Which Sume video ids turn audio off, and which add sound

On Sume, Gemini Omni Flash 1.1, MiniMax H3 and MiniMax H3 Max always return audio, and a generate_audio: false request is rejected. Kling 3 has a sound-on and a sound-off price, so it is the clearest choice when you want a silent clip. Wan 3.0 and the Seedance 2.x ids list audio in their catalog capabilities, and Sume reads a generate_audio flag for Wan.

If you need silence and the id has no switch, you remove the sound after generation or you pick another id.

By id

This is how the catalog and docs describe each row.

Removing sound you did not want

When a clip arrives with sound that does not fit, the fix is on the render side. Build a timeline with your own audio spine, or use a silence mode, so the generated track does not reach the final file. This costs a render, but it keeps the visual you liked. For a batch of silent clips, route them to the id with the switch and skip the extra step.

The opposite case also matters. If you want dialogue or ambient sound, an always-on id saves a TTS and mixing step, and the audio is generated in sync with the picture.

A cost note

Sound is free on the always-on ids in the sense that there is no separate price line for it, but Kling's sound adds 50% to the per-second rate. If you do not need sound from Kling, turn it off. If you need it, compare Kling on sound against an always-on id at the same resolution before you commit.

Decision rules

Match the id to what you plan to do with the sound.

Whichever way you go, say so in the prompt. A line such as "no speech, ambient room sound only" does not switch audio off, but it does steer what the always-on models make.

• You want the generated sound: an always-on id, or Kling with sound on.
• You want silence: Kling with sound off, or a generate_audio: false request on an id that takes it.
• You will add narration: choose silence, then mix voice and music in a timeline.
• You want to check: inspect the file's audio streams after generation.