Your coding agent’s new side hustle: video production

What it does

OrkasVideoStudio is an MIT-licensed toolkit that turns coding agents like Claude Code or Cursor into video producers. You describe a clip in plain language—say, a 60-second vertical explainer with Chinese captions—and the agent authors a readable plan.json timeline that stitches together motion graphics, edited footage, or AI-generated segments. The project provides the deterministic render pipeline (thin wrappers around ffmpeg, whisper.cpp, and HyperFrames) and a delivery guard that verifies the output before anything ships.

The interesting bit

The whole video lives as a diffable, per-segment re-renderable JSON contract instead of a black-box timeline. Change one caption or narration line and only that piece rebuilds, while the agent-agnostic CLI and MCP server let you drive it from any shell-capable assistant.

Key highlights

• Zero-key trunk for composition, editing, and transcription; only the optional AI-generation layer requires your own provider keys.
• Four production lines—Compose (HTML motion graphics), Edit (cut/mix/subtitle), Generate (BYO cinematic footage), and Auto (end-to-end orchestration).
• Deterministic delivery guard audits the final cut against the plan before delivery.
• Self-hosted with no managed backend: no account binding, extracted from the Orkas desktop app into a standalone agent toolkit.
• MCP server and native skill packs for Claude Code and Codex that mirror the CLI one-to-one.

Caveats

• Still in early development: npm packages are not yet published, so it is currently source-only.
• A full local toolchain is required—Node 22+, ffmpeg, ffprobe, and a browser runtime for some tests; the test suite may skip video-heavy cases if any are missing.

Verdict

Worth exploring if you want your agent to own video workflows while keeping the plan auditable and the render pipeline local. Skip it if you need a polished GUI or fully managed SaaS today.