What Input Format Does the MiniMax H3 API Use for Multimodal Prompts?

Overview

Modern multimodal video APIs usually accept a structured content array rather than a single text string, and MiniMax H3 fits that pattern where it exposes image, video, or audio references. Confirm the exact schema in the current documentation, because the field names and accepted types vary by provider and version, and a wrong shape is the most common first-call failure.

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

The reason a content array matters becomes obvious once you go beyond plain text-to-video. A single prompt field can carry a description, but it has nowhere clean to attach a reference image, a motion clip, or an audio track. A content array lets each input declare its type and role, so the model knows which item is the character reference, which is the scene, and which is the text instruction. That structure is what makes reliable reference-guided generation possible instead of hoping one blob of text does everything.

Build your request from the documented schema, validate it before sending, and keep a small example payload for each generation mode you support. That reference saves real debugging time.

When you reach the model through a unified provider, the input shape is often normalized across models, so the same array structure works whether you call MiniMax H3 or another video model next.