MiniMax H3 takes up to 9 image, 3 video and 3 audio refs — directing, not prompting

LearnWithBishal · x · 2026-08-14

X creator @LearnWithBishal pitches the MiniMax H3 video model: it's currently 50% off on Magnific until September 1.

His point is that H3's biggest shift isn't quality but control — instead of hoping the model understands your vision, you guide it with up to 9 reference images, 3 videos and 3 audio clips, making generation feel "much closer to directing than prompting."

Related event: MiniMax H3 hits Magnific with multimodal reference control and 50% off 2K generation(6 posts)→

Original post →

More from Multimodal

Multimodal channel →