MuLaCover

MuLaCover

Makes an AI cover song from a reference track, your own lyrics and a one-line style description, in a Gradio demo. Its weights and outputs are for noncommercial use only.

Made by hugging-apps on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: MuLaCover is a Gradio demo for controllable AI cover songs and music remixes. You provide a track to follow, lyrics of your choice and a short line describing the style. The model reads the melody, chords and drums of the reference, then generates new audio and decodes it into a 48 kHz WAV file. It runs on GPU hardware at Hugging Face and carries an MCP label.

Who it suits: Music hobbyists and researchers who want to hear a song redone in a new style with new lyrics, for noncommercial use.

Industry | Who uses it | How they use it | Example
Music | Hobbyist musicians | Hear their lyrics sung over a reference tune | Give a reference clip, lyrics and a style line
Research | Music AI researchers | Try a controllable generation pipeline | Compare outputs for different style lines
Software | Developers | Study how transcription, style encoding and generation are chained | Read app.py
Education | Music technology students | See how melody and chords guide generation | Try the bundled example clips

Anyone who plans to sell or release the output commercially must skip it or obtain separate authorization.

Where to use it:

• A browser, by visiting the Space on Hugging Face
• A computer of your own that runs Windows, macOS or Linux and has Python 3.10
• An MCP client, since the Space carries the MCP label, although the source gives no connection steps

Use case | Who | What they get back
Make a cover song | Musicians | A 48 kHz WAV in the style you describe
Remix a track | Producers | A new take that follows the reference melody

How to set it up:

• Python 3.10 and pip on Windows, macOS or Linux
• The Python packages in requirements.txt
• The system packages in packages.txt, which Hugging Face installs with apt-get
• Gradio 6.17.3, the version the Space uses
• A GPU, since the Space is set to zero-a10g

1. Install the Hugging Face hub package.
2. Download the Space with the hf command.
3. Move into the new folder.
4. Install gradio at the version the Space uses, then the listed packages.
5. Start the app with Python.

pip install -U huggingface_hub
hf download hugging-apps/mulacover-demo --repo-type space --local-dir mulacover-demo
cd mulacover-demo
pip install gradio==6.17.3
pip install -r requirements.txt
python app.py

How to use it: Online, visit the Space on Hugging Face, add a track to follow and your lyrics, describe the style in a short line, then generate. On your own computer, run the start command and open the local page Gradio serves.

python app.py

Tool | What it does
Reference song | Supplies the melody, chords and drums to follow
Lyrics | The words you want sung in the new version
Style prompt | One line that describes the sound you want
examples | Holds the bundled reference excerpts

Example: You pick one of the bundled 45-second reference clips, write a few lines of lyrics, and describe the style in one line. The demo returns a new audio file that follows the clip's melody with your words.

For newcomers and experts: If you are new, begin with a bundled example clip so you know what a reference should look like. Then change only the style line and listen to how the result shifts.

If you know the field, the pipeline works in stages. YourMT3 and ChordNet turn the reference into notation, Qwen3-Embedding encodes the style text, and a 3B llama backbone produces HeartCodec audio tokens that are decoded at the end. The trade-off is heavy GPU use and a noncommercial license.

You give a song, lyrics and a style
Melody, chords and drums are read
The style is encoded
Audio tokens are generated
A 48 kHz WAV is decoded

Good to know:

• Hardware: the Space is set to zero-a10g
• Example clips: two 45-second excerpts by Josh Woodward, licensed CC BY 4.0
• License: cc-by-nc-4, with extra terms in the model repository
• Cost to run: GPU hardware, no API key or paid service is named
• Data it reaches: the songs and lyrics you give it, plus the models it fetches from Hugging Face
• Not for: commercial use of the weights or the generated output, unless MuLa Labs authorizes it separately

On Hugging Face, 13 November 2026

• Likes: 47
• Built with: gradio
• License: cc-by-nc-4.0

Links

• Space: https://huggingface.co/spaces/hugging-apps/mulacover-demo
• Files: https://huggingface.co/spaces/hugging-apps/mulacover-demo/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.