These weights are governed by the Stability AI Community License, not by mlx-audiogen's Apache-2.0 licence — that covers the conversion code only. The full agreement is in LICENSE.md; attribution and the statement of modifications are in NOTICE.

Key terms, summarised (the agreement governs):

Free for research and non-commercial use.
Commercial use requires registration at https://stability.ai/community-license.
The grant terminates above USD $1,000,000 annual revenue.
Stability AI's Acceptable Use Policy applies to outputs.
You may not use the model or its outputs to create or improve another foundational generative AI model.

The upstream repository is gated. If you have not already accepted the licence there, please do so — this mirror exists for MLX convenience, not to route around Stability AI's terms.

Modifications

These weights are a Derivative Work: the upstream checkpoint converted to the MLX array format for inference on Apple silicon. Specifically, convolution weights are transposed where MLX's layout differs from PyTorch's, tensors are re-serialised as safetensors, and the model is re-implemented against MLX rather than PyTorch.

No weights are retrained, fine-tuned, or otherwise altered in value.

Usage

bash

pip install "mlx-audiogen[server]"
mlx-audiogen --model stable_audio --prompt "ambient pad" --seconds 10

The weights download automatically on first use.

jasonvassallo/mlx-stable-audio

mlx-stable-audio (Stable Audio Open small, MLX)

Licence

Modifications

Usage

jasonvassallo/
mlx-stable-audio