Higgsfield just released one of the most convincing AI-generated films of the year, and the most interesting thing about it is how much of its quality traces back to actual people. According to The Verge AI, the movie is called Cully Hill Boys, a direct-to-video-style action comedy about three struggling East London rappers who accidentally steal a crime lord’s cash. It looks and moves like something Edgar Wright or Guy Ritchie might make. That’s not a coincidence. Higgsfield published many of the prompts behind the film, and several name Wright and Ritchie directly.
This is significant because it exposes the real supply chain behind “AI filmmaking.” The tools generated the pixels. Humans supplied the taste.
What actually made it work
The Verge AI reports that the film’s strongest asset is its script, written by screenwriter Tim Planagan, who says he doesn’t use AI tools at all. His screenplay came off the Black List, the industry’s home for unproduced work, before Higgsfield acquired the AI production rights. Planagan kept the rights to produce it traditionally.
That pacing shows. The film sets up its three leads, their families, and their problems in the first few minutes with a tonal smoothness you rarely see in AI video. As The Verge AI puts it, the story “was crafted by a human with an understanding of proper pacing.”
The production stack tells the rest of the story:
- Anthropic’s Claude wrote the prompts.
- ByteDance’s Seedance 2.5 generated video and speech, with clips running up to about 30 seconds, far longer than most models.
- ByteDance’s Seedream and Google’s Nano Banana handled image and video editing.
- A team of eleven directors, led by Aitore Zholdaskali and Adilet Abish, followed the script’s beats closely.
The leads, streamer N3on, creator Matt Kiatipis, and UFC fighter Israel Adesanya, never shot a scene. They licensed their likenesses, and Higgsfield built the avatars from photos and audio they provided.
Where the seams still show
The machine parts are still visible. The Verge AI points to gibberish text in books on screen and a distracting lack of chemistry between characters. The team also had to split work across models because, in their own words, “the rules of one job poison the other.” One Claude instance did nothing but write prompts in a specific format. That’s not a polished pipeline. That’s a workaround stitched together by people who understood the limits.
Why it matters now
Cully Hill Boys isn’t really a movie release. It’s a product demo. Higgsfield uploaded it to showcase its Cinema Studio suite and its new Seedance 2.5 feature, and to pitch studios on collaborating with digital creators. What stands out is the honesty of it. By publishing the prompts, Higgsfield accidentally documented how dependent these outputs are on human writing, human direction, and human style references.
For anyone tracking where generative video is headed, that’s the real signal. The technology is closing the gap on look and length. It has not closed the gap on story, chemistry, or judgment. The value is migrating, not disappearing.
Practical takeaways
- If you’re a creator or studio: the leverage is in the script and direction, not the render. A strong screenplay plus tight prompt engineering beats raw model access. Treat AI video as a finishing tool, not an author.
- If you’re a performer: likeness licensing is becoming a real revenue line. Read those agreements carefully, because your face and voice can now “star” in a film you never shot.
- If you’re a writer: your rights are the asset. Planagan handed over AI adaptation rights while keeping traditional production rights. That kind of split is worth structuring deliberately.
- If you’re evaluating the tech: longer clip generation is the quiet breakthrough here. Thirty-second takes cut the visible stitching that gives most AI films away.
The next competitive question isn’t whether AI can generate a watchable feature. It’s who owns the human craft that still makes it good. You can find the full breakdown at the original source.