How Do I Use Reference Audio with the MiniMax H3 API?

Overview

Where supported, an audio reference guides the sound of a generated video — a soundscape, a dialogue tone, or a voice. In the MiniMax H3 API you attach it through the content array, so confirm the accepted formats and how the audio field is declared in the current schema before you plan around it.

Source note: Verified 2026-08-06 against the MiniMax official H3 blog, MiniMax Video Generation API docs, and Hugging Face MiniMax-H3 model page.

Match the input to the job, and mind the rights. A soundscape or ambience reference is low-risk and shapes mood. A voice reference is more sensitive, because using a real person’s voice raises consent and rights questions that ambience does not. If you reference a real voice, get clear, documented permission first and confirm the provider allows it; that step is a legal requirement, not a nicety, and it does not get cheaper to skip. Prepare the audio like any reference: trim to the relevant segment, cut background noise, and match it to the scene.

Keep the audio short and focused so the signal you care about is not buried, and pair it with a prompt that describes the visual while the reference handles the sound.

The call runs asynchronously like the others: submit with the audio attached, store the task ID, poll for completion, and retrieve the result. Keep permission records for any real-person voice in case anyone asks later.