Mix your stems in the browser

Mixing stems in the browser: balance, loudness, and renders you can trust

Stems are the most useful format a song comes in — vocals, drums, bass, and music as separate files — and the least convenient to audition. Opening a DAW to answer 'how does it sound without the vocal?' is a heavy tool for a light question.

This page is the light tool: drop the stems, balance them with gain and pan, measure what you made, and render the versions a working set actually needs.

§How to mix stems without a DAW

1. Drop 2–8 stem filesEach file becomes a channel with mute, solo, gain, and pan. Nothing uploads.
2. Balance while playingSolo to inspect, mute to test arrangements, pan to open space. Changes apply live.
3. Measure the blendOpen Meters and run the loudness pass — per-stem LUFS shows which stem really dominates.
4. Save two candidatesStore mixer states as A and B in Snapshots, then flip between them mid-playback.
5. Render what you needExport the mixdown, or open Variant renders for no-vocal and drums-only WAVs in one ZIP.

01Four controls are most of a mix

Gain, pan, mute, solo — the classic channel strip survives because rough balance is most of what a blend needs. Gain decides the hierarchy, pan opens space between instruments that share a register, and mute answers arrangement questions faster than any fader ride.

Everything here applies live during playback and identically in the offline render, so the mixdown you export is the blend you heard, not an approximation of it.

02LUFS: loudness you can compare

The Meters mode measures integrated loudness to the BS.1770 recipe: the signal is filtered to weight the frequencies ears care about most, energy is averaged in 400 ms blocks, and blocks quieter than a gate are discarded so silence doesn't dilute the number. The result, in LUFS, is a loudness figure you can compare across stems and across songs.

Worked example: drums at −20 LUFS and a music bed also at −20 LUFS don't blend to −20 — two roughly independent signals add on power, so the pair lands near −17 LUFS, three louder. Measure the master after balancing: if the blend reads −14.6 LUFS and you want −16 for an even set, pull the whole thing down 1.4 dB and re-render.

03One session, every version

A stem set's real value is the versions hiding inside it. The Variant renders mode renders mute-combinations as separate WAVs — no vocals for the live PA, drums only for an edit bed, no drums for a rehearsal loop — using your current gains and pans, packed into one ZIP.

The tool guesses which stems are vocals or drums from their filenames and enables the matching combos; anything it can't identify still renders in the current-mix variant. Name your stems plainly and the guessing disappears.

04A/B: flip faster than your ears reset

Ears adapt within seconds, which makes slow comparisons worthless — by the time you've re-ridden four faders, you've forgotten what version A sounded like. A/B snapshots stores the complete mixer state twice and flips between them instantly while the audio keeps playing.

The states are written into the page URL as plain numbers — gains, pans, mutes, in stem order, no filenames — so a copied link hands a collaborator both balances. They drop in the same stems, in the same order, and hear your A against your B.

05Where a browser mixer honestly stops

Stems only sum back to the song if they start at the same instant; files exported from different sessions or trimmed by other tools drift out of alignment, and no gain knob fixes time. Hot stems are the other trap: signals add, so a stack of near-full-scale files clips the sum — that distortion is arithmetic, and the fix is headroom, not a different mixer.

There is also no EQ and no compression here, deliberately. This page is for balance and versions; tone shaping belongs in a session.

06This mixer plays stems. The studio makes them.

Everything above assumes you already own the stems. Most songs you care about arrive as one finished stereo file — and splitting that file is not a mixing problem, it is a separation problem. SongStrip in the studio runs GPU stem separation that learned what a voice, a drum kit, and a bassline are, and hands you clean stems from any full mix. Split the song there, then balance it here or anywhere else.