How it's made
Methodology & transparency
Mi. isn't “type a prompt, get a song.” Every track runs through an integrated workflow of hundreds of AI-assisted tools — agents that diverge, score, challenge, and refine — explored and evaluated fresh for every Mi. session. We study structure, not expression, and we show our work. Here's what happens between your story and your song.
The workflow
- 01
Squeeze the Lemon — one story, many auditions
Your recorded session is the starting point. The Squeeze the Lemon agent wrings it for every way your story could be told — a wide spread of creative angles instead of stopping at the first good one. Your single idea becomes a set of auditions: different sounds, tones, and lyrics, so you can hear what lands instead of guessing.
- 02
The CIS challenge — break the formula
Before a single note, those angles run through the CIS method from the BMAD method — a creative challenge that interrogates what actually makes a song work. It questions the lyric, the structure, even the goal of the track itself, so your song never settles into a familiar formula. This is where the art gets pushed further than a template ever could.
- 03
The Emotion Universe — feeling, in full resolution
Most tools flatten a lyric into a sentiment — happy, sad, glad, love, hate. Feeling doesn't work that way, and neither do we (aka “Wi.”). Mi. places every line and section into the Emotion Universe: a vector database of thousands upon thousands of distinct human emotions. Not happy, but the giddy disbelief of a long-shot win. Not sad, but saudade — the ache for something you're not sure you ever had. Not glad, but the slack-shouldered relief after dread finally breaks. Not love, but limerence, the obsessive vertigo of its first days. Not hate, but the cold, settled contempt that's stopped bothering to shout. A song's emotion shifts line to line — we map the whole arc in high resolution.
- 04
Color theory for the ear — one feeling, a palette of sound
Once a section's true emotion is located, we don't just grab the obvious matching sound. We borrow from color theory — the harmony models a designer uses to build a palette. Analogous: neighboring feelings, for a cohesive, deepening mood. Complementary: the true opposite, for deliberate tension — a tender lyric carried by a hard, defiant sound. Triadic and split-complementary: three- and four-point palettes that give a track real range. Applied to emotion, these models surface sound directions a human would struggle to map by ear — and that spread becomes your auditions. A soft, melodic reading and a hip-hop banger reading of the very same words can both go on the table; the moment you hear them back, the winner announces itself.
The sounds come from FingerPrints — the architectural DNA of music (tempo, key, form, mood, production aesthetic), studied as structure, never sampled, never named. - 05
Keep what lands — you're not limited to one
Most people can't pick just one audition. Good — we (aka “ Wi”) don't make you. Every take that lands is yours to keep. (The revision round included in your package focuses on one track; extra tracks can be polished with a Mi Revision.)
- 06
The Tuning Fork — A&R that refuses “good enough”
Once you've chosen, the Tuning Fork agent steps in as your A&R. It challenges the track — pushing the revisions and edits that take it from “good” to fully landed, past what you'd expect a song built from your own words could do.
- 07
Finish & publish — your track, eight ways
Every finished track is mastered into eight level-matched sound profiles — Fire, Clarity (the house default), Universal, Tape, Natural, Spatial, Cinematic, and Punch — so you and every listener can pick the master that fits the moment, whether that's studio headphones, a phone speaker, or a car. Because they're all matched to the same loudness, switching never jumps the volume. Then it's given cover art, aligned for karaoke, and published with its methodology in full — the same transparency you see on this page.
That's the product: not a song from a prompt, but an integrated, ever-evolving workflow of hundreds of AI-assisted tools — all aimed at the one track that's unmistakably yours.
About the inspiration
Every Mi. track is generated with AI from a documented analysis of a genre's structural elements — tempo, key, time signature, vocal range, mood, production aesthetic. Mi. publishes that analysis with each track. Any resemblance to a specific recording artist's voice or composition is a coincidence of methodology, not imitation.
Mi. commits to publishing the methodology with every track, and to responding in good faith to any rights-holder claim.
Why this matters
AI music carries real legal questions. Mi. doesn't hide from them — we lead with the methodology so listeners, artists, and the legal system can see the work. Receipts of study, not theft.