MiniMax

Hailuo 02 and Hailuo 2.3 support text-to-video and image-to-video workflows, while 2.3 Fast is intended for faster image-driven iteration. The models accept prompts for action and camera direction; Hailuo's API also supports bracketed camera commands for compatible image-to-video variants.

MiniMax H3 is a multimodal video model for building a shot from text, animating a first frame, targeting a last frame, interpolating between both endpoints, or guiding the result with up to nine images, three videos, and three audio clips. It supports explicit 4–15 second durations and 768P or 2K output; text-to-video uses a selected aspect ratio, while frame-driven generation follows the supplied image through adaptive sizing.

MiniMax Speech 2.8 HD turns text into speech using a selected voice ID. Omnipix exposes emotion, speed, volume, pitch, language assistance, channel layout, sample rate, bitrate, file format, and optional subtitles. It is suited to workflows that need more explicit control over both performance and the delivered audio file.

MiniMax Voice Cloning creates a reusable private voice from a reference audio file. Name the voice, enter a short preview script, and optionally enable noise reduction or volume normalization. Clean, single-speaker source audio makes evaluation easier; use only recordings you are authorized to submit and reproduce.