Cully Hill Boys, the latest project from the AI filmmaking platform Higgsfield, is being presented as a proof point for what generative video tools can now do. It follows three struggling rappers from East London whose lives change after they accidentally steal a crime lord’s money.
The film is also a revealing case study. Its smoother pacing, clearer character beats, and more traditional feel do not come from AI alone. They come from a mix of human script work, human direction, licensed human likenesses, and prompts that openly draw on established film styles.
A showcase for AI movie production
Higgsfield did not position Cully Hill Boys as a conventional theatrical release. Instead, the company uploaded the film online along with many of the prompts used to make it, turning the project into a demonstration of its Cinema Studio suite of AI filmmaking tools.
The film also serves as a large-scale showcase for Higgsfield’s newly added Seedance 2.5 functionality. According to the source article, the project highlights how studios might work with digital creators while using AI models to generate video, speech, images, and edits.
The setup is deliberately familiar: Cal, Horace, and Oli are longtime friends who want to become known musicians but cannot quite make that ambition real. Cal is played through the likeness of Matt Kiatipis, Horace through Mikyle “N3on” Rafiq, and Oli through Israel Adesanya.
None of those performers filmed scenes on a set for the project. Each signed licensing agreements allowing Higgsfield to use their likenesses. That detail matters because it shows one of the central tensions in AI filmmaking: even when the images are generated, the value often still depends on real people, their identities, and the agreements that allow those identities to be used.
The story starts with a human script
Cully Hill Boys is based on a script first written by Tim Planagan before it appeared on the Black List screenplay submission service for unproduced projects. Higgsfield acquired the AI production rights, but Planagan retains the rights needed for a traditional production.
That arrangement makes the film more than a technical demo. It is an AI adaptation of a human-written story. The source article notes that Planagan has said he does not use AI tools in his writing process, and the film’s relative coherence appears closely tied to that foundation.
The early scenes give the characters simple, readable pressures. Oli’s mother wants him to get a real job. Cal’s ex doubts he can be a good father to their daughter. Horace has little happening beyond serving as the group’s designated driver. Those details give the story a basic emotional map before the crime plot takes over.
For AI-generated films, that structure is important. A model can generate shots, motion, voices, and fragments of performance, but a feature-length narrative still needs pacing, character logic, and an understanding of why one scene should follow another. In Cully Hill Boys, those elements appear to come from the script and the directors shaping it.
Longer clips help the film feel more continuous
The project was codirected by Adilet Abish and Aitore Zholdaskali, who also helmed Higgsfield’s Hell Grind feature. According to Higgsfield, they led a team of nine other directors on Cully Hill Boys.
One technical reason the film can feel closer to a traditionally produced feature is Seedance 2.5. The source article says the model can generate much longer clips, about 30 seconds, than other video generation models. That gives scenes more room to breathe before the next cut arrives.
Short AI video clips often reveal their construction. When a film is assembled from very brief generated segments, the joins can be obvious in the image, the sound, or the rhythm of the edit. Cully Hill Boys still has some of that, but many scenes move with a more natural pace.
That improvement does not erase the artificiality. The source article points to incomprehensible text in books and a distracting lack of chemistry between characters as signs that much of the movie was built through prompts and generative models. The film can resemble a conventional feature, but close viewing still exposes the machinery behind it.
The prompts reveal a complicated workflow
Higgsfield published many of the prompts used in the production, and that transparency is one of the most useful parts of the project. Several prompts mention Edgar Wright and Guy Ritchie by name, making clear that the movie’s energy is not emerging from nowhere. Its style leans on recognizable human-made references.
The workflow also involved several tools. Anthropic’s Claude was used to create prompts, which were then sent to Seedance 2.5 for video and speech. ByteDance’s Seedream and Google’s Nano Banana were used for image and video editing.
Higgsfield’s own materials explain that the creative leads split the workflow across models “because the rules of one job poison the other.” In practice, that meant one Claude instance could focus on prompts meant to create workable images while another focused on prompts intended to become video clips.
Most of the prompts were in English, while some were in Simplified Chinese. The level of detail in those prompts shows that AI filmmaking is not simply a matter of typing a rough idea and receiving a finished movie. It involves iteration, model selection, prompt engineering, editing, and direction.
What Cully Hill Boys really demonstrates
Cully Hill Boys shows that AI filmmaking tools are becoming more capable, especially when they can generate longer clips and when a production team can coordinate multiple models. It also shows that the most convincing parts of this kind of project are still anchored in human work.
The film depends on a human-written script, human directors, licensed likenesses, and the borrowed grammar of filmmakers named directly in the prompts. Its existence may point toward new production workflows, but it does not support the idea that AI systems are independently replacing the creative process from end to end.
For Higgsfield, the project is a public demonstration of its platform. For viewers, it is a clearer look at the current state of AI movies: increasingly watchable, technically ambitious, and still strongest when people provide the structure, taste, rights, and references that hold the generated pieces together.