Frame In-N-Out: Unbounded Controllable Image-to-Video Generation

TL;DR

This paper introduces a novel controllable image-to-video generation method leveraging the Frame In and Frame Out technique, allowing user-guided object manipulation with improved coherence and detail.

Contribution

It presents a new dataset, a motion-controllable video Diffusion Transformer architecture, and an evaluation protocol for enhanced controllability in video synthesis.

Findings

Our approach significantly outperforms existing baselines.

The new dataset facilitates better training and evaluation.

The method maintains identity and temporal coherence effectively.

Abstract

Controllability, temporal coherence, and detail synthesis remain the most critical challenges in video generation. In this paper, we focus on a commonly used yet underexplored cinematic technique known as Frame In and Frame Out. Specifically, starting from image-to-video generation, users can control the objects in the image to naturally leave the scene or provide breaking new identity references to enter the scene, guided by a user-specified motion trajectory. To support this task, we introduce a new dataset that is curated semi-automatically, an efficient identity-preserving motion-controllable video Diffusion Transformer architecture, and a comprehensive evaluation protocol targeting this task. Our evaluation shows that our proposed approach significantly outperforms existing baselines.