What 2.2 open models emit
- T2V-A14B / I2V-A14B / TI2V-5B — silent video.
- S2V-14B — you bring the audio; the model animates.
- Animate-14B — motion transfer; audio not generated.
What 2.5+ products add
Public 2.5 write-ups (fal, DashScope-style APIs, aggregators) describe joint generation of dialogue, ambience, and music, 1080P, and clips longer than the 2.2 ~5 s recipes. Later 2.6/2.7/3.0 keep native AV and add multi-shot or reference controls. None of those are published as Wan2.2-style GitHub checkpoints.
Open-weight substitutes
To stay on Apache-2.0 / public weights: generate silent Wan2.2 video, then ThinkSound (or another V2A) for foley; or drive S2V with CosyVoice for talking heads. That is two (or three) models, not native AV.
FAQ
- Will a 2.2 GGUF magically add audio?
- No. Quantization changes weight dtype, not the output modality.
- Does S2V count as native AV?
- No. Native AV synthesizes sound. S2V consumes sound.
Primary sources
Checked against Wan-Video/Wan2.2 README · Aug 28, 2026
