Ovi: Twin Backbone Audio-Video Generation by Character.AI
Ovi by Character.AI is a twin-backbone audio-video generation model built on Wan2.2. Generates synchronized audio and video bidirectionally with MMAUDIO conditioning.
O
Ovi
Audio-Video GenerationVideo GenerationCross-ModalWan2.2Character.AIOvi by Character.AI is a twin-backbone cross-modal model that generates synchronized audio and video bidirectionally. Built on Wan2.2 (A14B MoE architecture), Ovi uses separate diffusion backbones for audio and video with cross-attention conditioning between modalities. The audio pipeline uses MMAUDIO VAE and BigVGAN vocoder for 16kHz audio synthesis. Available in bf16 precision, optimized for ComfyUI via Kijai's WanVideoWrapper.
| Developer | Character.AI |
| Release Date | 2025-10 |
| Architecture | Twin-backbone Diffusion Transformer (Wan2.2-based, A14B MoE) |
| License | Apache-2.0 |
| Model Sizes | Audio ~11.4 GB / Video ~10.4 GB (bf16) |
| Capabilities | Video→Audio, Audio→Video, Video+Audio generation |
| Audio | 16kHz MMAUDIO VAE + BigVGAN vocoder |
| Inference | ComfyUI via Kijai WanVideoWrapper |