Sidechain Ducker

Ducking is the audio move you've heard a thousand times without knowing its name: when a voiceover comes in, the music behind it quietly dips out of the way, then swells back when the voice stops. In every DAW it's a routing and sidechain affair. MKRSidechainDucker does the same thing as a single ComfyUI node - two audio inputs in, one ducked mix out - which means you can build it into a generation pipeline instead of babysitting it in another app.

What it's for

You've generated a music bed and you've generated (or recorded) a voiceover, and you want the final audio to sound like a real edit rather than two tracks fighting. Wire the voice into main_audio, the music into sidechain_audio... wait, flipped: main_audio is the track that stays prominent and sidechain_audio is the one that triggers the ducking. Common setup: main_audio = music bed, sidechain_audio = voiceover, so the music ducks under the voice. The voiceover itself stays untouched.

How it works

Both inputs are wildcard (file, waveform, or MKR_AUDIO payload). It decodes the sidechain, resamples it to the main track's sample rate, and pads/truncates to the same length. The sidechain envelope drives the gain: whenever the sidechain level crosses threshold_db (default -40 dB), the main track's gain drops by duck_db (default -12 dB), with attack_ms (15) and release_ms (180) shaping how fast it dives and recovers. Slower release = smoother, less "pumping" sound.

Inputs and outputs

• main_audio - the track that gets ducked (your music bed).
• sidechain_audio - the trigger (your voiceover).
• duck_db - how much gain reduction, default -12 dB.
• threshold_db - how loud the trigger must be before ducking starts.
• attack_ms / release_ms - envelope shaping.
• output_format, filename_prefix, subfolder, overwrite, optional filename_label.

Outputs: audio (MKR_AUDIO payload), output_path, duration_sec, summary.

Installing it

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/criskb/MKRShift_Nodes

restart ComfyUI, or Manager → MKRShift_Nodes. WAV output works without extra tools; mp3/flac/ogg encoding needs ffmpeg on PATH.

Common issues

The classic beginner stumble is mixing up which track ducks - if the voiceover is the one getting quiet, you've swapped the inputs. If the duck sounds like a pumpy mess, raise attack_ms a little and increase release_ms; if the music never ducks at all, the voice isn't crossing threshold_db - lower it. And remember the sidechain is resampled to the main track's rate and clipped to its length, so wildly different-length inputs just get truncated rather than aligned.