At first glance, Cully Hill Boys appears to prove that a film can be constructed without a set and without actors performing before a camera. Its leads authorized the use of their likenesses but did not act out the scenes. The images were produced through Higgsfield and Seedance 2.5. Yet the element that makes the film more coherent than many generative works does not come from the model: it is a screenplay Tim Planagan wrote before the project became an AI film.
Higgsfield acquired the rights to produce a generative adaptation of the text, while the writer retained the possibility of a traditional production. That distinction matters. The screenplay did not enter a system as anonymous raw material; it entered the project through a specific agreement. The personalities lending face and identity to the characters also signed licenses for the use of their likenesses.
The film follows three young Londoners dreaming of music stardom who become entangled in crime. Comedy, action and fantasies of celebrity are connected through rapid editing. The Verge found synthetic defects and uncertain chemistry between the characters, but also a more legible progression than in many AI features. The reason appears to be the directing team’s close adherence to beats already built into the screenplay.
The case separates two operations that are often confused: generating a scene and constructing a story. A video model can produce a convincing chase, argument or party. That does not mean it knows which event should precede another, what information must be withheld, when a character should change or why a consequence must return forty minutes later. Visual continuity is a technical problem; causality is a narrative problem.
Seedance 2.5 promises longer clips, synchronized audio, numerous references and stronger consistency across characters, wardrobe and locations. Those advances reduce fractures between segments. But thirty coherent seconds do not automatically become ninety meaningful minutes. As the model extends the duration of one output, the need for a structure that orders scenes beyond each generation becomes more visible.
The screenplay performs that function. It establishes goals, obstacles, relationships, reversals and transformation. It gives direction, editing, sound and generation a shared map for testing whether every part truly serves the film. Without that architecture, production risks accumulating spectacular moments that change nothing: images capable of surprising in isolation but without memory when placed in sequence.
In a generative workflow, the text may become even more operational. Every scene must be translated into environments, characters, actions, light, camera and sound. Yet a good screenplay is not a collection of detailed prompts. A prompt describes what should appear in an output; a screenplay defines why a scene exists, what changes and what tension it delivers to the next scene. Confusing the two replaces dramaturgy with a visual inventory.
Cully Hill Boys also introduces a new kind of adaptation. The production did not acquire only a story to realize by different means; it acquired specific rights for an AI version while preserving the possibility of a traditional film. Future contracts may distinguish live action, animation, generative production, likeness use, synthetic localization and interactive versions. Production method itself becomes an explicit part of the rights granted.
That separation protects the writer but raises questions of value. If one screenplay can feed many rapidly produced versions, who controls their number, duration and territories? A generative work can be re-edited, translated or personalized without another shoot. A contract must therefore describe not only the first film, but allowed transformations, reusable materials and the author’s participation in later versions.
The director does not disappear between text and model. The role multiplies. The team must choose which screenplay beats to show, build a visual grammar, maintain synthetic performances and reject technically successful outputs that betray a scene’s tone. Generation produces alternatives; direction creates continuity through many decisions that remain invisible in the result.
Editing also becomes structural. Clips originate as relatively autonomous units with different rhythms, light and intensity. The editor must transform them into a common time, reconstruct eyelines, anticipate sound, remove repetition and give consequences weight. In traditional film, editing discovers the story inside footage; in AI film, it can also identify which images must be regenerated for the story to exist.
Cully Hill Boys also functions as a commercial demonstration of Higgsfield, which complicates its evaluation. It must tell a story while displaying the platform’s capabilities. That dual purpose can encourage more environments, camera movements and spectacular situations than the narrative needs. A strong screenplay then becomes a form of resistance: it reminds the technology that not every possibility belongs in the film.
The lesson is not that only a person can write a good story. AI tools can assist research, analysis, alternatives and revision. The point is that narrative requires intention, memory and responsibility. Someone must decide what the work wants to observe, which experiences it is using and which consequences it assigns to its characters’ actions. A model’s fluency does not replace that authorial position.
As image quality improves, writing stops looking like the least spectacular stage and returns as infrastructure. AI cinema does not need a screenplay less because it can generate anything. It needs one more because it can generate too much. Before the image, before the model and before the prompt remains a question no frame can answer alone: what story are we actually trying to tell?