Film — 02

Film & Cinematic AI

Work from Fliction — a research practice in synthetic cinema. Films are built frame by frame from story-state graphs; cuts are treated as reframings of attention, not as boundaries. The system prepares coverage. The director chooses the cut.

What it is

A structured adaptation engine. It reads a work, decomposes it along its natural fault lines, plans coverage in professional film grammar, and generates keyframes, motion, dialogue, score, and ambience — with character continuity locked across every scene.

What it isn't

Not a "make me a movie" button. Not a replacement for the director, the editor, or the cinematographer. Speed without structure is negligence — the system slows down wherever craft is at stake.

MFA Thesis · 2026

The Narrative Data Layer — A 472+-Parameter Taxonomy for Cinematic Syntax Translation

The written architecture behind Fliction. Massachusetts College of Art and Design.

Read the thesis (PDF) ↗

Now Playing — Flagship Pilot

L.U.L.I.T.A.

Watch on lulita.net ↗

Starring Luz Ginel · Directed by Allen Redwing · A series in VIII episodes

An operative raised by the state discovers her file was never sealed — it was written.

Anitha is the state's most precise instrument — until a dead terminal in a Yale law library greets her by name. L.U.L.I.T.A. is not just a program. It is her paper trail. And every door she was trained to breach was built to keep her in.

"I've always dreamed of shooting a scene in a real panopticon. When Luz stands in that circle, I envisioned four hundred cameras pointed at her — and my dream became an instant memory feedback loop."
— A. Redwing, Director

Pilot Season — Fall 2026

Case Studies

Adaptation demos — scene reconstruction

I

Dorian

Wilde. A portrait that ages so its subject does not — rendered as continuous coverage of a face refusing time.

II

Joe Turner's Come and Gone

August Wilson. Bertha's boarding-house kitchen and Bynum's binding songs — coverage built to hold Black interiority without cutting away from it.

III

That Will Never Work

Marc Randolph's Netflix origin memoir. Coverage study of four beats — the commute, the founding table, the Scotts Valley HQ, the VSDA trade-show floor. Business memoir treated as cinema: a garage is a location, a whiteboard is a monologue, a trade show is a set piece.

IV

Basquiat: A Quick Killing in Art

A 128-scene adaptation. Four scenes from Act 2 (Emergence), each planned against a documented narrative function from the canonical scene index — pressure, obsession, rivalry, thesis.

Also in the library

Amor e Sombras — Scene 7 · Barganha

Portuguese-language adaptation. A bargain lit like a confession. Full coverage study forthcoming.

Method

Coverage · Continuity · Choice

Every scene is planned as multiple coverage options. Nothing is auto-selected. The human chooses the cut.

A single anchor frame propagates identity and continuity across every coverage variant — character, wardrobe, and set stay locked while the camera language changes.

Casting & Character Identity

CIP · Sticky context · 0.75 pass threshold

Before a single frame is generated, every named character is compiled into a Character Identity Profile (CIP) — face, wardrobe, voice, and style, all locked to a reference set. Casting is not a vibe. It is a contract the pipeline is legally bound to honor at the edge layer.

  • 1 · Identity profile. Face, ethnicity, age range, gender presentation, wardrobe canon, voice signature.
  • 2 · Casting gallery. Curated reference images anchor the CIP. Actor plates, mood plates, wardrobe plates.
  • 3 · Sticky context. Location, wardrobe, and emotion propagate scene → scene at the edge layer. Prevents drift.
  • 4 · CIP enforcement. Every rendered shot is scored. Below 0.75 match, the shot is rejected and re-cast. Every enforcement is logged.
  • 5 · Voice casting. Voice is a first-class identity dimension. ElevenLabs stems bound to CIP, not to scene.

Result: an actor can appear in 400 shots across 8 episodes and still be recognizably the same person — no ethnicity drift, no wardrobe accidents, no fourth-episode face.

Roster · Pilot slate

Range

One system · Many cinematic dialects

The same pipeline speaks techno-noir, gothic portraiture, August Wilson interiors, colonial epic, Portuguese romance, sci-fi, fantasy, and Victorian romance — the dialect shifts, the grammar does not.

Frames from the pipeline

Not a highlight reel · A working audit trail

A working record of what the editor was given to choose from — coverage retries, character continuity locks, and establishing candidates paired side by side. The machine prepares options; the human commits.

Coverage retries

Prepare, don't decide.

Every angle is generated as a pair — original and retry — so the editor cuts from options, not from a single take.

Character continuity

CIP enforcement across takes.

Character Identity Profiles lock face, wardrobe, and props across generations — same person, new light, held above the CIP threshold.

Establishing + scene compare

Two candidates, one choice.

Establishing wides are always generated as alternates. Scene-level compares are the continuity evidence the editor uses to lock a take.

Position

"A cut is not a semantic boundary. It is a reframing of attention."
"Machines prepare. Humans decide. Everything is recorded."
"The system speaks film language. It does not author films."

L.U.L.I.T.A. © 2026 Fliction Originals. Case-study stills are directorial references from Fliction's research pipeline — not final frames. Some works are unreleased.