PipeNetwork/minimax-h3-mlx

MiniMax-H3 is an omni-modal generative system that accepts text, images, audio, and video inputs to generate video clips up to 15 seconds long with matching audio. A new Python package named PipeNetwork/minimax-h3-mlx ports this model to MLX so it can run locally on Apple Silicon hardware.

Running the model requires downloading roughly 115 gigabytes of model files from Hugging Face, including specific components from MiniMaxAI and the 8-bit MLX port by PipeNetwork. Using an M5 Max MacBook Pro, generating a video from a text prompt takes just under 45 minutes of processing time.

The generated video output is visually impressive, though the accompanying audio can sound like strange speech-like noise if the user does not supply specific prompt guidance for the sound. The official prompting guide contains detailed instructions on how to properly configure these audio and video generation parameters.