🎛 W1 Hit

An AI-powered MIDI rhythm generator with Ableton Live integration

Music is defined as a sound that repeats.
Rhythm shares the same definition and thus is the basis of all music.

🎥 Watch W1 Hit in Action

🇯🇲 Example songs created with W1 Hit:
Riddim 1 Riddim 2 Riddim 3 Riddim 4 Riddim 5 Riddim 6

W1 Hit is an AI system for generating expressive single-voice drum patterns. It trains on MIDI rhythms and produces musically coherent variations, fills, and new sequences that can be inserted directly into Ableton Live using a Max for Live device.

Designed for producers, sound designers, and researchers, W1 Hit focuses on controllable generation rather than randomness — allowing users to shape output using parameters such as variation and temperature.

✨ Features

🧠 Trained on single-voice MIDI drum patterns
🎚 Generate new rhythms and variations
🥁 Velocity-aware output (not just note placement)
🎛 Parameter-controlled generation
⚡ Real-time Ableton Live integration
🎹 MIDI-native workflow (no audio preprocessing)

🧠 How It Works

W1 Hit trains a Temporal Convolutional Network (TCN) on quantized MIDI grids of 16th notes, 8 bars each, representing a single drum voice. During inference, the model generates new patterns conditioned on:

• training data distribution
• input samples
• user-controlled parameters

This enables musically coherent results suitable for production workflows.

🎛 Use Cases

• Drum pattern generation for producers
• Creating variations of existing grooves
• AI-assisted composition tools
• Symbolic music generation research
• Max for Live instrument development

📂 Project Structure

W1_Hit/
│
├── Max MIDI Effect/                  # Ableton integration. If you're an artist, you only need this folder
│   └── W1_Hit/                       # Root Directory for W1 Hit. Copy this to the Max MIDI Effect folder on your computer
│       ├── w1_hit.amxd               # Max for Live device
│       ├── w1_hit.js                 # Node for Max interface
│       ├── clipcmd_to_live.js        # MIDI insertion logic
│       ├── install/                  # Folder containing dependencies and utilities to install them
│       │   ├── install_mac.command   # Installs python dependencies for mac
│       │   ├── install_win.bat       # Installs python dependencies for windows
│       │   └── requirements.txt      # List of python dependencies
│       │
│       └── w1_hit_infer/             # Inference engine
│           ├── inference.py          # Runs inference on the model in the Loader dir. The output is sent back to a MIDI clip in Ableton Live
│           ├── hit_generator.py      # Contains the model definition. Basically a stripped down version of hit_generator.ipynb
│           ├── mid_to_velocity.py    # Used to load MIDI files used as seed data for inference
│           ├── Loader/               # Dir to store the model. inference.py loads the first model it finds in this folder
│           ├── MIDI/                 # Stores MIDI files to be used as training data and as seed data for inference
│           └── random_midi_input.py  # Generates MIDI data for use as the seed input to inference, in place of MIDI files
│
├── HitGenerator/                # Training pipeline. For the more technical minded
│   ├── hit_generator.ipynb      # Jupyter notebook containing model definition. The main function can be run to train the model
│   ├── mid_to_velocity.py       # Used to load MIDI files into the training set. Each voice in the MIDI is a new sample
│   ├── midi_export.py           # For testing inference. Generates a MIDI file from the models output data
│   ├── Models/                  # The output directory for trained models.
│   ├── rename_midis.ipynb       # A Jupyter notebook used as a utility for renaming midi samples to uuids
│   └── generated_midis/         # Output directory for MIDIs generated by inference
│
├── requirements.txt
└── README.md

🛠 Installation

1. Clone the repository

git clone https://github.com/david-a-campbell/W1_Hit.git
cd W1_Hit

2. Install Python dependencies

pip install -r requirements.txt

(Recommended: use a virtual environment)

3) Ableton Live Setup

1. Open Ableton Live
2. Go to User Library → Presets → MIDI Effects → Max MIDI Effect
3. Copy the entire folder: Max MIDI Effect/W1_Hit/ from this repository into that directory

⚠️ Important:
Copy the folder itself, not the files inside it.
The final path on your system should be: User Library/Presets/MIDI Effects/Max MIDI Effect/W1_Hit/

📌 Required Folder Structure:
After installation, the device file should be located at: .../Max MIDI Effect/W1_Hit/w1_hit.amxd

Thats all you need to run the Ableton Plugin if you want to use this as an artist. Below are steps to train your own AI model:

🚀 Quick Start Train a model

Place MIDI files into the training dataset folder (called MIDI) and run:

python HitGenerator/hit_generator.py

Trained models will be saved in:

HitGenerator/Models/

Copy your trained model to:

Max MIDI Effect/w1_hit_infer/Loader

Generate patterns: Inference is handled by:

Max MIDI Effect/w1_hit_infer/inference.py

This script loads a trained model and generates MIDI patterns.

Use inside Ableton Live:

1. Insert the Max for Live device on a MIDI track
2. Trigger generation
3. Generated patterns are inserted into the MIDI clip

🎚 Generation Parameters

W1 Hit exposes controllable parameters to shape output:

• Variation — degree of deviation from input pattern
• Temperature — randomness vs determinism
• Iterations — Refinement passes
  These allow producers to dial in results that fit their groove.

🤝 Contributing Contributions, issues, and suggestions are welcome.

🎧 About
Created by W1NGY — a project exploring AI-assisted music production tools.