SuanNi

Mar 23, 2026 · Artificial Intelligence

How daVinci-MagiHuman Achieves Ultra-Fast, High-Quality AI Video Generation

The open‑source daVinci‑MagiHuman model introduces a single‑stream Transformer with 150 billion parameters that generates synchronized audio‑video content in just 2 seconds on a single H100 GPU, simplifying architecture, supporting multiple languages, employing novel attention gating, latent‑space super‑resolution, and achieving state‑of‑the‑art visual and audio quality compared to leading closed‑source systems.