ComfyUI-HunyuanVideoFoley

English | 简体中文

A ComfyUI custom node integrating the inference pipeline of HunyuanVideo-Foley, enabling audio generation from video + text prompts, plus audio-video merging utilities.

Features

• 🎵 Auto model download on first use (from Hugging Face)
• 🎬 Video-aware audio generation guided by text prompts
• 🔧 Audio-video merging to produce a final video with an audio track

Nodes

1) HunyuanVideo-Foley Generate Audio

Generate an AUDIO tensor from a VIDEO input and a text prompt.

Inputs:

• video: VIDEO (LoadVideo / CreateVideo output)
• prompt: Text prompt describing the desired audio
• model: Model setup (auto-detected under models/hunyuan_foley)
• guidance_scale: Float, default 4.5
• num_inference_steps: Int, default 50
• device: auto/cpu/cuda/mps
• gpu_id: CUDA device id

Output:

• audio: { "waveform": [B=1, C=1, T], "sample_rate": int }

2) Video Audio Merger

Merge AUDIO and VIDEO into a single video file with an audio track.

Inputs:

• video: VIDEO input
• audio: AUDIO input
• audio_sync_mode:
  • stretch | loop | truncate | pad_silence
• video_codec: copy | libx264 | libx265
• audio_codec: aac | mp3 | copy
• quality: high | medium | low

Output:

• merged_video: VIDEO

Quick Start

LoadVideo → HunyuanVideoFoleyGenerateAudio → VideoAudioMerger → SaveVideo
               ↑                                  ↑
           text prompt                         audio input

1. Load a video (LoadVideo)
2. Generate audio via "HunyuanVideo-Foley Generate Audio"
3. Merge audio + video via "Video Audio Merger"
4. Save or preview the final video

Dependencies

# Use your ComfyUI Python environment
D:\AI\ComfyUI-aki-v1.3\.ext\python.exe -m pip install ffmpeg-python

Notes

• On first run, models will be downloaded automatically (internet required)
• FFmpeg is required for audio/video processing
• CUDA is recommended for best performance