Skip to content

Media · Generative Film

Concept specification

Unbound

Any book becomes a film while you read it — cast yourself, step in, and say what happens next

A living room at night: four friends, the Reader on the table, Moby-Dick on the wall. The Reader reads the book, scans the friends and casts them in the whaleboat, Starbuck speaking in Sam's voice. Then each mode, named as it appears: sound packs, style packs, transitions, a branching book, Sam stepping in to pry the doubloon off the mast, Harker set loose by a spoken line, a game played from four places, a pause while the cat takes Priya's seat — and all of it at once.

What this is

A colour reader that turns any book into a movie: every scene on screen, the characters' lips moving to their lines, movie sound and, if you like, a score. Before anything plays it reads the whole book — every character, place, prop and spoken line — and renders reference plates of each, so the film stays the same film from shot to shot. You and your friends can be scanned into likeness profiles and cast as any character, in your own faces and your own voices. Three agents work on every shot while the film plays and the rest is made ahead of it. And the page doesn't have to be the end: step into the story as a character, say anything, and it keeps writing and filming — books, branching books and live tabletop games alike.

The technical shape

Reader
a colour tablet-class device · plays on its own screen or casts to a TV or projector · every part of the film monitored and controlled from it
Any book
published books, anything you or a friend writes, the transcripts of your own games
Cast
likeness profiles each person records and owns — face, build, voice, how they stand, walk and sit
Voices
a narrator you select · characters voiced as-is, by the narrator modulated to fit, or in their own voices
Look & sound
a base look set before it's made · style packs that change every frame at once · transition, effects, ambience and score packs
Free choice
step into any book, say anything, rejoin the book when you like · branching books with far more ways on than the page had
Games
tabletop role-playing with a human or AI game master · dice, map and pieces rendered for every player, anywhere
Stage
concept specification · pipeline, lookahead and cache sizing, casting, free-choice engine and response ladder

The deep end · full technical outline

Outline v2 · expanding

Read first, then film

There is no instant start. When a book is chosen the system reads all of it and builds registries: every character with their aliases and how the book describes them, every location with its layout, era and light, every prop a character handles. It attributes every spoken line to its speaker, nominates a look from the book and a web lookup of study guides, and renders reference plates — an establishing shot of each place in each of its lights, each character turned all the way round. Every shot after that is built against those plates, which is what keeps a face, a room and a time of day the same from scene to scene.

Three agents on every shot

A shot is one segment of narration. An Author reads it and writes down what happens: who is there, what they do, who speaks, what is heard. A Director keeps continuity — the location, light and look carry over unless the text changes them — and chooses the framing, the movement, the transition and the score cue; a continuing shot starts from the last frame of the one before. A Reviewer checks the request before it is sent and the clip when it comes back, for faces, hands and mouths, and retries a failed clip at most twice so one stubborn shot can't drain the buffer.

Playing while it's made

One finished clip takes 40–120 seconds to generate, so the film plays while the rest is being made ahead of it. How far ahead is measured, not guessed: the time to deliver a clip end to end and the rate of finished film per second of real time are re-measured on every clip, and the lookahead and the cache grow on a slow or jittery network and shrink on a fast one. If production can't keep up, the Reader says so before play and offers the quality that would.

Cast yourself

Each person records a short video on the Reader — face from several angles, body, a spoken phrase — and gets a likeness profile they own; they decide who may cast them and whether their voice may be used. Each profile also carries how that person stands, walks and sits, and real motion-captured movement is retargeted to it, so a cast character moves like the friend who plays them. Wherever the character appears — by name, by alias or as 'he' or 'she' — their description is replaced by the profile.

Packs, like a video editor

A style is applied to every frame of the finished film, so changing it is instant, even on a film already made. Transitions — flashbacks, dreams, time jumps, a letter on screen — come from packs the Director picks from. Sound does too: an explosion has one right sound, and it is pulled for the frame where the blast happens; ambience is laid in layers that carry straight through a cut; a score pack is placed cue by cue, or there is no score at all.

Free choice

The book or the adventure fixes the world — its places, people and history. What you do inside it is open. Step into any book as a character and say what they do; a story engine that already knows the whole book writes what happens next in the book's voice, and the same pipeline films it. Consequences stay, characters remember, and the story is steered toward its goals through the world itself, never by refusing. Rejoin the book whenever you like; your version is kept beside it.

Play from anywhere

A live tabletop game is a book written as it is played. The game master — a person, or the story engine itself — prepares the adventure, whose set pieces are filmed ahead. Every player joins from any device and sees the same moment on a shared clock; the dice, the map and the pieces are rendered by the system, so a roll on one screen lands on all of them. Each session becomes a film afterwards, and a campaign plays as a series.

What is still open

This is a specification, not a built device. Generation speed is the hard limit: a full shot takes far longer than a player will wait, so live play climbs a ladder — narration within about 2 seconds, an illustrated still within 5, a moving loop within 15–30, then the film — and only gets fully cinematic as providers get faster. Cost scales with every second rendered, including speculative openings never watched. And what a provider will render is that provider's policy; the system works with all of them and moves a refused clip to the next.

Related in the lab

Built at this bench

The machine, the instrument, the film that explains it — all designed and built in-house. The same bench works for hire.

Services →