Music Flamingo on Replicate

Cog wrapper for NVIDIA Music Flamingo - a state-of-the-art Large Audio-Language Model for music understanding.

Features

• Deep music understanding across songs and instrumentals
• Theory-aware Q&A (harmony, structure, timbre, lyrics, cultural context)
• Chain-of-thought reasoning
• Long-form song reasoning (up to ~10 minutes)

Usage

import replicate

output = replicate.run(
    "jhurliman/music-flamingo:latest",
    input={
        "audio": open("song.mp3", "rb"),
        "prompt": "What genre is this track? Describe the instruments and production style."
    }
)
print(output)

Local Development

# Install Cog
brew install cog  # or: curl -o /usr/local/bin/cog -L https://github.com/replicate/cog/releases/latest/download/cog_$(uname -s)_$(uname -m)

# Run a prediction locally (requires NVIDIA GPU with ~80GB VRAM)
cog predict -i [email protected] -i prompt="Describe this track"

# Push to Replicate
cog push r8.im/jhurliman/music-flamingo

License

Model weights are under NVIDIA OneWay Noncommercial License (research use only).