FireRedAudio 保存音频 · T8star-Aix

The basic flow in this pack ends at exactly one node: SaveAudio. You generate a WAV somewhere upstream, this saves it to ComfyUI/output/fireredaudio/ in WAV, FLAC, MP3 or OGG, and - the part other save nodes don't give you - it registers a native audio player and download entry in ComfyUI's results area. That last bit matters more than it sounds: for a pack that lives and dies on listening to output, a clickable in-canvas player beats hunting through the output folder every time. It's the output node the whole "TTS → save" chain is built around.

How it works

Nothing fancy under the hood - a safe file write into the output directory with a filename_prefix, subfolder, and a audio_format combo. The audio input is any standard AUDIO socket (from the pack's TTS, SeedAudition, SpeechEdit, TimelineRender, or anything else producing AUDIO). Two things make it nicer than a bare file dump:

• The player/download UI: 已保存音频 passes the audio back out (so you can chain multiple save targets off one generation), and the node's UI shows a native player for the file it just wrote.
• delivery_preset (optional): wire in one of the pack's delivery presets and it overrides audio_format. That's how you get "save this as FLAC because the audiobook preset says so" without setting the format twice.

The inputs you'll actually touch

• audio_format - wav/flac/mp3/ogg, default wav. FLAC for lossless archiving, MP3/OGG when the file is going somewhere that doesn't like big WAVs.
• filename_prefix - default fireredaudio. A timestamp and unique suffix get appended, so you won't silently overwrite earlier saves.
• subfolder - default fireredaudio, under the output root. Keep projects in their own subfolders and your output directory won't become a landfill.

Outputs: 保存路径 (the full path, useful for wiring into anything that needs to know where the file lives) and 已保存音频 (passthrough).

Where it fits

The README's basic workflow is Load Audio → 零样本声音克隆 (reference) + 模型/隔离运行时 (model) → 生成参数 → TTS → this node. It's also the shared sink for A/B comparisons - the 16_synchronized_ab example sends the two synced candidates to two SaveAudio nodes so you can listen to both side by side and keep both files. If you're comparing, give each save node a distinct filename_prefix or you'll be sorting timestamps.

Installing

Pack-wide: ComfyUI Manager search comfyui-fireredaudio-T8, or clone + python scripts\setup_runtime.py. This node itself is pure file I/O - no model, no worker - so it works even on a broken setup, which makes it the wrong place to debug a "save did nothing" complaint: if upstream never generated audio, the save node just has nothing to write. Check the generation node first.