Higgsfield has released the official trailer for The Cully Hill Boys, the 110-minute AI feature film produced for USD 2m in four weeks using ByteDance’s Seedance 2.5 model inside Higgsfield Cinema Studio.
The trailer shows the same voice-and-likeness pipeline the company published as a study-only production build on August 7, 2026, and it documents how a feature-length AI film is generated from text prompts, reference assets, and two-minute voice recordings.
The build is the first time a studio has exposed the AI production pipeline behind a licensed feature film. The trailer is the most visible piece of that build: the first public look at what the voice cloning architecture actually produces on screen.
What the trailer shows
The trailer is a 110-minute feature generated as 137 scenes in 4K, every shot and every line of speech produced by AI from written prompts and reference assets. The film is an action comedy about three East London rappers whose attempt to shoot a music video ends in a drug war.
The four lead likenesses are licensed AI performances of two-time UFC middleweight champion Israel Adesanya, MMA veteran Quinton “Rampage” Jackson, livestream creator N3on, and streetball athlete Matt “MK” Kiatipis.
The headline technical number is the input-to-output ratio: each licensed performer supplied roughly two minutes of voice recording and a set of high-resolution photographs. From that input the AI generated 110 minutes of speech, action, and on-screen performance.
How voice cloning scaled a feature-length performance
The voice pipeline runs through two Higgsfield products. The voice cloning product lets a creator “Clone, Read, Dub & Export Any Voice” from a short reference recording; the Soul Cast feature inside Cinema Studio 2.5 then lets the production team call saved or newly generated AI actors as if they were voice talent.
For The Cully Hill Boys, that meant a 28-person team could produce the four lead performances from a single two-minute voice session per actor, plus photo references. Source assets are permanently deleted within 30 days of wrap.
N3on told Variety in a video interview after the premiere that the project “needed only a two-minute recording and some photos” to bring his likeness to life, and that the compensation was “a pretty good check with very little work.”
Alex Mashrabov, CEO and Co-Founder of Higgsfield, framed the deal in the premiere statement: “Building brand equity takes years of hard work for talent, and we are providing creators and studios a powerful new extension of it. By removing physical constraints and reducing production costs, we’re empowering filmmakers, creators, and studios to unlock a whole new talent pool and tell stories that were previously out of reach.”
Seedance, text-and-reference prompts, and the 137-scene pipeline
The visual pipeline ran on ByteDance’s Seedance 2.5 model, hosted inside Higgsfield Cinema Studio. The production did not use an image-to-video workflow; every shot was generated from written prompts and reference assets, with a structured prompt system called CINEDANCE sitting alongside the project on the production canvas.
A 137-entry production log sits next to the project under the HOL-RO-1.0 study-only license. The log documents successive revisions, version history, and internal assessments across the 137 scenes and roughly 1,000 production-ready assets; the license permits reading and analysis but prohibits copying, generative reuse, derivative works, and commercial use.
About 40% of the 28-person team had no prior AI experience, with Higgsfield Academy handling their onboarding.
What the build means for dubbing and voice cloning
The same architecture produced Kok Boru, a 15-minute Central Asian folktale short with native Kazakh voice performances that Higgsfield released at the SIGGRAPH conference in July 2026. A 10-person team completed it in two weeks.
The trailer is the proof point: two-minute voice recordings can be the input of a feature-length AI film; a feature-length AI film can be the input of a dubbed localization pipeline. The dubbing studio is, in effect, becoming a single recorded voice session, and that is the architecture the trailer documents.
