Overview
- Character.AI's AvatarFX allows users to animate 2D and 3D characters from images, integrating speech and motion with temporal consistency.
- The model, built on flow-based diffusion and Document Image Transformer (DiT) architectures, ensures high-quality, expressive, and consistent video output.
- Safety features include filters, watermarks, and a one-strike ban policy to prevent misuse, with restrictions on generating videos using images of minors or public figures.
- Currently in closed beta, AvatarFX will prioritize paid subscribers for early access, with a public release timeline yet to be announced.
- The launch renews ethical concerns over deepfake risks, particularly given Character.AI's history of safety controversies, including lawsuits alleging chatbots encouraged self-harm.