A config-driven documentary render engine (Remotion + ElevenLabs + whisper + ffmpeg)
A config-driven documentary render engine: Remotion + ElevenLabs + whisper + ffmpeg. It turns a written script into a finished 25-minute documentary with animated maps, archival footage, quoted pages, a mixed soundtrack, dubs and subtitles, and renders it in a way that survives the laptop running out of battery.
Source on GitHub · MIT licence · v0.2.0 · by Bohdan Oximets (heisssen) · built for one channel, works on Windows, expect rough edges.
Everything is indexed by paragraph. Each spoken paragraph of the script is one voice take and one visual beat. The voice decides the time; whisper finds every sentence inside every take; a planner turns a prose editing bible into frame numbers (cuts on sentence boundaries, rationed transitions, maximum shot lengths) and grades its own result. The picture and the sound read the same plan, so they cannot drift apart, and a dub is just another soundtrack fitted into the same slots.
Resumable minute-part rendering with a power and battery guard. A crash or a dead battery costs at most one minute of rendering; a supervisor restarts the job and a guard freezes it before the battery dies.
ledger_root hash lets anyone check a published film against it.