DreamID Omni now official: ByteDance’s multi-character deepfake engine

ByteDance is working on DreamID Omni, a powerful video generator that can create deepfake-style content by combining text, images, and voice samples. The model can generate talking-head videos, edit existing footage, and even handle scenes with multiple characters in one shot.

DreamID Omni supports voice cloning and face editing, allowing it to synchronize lip movements with arbitrary speech and swap faces with convincing results. The project is targeting a March release for version 1, with the team planning to open source the model.

However, this combination of multi-character support and planned open access is already raising questions about practical uses, as well as potential misuse in impersonation and misinformation.