kling-3 on Sume: generate_audio true or false, and what changes

On Sume generate_audio is optional and defaults to the model's audio capability, so a plain kling-3 request produces audio. Send generate_audio: false for a silent clip. The catalog lists two list rates for kling-3, one with audio on and one with audio off, so the flag is also a cost choice. For MiniMax H3 and H3 Max and for Gemini Omni Flash 1.1 the audio is always on and generate_audio: false is rejected with a 400.

Kling's page describes Kling 3.0 audio as single-channel and Kling 4.0 as two-channel stereo (kling.ai, read 2026-10-03); Kling 4.0 is not callable on Sume. The Sume behaviour is in the Video generation docs and the Video Router docs.

Which ids let you turn audio off

Sume does not print a single audio table in the docs, so the safest read is the live model list: each entry has generate_audio: true when the model can produce audio. Beyond that flag, three ids reject generate_audio: false outright because their audio is always produced.

If you request audio from a model that cannot make it, the answer is also a 400 ('does not support generate_audio'), not a silent clip.

When to turn it off

Turn audio off when you will lay your own voice or music over the clip: an avatar voice-over, a licensed track, or a captioned social cut. Generated sound would only be muted later, and with kling-3 you avoid paying for it.

Leave it on for dialogue or sound-effect clips, and check the result: the video tools can inspect the file, and the audio track tells you whether the model produced speech or ambience.

Do not assume the price

Sume bills list times 1.25 and the list price is shown per model in the models endpoint. This post does not quote a figure because list prices change; read pricing for kling-3 at request time and compute your own total for the number of seconds you will generate.

A worked habit: store the list rate with each job, so later you can explain a bill without guessing which flag was set.