[Submitted on 23 Jun 2025]
View HTML PDF (experimental).
Abstract.AI content creation has shown promise in film production. Existing film generation systems are unable to implement cinematic principles, and therefore fail to produce professional-quality films. This is due to the lack of diverse camera language and cinematic tempo. This leads to templated visuals, and unengaging stories. FilMaster is an AI system that uses real-world cinematic principles to create professional-grade films. It produces editable, industry standard outputs. FilMaster is based on two key principles, namely (1) learning cinematography by analyzing real-world film data, and (2) simulating professional post-production workflows that are audience-centric. FilMaster is built on these principles and incorporates two stages. A Reference-Guided Generating Stage that transforms user input into video clips and a Generative post-production stage that orchestrates visual and auditory components for cinematic rhythm. Our generation stage features a Multishot Synergized RA Camera Language Design Module to guide AI in generating professional cameras by retrieving reference clips. Our post-production stage emulates workflows of professionals by designing an Audience Centric Cinematic Rhythm Control Module, including Rough Cut, Fine Cut, and other processes informed by simulated feedback from the audience, to ensure effective integration of audiovisual components for engaging content. The system is powered by generative AI models such as (M)LLMs, and video generation models. FilmEval is a comprehensive benchmark to evaluate AI-generated films. Extensive tests show FilMaster’s superiority in camera language design, cinematic rhythm control and advancing generative artificial intelligence in professional filmmaking.
Submission history
From: Kaiyi Huang [view email]
[v1] Mon, 23 Jun 2025 17:59:16 UTC (21,617 KB)
