ComfyUI custom nodes for multi-person digital human audio processing, designed for InfiniteTalk workflows.
When creating multi-person talking head videos, each person needs to speak in sequence. These nodes process multiple audio tracks into equal-length outputs with proper silence padding.
Example: Person A speaks for 3s, Person B speaks for 2s:
- Output A: 3s audio + 2s silence
- Output B: 3s silence + 2s audio
- Merged: 5s continuous audio (A → B)
default.mp4
For two-person dialogue.
Inputs:
| Name | Type | Description |
|---|---|---|
| audio_left | AUDIO | First speaker's audio |
| audio_right | AUDIO | Second speaker's audio |
Outputs:
| Name | Type | Description |
|---|---|---|
| left_padded | AUDIO | First audio + silence padding |
| right_padded | AUDIO | Silence padding + second audio |
| merged | AUDIO | Concatenated audio (no silence) |
For 2-4 person scenarios.
Inputs:
| Name | Type | Required |
|---|---|---|
| audio_1 | AUDIO | Yes |
| audio_2 | AUDIO | Yes |
| audio_3 | AUDIO | No |
| audio_4 | AUDIO | No |
Outputs:
| Name | Type | Description |
|---|---|---|
| padded_1~4 | AUDIO | Each speaker's padded audio |
| merged | AUDIO | All audio concatenated |
cd ComfyUI/custom_nodes
git clone https://github.com/ALIoneTreE1119/ComfyUI-DualAudioProcessor.gitRestart ComfyUI.
See workflow/InfiniteTalk双人数字人对口型.json for a complete example.
- torch
- torchaudio
Both are included with ComfyUI by default.
Based on the workflow by YZ_金鱼 on Bilibili.
MIT