MiniMax H3 Test: Generates 15s High-Consistency Video from Single Prompt

Div_pradeep · x · 2026-08-04

A test of the MiniMax H3 model demonstrates its ability to generate a continuous 15-second video from a single text prompt containing references.

Throughout the uncut sequence, the model successfully maintained character consistency (featuring a Gothic sorcerer and a mechanical raven), lighting, and smooth camera motion. The model also supports a multimodal reference feature, allowing users to freely combine video, text, image, and audio inputs.

Related event: MiniMax H3 Video Model Impresses in Tests(3 posts)→

Original post →

More from Multimodal

Multimodal channel →