ParaScale: Scale-Calibrated Camera-Motion Transfer via a Gauge-Invariant Parallax Number

ParaScale introduces a geometric solution to camera motion transfer in video generation by identifying and preserving the Parallax Number (Pi), a scale-invariant metric that quantifies perceived camera movement independent of scene depth. The method enables creators to transfer cinematic camera movements between videos at vastly different scales without requiring retraining, improving transfer fidelity by over 3x compared to uncalibrated approaches.

ParaScale addresses a fundamental technical challenge in video synthesis: the inability to meaningfully transfer camera trajectories between scenes operating at different depth scales. When a reference video captures motion at one scale and a target scene exists at another, naive trajectory reuse produces either invisible or grotesquely exaggerated camera movements. The researchers identified the root cause as a geometric property where translation-induced image motion depends on the ratio of camera displacement to scene depth (||T||/Z), making monocular trajectories inherently ambiguous without depth-scale information.

The core innovation centers on the Parallax Number (Pi = ||Delta T|| / Zbar), a dimensionless quantity that captures the perceptual strength of camera movement independent of absolute scale. By proving that Pi—not raw trajectory data—must remain constant during transfer, the team transformed a scale-dependent problem into a scale-invariant one. This insight enables ParaScale to function as a plug-and-play module compatible with existing pose-conditioned video generators without retraining requirements.

For the video synthesis and AI-driven content creation industries, this development removes a significant practical barrier to workflow efficiency. Cinematographers and animators can now confidently reuse camera moves across projects with completely different spatial contexts—from galaxy-scale movements to microscopic object manipulation. The introduction of Parallax Consistency Error (PCE) as a scale-symmetric evaluation metric provides practitioners with appropriate diagnostic tools, replacing similarity-aligned metrics that mask fundamental mismatches.

The results demonstrating effectiveness across four orders of magnitude in scale variation suggest robust applicability across diverse creative domains. Future developments might extend this framework to handle rotational components or automate Parallax Number extraction from diverse source materials.

• →ParaScale solves scale-mismatched camera motion transfer using the Parallax Number, a dimensionless metric independent of scene depth.
• →The method integrates as a plug-and-play module between pose extraction and injection without requiring model retraining.
• →Parallax Consistency Error (PCE) provides scale-symmetric evaluation that exposes scene-scale mismatches invisible to traditional metrics.
• →Testing across four orders of magnitude in scale variation demonstrates robust performance for diverse creative applications.
• →The framework preserves rotation information while calibrating translation, maintaining visual fidelity during scale-aware transfer.