FireRedAudio 保存字幕/文本 · T8star-Aix

Text comes out of this pack all over the place - ASR transcripts, long-audio subtitles, understanding answers, QA reports - and most of it is only useful once it's a file. SaveSubtitle writes any string to ComfyUI/output/ as SRT, VTT, TXT, or JSONL. It's the smallest node in the family and arguably the one you'll use every single time: caption a long recording, keep a QA log, or turn a MultiUnderstand answer into a shareable text file. SRT/VTT for subtitles, TXT for transcripts, JSONL when you want line-by-line machine-readable output.

How it works

Simple and safe: take content (any multiline string), pick text_format, name it with filename_prefix, drop it in subfolder. The output saved_path gives you the full path back. No model, no worker, no RPC - pure file I/O, which means it works even before you've installed anything heavy, and it's a good smoke test for whether the pack is on disk at all.

The inputs

• content - the string to write. Forced as an input (you can't just type it in), because the node exists to save what other nodes produce. Wire the 回答/自动逐字稿/long-ASR output, or any text.
• text_format - srt (default), vtt, txt, or jsonl. If you're feeding subtitles to a video tool, SRT and VTT are the interoperable ones; VTT has the WEBVTT header some players insist on.
• filename_prefix - default fireredaudio, gets a timestamp suffix so you don't overwrite earlier saves.
• subfolder - default fireredaudio. Keep long-ASR output in its own subfolder and delivery subtitles in theirs.

Where it fits

The long-audio chain is the natural home: 长音频 ASR → this node as SRT/VTT, so a multi-hour recording becomes a subtitle file you can drop into any player or editor. It's also the natural tail for the understanding and QA nodes - paste a MultiUnderstand comparison into a TXT, or archive a SpeechQA report to JSONL for later diffing. And in a delivery workflow, note that ProductionPackage generates its own SRT/VTT from actual timeline placements - if you need subtitles that match the rendered timeline, use that; this node is for saving raw text from upstream nodes.

Installing

Same pack story: ComfyUI Manager search comfyui-fireredaudio-T8, or clone + python scripts\setup_runtime.py. Zero dependencies beyond the pack itself. The only real gotcha is content discipline: this node writes exactly what you give it, so a transcript saved as SRT will not magically get timestamps - if the upstream node output plain text, save it as TXT and expect plain text. Save SRT output (timed) as SRT, and you're golden.