凤希AI - 多音频加载

FxAiMultiAudioLoad (凤希AI - 多音频加载) is a routing node with one trick: take up to seven audio inputs, choose one by number, and pass it through. That's the whole job, and it's the kind of plumbing node that looks pointless until you're mid-workflow and need to A/B two voice lines or drive a multi-shot render from a single selector without redrawing wires.

Here's the pattern it exists for. In a long-video workflow you often have several candidate audio tracks - a narrator, a character voice, a music bed, a source clip's audio - and you want to switch between them per segment or per experiment. Rewiring audio into the sampler each time is tedious and error-prone; instead you wire all seven into this node once, then drive the 编号 input. Because 编号 is forceInput, you can feed it from a loop counter or another node's output, which turns this from a manual switcher into an automated per-segment audio selector. The tooltip says it plainly: "from 1, choose the audio number to output."

The inputs: 编号 (INT, starts at 1) and 音频一 (required - it's the guaranteed fallback), with 音频二 through 音频七 optional. Outputs: 音频 (the selected track) and 编号 (the number, passed through). If the number is out of range or garbage, it silently falls back to 音频一, which is a deliberate "never crash the graph" choice - but it means a typo gives you the wrong audio rather than an error, so check the number if the wrong voice comes out.

Mechanically there's nothing to get wrong: it's a seven-way match on the index, pure and simple. No resampling, no trim, no level adjustment - what goes in comes out, byte for byte. If you need a real audio mixer (multiple tracks at once), this isn't it - that's the pack's audio merge node. This one is a selector, and it's honest about it.

Install: ComfyUI Manager → search "fxai-toolkit", or cd ComfyUI/custom_nodes && git clone https://github.com/fxai666/fxai-toolkit, restart. The pack auto-installs soundfile and psutil at startup and wants FFmpeg on PATH for the audio paths that actually process audio. All labels are Chinese (音频一…音频七), because 凤希AI builds for a Chinese-speaking community - support lives in QQ group 775649071 and on Bilibili, and there's no English Reddit presence to lean on.

It's a two-minute node with a narrow job, so don't overthink it: wire your candidate tracks in, wire a counter or a widget into 编号, and you've got per-segment audio switching with zero graph edits. If your number is always picking the same track no matter what you set, check that you're feeding 编号 rather than leaving it as the fallback - and remember it's 1-based, which trips up everyone raised on zero-indexed arrays.