Playmate: Flexible Control of Portrait Animation via 3D-Implicit Space Guided Diffusion

Figure 1. Playmate can generate lifelike talking faces for arbitrary identity, guided by a speech audio clip and a variety of optional control conditions. (a) shows the generation results under different emotional conditions using the same audio clip. The top row in (b) shows the driving images, while the bottom row shows the generated results. The poses in the generated results are controlled by the driving images, and the lip movements are guided by the driving audio. (c) demonstrates highly accurate lip synchronization and vivid, rich expressions across different style images.