AI MIDI generators are an emerging software category that cater specifically to songwriters. They produce MIDI material based on user parameters, using neural networks instead of hardcoded algorithms. They are generally available on web browsers, standalone apps, and as DAW plugins.
Music producers favor plugins and native DAW features to web apps. Toggling between a web browser and your DAW rarely makes for a good experience. That's why we've seen a growing number of native applications in this niche in 2024.
YouTube artist Nobody and the Computer created the poetic video below exploring AI powered MIDI generation as a theme.
No generative model? Then it's not AI.
The "AI" prefix in the phrase AI MIDI generator is an important one.
What does it mean for a MIDI generator to be powered by artificial intelligence in 2024? There was a time where the term AI could be used loosely.
Today, many high quality AI audio models exist. The bar has been raised and companies using the expression "AI" owe it to users to offer some transparency.
Here are some of the most popular products in this space today. We were not paid by any of the following companies to promote their service in this article.
Table of Contents
Lemonaide VST: AI MIDI seed generator
Lemonaide is the most musical AI MIDI generator we've seen. As in, the output actually sounds good. During August 2024 they launched a new AI marketplace, where they sell fairly trained models in partnership with hand-selected artists.
The plugin lets you create and collect seed ideas (chord and melodies in MIDI) to assist in the beat making workflow. These seeds might sprout into a new idea or help you build new verses, choruses, bridges, and so forth.
The Lemonaide MIDI generator is coupled with a “seed bank” that stores all of the MIDI creations, so you can refer back at any time in your future DAW sessions.
Lemonaide's founder and team
Lemonaide's founder MJ previously worked as a Google cloud engineer. He set off on a journey and started his company, where he now manages a team of programmers, designers and marketers.
Near the end of 2023, AI researcher Julian Lenz joined the team. Lenz is formerly of prominent AI music companies Okio and Neutone. He brings prior experience and AI expertise to their team. Machine learning engineers Anirudh Mani and Saif Abid are the other two people working on the team.
The updates in 2024 introduced a new suite of models called the Collab Club. Each AI model is trained on hundreds of projects from a single artist, capturing their unique melodic tendencies, chord voicings and expressive nuance.
They include a hefty 40% revenue share with each of the artists, every time a customer purchases that models. In this way, they're not only solving a problem in the DAW but showing the way for other industry players to use AI equitably.
Lemonaide's pricing is currently set at $9.99 for 150 credits and an additional $9.99/mo for each rented model. 8-bar music output in audio & MIDI Loop format. Available as a web browser app or as a DAW VST
Check out Lemonaide's website here to learn more.
HookPad's Aria: Browser + standalone options
We've got a full article dedicated to HookTheory's new Aria feature in HookPad that you can read here. For now we'll share some of the most important details.
HookPad is a songwriting app that loads in browsers or can be downloaded for standalone use on MacOS and Windows devices. In June 2024, the company announced a new feature called Aria that uses a state of the art AI model to generate MIDI compositions based on your existing song.
HookTheory partnered with a research scientist from Google Deepmind, whose open source Anticipatory Music Transformer was trained on ~176k MIDI files under a creative commons license. They fine-tuned that base model on their own database of 50,000+ MIDI transcriptions of popular songs from major artists and composers in every genre.
It's unclear whether they licensed the music that they trained on, since it's user generated content developed by their community over several years. However, the quality of those songs may very well carry over into the quality of the ideas and suggestions that Aria makes when analyzing your own music.
Visit HookTheory's Hookpad page here to learn more.
AIVA: Browser + standalone AI MIDI generator
Skrrrt. Alright let's turn back the clock a few years.
AIVA is one of the most powerful multi-track AI MIDI DAWs online, and they were arguable one of the first to do it. They serve up multitrack MIDI with mixing and effects modules for handling most composition workflows.
AIVA's simplest interface is the one you land on when you first sign up. It's a kind of song panel, where music parameters like key signature and BPM dictate the music you get from the machine.
However, if you dig a bit deeper into the product, it's easy to discover their MIDI DAW. The workstation runs in most modern web browsers (Chrome, Firefox) and also as a standalone DAW on Mac and Windows devices.
AIVA brings a lot to the table, and if they can improve the default instrument quality, it would give them a huge competitive advantage in this space.
Also, a VST would be a nice alternative to the full DAW, as many people prefer to work in their existing audio workstation.
Unison MIDI Wizard 2.0: MIDI generator
Unison’s MIDI Wizard is a chord and melody generation app with several of the core features people look for in a MIDI generator. It can create chord progressions and melodies in a given key signature and genre, at your tempo of choice.
Unison's reputation precedes them. Seasoned DAW producers recognize the Unison brand. Their MIDI chord packs have been online for years, attracting more vocal criticism than praise.
One of the main complaints we've heard about MIDI Wizard is the cost. It retails at $347. By comparison, Captain Plugins (one of the most advanced non-AI MIDI generators) costs $99. Scaler 2 retails for $59 and AudioCipher is $29.99.
So acting when taken in the context of the greater product ecosystem, $347 for a chord and melody generator is very steep.
