What Wan2.2-S2V documents about speech

Wan2.2-S2V (huggingface.co) is an audio-driven model with open weights: a track, a reference image and an optional prompt go in, 480P or 720P comes out, and the weights carry an Apache 2.0 licence. As of 2026-09-22.

Inclusion rule. Fields are the same five for every model here; a field the vendor does not document is recorded as not documented rather than inferred from generated output. Order. Fixed field order, identical on every model page.

1Open weights change what a documented field can mean

This is the one entry here whose documentation is a model card rather than a product page, and the difference shows in what gets settled. Resolutions, the licence, the arguments and the inputs are all stated; voices, languages and a cast are not, because a card describes a model and not a service wrapped around one.

Apache 2.0 also moves the compliance question. A production running the weights itself decides what audio goes in, which is a stronger position than a terms page and a heavier obligation, since nothing upstream is checking the consent behind a voice.

2A pose track beside the audio

A pose video argument lets the result follow a pose sequence while staying synchronised to the audio. Two drivers at once is a more filmic arrangement than audio alone: the body can be directed while the mouth follows the take, which is roughly how a performance-capture pipeline is organised.

For a serial that matters more than it sounds. Most audio-driven entries here animate a portrait; a pose track is what makes a shot rather than a bust, and it is published as an argument rather than as a promise.

3Two fields empty by construction

Languages and voice source are blank here, and not because the card is thin. The audio is supplied, so the language is whatever was recorded and the voice is whoever recorded it, neither of which the model has reason to describe. The register still marks them unstated, because a reader comparing rows needs to know the answer is not on the page.

What is left is a clean division of labour. Casting, language and consent are settled upstream in a recording session, and the model is asked only to make a face agree with a waveform.

4This entry, one column at a time

Each of these stays inside a single field: what Wan2.2-S2V puts there, what the wording settles, and what it leaves for a take to answer.

• Audio source — supplied; the model is audio-driven.
• Voice source — the audio track itself.
• Languages — not documented by the vendor.
• Lip-sync — the audio input.

5Read against another entry

Each of these puts Wan2.2-S2V beside one other entry on a column where the two land at opposite grades of answer.

• Wan 3.0 and Wan2.2-S2V — on audio source.

6Where this model sits against the rest

Inclusion rule. Models whose vendor documents something about generated speech. A model with nothing published on the point is still listed, with the cell marked hollow. Order. Alphabetical by model name.

Inclusion rule. Models whose vendor documents something about generated speech. A model with nothing published on the point is still listed, with the cell marked hollow. Order. Alphabetical by model name.

Inclusion rule. Models whose vendor documents something about generated speech. A model with nothing published on the point is still listed, with the cell marked hollow. Order. Alphabetical by model name.

7Sources

Taken from the model card at huggingface.co on 2026-09-22. Every entry against the same five fields sits on the speech controls page; the rule behind each field is on how read. The entries either side of this one: D-ID, Hedra.

• Audio source
  The card describes audio-driven cinematic video generation from an audio input with a reference image and an optional text prompthanded to the model rather than made by itWan-AI, Wan2.2-S2V-14B model card / recorded 2026-09-22
• What the card does settle
  The card states support for 480P and 720P and licenses the weights under Apache 2.0Wan-AI, Wan2.2-S2V-14B model card / recorded 2026-09-22
• A second control alongside the audio
  A pose video argument lets the result follow a pose sequence while staying synchronised to the audioWan-AI, Wan2.2-S2V-14B model card / recorded 2026-09-22
• Languages
  Not documented by the vendor (as of 2026-09-22)nothing published about the spoken performanceWan-AI, Wan2.2-S2V-14B model card / recorded 2026-09-22