小珠光音频保存

Core ComfyUI still doesn't have a first-party audio save node worth the name - you're usually hunting through whatever VHS or audio pack you happened to install. 小珠光音频保存 (Xiaozhuguang Audio Save) is the tidy answer from the 小珠光 suite: it takes an AUDIO tensor, writes it to output/ as MP3, WAV, or FLAC, and - the genuinely nice part - has a preview mode that encodes to temp/ instead, so you can audition audio in a pipeline without polluting your output folder with 40 test files.

It's the natural end of the line after the pack's audio loader or its AudioDiT TTS nodes: generate speech, wire the AUDIO in, pick a format and bitrate, run.

How it works

The node hands the waveform to FFmpeg and lets it do the encoding. MP3 gets the bitrate from the 质量 (quality) dropdown (320 / 192 / 128 kbps); WAV and FLAC are lossless, so the quality setting is ignored. In save mode (模式 = 保存) the file lands in output/ with your 文件名前缀 plus a counter; in preview mode (模式 = 预览) it lands in temp/ with a random tag and is flagged as preview. Either way it returns waveform peak data so the on-node display can draw the result - the front end shows you the actual encoded audio with a playable waveform, and has a "send to audio loader" right-click that routes the file straight back into the pack's loader node for round-tripping.

The node needs a working ffmpeg, same as the loader: the pack depends on imageio-ffmpeg, which ships a static binary, with a system ffmpeg as fallback. If neither exists you'll get a clear "FFmpeg not found" error rather than a silent failure.

The inputs that matter

• 音频 (Audio) - the AUDIO tensor to save. No connections, nothing happens.
• 格式 - mp3, wav, or flac.
• 质量 - bitrate for mp3 only.
• 文件名前缀 - filename stem, default xzg-audio.
• 模式 - 保存 (save to output) or 预览 (preview to temp).
• 音量 - preview-only gain; it doesn't touch the file you save.

It's an output node (no outputs), which means it's one of the endpoints the ComfyUI cache works backward from - if you remove it, nothing downstream of it runs.

Install

From the ComfyUI-xiaozhuguang pack - Manager (search "ComfyUI-xiaozhuguang"), or:

cd ComfyUI/custom_nodes/
git clone https://github.com/xiaozhuguang/ComfyUI-xiaozhuguang.git

Restart after. If the save fails with an ffmpeg error, pip install imageio-ffmpeg in your ComfyUI environment.

Gotchas

The two things that bite people: preview files in temp/ get cleaned up - if you close ComfyUI after a preview, the file is likely gone, so flip to save mode before you rely on a file being there. And the preview-only 音量 widget doesn't affect the saved file; a workflow that previews at 0.5 and saves at full volume will surprise you if you forget. Save mode is the one that writes what you hear at 1.0.