For several days after leaving Nantucket, nothing above hatches was seen of Captain Ahab. The mates regularly relieved each other at the watches, and for aught that could be seen to the contrary, they seemed to be the only commanders of the ship.
VoiceidoAI StudioAny source,
directed_
like film.
The Voiceido engine reads an entire source before it draws anything — a manuscript, a technical article, a research report, a set of internal documentation. It resolves the entities, locks a canonical look for each one, times the narration, directs every shot against those locked references, and returns a finished video with the lineage behind it.
- Category
- Source-to-video infrastructure
- Inputs
- Manuscripts, articles, reports
- Defensibility
- Persistent entity registry
- Unit economics
- Per-asset cost and lineage
LIVE STORY SYSTEM
SOURCE01ingest(source)02trace.pages()
WORLD01map(world)02resolve(beats)STORY GRAPH
01const graph =02await read(source)03graph.connect()
CAST 0101cast.lock(id)02memory.keep()
CAST 0201identity.ref02continuity: 1SCENE LOGIC
01batch([1, 3])02direct(beat)03camera("9:16")STORY
MEMORY08 / 08 systems
MEMORY08 / 08 systems
MOTION01seedance.run()02motion.track()
SCENE 0401compose(shot)02focus(beat)VOICE CLOCK
01voice = rex()02await measure()03sync(captions)REX · 02:48
VOICE01timeline.fit()02words.sync()
MASTER01render(720)02verify(master)DURABLE RUN
01checkpoint(job)02retry(failed)03resume(refresh)PUBLISH
01export({02 codec: h26403})STORY IS FORMINGAUTO · 03.2S
Technical articleSystems paperFigures become directed shots. Every term is resolved once, then reused.
ManuscriptSleepy HollowEpisode 02 · publishedA finished episode, produced end to end by this engine.Watch on YouTube 
Research reportAnnual reportCharts keep their meaning; narration sets the clock for every cut.Any source, four irreversible steps
PARSING SOURCECanonical identity
BuildWeathered, broad through the shoulders
WardrobeOilskin coat, storm-worn cap
PeriodMid-19th-century whaling
ReferenceApproved once, reused downstream
Identity locked“For several days after leaving Nantucket, nothing above hatches was seen of him.”
- Planned against source units 118–124
- Cast reference attached, not re-imagined
- Narration measured first, picture cut to it
Watch on YouTube Published episodePOV: the schoolmaster arrivesSleepy Hollow · Episode 02
A different source, run through the same four steps. Captions burned, audio mixed, title and description written, lineage recorded for every asset.
H.264 · 720×1280CaptionsPer-asset costResumable
The reel
Eight genres. One engine.
0102030405060708Eight departments, one continuous run.
Every department hands the next one a checked artifact. Nothing downstream is allowed to invent what an earlier department already decided.
- 01IngestReads the source end to end — manuscript, article or report — and keeps every page addressable.page map
- 02UnderstandBuilds the graph: entities, places, systems, chronology, relationships.story graph
- 03CastingFreezes one canonical look per entity and refuses to redraw it later.entity registry
- 04ScenesTurns beats into directed shots with framing, focus and continuity rules.scene packs
- 05SoundRecords narration first so the voice becomes the clock every cut lands on.voice timeline
- 06MotionAnimates each shot against the approved references, never a fresh guess.shot batch
- 07PostMixes audio, burns captions and assembles the episode master.episode master
- 08PublishWrites the title, description and lineage record for the finished upload.release kit
Runtime
A production run you can walk away from.
Runs execute on Cloud Run behind a durable queue. Progress is written down as it happens, so a refresh, a dropped connection or a closed laptop costs nothing.
- Checkpointed
- Close the tab. The run keeps going on Cloud Run.
- Idempotent
- A retry never buys the same paid asset twice.
- Resumable
- A failed shot restarts alone, not the whole episode.
- Accounted
- Every asset carries a checksum, an attempt history and a cost.