Higgsfield releases AI film 'Cully Hill Boys' to demonstrate generative video capabilities

AI filmmaking platform Higgsfield has released the full AI-generated film 'Cully Hill Boys' alongside its underlying prompts to demonstrate the capabilities of its Cinema Studio suite and Seedance 2.5 model. The project, based on a script by Tim Planagan, features digital avatars of Mikyle Rafiq, Matt Kiatipis, and Israel Adesanya. The release highlights the significant human input required for prompting and direction, with creative leads utilising multiple AI models to achieve a narrative cohesion that mimics traditional filmmaking styles.

Higgsfield acquired the AI production rights to Planagan’s script, which was initially submitted to the Black List screenplay service, while Planagan retains rights for traditional production. The film was codirected by Adilet Abish and Aitore Zholdaskali, who led a team of nine other directors. The project features digital avatars of streamer Mikyle “N3on” Rafiq, basketball content creator Matt Kiatipis, and UFC fighter Israel Adesanya, who signed licensing agreements for their likenesses.

The production workflow utilised multiple AI models: Anthropic’s Claude for prompt generation, Seedance 2.5 for video and speech, ByteDance’s Seedream for image generation, and Google’s Nano Banana for editing. Prompts explicitly referenced the stylistic influences of directors Edgar Wright and Guy Ritchie, such as “GUY RITCHIE COVERAGE” and “Edgar Wright snap.”

Higgsfield released the film and prompts online as a demonstration of its tools and a marketing effort for Seedance 2.5, rather than for theatrical release. The film also functions as a demonstration of new ways studios can collaborate with digital creators, showcasing the platform's ability to handle complex narrative structures and character dynamics.

While the film achieves a level of visual smoothness and pacing that distinguishes it from many other AI-generated clips, the release underscores the extensive human curation required to guide generative models. The creative team had to split their workflow across different models to ensure compatibility, with specific instances of Claude dedicated to generating prompts for workable images versus video clips.

The transparency regarding the prompts and the reliance on human direction for narrative cohesion suggests that current generative video technology still requires significant human oversight to produce feature-length content. The project serves as a proof of concept for the potential of AI filmmaking tools, normalising the idea of AI-generated features in the market.