Why API documentation answers the wrong questions
Most of the silence in this register is not evasion. API references are written for a developer calling an endpoint and describe parameters and return values; the fields this site keeps are about behaviour across a series, which those documents have no reason to cover. As of 2026-09-12.
| Model | What is documented | Detail |
|---|---|---|
| API reference | Parameters and returns | Veo, Vidu, D-ID |
| Product page | Features and plans | Most entries |
| Prompting guide | How to write the request | Sora 2, the dialogue block |
| Model card | Inputs, arguments and a licence | Wan2.2-S2V, open weights |
| Compliance guide | Boundaries and obligations | The one preset-library statement |
Inclusion rule. Document types actually read for this register. Marketing announcements and launch posts are not read, because a capability shown once is not a documented behaviour. Order. By how many entries were read from that kind of document.
1The audience explains most of the empty cells
A developer wants to know what to send and what comes back. A producer wants to know whether a character will sound the same in episode forty. The first question is answered thoroughly across this register and the second is barely addressed, and those are the same documents.
Reading the emptiness as a product deficiency would be a mistake. It is a statement about who these pages were written for, and the register says so rather than letting blanks imply inadequacy.
2The most useful statement here came from a document about risk
The clearest description of where voices come from sits in a compliance checklist. That is not where anyone would look, and it exists because a different obligation forced the vendor to write down something product documentation had not.
It suggests where the next useful statements will come from: obligations rather than marketing. Pages written to satisfy a rule tend to be specific in the way this register needs.
3What this site would read if it existed
A behaviour note: what the model does with two speakers, whether a voice persists, what happens when a supplied sample conflicts with a bound voice. None of that is exotic and none of it appears in either kind of document today.
Until it does, the register reads what exists, records what it finds, and marks the rest not documented rather than filling it from how the products appear to behave.
- Audio sourceAudio is generated with the video rather than added afterwardsgenerated with the picture
- Audio sourceQ3 outputs speech, sound effects and music natively, on by default in the API, with a speech-only modethree tracks in one pass
- Voice sourceThe guide tells users to use a preset voice library or their own recordingstwo routes named
- What the card does settleThe card states support for 480P and 720P and licenses the weights under Apache 2.0
- How a spoken line is written into a promptDialogue belongs in a dialogue block below the prose description, with exchanges limited to a handful of sentences
- Audio sourceA script is either text of up to forty thousand characters or an audio url, with audio limited to five minutes for clips and ten for talksa script read aloud or a file
4Sources
Every line is read from the vendor documentation linked on the speech controls page, checked 2026-09-12. What counts as documented is on how read. Nearby: How long a shot can be, Fifteen seconds of voice, Answers software can check.