Alibaba's Wan2.7-Video Model Launches: Supports Multi-modal Input and Refined Video Editing

Alibaba officially launched the video generation model Wan2.7-Video, which supports multi-modal input including text, images, videos, and audio. It covers the entire creation chain—from generation and editing to replication—and is said to be more controllable, with strong directing and performance capabilities.

This model allows videos to be edited like documents: users can make local adjustments to the画面 via commands, add/remove/replace elements (such as deleting a train or replacing film with a plate), change the environment style (like turning summer into late autumn), or enhance video quality (e.g., colorizing black-and-white footage).

Moreover, it supports secondary creation: modify character lines (while keeping lip sync and tone consistent), change actions (e.g., from sitting on the sofa to standing and playing games), replace characters (like swapping a player with a medieval knight), adjust camera settings (such as raising the lens from the ground); it also controls the plot through first and last frames, and multi-modal references ensure consistent character features across scenes.