RIU Music Buy me a coffee DE

Reading · Guitar · Lesson 29

AI Tools and Stems for Guitarists Who Write Their Own Songs

Which apps and AI tools help with songwriting on guitar, where they are unreliable, and in what order to use them without handing over the craft.

776 words4 min read

A stem splitter takes a finished recording apart into its tracks: vocals, guitar, bass, drums, each playable on its own. Play along with the isolated bass and guitar and you hear what the bassist actually plays, where the rhythm guitar drops out, and how little really happens in a verse. Around tools like this a whole field of apps and AI services has grown, and it changes month by month. What stays constant is the question of what they are good for and what they are not.

Six categories, one yardstick

The basic kit needs no AI and remains the most important: a tuner and metronome app, the voice memo for every phrase that sticks, a notation and tab program such as MuseScore or Guitar Pro, and a DAW like GarageBand or Reaper for the demo. The only danger in this category is the most popular way of not finishing a song: the perfect demo.

Generative audio services such as Suno or Udio turn a text into a complete recording. That works for quick reference demos and style sketches, say to hear how your own lyrics would sound as a country song. The results are generic, and the copyright situation is unresolved, with label lawsuits under way; before any release, read the terms of use. Language models such as Claude, ChatGPT or Gemini help with lyric brainstorming, rhyme alternatives, lists of metaphors, translation, looking up theory and analyzing your own progressions. They lean toward cliché lyrics, will not reproduce other people's lyrics for copyright reasons, and their chord information about real songs needs double-checking.

Harmony assistants such as Hooktheory with TheoryTab and Hookpad, Scaler or Captain Chords search progressions in real songs, tell you which chord statistically comes next, and suggest voicings. But statistics is not aesthetics; the most common next chord is rarely the most interesting. Transcription and stem tools such as Chordify, Moises, Demucs or Chord ai recognize chords and separate tracks. With inversions, seventh chords and dense production the recognition is unreliable, and that is exactly what guitar accompaniment is made of: a slash chord with a different bass note, an open string ringing as an add9, a capo that separates the shapes you play from the chords that sound. The ear remains the referee. Backing apps such as iReal Pro, Band-in-a-Box or Moises with tempo and key change deliver backing tracks in any key; their risk is not an AI risk but the temptation to practice chords instead of songs.

Guitar first, then the screen

The order decides whether the tools complement the craft or replace it. Hook and progression come first on the guitar, with no screen, because whatever comes from outside at this stage becomes the song rather than a tool. Then analysis tools serve as a cross-check: look up your progression in Hooktheory and see who has used it before and what they did differently, or ask a language model for alternatives to the third chord and play every suggestion on the guitar before believing it. For lyrics the rule is: ask for twenty images on a theme, not for lines, and ban the clichés explicitly, no heart and pain, no stars. You write the line yourself. A generative tool may show how an arrangement concept sounds; after that it gets built by hand. And stems are for learning: split a familiar song into tracks and play the bass and guitar parts in isolation.

What actually changes

No current tool writes a better song than someone who has understood scale degrees, modal interchange, melody and arrangement. What changes is the speed of the feedback loop: an idea can be heard, compared and discarded within minutes. Use that to test more ideas and you write better. Use it to skip the idea and you write mediocrity with a high gloss. I rarely reach for stem splitters or chord recognition myself: working out a bass line by ear, or at least trying to, is harder and teaches more than the finished track. Backing tracks to play along with are a different matter. And I find the AI tools more interesting as theory than in practice, because on the guitar I mostly want to play, a hundred times rather with real people than with a model; that is what the practicing is for. The lesson lays out the six categories with examples, uses and cautions in a table and sets two tasks: have three reharmonizations of your chorus progression suggested, play them on the guitar and judge them in the course's terms, and split a song into stems, play along with the bass track and compare it with your own degree analysis.

The course lets you hear it: sound examples, diagrams and trainers on this topic are in Lesson 29 · Tools: apps, stems, AI.

Open lesson 29 in the Guitar course

Same topic on the piano: Apps, Stems and AI for Songwriting at the Piano