MiniMax H3 MiniMax

MiniMax H3 Max is a high-speed video generation model built by fal as a post-trained variant of MiniMax's open-weight H3 video architecture. Optimized specifically for rapid inference, the model trades away heavy multi-stage upscaling loops to maximize generation throughput, delivering standard 768p clips with synchronized stereo audio significantly faster than real-time wall clock speed.

Designed for iterative creative workflows, MiniMax H3 Max supports text-to-video, image-to-video, and reference-to-video modalities. By streamlining the native context understanding of the underlying H3 foundation, it achieves strong prompt adherence and fluid aesthetics while enabling rapid prototyping for short-form media, real-time interactive streams, and dynamic visual content.