PipeNetwork/minimax-h3-mlx
MiniMax-H3 omni-modal model now runs locally on Apple Silicon via MLX port
“a rainbow colored skunk leaps over a mossy log in a supermarket”
MiniMax released H3, an omni-modal model accepting text, images, audio, and video that can generate 15-second video clips with audio. A community port now enables it to run locally on Apple Silicon via MLX, though it requires ~115GB of model downloads and took 45 minutes to generate a single clip on an M5 Max. The local inference capability is notable but the hardware demands and generation time limit practical accessibility.