oma-video

Run oma video generate with your brief and --mode shorts to produce a 9:16 vertical clip. The pipeline writes a script, synthesizes narration, generates stills with Ken Burns motion, builds captions, and renders the mp4 via Remotion into a run directory.

Use --mode explainer with a brief pointing at the README, code, or data file. The Skill combines slide frames, generated diagrams, and code frames with narration, rendering a 16:9 or 9:16 explainer video deterministically from render-spec.json.

Yes, the router is key-optional with key-free defaults for every capability: local voice synthesis, oma-image stills, key-free captions, and Remotion rendering. Paid providers like Pexels or Pixelle activate only when their environment keys are present, otherwise the chain falls back automatically.

The demo mode supports --source web with a --url, which opens a headed Chromium browser and records while a human drives the on-screen flow. It performs no credential automation, masks URLs in logs, and falls back to a guided protocol when no interactive terminal exists.

Fallbacks occur when a preferred provider is unavailable, such as a missing Pexels key or an unreachable Voicebox MCP server. Every fallback is recorded in the manifest's warnings array, and the run still completes with the key-free alternative.

Videos are capped at 180 seconds duration and 40 scenes per run. Outputs must stay inside the project directory unless --allow-external-out is set, and estimated costs above the $0.20 guardrail require confirmation before proceeding.