Magnific adds MiniMax's reference-driven H3 video model
By Steven Van ·
H3 takes up to 9 images and 3 videos as references and outputs 2K clips with native stereo audio, 5 to 15 seconds long.
Magnific added MiniMax H3 by Hailuo AI, and the pitch is refreshingly simple: add characters, motion, and the beat, and the model handles the rest. H3 is MiniMax's new video generation model, unveiled at WAIC 2026, built around reference-driven control rather than pure text prompting.
Reference-driven, not just prompt-driven
On Magnific, H3 accepts up to 9 images for character and style plus 3 videos for motion and camera control, then renders 2K output running 5 to 15 seconds. That's a materially different workflow than typing a description and hoping the model interprets it correctly, you show it the character, the look, and the camera move, and it composites a coherent clip from those references rather than guessing from text alone.
Native audio, not an afterthought
H3 also generates native stereo audio alongside the video rather than requiring a separate sound pass, part of why MiniMax is positioning it against Seedance on both quality and cost. For Magnific users building ad concepts, product videos, or character-led content, that closes a step that used to mean exporting to a separate audio tool.
Part of a broader rollout
H3 landed on Magnific the same day it showed up on Krea, Leonardo, Lovart, Pika, and Runway, a coordinated push that signals MiniMax is prioritizing distribution through existing creative platforms over building its own destination app from scratch.
Sources: Magnific on X.