hybrid face synthesis

Hybrid face synthesis is a method for keeping a real person's face looking like the same person across different scenes when it's placed into AI-generated video.

If you asked a painter to paint a portrait of the same person in several different scenes, the face might come out slightly different each time, eventually looking like a different person. Hybrid face synthesis is like handing the painter several reference photos of that person's face beforehand, and having them check back against those photos every time they paint a new scene. Instead of describing the person from scratch in each scene, the same set of pre-set facial references gets pulled up again and again.

This method exists because when a real person's face is fed directly into AI video generation, the face often shifts subtly from scene to scene, ending up looking like a different person. Hybrid face synthesis is closer to a practical production trick used to prevent this problem.

The article reports that "to avoid the common 'likeness collapse' problem that occurs when putting a real person's face into AI video, a hybrid face synthesis approach was used." It's worth not mistaking this for a new AI model or feature — it refers to a production method that combines existing image and video generation tools to keep a face consistent.

If you're planning to create multiple scenes featuring the same person, you can try the following steps.

1. Prepare several photos of the person from different angles, such as front-facing and side profile.
2. Whenever you generate a new scene, feed these photos in together as reference images.
3. Compare the face in each finished scene against the reference photos to check whether it still looks like the same person.