Motion Previs Studio v4 is an open-source desktop app for AI-film previsualization, offering filmmakers precise control over AI video generation. It extracts key elements like pose, depth, and camera movement from source video, creating "production packs" for downstream AI workflows, thereby streamlining preproduction and enhancing creative iteration.

How It Works

Built with Electron, React, and TypeScript, the app processes video (local or URL via yt-dlp) to extract detailed control layers. It employs MediaPipe Pose Landmarker for pose extraction and Lucas-Kanade optical flow with RANSAC for subject-independent camera solves, recovering movement from background elements. FFmpeg ensures deterministic frame encoding. Users select shot ranges and reference modes (e.g., Camera only, Actor motion) before exporting bundles for AI tools like Seedance and ComfyUI.

Quick Start & Requirements

• Install/Run: Development: npm install, then npm run dev. Build: npm run build, npm run dist:dir.
• Prerequisites: Node.js, npm. Requires FFmpeg and yt-dlp. MediaPipe assets generated via npm run prepare-assets.
• Environment: macOS build is not signed/notarized.
• Links: Official website: WassermanProductions.com, Wasserman.ai.

Highlighted Details

• OpenPose Export: Deterministic openpose_pose.mp4 skeleton videos and openpose_keypoints.json (BODY_25).
• Camera Solve: Background-based optical flow recovers camera movement for subject replacement.
• Agent Control (MCP): Localhost HTTP server enables programmatic control via AI agents (Claude Code, Codex, Hermes).
• Direct Export: One-click export to Blockout app as a ghost underlay.
• Reproducibility: Features deterministic encoding and project save/restore.

Maintenance & Community

Developed by Sam Wasserman. No specific community channels (Discord, Slack) or public roadmap are detailed beyond a "Future Ideas" list. Attribution requirements are outlined under licensing.

Licensing & Compatibility

• License: Apache License 2.0.
• Compatibility: Permissive, suitable for commercial use and integration into closed-source projects. Requires NOTICE file preservation and citation.

Limitations & Caveats

The current macOS build requires signing/notarization for public distribution. Several planned features (e.g., shot-board view, batch processing, direct Blender file export) are not yet implemented. The project is primarily for local testing and development, with public releases pending further packaging.