MAGI 2 Blog
Blog
Guides, comparisons and release notes for MAGI 2 - the unified audio-video generation model from Sand.ai.
How to Use MAGI 2: A Practical Guide to Unified Audio-Video Generation
A practical walkthrough of MAGI 2: how to prompt for dialogue, music and camera movement, how to keep long-form scenes coherent, and how credits are consumed per generation.
MAGI-2 Preview: Release Date, Availability and What Comes Next
What MAGI-2 Preview is, what Sand.ai has released so far, how to get access today, and how to tell a preview milestone apart from a general availability release.
MAGI 2 vs MAGI 1: What Changed in Sand.ai's Video Model
MAGI 2 replaces MAGI 1's autoregressive temporal chunking with a single-stream MagiMoE backbone: 114B parameters, about 6B activated per token, and audio plus video in one token sequence.
What is MAGI 2
MAGI 2 is a unified audio-video generation model — text, visuals and sound live in a single token sequence, with 114B parameters and only 6B activated per token.