MiniMax-H3 Omni-Modal Model Runs on Apple Silicon with MLX Port
MiniMax
MiniMax released MiniMax-H3, an omni-modal generative system accepting text, images, audio, and video to generate up to 15-second video clips with audio. A Python package ports it to MLX for Apple Silicon; Simon Willison got it running on his M5 Max MacBook Pro, producing a video for a prompt but noting the audio was garbage without proper guidance.
MiniMax released MiniMax-H3, describing it as a general-purpose, omni-modal generative system that accepts text, images, audio, and video, and can generate up to 15-second video clips including audio. A Python package ports this model to MLX, enabling it to run on Apple Silicon. Simon Willison successfully ran it on his M5 Max MacBook Pro by cloning a repository and executing a script. The process involved downloading approximately 115 GB of model files, and the video generation took just under 45 minutes. The resulting video was impressive, but the audio was described as weird speech-like garbage, because Willison did not provide prompt guidance for the audio. He noted that the prompting guide contains extensive information on how to achieve proper audio generation, which he had not read prior to the experiment.
- Abbreviations
- MLX = Machine Learning eXchange — формат обмена моделями машинного обучения
Source: Simon Willison —
original