Load Audio from Video

This is the node you reach for when your source material is a video and all you actually want is the words someone said in it. ComfyUI-TranscriptionTools is a small, focused pack built around one job - turn spoken audio into text, and hold up on long recordings while doing it - and this node is one of its two front doors. If your source is already a bare audio file, you want its sibling, Load Audio. If it's a video - a screen recording, a downloaded clip, a capture off a VHS tape - this is the one that strips the soundtrack out and hands it to the rest of the pack in the shape it expects.

How it works

There's exactly one field, and it's the same upload-and-pick pattern you already know from ComfyUI's native video loaders: click upload (or drop a file straight into ComfyUI/input), then choose it from the dropdown. Under the hood the node demuxes the file and pulls out just the audio track, discarding the picture - you're not paying for a video decode you don't need on the frames side. What comes out is WAV_BYTES, a type that belongs to this pack specifically. It's not ComfyUI's native AUDIO type, so it only plugs into other TranscriptionTools nodes.

The input and output that matter

• video - the only field. A dropdown populated from files already sitting in ComfyUI's input folder; upload through the node or drop the file in yourself first.
• wav_bytes (output, WAV_BYTES) - wire it into Whisper Transcribe to get text out, or into Audio Sink if you just want the extracted audio track saved as its own file without transcribing anything.

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. There's no model download tied to this specific node - the heavy download (an actual Whisper checkpoint) doesn't happen until you get to Load Whisper Transcription Model, further down the graph.

Common issues & troubleshooting

The video dropdown is empty or doesn't show your file. Same fix as every ComfyUI file-picker with this shape: upload it through the node's button, or drop it into ComfyUI/input and refresh.

A video fails to load with no obvious error. Demuxing audio out of an arbitrary video container is ffmpeg's job in this ecosystem - VHS's own loaders need it for the same reason, and this pack's readme credits VideoHelperSuite as instrumental to its development, so it's reasonable to expect the same requirement here. If you're self-hosting, confirm ffmpeg is actually installed and on PATH before you go looking for a stranger bug. This isn't something you'll hit on comfy.icu - ffmpeg ships in the executor image already.

wav_bytes won't connect to a native Save Audio node, or another pack's audio input. That's expected, not a bug - WAV_BYTES is TranscriptionTools' own internal type, not interchangeable with ComfyUI's AUDIO. Stay inside the pack (Whisper Transcribe, Audio Sink), or route through Convert VHS Audio to WAV bytes first if your audio actually originated from a VideoHelperSuite node instead of this one.