Keep whatever carries the performance, and re-record whatever carries fidelity. In practice that means the lead vocal moments, the tempo as it was actually sung, and any printed sound the song grew around usually survive, while drums, bass and the main harmonic bed get cut again properly. I produced and mixed Stefania for Kalush Orchestra and I spend a large share of my production weeks in Burbank rebuilding songs that arrived as a voice memo or a writing-session bounce. The job is to find the four bars of it that made everyone sign up for the song, and build the record under them.
Why does the demo keep winning?
Because the artist has heard it two hundred times and the new version twice. Engineers call the condition demoitis, and Production Expert's piece on fixing demo-itis without losing the artist's trust describes the mechanism honestly: repetition turns every accident of the writing night into a feature, so the cleaner take reads as a wrong take. If a flat guide vocal beats a tuned final for everyone in the room, the guide contains information, and my job is to find out what it is. Usually it is one of three things: a tempo that breathed, a phrasing choice the singer could not repeat on command, or a cheap sound that glued the arrangement together.
So the first hour of a rebuild is an audit, with the demo printed on a reference track at matched loudness, because a demo bounced 6 dB hotter wins every comparison for reasons that have nothing to do with the arrangement. I level both to the same integrated loudness before anyone votes.
What usually deserves to survive?
Vocal moments first. Lead fragments, ad-libs, a cracked word in the last chorus, the breath before the first verse. These are the hardest things to re-create, because the singer on writing night was discovering the melody and the singer at the final session is reciting it. The famous version of this decision is Strawberry Fields Forever: the released master joins two different takes, recorded in different keys at different tempos, because Lennon wanted the start of one and the body of the other. If that splice was worth the varispeed gymnastics in 1966, importing a demo vocal into a new session needs no apology.
Printed sounds second. The detuned synth from a stock plugin, the phone-recorded piano, the loop the topline was written over. A sound the melody was composed against is part of the composition, and replacing it with a better-recorded cousin often unglues the hook. I keep the original print, clean it, and layer the upgrade underneath instead of swapping.
Tempo and key third, as data. The demo's BPM, its drift, and the key the singer actually sat in are decisions someone already made with their body. I extract them before judging them.
How do I set the new session up around the demo?
The demo goes in first, at bar 1, and the session grid bends to it rather than the other way around. I tempo-map the new session to the demo, so it plays in sync with the grid from minute one. If the writing take drifted from 96 to 99 BPM into the choruses, the map keeps the push, and the band I record later plays to a click that breathes the way the writing night did. Ableton's manual on audio clips, tempo and warping covers the mechanics of pinning audio to a grid, and every major DAW has an equivalent.
Then the demo becomes two things in the session: a muted reference track at the top, leveled to the working mix, and a folder of extracted parts, vocal fragments, printed sounds, labeled and colored as demo material. Every milestone, end of production, end of mix, gets an A/B against the reference at matched loudness. The question in the A/B is whether anything the demo did well has gone missing. The folder discipline that makes this painless later is the same one from how I prepare a mix folder.
When does the demo vocal become the final vocal?
When the delivery is unrepeatable and the recording is rescuable, and I test both instead of feeling it. Repeatability gets tested at the re-record session: I ask for three passes of the line with the demo in the singer's ears, and if the third pass still reads as a performance of the demo rather than the song, the demo take wins. Rescuability is a technical checklist. A demo vocal survives if the noise floor sits low enough that 6 to 8 dB of broadband cleanup leaves no pumping, if nothing clipped on the way in, and if the room on it is small enough that the new space I print around it, the approach from how much space a lead vocal needs, can own the depth. Tuning is the least of the problems; the amount of correction a lead takes is its own decision, covered in how much pitch correction a lead vocal actually needs.
What I do not do is keep a damaged take out of sentiment and then spend the mix hiding it. A kept demo vocal has to sit in front of the new production, at full level, in the first chorus. If it only works tucked and drowned, it did not survive, and the honest move is a comp: demo words where the magic lives, new takes everywhere else, matched with the same chain, the way I build any comp in vocal comping in two passes.
What gets re-cut every time?
The rhythm section. Demo drums are almost always a loop or a two-mic afterthought, and drums carry more of the fidelity impression than any other element, so they get recorded or programmed properly against the tempo map. Bass gets re-cut with it, because the demo bass was played to the old drums and the pocket moves. Stacked harmonies get re-sung even when the lead survives, since a stack multiplies every noise and tuning artifact by eight. And anything with timing smear, a pad that enters late, a double that flams, gets replaced, because repair tools that could fix it would dull it. The demo stays in the session till mastering, muted, as the record of what everyone fell in love with. I have unmuted it in the last week of a mix more than once, taken one breath or one word from it, and shipped that.
Frequently asked
What is demoitis? The attachment an artist forms to a rough demo after hundreds of listens, which makes every polished re-recording feel wrong. The fix is to treat the demo as source material and carry its best parts forward, instead of asking the artist to forget it.
Can a demo vocal be used on a final record? Often, yes. If the take was recorded reasonably clean, a demo vocal with an unrepeatable delivery can be tuned, de-noised and re-spaced to sit in the new production. Heavy bleed, clipping or a loud room are the usual disqualifiers.
Should the final session copy the demo tempo? Copy the feel, question the number. I tempo-map the new session to the demo first, so everything lines up, then decide deliberately whether the drift and the average BPM were part of the song or an accident of the writing night.
If your demo is better than your mix
Send me the demo and whatever multitracks exist. I will send back an audit: what I would keep, what I would re-cut, and what the kept parts need to survive, with the first chorus rebuilt both ways so you can hear the difference before committing to a production. Credits are at /credits, background at /about.