PipeNetwork/minimax-h3-mlx

MiniMax recently released MiniMax-H3, a multimodal AI system that ingests text, images, audio, and video to generate up to fifteen-second video clips—with audio included. Per Simon Willison, a developer successfully ported it to MLX, enabling it to run on Apple Silicon, and tested it on an M5 Max MacBook Pro. The experiment required downloading roughly one hundred fifteen gigabytes of model files, and generation took just under forty-five minutes. The resulting video looked impressive, but Willison notes the audio output was disappointing—the system needs explicit audio guidance to produce usable sound.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton