2024/trailer-faces/