Some ComfyDeploy server images ship a broken torchcodec (missing libnppicc.so.13), which makes the default LoadAudioFromURL / ComfyUIDeployExternalAudio flows crash with:

TypeError: 'NoneType' object is not subscriptable

This node bypasses torchcodec entirely. It downloads the file via requests, detects the format (mp3 / wav / m4a / ogg / flac / aac / opus), then tries 3 backends in order until one succeeds:

torchaudio (disk file)
soundfile (in-memory bytes)
ffmpeg → wav (hard fallback)

Node

Name	Display	Category
`URL_To_Audio`	`URL → Audio`	`🔗ComfyDeploy`

Inputs

Field	Type	Notes
`url`	STRING	Public URL of the audio file
`timeout_sec`	INT (optional)	HTTP timeout, default 90

Output

Name	Type
`audio`	AUDIO

The AUDIO dict format is the standard ComfyUI one:

{"waveform": torch.Tensor [batch=1, channels, samples], "sample_rate": int}

Wiring example (Kling+Sync workflow)

Replace the broken audio loader with:

ExternalText (input_id="audio_url")  ─→  URL → Audio  ─→  FalLipsyncV3.audio

From your client, send the audio URL as a string instead of uploading the WAV file:

{
    "deployment_id": "...",
    "inputs": {
        "input_image": image_url,
        "audio_url": "https://your-cdn.com/voice.mp3",
        "motion": "...",
        ...
    }
}

Install (ComfyDeploy)

Add this repo URL to your ComfyDeploy machine's custom nodes list, or in a self-hosted ComfyUI:

cd ComfyUI/custom_nodes
git clone https://github.com/PauldeLavallaz/comfyui-url2audio
pip install requests soundfile

License

MIT