Which column every entry manages to fill

One column is almost always answered and the rest are mostly not. Where the sound comes from is filled by fifteen entries. Languages, a voice source, a persistence mechanism and a mouth driver are all minority answers. As of 2026-09-12.

Inclusion rule. Every column this register keeps is listed, with the number of entries whose documentation fills it. A column is not added unless enough vendors publish something comparable. Order. Fixed field order, the order the questions arrive in.

1The easiest question is the one everybody answers

Whether sound comes out of a model is a single binary fact, easy to state and hard to get wrong, so almost every document manages it. The columns that stay empty are the ones requiring a vendor to describe a workflow rather than a capability.

That pattern is the most consistent finding in this register. Documentation gets thinner exactly as the questions get more useful, and it thins in the same order across products that otherwise share nothing.

2The two blanks in the filled column are instructive

One is a video reference that never mentions sound; the other is a voices reference that never mentions a shot. Between them they cover both halves of this subject and neither covers the join, which is where every question here lives.

Neither is a weak product. The blanks describe what each document was written for, and a production is the party left to establish the rest by test.

3A filled cell still says less than it appears to

Counting filled cells would flatter the vendors who write most, so this register keeps the wording instead. Ten entries fill the mouth column and only five of those name anything as the driver; four name the feature and stop, and one argues the step away.

Grades inside a column matter more than the tally across it. A list beats a total, a driver beats a capability, and a stored voice beats a clip supplied on every call, regardless of how many cells are filled on either row.

• Audio source
  Not documented by the vendor (as of 2026-09-22)no mention of sound in the video referenceLuma, video generation reference / recorded 2026-09-22
• Lip-sync
  Not documented by the vendor (as of 2026-09-22)not addressed where voices are definedRunway, custom voices reference / recorded 2026-09-22
• Per-character binding
  Not documented by the vendor (as of 2026-09-22)no public descriptionSceneMixer, compliance guide / recorded 2026-09-12
• Languages
  Not documented by the vendor (as of 2026-09-22)neither a list nor a countOpenAI, video generation guide / 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

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: Speech without the bed, Regenerating a shot, Consent in the docs.