If you can look past the plugin's price, the MIDI Wizard is nevertheless a visually dynamic, fun and comprehensive product. Here's how it works:
A control panel sits at the top left quarter of the design, with input variables like genre, key signature, chord/melody mode, leads/pluck articulation, and length of the MIDI output.
A belt of audio controls divide the interface. Here you'll find synth effects, volume knob, glide, and several other musical parameters.
A MIDI editor piano roll occupies the bottom half of the app. This is where your MIDI generations appear. Edit the notes as needed, before dragging them into their DAW for further refinement and sound design.
Is MIDI Wizard 2.0 generative AI MIDI?
There has been speculation that MIDI Wizard 2.0 might not be running real AI, however at this time it's still being determined. On the MIDI Wizard’s marketing page, they explain that:
“The software leverages advanced AI tools and machine-learning technologies to analyze and create MIDI data that is technically proficient and musically expressive.“
The MIDI Wizard plugin doesn’t require an internet connection.
Maybe they've built a lightweight AI model that consumes minimal CPU. After all, MIDI file size is considerably smaller and less complex than generative AI audio.
As for training data, my guess is that Unison trained on their own MIDI pack library. This takes a pretty big leap of imagination to believe, as someone who monitors Twitter and LinkedIn closely.
If a company creates new AI MIDI generation models, on par with Google Magenta or Microsoft's MuseCoco, one would expect to see an FAQ page bragging about the tech, their ethically sourced training data, and if not open sourcing the model, at least giving a basic description of what it is.
There are no public records on this matter, so customers are left in the dark. I think this is why people are doubting that MIDI Wizard is truly generative AI.
On the other hand, it sounds like Unison has hired AI experts. That's what their founder Seb claimed in a recent promotional video.
So maybe Unison is just playing it cool and keeping their cards close to their chest. They don't have any legal obligation to be transparent, and it may work to their advantage to say as little as possible while they're still developing the tech.
We now have over 20 people on our team, including many more plugin developers, many more music theory experts, AI experts, and a creative director. - Unison MIDI Founder
At the end of the day, MIDI Wizard is a fun AI MIDI generation app. Let's see how they continue to build on it in the coming years.
WavTool: An AI text-to-MIDI browser DAW
We’ve reported at length about WavTool here on the AudioCipher blog.
This AI DAW runs in a browser, accompanied by an OpenAI ChatGPT assistant. You can talk to the bot as if it's a bandmate or teacher, asking questions and sharing ideas.
The really impressive thing about WavTool is the assistant's ability to take direct actions in the DAW. Ask it for musical compositions in MIDI, or have it build custom wave table synthesizers to achieve a certain sound you want.
Beginners will learn how to create effects chains, through ordinary conversation with the bot, and be able to generate AI audio samples to go beyond MIDI.
Find tutorials in the app to guide you through all of the presets. Check out our full write up about WavTool here.
AudioCipher V4: Text-to-MIDI without AI
AudioCipher is an AI emulator. It imitates artificial intelligence through a home brewed text-to-MIDI algorithm
We released our V1 in 2020, a few years before AI text-to-music took off with MusicLM and Suno. Our original mission was to promote experimentation through a fun and kinda weird DAW plugin.
We were inspired by a centuries-old musical cryptogram technique where words were converted into melodies. Our team built out a custom algorithm that incorporates all keys and chord extensions.
Composers and secret agents used this technique through the 18th and 19th century, but it lost steam during the 20th century. Audiocipher brought musical cryptograms to the DAW, as a VST and standalone app.
Our text-to-midi generator lets your type in a word or phrase and choose melody or chord mode. Pick from nine scales, chord extensions and rhythm settings. Drag MIDI directly into your DAW of choice, including FL Studio, Ableton, and Logic Pro.
AudioCipher version 4.0 was released in September 2024 with a slew of new text-to-MIDI features including note joining and chord inversions for smooth voice leading. Choose your BPM, Key signature, and set it to melody or chord output.
Visit the AudioCipher website to learn more.
AI MIDI models from Google, OpenAI, Microsoft
Google published an early MIDI generator prototype called Magenta Studio, but the quality was not great. OpenAI's MIDI generation at MuseNet was better in our opinion. When ChatGPT launched, the company stopped hosting Musenet as a service. It has since faded from the public eye.
Microsoft Asia's Muzic team built an AI MIDI generation model and published research about it in 2023. The MuseCoco model seems to produce sheet music inspired by classical harmony. No user interface has been deployed by the team.
2023 was a good year for companies like this. Suno was brought into the fold at Bing-Microsoft and Riffusion raised a $4M round.
As recently as March 2024, AI song playback is at near real-time streaming on Suno’s V3 model.
The creative experience is becoming addictive. These AI song apps will become like Midjourney for music. We're on the crest of that wave in the first half of 2024.
Market conditions have favored AI song generators for a few reasons. They are remarkably easy to use. They play into people's desire to create original music, without necessarily knowing how to play an instrument.
These companies sprang forward with low regulatory barriers. Some companies may have scraped copyright protected music for their models, and if that comes to light, lawsuits are certain to eventually follow.
Despite the question of ethics, everyday users are adopting and enjoying the new high quality AI music models becoming available. Apps that turn text prompts into audio and MIDI represent an exciting new development in the industry.
We'll likely to see more developments here over the next 4-12 months.
Comments