FireRedAudio 交付预设 · T8star-Aix

Every delivery platform has a loudness spec, and they're all different - audiobooks sit around −20 LUFS, podcasts −16, broadcast video dialogue −23, with True Peak limits to match. If you've ever typed -16 LUFS into a settings panel and hoped, you know why this node exists. T8_FireRedAudio_DeliveryPreset packages the whole loudness-and-format decision into one dropdown.

What it is

One input, preset_name, with three choices: audiobook, podcast, and video_dialogue. Each one bundles a coherent delivery spec that the pack applies downstream - the README spells them out: audiobook defaults to −20 LUFS / −3 dBTP / FLAC, podcast to −16 LUFS / −1 dBTP / WAV, video_dialogue to −23 LUFS / −1 dBTP / WAV. The preset also carries arrangement (how clips are laid out), crossfade behavior, sample rate, and save format. You get two outputs: the preset object (wire it into TimelineRender or a save node) and a JSON description of what it actually contains, because "just trust me" is not a delivery spec.

The two-pass EBU R128 normalization with True Peak limiting is applied by the downstream renderer - this node is the configuration object, not the effect itself. The report records the actual parameters used, so your delivery package has the numbers in writing.

When to reach for it

This is the node that sits between your finished audio and the platform. If you're making an audiobook chapter, pick audiobook and stop thinking about loudness. Podcast - podcast. Voiceover laid under video - video_dialogue. It's also the consistent choice: when you render fifty episodes, using the same preset object every time means they all come out at the same loudness, which is what actually makes a series listenable.

Inputs and outputs that matter

• preset_name - the only input. Defaults to audiobook.
• Outputs: delivery_preset (the T8_FIREREDAUDIO_DELIVERY_PRESET object) and preset_report (details as JSON).

That's the entire interface, which is the point. It connects to the pack's TimelineRender node, which accepts the preset and applies arrangement, crossfades, R128 normalization, and format on export.

Install

Standard pack install, once for the whole family:

cd ComfyUI\custom_nodes
git clone https://github.com/T8mars/comfyui-fireredaudio-T8.git
cd comfyui-fireredaudio-T8
python scripts\setup_runtime.py

No model needed - this is pure configuration. You only need the FireRedAudio model when something upstream generates audio.

Gotchas

The preset is a starting point, not a magic fix: loudness normalization (EBU R128) makes your average level consistent, but a badly clipped take stays clipped. Run quality QA on the individual lines before you normalize the master - you can't polish a clipped peak into existence. And if your client hands you a spec that's not one of the three (some platforms want −14 LUFS for web), you'll have to pick the closest and adjust downstream; the preset doesn't do custom targets. For the three mainstream delivery jobs it's right, and right is worth a lot when you're exporting at 2 a.m.