FireRedAudio 音色库（1–8）· T8star-Aix

This is the unglamorous node that makes the whole pack's multi-voice pipeline work. FireRedAudio 音色库 doesn't generate anything, doesn't load a model, doesn't touch your GPU. It just gathers between one and eight voice profiles into a single VoiceBank object that every other character-aware node can consume.

If you've ever fought with a TTS workflow where every character is a separate hard-coded branch, this is the node that kills that mess. One bank object in, and the script parser, batch dubbing and delivery chain all know who's who.

How it fits

The flow looks like this:

VoiceProfile (1–8 of them) → VoiceBank → ScriptParser → BatchDubbing

Each VoiceProfile carries a reference clip, its transcript, language and a name - the "character." VoiceBank bundles those into one named roster, and downstream nodes route script lines to characters by name. It's the roster, not the cast. The actual voice cloning happens elsewhere (the TTS node); this is the data structure that keeps eight characters from turning into spaghetti.

Inputs and outputs

The single input, profiles, is an auto-growing set of 1–8 VoiceProfile connections - the node adds input slots as you drag more profiles in. There's a rule: every profile in the bank must have a unique name, because downstream script routing matches on exactly that name. Two profiles both named "旁白" and the bank will reject the second.

Outputs are 音色库 (the T8_FIREREDAUDIO_VOICE_BANK object you wire forward) and 音色库 JSON (a STRING of the same thing). The JSON output is more useful than it looks - it's how you save a cast to a file and reuse it in another session or even another workflow, instead of re-dragging eight reference clips.

Installing it

Same pack, same two steps as the rest of comfyui-fireredaudio-T8:

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

Technically you don't need the model download for this node - it does no inference - but the pack is useless without models, so grab them now if you haven't:

python scripts\download_models.py --target "D:\ComfyUI\models\TTS\FireRedAudio" --profile full --variant int8-wo-safe-v1

Gotchas

The honest version: this node is a container, and beginners sometimes wire it up, hit run, and wonder why nothing happened. Nothing happens here - the bank is inert until it reaches the script parser and batch dubbing chain, and those are where the actual generation happens. Also: it's capped at eight profiles. For a bigger cast you'd have to build a second bank or get creative, and honestly eight voices covers a lot of narration before you should question your life choices.