Higgsfield's AI feature film leans heavily on human craft—and it shows

Higgsfield, an AI filmmaking platform, recently released Cully Hill Boys, a feature-length comedy that explores how far generative video technology can push toward traditional cinema. The film follows three struggling East London rappers who accidentally steal from a crime lord, blending comedic character dynamics with action-comedy set pieces. Starring AI-generated likenesses of streamer Mikyle "N3on" Rafiq, basketball creator Matt Kiatipis, and UFC fighter Israel Adesanya, the project stands out among recent AI films for its structural coherence and visual polish—but examining how it was made reveals something crucial about what actually drives quality in AI cinema.

The human architecture beneath the tech#

The foundation of Cully Hill Boys' narrative strength lies not in generative models, but in screenwriter Tim Planagan's original script, acquired from the Black List screenplay marketplace. Director duo Adilet Abish and Aitore Zholdaskali, working with a team of ten additional directors, followed Planagan's story beats meticulously throughout production. This disciplined adherence to a human-authored screenplay—structured with proper pacing, character development arcs, and logical dialogue—creates a viewing experience that feels substantially more organic than most AI-generated content. The film's opening minutes efficiently establish three distinct characters and their personal stakes before accelerating into plot momentum, a feat that speaks to experienced storytelling rather than algorithmic assembly.

Technical capabilities and creative work#

Higgsfield's deployment of multiple generative models underscores how complex even AI filmmaking demands specialized human orchestration. Claude, Seedance 2.5, ByteDance's Seedream, and Google's Nano Banana each handled distinct tasks—prompt generation, video synthesis, and image editing—because, according to the company, "the rules of one job poison the other." The ability to generate longer clips (roughly 30 seconds) than many competing systems reduces visible seams between stitched segments. Yet this technical fluency masks a deeper reality: Higgsfield published the prompts used to create individual shots, and comparing them to final scenes reveals substantial human refinement layered atop model outputs. The most polished sequences—particularly the action-comedy moments—explicitly reference directorial signatures, with prompts requesting "GUY RITCHIE COVERAGE" and "DYNAMIC clip-style (Edgar Wright snap.)" These cues suggest the underlying models were trained on those filmmakers' work, raising questions about how existing creative IP anchors even "original" AI production.