How I build Keyscape piano beds that leave room for breathy dark-pop vocals

Horia Stan is a music producer and sound engineer at The One Records, Bucharest.

I make dark-pop piano beds that don't fight the vocal. I want intimacy and impact in the same take. Sparse arrangements cannot hide sloppy frequency masking. Your piano will either compete with the vocal or support it. I choose support. Here is the exact chain and settings I use in 2026, with Logic Pro, Keyscape, FabFilter, Waves and Valhalla.

Why this matters in 2026

Streaming normalization sits near -14 LUFS for singles. That flattens peaks and makes midrange clarity crucial. Dark-pop vocals sit forward and breathy. If the piano bleeds into 700 Hz to 2 kHz, the vocal loses personality after loudness normalization. Modern listeners have earbuds and ANC headphones. That exposes masking problems. Fix them in the mix, not during mastering.

The sound I aim for

A warm, percussive Keyscape piano close enough to feel breath. A short room tail for intimacy. A parallel saturated low layer for body that never muddies 200-800 Hz. A dynamic carve that ducks only when the vocal needs space. No swimming midrange.

What I use (exact tools)

• DAW: Logic Pro 11 at 48 kHz, 24-bit.
• Piano: Keyscape "Studio Grand" preset with 3 velocity splits. I load one natural, one with EQed top-end removed, one low-shelf enhanced for sub.
• EQ / Dynamic EQ: FabFilter Pro-Q 3.
• Multiband ducking: FabFilter Pro-MB or Pro-Q 3 dynamic band with external side-chain from vocal bus.
• Transient shaping: Waves Trans-X or SPL Transient Designer.
• Saturation: Soundtoys Decapitator or Waves J37 for tape-like glue.
• Resonance control: Oeksound Soothe2.
• Reverb: Valhalla VintageVerb and Space Designer IR for micro-room.
• Stereo imaging: Waves S1 or Ozone Imager.

Recording and instrument setup

I rarely use a single Keyscape instance. I use three instances with different roles.

1. Main - close, crisp, full spectrum. Velocity split 1-64.
2. Texture - rolled-off highs, heavy mid to add motion. Velocity split 65-120.
3. Sub - low-focused, low-pass at 600 Hz, for presence on small speakers.

I set each instance to a different MIDI velocity curve. Small changes in velocity create timbral distance without layering different samples. That yields an organic bed without chorus-like phasing.

Layer processing - exact chain per instance

Main instance chain (insert on channel):

• FabFilter Pro-Q 3: high-pass at 28 Hz. Surgical cut at 3.8 kHz if clash occurs. Add 1.8 dB shelf +8 kHz for sparkle when needed.
• Waves Trans-X: reduce attack 12-18% on percussive notes to make space for consonants.
• Oeksound Soothe2: target 1.2-2.2 kHz with depth 40%, select 'instrument' curve.
• Valhalla VintageVerb: small plate, decay 1.2 s, mix 9%.

Texture instance chain:

• Pro-Q 3: low-pass 6 kHz, boost 200-350 Hz by 1.5 dB to add warmth.
• Decapitator: Drive at 2-3%, Tone set to 'A' for smoothness. Mix 30%.
• Space Designer IR: micro-room impulse, gain -6 dB, mix 12%.

Sub instance chain:

• Pro-Q 3: low-pass 600 Hz, boost 70-120 Hz +2.5 dB.
• Gentle compressor 2:1 ratio, 15 ms attack, 120 ms release.
• Send to the main bus only when below -18 dB on RMS to avoid muddiness.

Dynamic carving - the non-obvious move

Most producers static-EQ the piano away. I duck dynamically with frequency precision.

• Route the vocal bus as an external side-chain to FabFilter Pro-Q 3 on the Main piano channel.
• Create a dynamic bell at 900 to 1.5 kHz. Set range -6 dB. Set threshold so the band reacts on consonants. Attack 3 ms, release 60 ms.
• Duplicate that band on the Texture instance but opposite polarity for rests - this is subtle and optional.

This frees vocal presence during words and lets the piano breathe between lines. It prevents over-EQing the piano while keeping its energy.

Transient and consonant strategy

Consonants live in 2.5-6 kHz and need transient space. I reduce piano attack around 3-6 kHz by 1.5-3 dB with a narrow Pro-Q 3 dynamic band, triggered by the vocal side-chain. Then I add transient reduction with Waves Trans-X on the main piano to soften the initial hit.

Do not kill the piano's percussive identity. The goal is to reshuffle attack energy away from consonants, not remove it.

Stereo and depth

Keep the sub instance mono. Widen the main and texture highs by 20-40 ms Haas with very low level using a stereo delay or Ozone Imager.

Insert FabFilter Pro-Q 3 on the piano bus in M/S mode. Reduce mid 300-800 Hz by 0.8-1.2 dB, leave sides intact. That moves body to mono while giving side information high harmonic content. The vocal stays centered and untouched.

Reverb and tails

I use two sends.

• Short send to Valhalla VintageVerb plate - decay 1.1-1.6 s, pre-delay 18-22 ms, mix 12%.
• Long send to Space Designer IR with an isolated small concert hall impulse, decay 2.4 s, low-pass at 3.2 kHz on the return. Mix 4-6%.

Automate the long send up in instrumental breaks, down under verses.

Automation and micro-moves

Automate piano level -3 dB under main choruses if the vocal doubles. Add a micro gain dip -0.8 dB over consonants when the vocal is doubled and heavy.

Use Logic Pro's automation write briefly during a comped run. I do not rely on static processing alone.

Final bus processing and loudness

On the piano bus I use:

• FabFilter Pro-MB: gentle compression on 60-200 Hz to keep dynamic low-end neat. Band gain -1.5 dB when peaks at bus exceed -6 dBFS.
• Decapitator: 2% drive on bus for glue.

Export stems at 48 kHz, 24-bit. Mix target: -14 LUFS integrated for streaming. Keep peak ceiling -1 dBTP.

Quick checklist

Common mistakes I see

• Over-EQing the piano across the entire midrange. You kill tone. Use dynamic carving instead.
• Widening low frequencies. That ruins mono compatibility.
• Relying on a single Keyscape patch. Small timbral offsets make space without EQ.

Concrete takeaway

Next session, load Keyscape Studio Grand into three slots. Set velocity splits at 64 and 121. Insert FabFilter Pro-Q 3 on the main channel. Add a dynamic band at 900-1.5 kHz with external side-chain from your vocal bus and range -6 dB. Play the track. If the vocal reads clean on earbuds at -14 LUFS, you are done.