Can an LLM make a feature-length movie on its own?

Per LessWrong blogger Josh Snider, he created an entire feature-length movie using large language models, adapting William Hope Hodgson's science fiction novel The House on the Borderland for video. The film was completed—and technically works—but Snider is frank that it falls short of professional standards, though it's impressive for a personal YouTube project. The experiment revealed three major obstacles: LLMs struggle with timing and pacing; video generation remains expensive and error-prone, with recurring issues like faces and character consistency; and coordinating across twelve different models from five companies created integration nightmares. Snider expects these problems—particularly LLM timing sense and multimodal coordination—to be mostly solved by end of twenty twenty-seven, and he's planning to apply the lessons to adapt public-domain science fiction classics into film.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton