In-Context Sync-LoRA for Portrait Video Editing

TL;DR

Sync-LoRA is a novel portrait video editing method that ensures high-quality visual modifications while maintaining exact frame synchronization and identity consistency across diverse edits and unseen subjects.

Contribution

The paper introduces Sync-LoRA, a new in-context LoRA training approach that leverages automatically curated synchronized video pairs for precise, identity-preserving portrait video editing.

Findings

Achieves high visual fidelity in edited videos.

Maintains frame-accurate synchronization and identity consistency.

Generalizes well to unseen identities and diverse edits.

Abstract

Editing portrait videos is a challenging task that requires flexible yet precise control over a wide range of modifications, such as appearance changes, expression edits, or the addition of objects. The key difficulty lies in preserving the subject's original temporal behavior, demanding that every edited frame remains precisely synchronized with the corresponding source frame. We present Sync-LoRA, a method for editing portrait videos that achieves high-quality visual modifications while maintaining frame-accurate synchronization and identity consistency. Our approach uses an image-to-video diffusion model, where the edit is defined by modifying the first frame and then propagated to the entire sequence. To enable accurate synchronization, we train an in-context LoRA using paired videos that depict identical motion trajectories but differ in appearance. These pairs are automatically generated and curated through a synchronization-based filtering process that selects only the most temporally aligned examples for training. This training setup teaches the model to combine motion cues from the source video with the visual changes introduced in the edited first frame. Trained on a compact, highly curated set of synchronized human portraits, Sync-LoRA generalizes to unseen identities and diverse edits (e.g., modifying appearance, adding objects, or changing backgrounds), robustly handling variations in pose and expression. Our results demonstrate high visual fidelity and strong temporal coherence, achieving a robust balance between edit fidelity and precise motion preservation.