song-ace-step-1.5 / README.md
arraypress's picture
Model card
f7c7615 verified
|
Raw History Blame Contribute Delete
1.49 kB
---
license: mit
base_model: ACE-Step/Ace-Step1.5
pipeline_tag: text-to-audio
tags:
- music-generation
- mlx
- ace-step
---
# ACE-Step 1.5 for `song`
An unchanged copy of the files from
[ACE-Step/Ace-Step1.5](https://huggingface.co/ACE-Step/Ace-Step1.5) at revision
`19671f406d603126926c1b7e2adc169acbcade22` that the on-device `song` tool reads
(Swift + MLX, [swift-music-generation](https://github.com/arraypress/swift-music-generation)).
It is kept here so the tool keeps working if the upstream repository moves or changes.
Every file is byte-identical to upstream; `song` checks SHA-256 on download.
Only the 23 files the engine loads are included; the Python modelling code is not.
## What is inside
| Folder | Part | Upstream licence |
|---|---|---|
| `acestep-v15-turbo/` | DiT (turbo, 8 steps) and its silence latent | MIT (ACE-Step) |
| `acestep-5Hz-lm-1.7B/` | 5 Hz planning language model | MIT (ACE-Step) |
| `vae/` | Oobleck audio VAE | MIT (ACE-Step) |
| `Qwen3-Embedding-0.6B/` | text encoder | Apache-2.0 (Qwen) |
Licence texts are in `licenses/`. All credit for the model goes to the ACE-Step
team and, for the text encoder, the Qwen team.
## Citation
```bibtex
@misc{gong2026acestep,
title={ACE-Step 1.5: Pushing the Boundaries of Open-Source Music Generation},
author={Junmin Gong, Yulin Song, Wenxiao Zhao, Sen Wang, Shengyuan Xu, Jing Guo},
howpublished={\url{https://github.com/ace-step/ACE-Step-1.5}},
year={2026},
note={GitHub repository}
}
```