Whisper Transcribe Batch

This is Whisper Transcribe's batch sibling, built to pair with Load Audio from Batch - feed it a whole folder's worth of files at once and it transcribes every one. But there's one thing about this node that catches people out immediately: it has no output sockets. None. Not a transcription string, not a chunks string, nothing you can wire downstream. The only way anything comes out of this node is by saving it to disk.

Why it works this way

This looks like an oversight the first time you see it, but it's a real design constraint, not a bug. Whisper Transcribe returns a single STRING per file because it only ever handles one file. Batch transcription handling N files would need N sockets, or some packed structure the rest of the graph would have to know how to unpack - ComfyUI's graph model doesn't make that pleasant. So instead of forcing that, this node writes straight to files, one per source, and you go read them off disk afterward. If you want per-file text living inside the graph for the current run, this is the wrong node - use Whisper Transcribe on individual files instead.

The inputs that matter

• pipeline (TRANSCRIPTION_PIPELINE) - from Load Whisper Transcription Model, the same loaded model reused across the whole batch.
• wav_bytes_batch (WAV_BYTES_BATCH) - from Load Audio from Batch.
• save_transcription / save_chunks - both off by default, and this is the trap: leave both off and this node does real GPU work transcribing every file, then throws every result away. Turn at least one on.
• filename_prefix / filename_suffix (both blank) - tag the batch's output filenames so you can tell which run produced them, and avoid one batch overwriting another's files.
• format_newlines_on_punctuation (default on) and overwrite_existing (default on) behave exactly like they do on the single-file node, just applied per file across the batch.

No outputs - nothing to wire downstream from this node at all.

Installing it

ComfyUI Manager: search ComfyUI-TranscriptionTools, install, restart. Or by hand:

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

then restart.

Common issues & troubleshooting

"I ran it and got nothing." This is the single most likely thing to trip you up on this specific node. Turn on save_transcription (and save_chunks if you want the segment breakdown too) before you run it - otherwise the whole batch transcribes and the text goes nowhere you can see.

Can't find the output files. These typically land in ComfyUI's output/ folder, the same convention as any other Save-style ComfyUI node. Use filename_prefix/filename_suffix if you're running multiple batches back to back and don't want to hunt through generically-named files - or lose one batch's results to overwrite_existing.

A long batch looks like it's hanging. It's genuinely slow on many long files - it runs the full Whisper pipeline once per file, not in parallel. count, from Load Audio from Batch, tells you how many files you actually queued going in, which is your rough sense of how long the wait should be.