VideoReTalking: 7.2K+ Stars — AI 唇形同步视频编辑完整搭建指南 2026

git clone https://github.com/OpenTalker/video-retalking.git
cd video-retalking

conda create -n video_retalking python=3.8 -y
conda activate video_retalking
conda install ffmpeg -y

# CUDA 11.1（项目默认）
pip install torch==1.9.0+cu111 torchvision==0.10.0+cu111 -f https://download.pytorch.org/whl/torch_stable.html

# CUDA 12.1（现代 GPU，2026 年）
pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cu121

pip install -r requirements.txt

# 目录结构应如下：
# ./checkpoints/
#   ├── 244000.pth          (D-Net 表情编辑)
#   ├── wav2lip.pth         (L-Net 唇形同步)
#   ├── GFPGANv1.3.pth      (GFPGAN 增强)
#   ├── GPEN-BFR-512.pth    (GPEN 增强)
#   └── ...

python -c "import torch; print('CUDA available:', torch.cuda.is_available()); print('Device:', torch.cuda.get_device_name(0) if torch.cuda.is_available() else 'CPU')"

# 第一步：使用 RVC 生成或转换音频
python rvc/infer.py --input input.wav --model weights/model.pth --output rvc_output.wav

# 第二步：将 RVC 输出输入 VideoReTalking
python inference.py \
  --face input_video.mp4 \
  --audio rvc_output.wav \
  --outfile output_rvc_synced.mp4

# gpt_sovits_videoretalking.py
import subprocess
import os

# 第一步：使用 GPT-SoVITS 生成 TTS
subprocess.run([
    "python", "GPT_SoVITS/inference_webui.py",
    "--text", "你好，这是视频配音的版本。",
    "--ref_audio", "reference_voice.wav",
    "--output", "tts_output.wav"
])

# 第二步：运行 VideoReTalking
subprocess.run([
    "python", "inference.py",
    "--face", "source_video.mp4",
    "--audio", "tts_output.wav",
    "--outfile", "final_dubbed.mp4",
    "--exp_img", "neutral",
    "--up_face", "surprise"
])

# 安装 Coqui TTS
pip install TTS

# 生成语音
tts --text "这是视频的新对话。" \
    --model_name tts_models/multilingual/multi-dataset/xtts_v2 \
    --speaker_wav reference.wav \
    --language_idx zh \
    --out_path coqui_output.wav

# 使用 VideoReTalking 进行唇形同步
python inference.py \
  --face original_video.mp4 \
  --audio coqui_output.wav \
  --outfile coqui_synced.mp4

# 启动 WebUI
python webUI.py

python webUI.py --server-name 0.0.0.0 --server-port 7860 --share

# Dockerfile
FROM nvidia/cuda:12.1.0-runtime-ubuntu22.04

RUN apt-get update && apt-get install -y \
    python3.8 python3-pip ffmpeg git wget \
    libgl1-mesa-glx libglib2.0-0

WORKDIR /app
RUN git clone https://github.com/OpenTalker/video-retalking.git .
RUN pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
RUN pip3 install -r requirements.txt

# 下载检查点（生产环境作为卷挂载）
RUN mkdir -p checkpoints

EXPOSE 7860
CMD ["python3", "webUI.py", "--server-name", "0.0.0.0"]

docker build -t video-retalking .
docker run --gpus all -p 7860:7860 -v $(pwd)/checkpoints:/app/checkpoints video-retalking

#!/usr/bin/env python3
# batch_process.py
import os
import subprocess
from pathlib import Path

INPUT_DIR = "input_videos"
AUDIO_DIR = "input_audio"
OUTPUT_DIR = "output"
os.makedirs(OUTPUT_DIR, exist_ok=True)

video_files = sorted(Path(INPUT_DIR).glob("*.mp4"))
audio_files = sorted(Path(AUDIO_DIR).glob("*.wav"))

for vid, aud in zip(video_files, audio_files):
    outname = f"{OUTPUT_DIR}/{vid.stem}_synced.mp4"
    print(f"Processing: {vid.name} + {aud.name}")
    subprocess.run([
        "python", "inference.py",
        "--face", str(vid),
        "--audio", str(aud),
        "--outfile", outname,
        "--exp_img", "neutral"
    ])

# 带结构化日志的生产包装器
import logging
import time
import torch

logging.basicConfig(
    level=logging.INFO,
    format='%(asctime)s - %(levelname)s - %(message)s',
    handlers=[
        logging.FileHandler('videoretalking.log'),
        logging.StreamHandler()
    ]
)

def inference_with_monitoring(face_path, audio_path, output_path):
    start = time.time()
    vram_before = torch.cuda.memory_allocated() / 1e9
    
    # 运行推理
    subprocess.run([...])  # 推理命令
    
    elapsed = time.time() - start
    vram_after = torch.cuda.memory_allocated() / 1e9
    
    logging.info(f"Processed {face_path} in {elapsed:.1f}s, "
                 f"VRAM: {vram_before:.1f}GB -> {vram_after:.1f}GB")