NAVA
NAVA is a 6.3B joint audio and video generator built from Wan2.2-TI2V-5B. Its diffusion design first aligns the two modalities and then combines them through shared processing.
Model design and access
The network uses ten dual-stream and twenty single-stream blocks. The supplied workflow supports text conditioning and image continuation. A reference WAV can bind a speaker’s voice to a marked speech span. The repository documents single-GPU operation and distributed inference.NAVA official model card
Practical use
Practical guidance. Review the image sequence, spoken words and sound timing as separate outputs. Voice conditioning needs appropriate permission for the reference recording. Keep generation settings and reference assets with the result so errors in speech or synchronization can be reproduced.NAVA official model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| License | Apache 2.0NAVA official model card |
| Ancestry | Wan-AI/Wan2.2-TI2V-5BNAVA official model card |
| Reference output | 720p-class video with synchronized audioNAVA official model card |