---
title: stable-audio-3-medium
canonical_url: "https://www.modelscope.cn/models/mlx-community/stable-audio-3-medium"
md_url: "https://www.modelscope.cn/models/mlx-community/stable-audio-3-medium.md"
repository: mlx-community/stable-audio-3-medium
last_updated: 2026-11-12
license: other
model_type:
  - stable_audio_3
base_model:
  - stabilityai/stable-audio-3-medium
base_model_relation: finetune
parameters: 1.9B
tensor_type:
  - F16
  - F32
  - BF16
library_name:
  - mlx
  - safetensors
  - pytorch
frameworks:
  - pytorch
downloads: 11
stars: 0
tags:
  - mlx
  - mlx-audio
  - text-to-audio
  - audio-generation
  - music
  - sound-effects
---

# stable-audio-3-medium

> stable-audio-3-medium - mlx-community 在 ModelScope 开源的模型。stable-audio-3-medium

mlx-community/stable-audio-3-medium 是 ModelScope 魔搭社区上的 1.9B 参数机器学习模型，采用 other 许可，基于 stabilityai/stable-audio-3-medium 构建。

- **Repository**: mlx-community/stable-audio-3-medium
- **License**: other
- **Parameters**: 1.9B
- **Base model**: stabilityai/stable-audio-3-medium
- **Tags**: mlx, mlx-audio, text-to-audio, audio-generation, music, sound-effects
- **Downloads**: 11
- **Stars**: 0
- **Last updated**: 2026-11-12

Source: https://www.modelscope.cn/models/mlx-community/stable-audio-3-medium

---

# stable-audio-3-medium

**Powered by Stability AI**

Stable Audio 3 medium in MLX format for [mlx-audio](https://github.com/Blaizzy/mlx-audio). It generates stereo 44.1 kHz music and sound effects from a text prompt, up to 380 seconds: a T5Gemma text encoder, a 24-layer diffusion transformer sampled in 8 ping-pong steps, and the SAME-L decoder. The diffusion transformer is stored in float16, the decoder in float32.

## Use

Stable Audio 3 support in mlx-audio is in review and ships in its next release.

```bash
pip install -U mlx-audio

python -m mlx_audio.music.generate \
    --model mlx-community/stable-audio-3-medium \
    --caption "Warm lofi hip hop with dusty drums, mellow Rhodes piano and deep bass, 85 BPM" \
    --duration 30 \
    --seed 11 \
    --output lofi.wav
```

```python
from mlx_audio.music import load

model = load("mlx-community/stable-audio-3-medium")

for result in model.generate(
    text="Warm lofi hip hop with dusty drums, mellow Rhodes piano and deep bass, 85 BPM",
    duration=30,
    steps=8,
    seed=11,
):
    print(result.audio.shape, result.sample_rate)
```

`--cfg-scale` defaults to 1 (guidance off). A larger value enables `--negative-prompt` and doubles the model calls per step.

## Source

Converted from the MLX export in [stabilityai/stable-audio-3-optimized](https://huggingface.co/stabilityai/stable-audio-3-optimized), revision `30cb2644da54bfa63275d1bc7f1fa67fc718680f`, with `model_config.json` and the text encoder from [stabilityai/stable-audio-3-medium-base](https://huggingface.co/stabilityai/stable-audio-3-medium-base). The conversion script, `mlx_audio/music/models/stable_audio_3/convert.py`, comes with medium support in mlx-audio. `config.json` and `model_config.json` are the medium-base model config, from which mlx-audio reads only the model sizes and sample settings. See `NOTICE` for what changed.

## License

The model is licensed under the Stability AI Community License (`LICENSE.md`). If you distribute it or a derivative work, you must provide a copy of that license, keep the attribution in `NOTICE`, and display "Powered by Stability AI". Use is subject to Stability AI's acceptable use policy, and the license sets terms for commercial use.

The T5Gemma text encoder in `t5gemma-b-b-ul2/` is subject to the Gemma Terms of Use (`LICENSE_GEMMA.md`, [ai.google.dev/gemma/terms](https://ai.google.dev/gemma/terms)), including the use restrictions in their Section 3.2 and the [Gemma Prohibited Use Policy](https://ai.google.dev/gemma/prohibited_use_policy).