Audio Master

Normalize loudness (LUFS) and trim leading silence. Video pass-through, audio re-encoded.

Try Audio Master in the playground

Last updated View as MarkdownAgent setup

Audio Master

Node type: utility:audio_master
Category: Editing

Description

Audio Master normalizes the loudness of a video's audio track (EBU R128) and can trim leading silence: connect your clip to the Video port, keep the Preset (default tiktok) or set Target LUFS, True Peak (dB) and Loudness Range to override it, and the result comes out of Mastered Video. The video stream passes through untouched; only the audio is re-encoded, at 48 kHz AAC 320k. Trim Silence acts only on an audio-only source. A source with video is never trimmed, because cutting the audio head would slide every word early against the picture. A source with no measurable audio is refused. When the result misses the target or the ceiling, the node still returns the file and shows a warning on the card, and the result carries measured values such as achievedLufs and converged.

Canvas ports

These appear as port handles on the left side of the node.

ID Label Details
video Video VIDEO (required)

These render as form fields in the right-side config panel when the node is selected.

ID Label Details
preset Preset TEXT · default tiktok. Named loudness preset applied by the mastering service
trimSilence Trim Silence BOOLEAN · default true. Cut leading silence before normalizing
lufs Target LUFS NUMBER · optional. Overrides the preset's integrated loudness target
truePeakDb True Peak (dB) NUMBER · optional. Overrides the preset's true-peak ceiling
lra Loudness Range NUMBER · optional. Overrides the preset's loudness range target

lufs, truePeakDb, and lra are passed through to the mastering service only when set; leave them empty to use the preset defaults.

Outputs

ID Label Type
video Mastered Video VIDEO

The result also carries preset (the preset that was applied) and appliedFilter (the ffmpeg filter string used), useful for debugging.

It also carries the MEASURED result: targetLufs, achievedLufs, achievedTruePeakDb, achievedLra, truePeakOk, converged (within 0.5 LU of the target), correctionDb (gain added on top of the plain normalization gain), ceilingDb (the limiter ceiling that shipped), limitedBy (when it missed: peak-reduction-cap, time-budget, write-passes or failed-pass), trimApplied, trimmedSeconds, and trimSkippedReason when the trim was skipped.

How it hits the target

  1. The source is measured (EBU R128: integrated loudness, loudness range, true peak). A source with no measurable audio is refused; the node never returns a silent file.
  2. Leading silence is trimmed ONLY for an audio-only source. A source with video is never trimmed (trimSkippedReason says so): the video is copied untouched, so cutting the audio head would slide every word early against the picture. For audio-only sources the threshold is 16 dB under the measured loudness but never above -50 dB (true silence, so a quiet intro survives), and the trim is skipped when it would remove more than 3 s or leave under 0.5 s. trimmedSeconds reports what was removed.
  3. The audio gets one linear gain from the measured values. When the source's loudness range is wider than the target's, a range-control stage runs first.
  4. A two-stage limiter holds the peaks under a ceiling below the true-peak target: a fast stage that lets loud targets (-9, -6 LUFS) actually get loud, then an oversampled true-peak stage.
  5. When the limiter is working, the gain is searched: a limiter turns 1 dB of gain into much less than 1 LU, so the node measures and steps until the target lands.
  6. The delivered file is measured. If the encode pushed the true peak over the target, the ceiling is lowered by what was measured and it runs again; if it is more than 0.3 LU off, the gain is searched again. Every try is kept as its own file, and after at most 4 writes the best verified one ships. A try that would not finish inside the time budget is not started; the best file so far ships instead.

When the result misses the target or the ceiling, the node still returns the file and shows a warning on the card, for example "Reached -7.4 LUFS of the -6.0 LUFS target: the audio has too much dynamic range to go louder without distortion."

Audio is written at 48 kHz AAC 320k.

Very loud targets on speech (around -6 LUFS) can need more limiting than the node will apply (24 dB of peak reduction at most). The node then ships the loudest clean result it can, says so with converged: false and limitedBy: peak-reduction-cap, and shows the warning above. A single loud click in otherwise quiet audio does the same.

Notes

  • The video input accepts a bare URL string or a { url } / { videoUrl } / { video } object.

Auto-generated from the Wireflow node registry.

For AI agents: the full documentation index is at /llms.txt, and most docs pages are available as Markdown by adding .md to the URL.

© 2026 Wireflow. All rights reserved.