AC3D: Analyzing and Improving 3D Camera Control in Video Diffusion Transformers

TL;DR

This paper introduces AC3D, a novel approach for precise 3D camera control in text-to-video models, improving quality and training efficiency by analyzing camera motion, optimizing conditioning, and curating a new dataset.

Contribution

We analyze camera motion from first principles, optimize conditioning strategies, and curate a new dataset, leading to the state-of-the-art AC3D model for camera-controlled video generation.

Findings

Camera motion is low-frequency, enabling better conditioning strategies.

Limiting camera conditioning to certain layers reduces parameters and improves quality.

A new dataset of 20K dynamic videos enhances model performance.

Abstract

Numerous works have recently integrated 3D camera control into foundational text-to-video models, but the resulting camera control is often imprecise, and video generation quality suffers. In this work, we analyze camera motion from a first principles perspective, uncovering insights that enable precise 3D camera manipulation without compromising synthesis quality. First, we determine that motion induced by camera movements in videos is low-frequency in nature. This motivates us to adjust train and test pose conditioning schedules, accelerating training convergence while improving visual and motion quality. Then, by probing the representations of an unconditional video diffusion transformer, we observe that they implicitly perform camera pose estimation under the hood, and only a sub-portion of their layers contain the camera information. This suggested us to limit the injection of camera conditioning to a subset of the architecture to prevent interference with other video features, leading to a 4x reduction of training parameters, improved training speed, and 10% higher visual quality. Finally, we complement the typical dataset for camera control learning with a curated dataset of 20K diverse, dynamic videos with stationary cameras. This helps the model distinguish between camera and scene motion and improves the dynamics of generated pose-conditioned videos. We compound these findings to design the Advanced 3D Camera Control (AC3D) architecture, the new state-of-the-art model for generative video modeling with camera control.