🎙️ Load Video (ASR)

Let's start with the most boring node in this pack, because it's the one everything else hangs off. VTT_LoadVideo (shown in the menu as 🎙️ Load Video (ASR)) doesn't transcribe, extract, or resize anything. It's a file picker. You get a dropdown of the videos sitting in ComfyUI/input/, you pick one, and out the other side comes a plain path string. That's it.

But here's why it exists: the rest of this video-to-text pipeline wants a real, absolute path on disk, not the preview-friendly file handle you get from dragging files around. The next node shells out to ffmpeg, and after that a multi-billion-parameter ASR model reads the audio. Everything downstream needs a path it can trust, and this node is what hands it over.

How it works

It's about as simple as ComfyUI nodes get. The node scans ComfyUI's input directory for anything with a video extension - .mp4, .mov, .mkv, .avi, .webm, .flv, .m4v, .wmv, .ts, .mpg, .mpeg - sorts the list, and populates the dropdown. If the folder is empty you get a placeholder entry telling you so, instead of a confusing blank widget.

One detail worth knowing: it reports a change check based on the file's mtime. Replace the video on disk, re-run, and the whole downstream chain actually re-runs instead of silently skipping. That's the kind of thing you only notice when it's not there.

Inputs and output

Only two inputs matter here:

• video - the dropdown. Points at whatever's in ComfyUI/input/.
• video_path_override - the one you'll probably use more. An absolute path to a video anywhere on the machine, say a lecture living on a big data drive. If it's non-empty, it wins over the dropdown.

The single output, video_path (a STRING), wires straight into VTT_ExtractAudio's video_path input. Then you're off.

How to install it

Installing this node is really installing the whole pack - they ship together. Easiest route is ComfyUI Manager: search for "ComfyUI-VideoToText", hit install, restart. Or do it by hand:

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

Then restart ComfyUI. The pack itself only needs torchaudio in your venv plus ffmpeg on PATH - but here's the catch that surprises everyone: the pack ships no ASR model at all. The actual transcribing comes from the author's companion pack, ComfyUI-MiMoASR, which wants a 24 GB GPU, the MiMo-V2.5-ASR weights, and a transformers==4.49.0 pin that will break other nodes like Sonic or Qwen3TTS. If you just want to see VTT_LoadVideo load, this pack alone is enough. If you want the workflow to produce text, install MiMoASR first and verify its nodes register.

Common issues

If the node shows up missing after install, check ComfyUI's startup log for IMPORT FAILED ComfyUI-VideoToText - a nested folder (custom_nodes/ComfyUI-VideoToText/ComfyUI-VideoToText/) from a botched tarball install is the usual cause. And if you're feeding it absolute paths, make them real: the node resolves and checks existence, so a typo raises FileNotFoundError instead of silently passing an empty string downstream. That's the one failure mode this boring node actually has, and it's a friendly one.