What Veo documents about speech

Veo (ai.google.dev) documents audio generated with the video rather than added afterwards. The other four fields this register keeps are not addressed in the API documentation. As of 2026-09-12.

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.

1One field filled, and it is the field everyone fills

Audio generated with the picture is the common position among the entries read first here, so the one documented field on this entry is also the least distinguishing. What separates the entries in this register is what comes after that sentence, and for this model nothing does.

The emptiness is not evidence of a weaker model. It is evidence about the documentation, which is written for developers calling an API and describes parameters rather than production behaviour. A field this site keeps is a field a series has to plan around, and those two audiences do not want the same page.

2What a drama production cannot plan from this page

Whether a character can be given a voice that persists, which languages the model will speak, whether a supplied recording is accepted, and what happens to mouths when two people talk. Four questions, none answered, and each of them decides a workflow rather than a preference.

The honest reading is that the answer may exist and may be good; it is not published where a developer or a producer would find it before committing. Recording that as not documented keeps the comparison on the same footing as every other row, which is the only way a register of this kind stays worth reading.

3Why the entry stays in the register

Dropping the sparse rows would leave a table that flatters the vendors who write the most. Seventeen entries have now been read the same way against the same five fields, and the shape of the result, one common answer and four blanks, is itself the finding for this entry.

If the documentation grows a paragraph about voices, this page changes and its date moves. Until then a reader can take one thing from it with confidence and take nothing else, which is more useful than a page that guesses at the other four.

4Read against another entry

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

• Veo and Hedra — on audio source.

5Where 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.

6Sources

Taken from the API documentation at ai.google.dev on 2026-09-12. 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: Vidu, Wan 3.0.

• Audio source
  Audio is generated with the video rather than added afterwardsgenerated with the pictureGoogle, Veo documentation / recorded 2026-09-12