Artificial intelligence is changing the entire art industry, it has even learned to write music based on a text request. AI systems analyze world classics, classify tracks and create something of their own on this basis. There are many questions in society about whether these tracks are original, whether they violate copyright or not. One thing is clear: the neural network opens up new opportunities for musicians, producers and just amateurs.
Let’s look at the most interesting AI services for generating music, their features and operating principles.
The most popular neural network for writing songs and music. Generates melodies based on a text request, supports many languages and even creates covers for tracks. You can fine-tune the structure of a song by writing verses, choruses, and instrumental parts in the prompt. With good settings, the AI can write a full-fledged 2-minute track.
New users get 50 free tokens per day. This is enough for 10 songs: 5 requests with two track options in each. Developers often give away additional tokens for various reasons, to celebrate the release of an update or to compensate for errors.
- Where to try : udio.com
- Difficulty : 🎵🎵🎵
Powerful AI for creating musical compositions, works with text queries. At the time of text preparation, the neural network is available in beta mode. Songs are created from words written in the prompt. Free users receive 10 credits daily and 100 additional credits monthly. One credit allows you to generate two versions of a track, similar to Suno.
The user can write detailed verses or make a request in the spirit of: “A song about how to become rich without doing anything.” The neural network itself will come up with the name of the track, lyrics and motive. In our example, everything is clear – “Broken Dreams.”
During the test, it seemed that Suno copes better with abstract requests, the text conveys the meaning and feelings more accurately, and selects the timbre and style better. But if you want to fine-tune the musical accompaniment to the finished words, then Udio will cope better.
In paid packages for $10 you can write tracks up to 2 minutes, the monthly limit is increased to 1200, priority queue, cover generation. For $30 everything from the previous plan, but the limits are higher.
Riffusion
- Where to try: com
- Difficulty: 🎵
Suitable for creating short 12-second musical compositions. To create a song, you need to enter a short text query. For example, specify the genre, style, female or male vocals. If this is not specified, the AI will select the settings on its own.
The service’s special feature is that you can split a finished composition into separate audio tracks: vocals, drums, bass and other instruments. Riffusion will give you 3 tracks, they can be downloaded in mp3 or mp4 formats with cover art.
Play
Mubert
- Where to try: com
- Difficulty: 🎵🎵
Writes background music, suitable for video content, for example, in conversational videos for the background, so as not to get a strike for copyright. You can also write an introduction to the podcast. The prompt must be written in English, you can set the duration, select the type of generation: track, loop, mix and jingle.
After registration, you are given free energy to create 25 tracks per month, up to 25 minutes long. To download, you must specify the channel where you plan to use the music.
Play
Soundful
- Where to try: soundful.com
- Difficulty: 🎵🎵
Easy-to-use AI for creating music. After registration, a short questionnaire of 2 questions will appear: what is your specialty, what genres are you interested in. After that, they will give you templates with genres and subgenres. You need to choose one, adjust the speed settings, tonality, give the track a name and the system will generate a melody.
Not the most flexible neural network, but if you don’t want to bother, Soundful’s template system is great. The free plan is limited to creating and downloading 10 tracks per month for personal use. A paid subscription for $10 per month removes these restrictions.
Screenshot CQ.ru / Soundful generation interface
Avia.AI
- Where to try: aiva.ai
- Difficulty: 🎵🎵🎵
The system has a library with genres, styles and user-created content. Based on this, the neural network learns and writes tracks in the user’s chosen key. It differs from other AIs in its built-in track editor. It can be used to locally adjust the sound of instruments, tempo, and effects.
For free, you can generate and download 3 tracks up to 3 minutes long. For 49 euros, the monthly limit increases to 200 tracks and as a bonus, you get the copyright to use the music for commercial purposes, such as advertising, films, video games.
Advertising
Play
Stable Audio 2
- Where to try: com
- Difficulty: 🎵🎵
Specializes in creating instrumental tracks without vocals. Works with text queries, you can upload a reference or choose a ready-made preset by genre. When training Stable Audio 2, licensed data was used, which means that all generations are legal and can be used for commercial purposes. In the free version, you can generate up to 10 compositions per month.
Play
Endel
- Where to try: io
- Difficulty: 🎵
A startup that generates sounds in real time to help you focus on completing tasks or music for sleep. Endel is more about relaxation and meditation, rather than writing tracks from scratch in a neural network. You can use it for free for 7 days, then for 280 rubles per month or 3490 per year.
The Endel algorithm takes into account various factors: time of day, weather, user’s heart rate. Based on this data, the system generates a personalized sound background.
Play
So-vits-svc
- Where to try: research.google.com
- Difficulty: 🎵🎵🎵🎵
So-vits-svc is a powerful tool from the Chinese for working with vocals. The main feature is the ability to realistically imitate the voices of famous performers, but the system does not create music from scratch.
The tool is very complex to work with, requires skills in handling code, knowledge of English and a lot of blanks. Among them: original a cappella, instrumental track and a sample of the selected artist’s voice. In So-vits-svc you can generate covers from the series “Dzhigurda sang a song about school”, but you should remember that using other people’s voices is unethical.
Play
MusicLM
- Where to try: google-research.github.io
- Difficulty: 🎵🎵🎵🎵
Google has created a music generation system MusicLM based on text descriptions. It can create tracks both for short queries, like “pop rock”, and for detailed descriptions “a song about how a neighbor in a dorm ate the last cutlet, I was very sad, hard rock, male vocals”.
Despite its impressive capabilities, MusicLM is not yet available to the general public. Google is concerned about copyright issues, as the system was trained on existing music. However, the company has opened up the database to developers who want to create their own music AI.