DomainShuttle: Precise Video Generation with Object Preservation

DomainShuttle is a new text-to-video (T2V) method based on Wan2.2-T2V-A14B that allows for maintaining object identity while changing visual styles.

Compiled by Sergey KostenchukPublished 2026-06-26Updated 2026-06-26

2026-06-26 Multimodal HuggingFace

🎨 DomainShuttle: Video Generation with Object Preservation

The new DomainShuttle method, based on Wan2.2-T2V-A14B, enables video generation that maintains high accuracy in reproducing specific objects when transferring them into any visual style.

🌍 The method solves the problem of preserving object identity in video, paving the way for flexible content personalization and domain transitions (e.g., from realism to anime).

👤 Users will be able to create videos where characters or objects retain their features while moving into completely new worlds or styles with high precision.

Source 1: https://github.com/HKUST-C4G/DomainShuttle Source 2: https://huggingface.co/CNcreator0331/DomainShuttle_weight

Sources