⭐ Star Video Sound Enricher

If you've generated video with LTX-2 or LTX-2.3, you know the sound. The audio is genuinely synced and genuinely thin - compressed, fizzy on top, scratchy in the 2–6 kHz region, with no low end to speak of. Community reviews have been calling it "heavily compressed" since launch, and the model's newer versions only partially fixed it. ⭐ Star Video Sound Enricher is the band-aid: a mastering-style filter chain, tuned specifically for the typical AI-video soundtrack, that you drop between your video node's audio output and your save/preview node.

It's part of the StarNodes pack, and unlike a lot of "audio fixer" nodes it's not a magic model doing something unknowable - it's a real EQ chain with real knobs, which means it behaves predictably.

How it works

The chain, in order, is a mini mastering session:

24 Hz highpass (rumble removal) → de-harsh bell cut at the scratchy region → low-mid warmth bell at 300 Hz → low-shelf bass boost → high-shelf cut to tame the fizzy top → gentle tanh saturation for harmonic richness → dry/wet blend → resample → peak normalize.

Two details are worth knowing. First, it never downsamples: a 48 kHz input stays 48 kHz, and anything below 44.1 kHz gets resampled up to 44.1 kHz via polyphase resampling. Second, every EQ corner is clamped below Nyquist automatically, so feeding it a 16 or 24 kHz track won't produce garbage - it just re-clamps your settings.

Presets and the one knob that matters

Pick a preset and you're basically done:

• Cinematic Warm (default) - balanced de-harsh + bass + warmth. The safe first try.
• Smooth & Soft - gentler across the board.
• Voice Clarity - focused on cutting harshness without sculpting bass; best for dialogue.
• Deep Bass Boost - for music, where the low end is the whole point.
• Custom - unlocks all the knobs: harsh_freq / harsh_cut, high_cut_freq / high_cut_db, bass_freq / bass_boost, warmth, and drive (the saturation amount, 0 = off).

But the single most useful control is intensity - a dry/wet mix that applies to every preset. Set it to 0 and you get the original sound back; nudge it to 0.5 for "fixed but not surgical." That's your "I don't want to over-process" dial, and it's why you shouldn't fear the presets. normalize (on by default) peak-normalizes to -1 dBFS so you don't clip on the way out.

The input is an AUDIO dict; the output is an AUDIO dict, so it slots between any video/audio node and your audio preview or save.

Installing it

ComfyUI Manager, search Starnodes, install, restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/Starnodes2024/ComfyUI_StarNodes
cd ComfyUI_StarNodes
pip install -r requirements.txt

It sits under ⭐StarNodes/Video. The only real dependency is torchaudio, which ships as part of ComfyUI's standard torch install - if your environment somehow lacks it, the console will tell you.

Gotchas

• It needs an audio input with an actual waveform - feeding it a node that didn't produce audio errors out with "audio input has no waveform."
• The presets are one-size-fits-most, not magic. Dialogue-heavy clips want Voice Clarity; a music bed wants Deep Bass Boost.
• If your audio comes out of the Star LTXV 2.5 All-in-One node, you don't need this node at all - its sibling, Star Video Sound Enricher Option, bundles these same settings and lets the all-in-one process its own audio internally. This standalone version is for every other source.

A mastering chain won't turn an LTX-2 soundtrack into a studio mix, but it will turn "unlistenable" into "fine," which is honestly most of the battle.