riff-api 0.1.2

Riffusion API

Riffusion API

This repo contains examples for crafting music from text prompts using the Riffusion API. The API is minimal and blazing fast, but also exposes powerful controllability features.

You can either use the typed Python client in this library or make requests directly.

This API is in a private beta and subject to change.

Contact [e-mail] for inquiries and access.

Getting Started

Set your API key:

export RIFFUSION_API_KEY="<your-api-key>"

Install the Python client:

pip install riff-api

Make some music:

import riff_api

riff_api.generate_from_topic(
    topic="Indie pop banger about my dog Boris",
    save_to="output.wav",
)

Have fun 💙

Here's a sample creation: output.wav

Topic Mode

The /topic endpoint is the easiest way to create music. The input is a single natural language description of both the sound and lyrical content you desire.

Sample inputs:

• "Indie pop banger about my dog Boris"
• "Explain the concept of time in French, piano chill"
• "Sing facts about Kansas history"

Run via Python client, this time saving the response:

import riff_api

response = riff_api.generate_from_topic(
    topic="Indie pop banger about my dog Boris",
)
print(response.lyrics)

riff_api.save_audio(response, "output.wav")

The audio, lyrics, and sound prompts are returned in the response. The audio bytes are base64 encoded in the response.

Alternatively, run directly via request:

import base64
import os
import requests

response = requests.post(
    "https://backend.riffusion.com/v1/topic",
    headers={
        "Content-Type": "application/json",
        "Api-Key": os.environ.get("RIFFUSION_API_KEY"),
    },
    json={
        "topic": "Indie pop banger about my dog Boris",
    },
).json()

with open("output.wav", "wb") as f:
    f.write(base64.b64decode(response["audio_b64"]))

Or via curl:

curl -X POST \
  -H "Content-Type: application/json" \
  -H "Api-Key: ${RIFFUSION_API_KEY}" \
  -d '{"topic": "Indie pop banger about my dog Boris"}' \
  https://backend.riffusion.com/v1/topic \
  | jq -r .audio_b64 | base64 -d > output.wav

If you don't describe a musical style in your prompt, the API will choose randomly for you. The output is not deterministic, so you can run the same prompt multiple times to get different results.

The schema of the entire API is defined in datatypes.py.

Custom Mode

The /custom endpoint gives lower level control over the generation. You specify custom lyrics and a list of sound prompts with individual strengths and time ranges.

Run via Python client:

import riff_api

lyrics = """
Hello from outer space
Can you hear me?
I'm a satellite
And I want to be by your side
""".strip()

response = riff_api.generate(
    prompts=[
        Prompt(text="chillstep pop"),
    ],
    lyrics=lyrics,
)

riff_api.save_audio(response, "output.wav")

Run via request:

import base64
import os
import requests

response = requests.post(
    'https://backend.riffusion.com/v1/riff',
    headers={
        'Content-Type': 'application/json',
        'Api-Key': os.environ.get("RIFFUSION_API_KEY"),
    },
    json={
        'prompts': [
            { "text": "chillstep pop" },
        ],
        'lyrics': (
            "Hello from outer space\n"
            "Can you hear me?\n"
            "I'm a satellite\n"
            "And I want to be by your side"
        ),
    },
).json()

with open("output.wav", "wb") as f:
    f.write(base64.b64decode(response["audio_b64"]))

Examples

The examples directory explains use cases of the API. To run, first pip install riff-api then execute with Python.

• 1_simple.py - Use lyrics and a sound prompt to create music.

• 2_instrumental.py - Pass no lyrics to create instrumental music.

• 3_variations.py - Hold a constant seed while modifying other parameters to create variations.

• 4_multiple_prompts.py - Use two sound prompts with different strengths to fine tune the output.

• 5_transition.py - Use two sound prompts with start/end times to create a transition.

• 6_audio_formats.py - Change the audio format of the response.

• 7_timestamped_lyrics.py - Print the timestamp of each lyric from the generation response.

• 8_topic.py - Use a single natural language sentence to generate music with the topic endpoint.

Streamlit Demo

demo_app.py is a basic web app that demonstrates using the api to create musical transitions using two sound prompts with time ranges.

Try it: https://riffusion-api-demo.streamlit.app/

Or run locally:

pip install streamlit
python -m streamlit run demo_app.py