MiniMax H3 Voice Library Load / 加载音色 (EXP/T8)

This is the other half of the voice-library pair, and it's about as simple as a node gets. MiniMaxH3VoiceLibraryLoadT8 takes a library_name string and returns the saved voice_profile - the same object MiniMaxH3VoiceLibrarySaveT8 wrote to your ComfyUI user directory earlier. Saved a voice once, and now you can drop it into any future workflow with a single node instead of rebuilding the profile and its reference audio from scratch.

It exists because workflow memory doesn't survive a restart. A Voice Profile node only exists in the graph while the graph is open; a library entry lives on disk. So the routine looks like: build the profile once, Save it by name, then in every subsequent workflow stick a Load node, type the same name, and the voice (and its cleaned 32kHz reference audio, if it had one) comes back ready to feed your dialogue chain.

How to use it

One input, library_name (default my_voice - match it to whatever you named the entry on Save). Two outputs: voice_profile and report_json. The profile feeds straight into whatever consumes a voice profile in the speech/dialogue workflow; the report is there if you want to confirm what was loaded and from where.

What you should know before you rely on it

• It's a lookup by name against local storage. Misspell the name and you get a load failure with a report, not a silent fallback. Good behavior - you want to know you grabbed the wrong voice.
• Local means local. The library lives in the ComfyUI user directory on this machine; it won't follow you to another install unless you copy it.
• There's no "refresh" concept - it reads the current saved state, so if you re-save the profile with a newer version, Load picks up the new one automatically (as long as you allowed the overwrite on the Save side).

Installing:

part of the MiniMax H3 Audio T8 pack - Manager, search "MiniMax H3 Audio T8", or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/comfyui-minimax-h3-audio-T8

Restart ComfyUI. No extra pip packages, no auto weight download. And the standing caveat applies: H3's local weights are license-restricted in the US/EU/UK/Korea, so make sure you're entitled to run them before building a voice pipeline around this pack.