IndexMapAnti-DoomscrollAbout
Model FamilyVideoImageCross-Modal

MiniMax & Hailuo Video

Also known as: Hailuo AI · Hailuo Video · Hailuo 02 · MiniMax H3 · Hailuo H3 · MiniMax Video-01 · MiniMax Speech-02

Plain English

MiniMax's family of video and speech foundation models—spanning the proprietary Hailuo 02 video generator, the Speech-02 TTS model, and the open-weight MiniMax H3 series (whose local license excludes the US, UK, EU, and South Korea).

Technical Overview

Developed by Shanghai AI lab MiniMax (part of Chinese AI Labs), the Hailuo video lineage (Video-01, T2V-01-Director, Hailuo 02, MiniMax H3) uses a spatiotemporal DiT architecture—upgraded in Hailuo 02 with Noise-aware Compute Redistribution (NCR). While MiniMax H3 shipped downloadable open weights on Hugging Face, its Community License Agreement legally excludes self-hosting in the US, UK, EU, and South Korea.

Timeline (5)

RSS
  1. 2024

  2. 2025

    Release

    MiniMax released T2V-01-Director and I2V-01-Director, embedding bracketed natural-language camera commands (such as [Truck left, Pan right, Zoom in]) directly into prompt conditioning for deterministic multi-axis camera choreography.

  3. Release

    MiniMax released Hailuo 02, introducing its Noise-aware Compute Redistribution (NCR) DiT architecture to improve token efficiency by 2.5x while rendering native 1080p video with complex acrobatics, gymnastics, and environmental physics.

  4. 2026

See also (9)

Open full graph
4 more connections