pacifickash/wav2lip-inference

Wav2Lip Inference Endpoint

Fast lip-sync video generation using Wav2Lip on Hugging Face Inference Endpoints.

Features

• T4 GPU acceleration for fast inference (~2-5 seconds)
• Auto-scaling with pay-per-use pricing
• Base64 API for easy integration
• Real-time processing with optimized model loading

API Usage

Input Format

{
  "inputs": {
    "face_image": "base64_encoded_image_data",
    "audio": "base64_encoded_audio_data"
  }
}

Output Format

{
  "video": "base64_encoded_video_data",
  "status": "success",
  "size": 312606
}

Python Example

import requests
import base64

# Encode your files
with open("face.jpg", "rb") as f:
    face_b64 = base64.b64encode(f.read()).decode()

with open("audio.wav", "rb") as f:
    audio_b64 = base64.b64encode(f.read()).decode()

# Make request
response = requests.post(
    "https://YOUR_ENDPOINT.endpoints.huggingface.cloud",
    headers={"Authorization": "Bearer YOUR_HF_TOKEN"},
    json={
        "inputs": {
            "face_image": face_b64,
            "audio": audio_b64
        }
    }
)

result = response.json()
if result["status"] == "success":
    # Decode and save video
    video_data = base64.b64decode(result["video"])
    with open("output.mp4", "wb") as f:
        f.write(video_data)

Deployment

This endpoint uses:

• Model: Wav2Lip GAN (139MB)
• GPU: T4 (recommended)
• Framework: PyTorch + OpenCV
• Container: Custom Docker image

Processing time: 2-5 seconds per video on T4 GPU.