Dual character fight with orbital camera (multi-video)

Cited / reported data

Use reference videos to reproduce camera language, action rhythm, and motion patterns with closer alignment. How it works: upload a reference video as @video1 in your prompt. The model analyzes the camera trajectory — including speed changes, focal-length shifts, and framing transitions — then generates new footage that follows the same path with your chosen subject and scene. You describe the scene; the reference video supplies the camera work. When to use this: recreating a signature camera move (dolly zoom, Hitchcock zoom, tracking shot) without expensive equipment; matching the visual rhythm of an existing ad or film scene; producing consistent camera language across a campaign; replicating event-coverage-style handheld movement for a more documentary feel. Tips and practical notes: shorter reference clips (3–8 seconds) with a single clear camera move tend to replicate more accurately than longer clips with multiple cut points. If your reference has both camera movement and subject action, the model prioritizes camera path — add explicit subject-action instructions in your text prompt. Combining camera-motion replication with character consistency lets you keep the same actor across scenes while varying the cinematography.

Getting the model to copy blocking, camera moves, or complex action from a film used to mean writing long prompt descriptions—or it simply didn’t work. Now you just upload a reference video.