New cast.
Same scene.
Choose the person, review the tracked selection, and give H3 your character photo. This take uses the original fictional reference plus a crop of its head, the source performance, and a tracked latent edit mask. The source retains its original cinematic aspect, and the woman and foreground railings remain in the shot. Facial articulation is generative and can differ from the source.
Compare the matched source and result
Inspect the actual tracked person
The green overlay shows the selected whole person across all 73 frames. Download the lossless mask or its executed tracking workflow.
Review every face frame and generation receipt
The source performance includes the open mouth and head turn. H3 follows that video performance; this example is not an audio-driven viseme system. Generative identity, articulation and occlusions still need review for each new shot.
