Ghosted: Brokers Gone Wild

Higgsfield served as the principal AI cinematic production platform for Ghosted. It was used extensively for the film's primary cinematic scenes, character performances, dialogue sequences, camera movement, environments, and visual effects.

Producing a feature this way was considerably different from generating isolated AI video clips. Characters had to recur throughout the story, locations needed to remain believable, dialogue had to fit the edit, and individual generations had to function as pieces of a much larger narrative.

That required extensive experimentation. AI-generated filmmaking remains unpredictable: a performance can change unexpectedly, a character can behave differently between generations, or an otherwise successful scene can contain a visual anomaly requiring another approach.

Those limitations ultimately became part of the filmmaking process itself. Scenes were rewritten, regenerated, reframed, shortened, extended, or reconstructed during editing. In some instances, the limitations of the technology even contributed to the intentionally chaotic character of the film.

ChatGPT served as an AI assistant director throughout the production. Its role extended beyond generating dialogue or writing isolated prompts: it was used as an ongoing production partner for screenplay development, scene construction, character development, shot planning, dialogue refinement, comedy timing, continuity analysis, trailer development, production troubleshooting, post-production planning, and the orchestration of prompts across different generative systems.

An important part of that role was determining which tool should create which shot. A dialogue-heavy character scene might be designed for Higgsfield, while an establishing shot could be assigned to Veo or Runway. A visual concept could be developed through Midjourney before being translated into motion. A scene that failed technically could be redesigned around editing, reaction shots, inserts, or sound rather than repeatedly attempting the same generation.

In that sense, ChatGPT functioned much like an assistant director sitting beside the filmmaker throughout production — helping organize an unconventional virtual production environment where the “crew” consisted partly of specialized AI systems.

ElevenLabs was used for AI-generated voices and vocal production. Voice generation provided another layer of control independent from image generation: performances could be developed around pacing, personality, timing, and delivery rather than being entirely dependent upon the audio produced during a video generation.

This was especially important for comedy. A pause of half a second, an intentionally flat delivery, or a character sounding completely serious while saying something ridiculous can determine whether a joke works.

Voice and picture therefore became separate creative elements that could be combined during post-production rather than treated as a single immutable AI generation.

Music and sound were treated as major components of the production rather than additions made after the picture was completed. FL Studio was used extensively for the film's musical score, audio production, mastering, mixing, and sound effects.

The soundtrack was developed to move between financial thriller, corporate drama, absurd comedy, tension, and outright chaos. Music frequently provides the seriousness against which the comedy operates: the image and score may suggest that an enormously consequential financial transaction is underway while the dialogue reveals that nobody actually knows who the buyer is.

Sound effects were similarly used to give generated imagery physical weight. Phones, doors, rooms, crowds, impacts, office ambience, environmental sound, transitions, and other details help connect independently generated shots into a coherent cinematic environment.

The final soundtrack required conventional post-production decisions involving dialogue levels, music placement, effects, dynamics, transitions, and mastering. AI could create source material, but the finished soundtrack still had to be mixed as a film.

Google Veo was used primarily for cinematic B-roll and supporting visual material. These shots helped establish locations, create transitions, expand the apparent scale of the production, and provide visual breathing room between dialogue scenes.

B-roll became particularly valuable because a feature film cannot consist entirely of characters speaking directly to one another. City environments, offices, streets, financial imagery, atmospheric shots, and other visual inserts helped create the larger world surrounding the characters.

Veo therefore became something similar to a specialized second-unit camera crew.

Runway was also used for B-roll as well as selected supporting scenes. Using multiple video-generation systems was intentional: no single generative model consistently excels at every type of filmmaking task, and different systems interpret movement, realism, camera direction, people, environments, and visual effects differently.

Instead of forcing the entire film through one technology, Ghosted used the available platforms as different production departments. The question became less “Which AI made the movie?” and more “Which tool is best for this particular shot?”

Midjourney was used for artistic renditions and visual development. Still-image generation provided a way to explore compositions, characters, environments, promotional concepts, and stylized imagery without immediately committing to video generation.

This also reflects a traditional filmmaking practice in a new form. Productions have long used concept artists, storyboards, reference photography, production paintings, and visual-development material before cameras begin rolling. Generative image systems dramatically accelerate that exploratory stage.