This is a portable version of Tencent's SongGeneration for Windows. It allows you to generate high-quality music using a pre-trained model, without needing to install dependencies globally.
You can download the latest portable .7z archive from the official release page:
🔗 EasySongGeneration_WinPortable.7z
-
Extract the Archive
- Unzip the
.7zarchive to a folder of your choice.
⚠️ Important: To avoid extraction errors, make sure you’ve installed the latest version of 7-Zip before opening the archive. - Unzip the
-
Clone the Model Repository
-
Make sure you have Git LFS installed.
-
Open a terminal and run only one of the following commands depending on the model you want:
-
Open a terminal and run:
# SongGeneration-base (10G/16G) git clone https://huggingface.co/tencent/SongGeneration- Copy the following folders from the cloned repo into the root of the extracted folder:
ckptthird_party
-
- If you want to use a different model, download one of these into the
ckptfolder and if you want to free up space you can delete thesonggeneration_basefolder inckpt:# SongGeneration-base-new (10G/16G) git clone https://huggingface.co/lglg666/SongGeneration-base-new # SongGeneration-base-full (12G/18G) git clone https://huggingface.co/lglg666/SongGeneration-base-full # SongGeneration-large (22G/28G) git clone https://huggingface.co/lglg666/SongGeneration-large
-
⚠️ Disable Flash Attention only if you experience issues with it- Open the file:
ckpt\songgeneration_base\config.yaml
⚠️ Note: if you are using a different model changesonggeneration_basewith the folder you downloaded.- Go to line 39 and change:
to:
use_flash_attn_2: true
use_flash_attn_2: false
- This disables Flash Attention, which may cause issues on some systems.
- Open the file:
-
Run the Application
- In the root folder, execute the following command:
.\tools\gradio\run_embedded.bat .\ckpt\songgeneration_base\
⚠️ Note: if you are using a different model changesonggeneration_basewith the folder you downloaded. - In the root folder, execute the following command:
SongGeneration is a text-to-audio model developed by Tencent, capable of generating songs with vocal-instrument harmony or dual-track separation.
- Hugging Face: SongGeneration
- GitHub: tencent-ailab/SongGeneration
For more details, check out the official demo.
Huge thanks to mingyi456 for helping resolve dependency issues and making this setup smoother!
- Make sure Git LFS is properly installed before cloning.
- If Flash Attention is enabled, the app may crash or behave unexpectedly.
- Run the
.batfile from the root directory to avoid path issues.
@article{lei2025levo,
title={LeVo: High-Quality Song Generation with Multi-Preference Alignment},
author={Lei, Shun and Xu, Yaoxun and Lin, Zhiwei and Zhang, Huaicheng and Tan, Wei and Chen, Hangting and Yu, Jianwei and Zhang, Yixuan and Yang, Chenyu and Zhu, Haina and Wang, Shuai and Wu, Zhiyong and Yu, Dong},
journal={arXiv preprint arXiv:2506.07520},
year={2025}
}