MiniMax-H3 Multimodal Video Generation Model Now Available for Apple Silicon via MLX

MiniMax's new MiniMax-H3 model, described as a general-purpose omni-modal generative system, is now available for Apple Silicon through an MLX port. The model accepts text, images, audio, and video as inputs and can generate video clips up to 15 seconds long with accompanying audio. A developer successfully tested the MLX implementation on an M5 Max MacBook Pro. The model files require approximately 115 GB of storage, and video generation on Apple Silicon takes roughly 45 minutes per clip. While the generated video demonstrated impressive visual quality, audio output was noted as unintelligible, suggesting the need for proper audio prompting guidance as documented in the model's prompting guide. The MLX port enables machine learning researchers and developers with Apple Silicon hardware to run this multimodal generation model locally, making advanced video generation capabilities more accessible outside of cloud computing environments.