comfyui-whisper-large-v3-repack

ComfyUI-WhisperLargeV3-Repack

MIT ComfyUI custom nodes for Whisper Large V3 transcription.

This package does not reuse code from non-commercial ComfyUI Whisper node packs. It wraps faster-whisper and defaults to large-v3, with optional large-v3-turbo for faster runs.

Nodes

• Whisper Large V3 Repack - Transcribe File
  • Selects an audio/video file from ComfyUI's input directory.
• Whisper Large V3 Repack - Transcribe Audio
  • Accepts a standard ComfyUI AUDIO input, such as the built-in Load Audio node output.

Both nodes output:

• transcript: full transcribed text
• language: detected or forced language code
• duration_seconds: audio duration reported by faster-whisper
• segments_json: segment timestamps and optional word timestamps

Install

Clone into ComfyUI/custom_nodes:

git clone https://github.com/endman100/ComfyUI-WhisperLargeV3-Repack.git

Install dependencies in the same Python environment used by ComfyUI:

pip install -r ComfyUI/custom_nodes/ComfyUI-WhisperLargeV3-Repack/requirements.txt

Restart ComfyUI.

Model Cache

Models are downloaded by faster-whisper into:

ComfyUI/models/whisper-large-v3-repack

The first large-v3 run can take time because the model weights are large.

Commercial Notes

OpenAI Whisper is MIT licensed. This package is also MIT licensed. Check the licenses and terms of any alternative model weights or hosting service you use.