Grok Imagine video sound: xAI audio vs Sume's silent row

Grok Imagine video sound is an xAI-side feature: its docs turn audio generation on and let you disable it with generate_audio=False. The Sume Grok row lists audio: false and rejects generate_audio, so it returns silent video.

xAI's facts are from its video docs and July 31, 2026 post, read 2026-09-30. Sume's are from its catalog and Video Router docs.

What does xAI offer for sound?

Its docs list audio generation, disabled with generate_audio=False, and up to 3 voices per request for reference audio. The July 31 post adds voice consistency through character images and voice references, with voice reference support in the API on request.

What does the Sume row do?

The catalog sets audio: false and reference_audios: false, and its constraints include "no aspect_ratio / bitrate_mode / generate_audio". Treat the output as silent and add sound separately.

Which Sume ids carry sound?

The docs say gemini-omni-flash-1.1 has native synced audio always on, and generate_audio: false is rejected there. MiniMax H3 Max is documented with native stereo audio. For a general walkthrough see AI video with sound.

What should I do for a Grok clip that needs sound?

Call xAI directly, or keep the silent clip and add music or a voiceover afterwards with a separate tool. Read audio in the catalog before assuming a track exists.