The method
The hand writes. The machine assists.
Every Ingvar Steinberg release follows one discipline: the song is decided before a note is rendered. Lyrics are human-authored and dated from 1992, arrangements are roughly ninety per cent manual, and mastering is done by ear. AI is a production tool, used under direction and disclosed — never the author.
Order of work
From the archive to the master.
Choose the verse
The starting point is the archive — a corpus of lyrics begun in 1992, alongside hundreds of arrangement and voice studies. The song is selected, not generated. Around ninety per cent of the lyric already exists; the work here is completing missing lines and refining.
Set the arrangement
Instrumentation is matched to the verse — grand piano, violin, guitar riffs, drums and rhythm — with deliberate attention to transitions, interludes, breaks and ad-libs, the moving parts that make a song hold. The draft is then revised, typically twenty to forty times.
Cast the voice
The right voice is found and shaped across as many as fifty passes — splitting, recasting and directing until the delivery matches the song as written. Each act carries its own register, from Romantik's saturated baritone to Vespertide's heavy attack.
Master by ear
The finished take is balanced and mastered to a fixed standard, so the record that leaves the room is the record that was intended — no more, no less.
The stance
The seams are shown, on purpose.
It would be easy to present a voice as a person and say nothing. This practice does the opposite. Every release states plainly what its voice is: a real performer whose recorded vocals are extended in production — as with Lexi and Rho recording as Vespertide — or a voice cast and directed under the Steinberg name.
The reasoning is simple. A dated, human-authored songbook is a rare asset, and honesty about the tools is what keeps it credible. Disclosure is not a disclaimer here; it is the reason to trust the work.
On iteration
Forty passes is discernment, not fishing.
The high pass-counts are often the first thing people notice. They read one of two ways — someone brute-forcing a result they can't picture, or someone rejecting take after take against a target that already exists. It is the second. The archive is the proof: the standard was set long before the tools arrived to meet it.
Questions
On the method.
How much of the arrangement is done by hand?
Roughly ninety per cent. Instrumentation, transitions, interludes, breaks and dynamics are decided by the composer. AI extends recorded vocals and speeds sketching — it never authors the song.
Why do the songs go through so many iterations?
Because the target already exists. Arrangements are revised twenty to forty times and voices cast across up to fifty passes — selection against a fixed standard, not a search for one.
Where exactly does AI fit in?
AI extends recorded vocal performances and accelerates arrangement sketches, under explicit direction. Lyrics, melody, arrangement and mastering are the author's own work. Each release states what its voice is — a real performer extended, or a voice cast under direction.