Vidu Q4 audio on or off vs Sume generate_audio by video row

Vidu Q4 Preview generates audio together with the picture, including dialogue and sound effects, and its reference-to-video endpoint exposes a boolean for audio-video output. On Sume the answer depends on the row. Kling 3 lets you turn audio on or off and bills more with it on. Gemini Omni Flash 1.1, MiniMax H3 and MiniMax H3 Max always produce audio and have no toggle. Sending generate_audio: false to Omni is rejected with a 400 rather than ignored.

That difference matters if you port a Vidu payload. A request that says audio off is valid for Vidu and for Kling 3 on Sume, and invalid for Omni.

What each side says

Vidu's image-to-video page lists direct audio-video generation with dialogue and sound effects. Its reference-to-video page lists a boolean toggle for simultaneous audio-video output.

Sume's video docs say Omni and the H3 rows have native audio that is always produced, and the design notes for sume/auto add that generate_audio: false fails for auto too, so you should not send it or should send true. Kling 3 is priced twice in the catalog: a silent rate and an audio-on rate.

Test the toggle on Sume

The safest way to learn a row's behaviour is a dry cost check and a short clip. Use the catalog first: generate_audio in each GET /v1/videos/models entry says whether the row can produce an audio track.

Steps:

• Fetch GET /v1/videos/models and read generate_audio for your target row.
• For Kling 3, submit the same 5 second prompt twice, once with generate_audio true and once false, and compare the cost: 5 s is $1.05 with sound and $0.70 without.
• For Omni and H3, leave the field out.
• Do not rely on silence from an always-audio row; mute it in editing if you need a silent cut.

{
  "model": "kling-3",
  "prompt": "A barista pours oat milk, close-up, cafe ambience",
  "resolution": "1080p",
  "duration": 5,
  "generate_audio": false
}

Why it affects your budget

For a batch of thirty 8-second Kling 3 clips, audio on adds $0.07 a second, which is $0.56 a clip and $16.80 across the batch. If your pipeline replaces the sound with a licensed track or a separate voice-over, switching it off is a straight saving. If you rely on the model's ambience and dialogue, the extra is the price of not building a sound step.

On the always-audio rows there is no such lever, so the audio is simply part of the rate.

What Sume does not do

Sume does not document any control over what the audio says or sounds like on Omni or H3, apart from your prompt and, on H3 and Wan, audio references. It also does not claim that Vidu's audio quality is matched by any of its rows.