Authors: Raz Sarusi · Tomer Dariel
10,000 original, synthetic movie records generated with a pretrained instruct LLM
(Qwen/Qwen2.5-1.5B-Instruct) using structured prompting and controlled variation over a
genre × decade × tone grid. Built for the PlotTwist app: given a movie idea, the app
recommends 3 similar loglines (embeddings + FAISS) and generates one new original logline.
v2 enum-validates every categorical field and bans cliché openers… See the full description on the dataset page:
https://huggingface.co/datasets/razsarusi/plottwist-movies.