Orkas-AI/Orkas-VideoStudio

What it solves

Traditional video production either requires hand-coding every frame (like Remotion) or using black-box SaaS editors that lock you into proprietary timelines. OrkasVideoStudio gives coding agents (Claude Code, Codex, Cursor, etc.) the knowledge, deterministic tools, and an editable intermediate representation (plan.json) to compose motion graphics, edit footage, generate AI media, and run an automatic end-to-end pipeline. Instead of a black-box agent, the video is expressed as a readable, diffable, re-renderable plan; change one line and only that piece re-renders.

How it works

The project wraps HyperFrames, ffmpeg, and whisper.cpp into thin deterministic capabilities: render, edit, transcribe, and generate. It adds a knowledge layer (skills) that teaches the agent what makes a good video and which production line to use. There are four production lines: Compose (script to HTML motion graphics to mp4, no paid keys), Edit (cut/join/trim silences/subtitle/dub, no paid keys), Generate (talking-head or cinematic footage via your own provider keys), and Auto (an end-to-end pipeline that routes across the other lines using a single plan.json). The plan is the contract: it contains segments, captions, narration, and every piece is independently re-renderable. A delivery guard (ovs plan promise-check) verifies the finished cut keeps its promise (e.g., real motion, not a silent slideshow) before anything ships.

Who it’s for

Developers and AI-agent users who want to produce videos programmatically via natural language instructions to a coding assistant. It’s especially suited for motion-graphic explainers, SaaS or app demos, product ads, and repurposing long recordings into highlight clips. Works with any agent that can run a shell or speak MCP.

Highlights

• Editable, diffable plan.json as the single source of truth — each segment re-renders independently.
• Three capability axes plus an automatic pipeline: compose, edit, generate, and auto (end-to-end).
• Zero-key trunk for compose/edit/transcribe; generation uses your own provider keys (OpenAI, Gemini, Doubao, MuAPI, etc.) — no managed backend.
• Delivery guard runs deterministic checks (motion, audio, captions) before a video ships.
• Cost-conscious: generation is treated as a budget to defend; plans aim for ~90% zero-billable-generation results.
• MIT-licensed, local-first, self-hostable.