1. Sample the take
While recording, the studio samples camera position, rotation, forward direction, field of view, elapsed time and—when present—the selected subject anchor.
METHOD
A camera path is more than a mouse path.
The analyzer works from camera geometry rather than trying to guess a move name from a cursor trace. Processing stays in the browser and does not require a model API or secret key.
2. Measure short windows
Short neighboring windows are compared for translation, rotation, lens change, speed, acceleration and distance to the anchor.
3. Classify the dominant change
Rotation without meaningful translation becomes pan, tilt or roll. Local sideways, vertical and forward travel become truck, pedestal and dolly. Field-of-view change becomes zoom.
4. Preserve subject relationships
When distance to an anchor stays stable while the camera travels around it and continues to face it, the path can be classified as an orbit rather than a generic sideways move.
5. Merge compatible windows
Adjacent windows with compatible direction and motion are combined into reviewable camera beats. Small corrective movements are compressed instead of being described as many invented actions.
6. Render a prompt
The selected beats are translated into plain cinematography language and model-oriented phrasing. Excluded beats disappear from the prompt and replay, while the raw JSON keeps the original samples.
What it does not know
The tool sees the virtual camera and optional spatial anchor. It does not understand story intent, emotion, physical set constraints or what an external video model will render.
How to review a result
Replay the detected path, select adjacent beats and remove any beat that does not match the shot you intended. Treat the generated wording as an editable starting point.