wassermanproductions/motion-previs-studio

Motion Previs Studio is a desktop app for filmmakers who want to plan out a shot before generating an AI video from it. It takes a source video clip, either a local file or a compatible web or YouTube link, and breaks it down into layers a filmmaker can reuse: pose, depth, camera movement, edges, masks, and other control layers. The idea is to give someone more control and precision when feeding a reference into AI video tools such as Seedance, ComfyUI, Blender, Runway, or Kling. After importing a clip, the user picks the exact range of frames to analyze, then chooses a reference mode. Camera only keeps just the camera movement and timing so the subject and setting can be replaced. Actor motion keeps the body movement along with the camera. Object motion keeps the path of an object or vehicle along with the camera. Full scene keeps the camera, blocking, subject motion, and depth all together. The app then runs the analysis, showing real progress at each stage, and lets the user cancel at any point. Under the hood it uses ffmpeg to process video and MediaPipe's Pose Landmarker, running locally, to detect body pose and skeleton points. Camera movement is calculated separately from the tracked subject, using optical flow and a fitting method, so the camera path can be reused even if the original actor or object is not. The tool can also export data in the OpenPose format used by many ControlNet-based AI pipelines. Once analysis is done, the app exports a production pack: a folder and zip file containing the reference video, depth video, edge and line-art videos, pose overlays, a contact sheet, and OpenPose data, ready to hand off to AI video generation or Blender. The app can also send a clip directly into a companion tool called Blockout for further scene staging. It runs on macOS and Windows, is open source under the Apache 2.0 license, and includes an optional local control server so AI coding assistants can drive the app directly.