Studio knowledge · AI in music production
AI in music production: a hybrid workflow for songwriters
A tool can suggest possibilities. You decide which ones serve the music.
Generative AI has changed how people sketch and produce songs. For songwriters and producers, the question is increasingly less about whether to use it and more about how to use it with musical judgement and control. Here’s a studio perspective.
Chapter 1
From the initial shock to practical use
Musicians’ responses to AI have shifted quickly. Here are four ways to understand that shift.
The shock
Suno and Udio made it possible to generate complete songs with vocals and arrangements in a click. Musicians responded with both curiosity and concern.
“Will musicians still be needed?”
The controversy
Attention turned to training data. The use of protected material prompted lawsuits and protests, followed by some licensing agreements.
“Who gave permission for our songs to be used in training?”
Reality in the studio
Phase issues, artefacts and predictable arrangements become apparent when you try to edit or mix the material. A quick export is only the start.
“It sounds great until you listen to the tracks separately.”
Hybrid production
Musicians can use AI where it helps and rely on their own skills where the output needs work.
“I use the sketch and make the song my own.”
A complete-sounding demo still needs production decisions
Reverb can hide rough consonants. A chorus that sounds full as a stereo mix can fall apart when you separate the tracks. Listen closely before building an arrangement around generated material.
The rights landscape is evolving
Debates over training data continue. Universal and Udio announced a licensing agreement in 2025, as did Warner and Suno. Check the specific tool, its terms and how you plan to use the material.
Key insightTreat the output as a draft and check whether it will hold up in your production.
Chapter 2
Five perspectives on AI in music
Musicians bring different priorities to the debate: expression, efficiency, income, fairness and experimentation. Many people share more than one perspective.
Traditionalists
They value lived experience, performance and the imperfections of an actual recording. They choose to work without generative AI.
Practical adopters
They use AI for specific jobs such as sketches, stem separation or variations, as they might use any other studio tool.
People concerned about livelihoods
Composers in advertising, television and stock music are especially concerned about price pressure from automated production.
Advocates for fairness
They call for transparency about training data, clear labelling and fair compensation for rights holders.
Experimenters
They explore code, open models and source separation as a way to test new musical ideas.
Key insightYou can experiment with a tool and still question how it was trained. Decide which uses fit your values and your music.
Chapter 3
The major debates
Money, rights and authenticity are all part of the discussion.
Permission to use training material
The use of protected recordings and songs to train models has raised questions of consent and compensation. Some licensing agreements are in place, and questions remain about individual models and uses.
Streaming at scale
Mass uploads and bot activity can put pressure on how royalties are distributed.
Deepfakes and cloned voices
A synthetic voice can make a real artist identifiable. Permission and transparency matter when a person’s voice is involved.
Pressure on paid work
Inexpensive generated background music can change the opportunities available to composers entering the industry.
Training rights and release rights are separate
Whether a model could use recordings for training is a different question from whether you can release a specific output. Consider similarities to existing songs and any rights attached to voices or recordings. Changing instruments or mixing again does not settle these questions.
A voice can identify a person
Work with documented consent and agreed usage terms for an identifiable voice. The EU AI Act addresses transparency for certain AI content and deepfakes. Check the requirements for the specific use.
Key insightKeep a record of the materials, tools and permissions behind the track before you release it.
Chapter 4
What’s changing in songwriting
When you can generate ideas quickly, deciding which ones belong in the song takes more of the work.
Recording-led
Generative
Choosing what matters
Generating sounds is easier. Selecting a musical idea that connects with people still takes judgement.
Directing the production
When routine tasks are automated, the writer’s choices about story, emotion and sound shape the result.
Arranging on your own
A solo songwriter can sketch arrangements that would once have required a group of session musicians to try out.
Give the song a clear direction
Twenty versions of a chorus won’t tell you which one works. Decide what the song should do. Should the verse feel close or distant? Where should the tension lift? Use those choices to compare versions.
Use guides to prepare a performance
A generated demo can help you hear arrangements and discuss them with musicians. You decide what to keep and what to record again.
Key insightJudge each idea by what it does for the song’s story.
Chapter 5
Six steps for a hybrid DAW workflow
Use AI to sketch or explore, then edit, arrange and refine the production in your DAW.
Generate an idea
Write a short brief covering the theme, mood, form and tempo. Use generation to explore specific ideas you can develop.
Separate and audition the stems
Split the stereo mix into drums, bass and vocals. Listen to each stem on its own for bleed and artefacts.
Convert audio to MIDI
Extract MIDI from a suitable track and play it with your own software instruments. Check the detected notes, particularly in dense material.
Check timing and feel
Find out where the timing actually drifts, then adjust what needs attention. Make timing decisions by ear and in the context of the groove.
Record the parts that matter
Use a guide vocal to explore the melody and phrasing, then record a performance that suits the track.
Give the song a recognisable identity
The story, choices and performances you bring to a song help give it a sound of its own.
Before releasing: Replaying a melody with your own instruments changes the recording. Check where the melody came from and what you’re allowed to use.
Key insightListen to every track, keep what serves the song and record the parts that need your own performance.
Chapter 6
Looking ahead
More hybrid tools
DAWs may integrate more generative tools into the same workflow as manual editing.
More attention to permissions
The EU AI Act sets transparency requirements for certain AI content. Licensing models are evolving. Keep track of the origin of your materials and the permissions you have.
Different uses for different music
Functional background music may become more automated. A distinctive voice and clear artistic choices can help a song stand out.
AI can give you material to work with. Your choices and performances shape the song.
What do you want the tool to contribute? Which decisions will you make yourself?
Which tracks help the song, and which are covering up a problem?
Can you trace the materials, contributors and permissions?
Key insightYour musical identity comes through in what you write, perform and choose to keep.
Frequently asked questions
Do I need AI in my workflow?
Choose the approach that fits your music, your standards and the people you work with. Working entirely by hand is a valid choice.
Is an AI-generated track ready to release?
An export can work as a demo. Before releasing it, check the sound, arrangement, artefacts and terms of use.
Does stem separation recover the original tracks?
It estimates the components of the mix. Bleed and artefacts may remain, so treat the results as working material.
Do I own the melody if I replay it?
A new sound alone does not settle copyright or third-party rights. Consider where the material came from, your own contribution and the applicable terms.
How I work
I compose on piano, guitar or in the DAW. I sometimes use AI for specific tasks such as rhyming or translation.
If you have a Suno or Udio demo, I can record a new production with musicians and vocals.
Would you like help producing your song?
Tell me about your idea and I’ll send a free preview and a quote.