Upload 2 files

Dockerfile.txt

@@ -0,0 +1,56 @@
+# 1. 使用与 README 要求一致的 PyTorch 版本 (Torch 2.4.1)
+FROM pytorch/pytorch:2.4.1-cuda12.1-cudnn9-devel
+
+# 设置非交互式安装
+ENV DEBIAN_FRONTEND=noninteractive
+
+# 2. 安装系统依赖 (README 要求安装 FFmpeg)
+RUN apt-get update && apt-get install -y \
+    git \
+    git-lfs \
+    ffmpeg \
+    libgl1-mesa-glx \
+    libglib2.0-0 \
+    && rm -rf /var/lib/apt/lists/*
+
+# 设置工作目录
+WORKDIR /app
+
+# 3. 克隆 InfiniteTalk 仓库
+RUN git clone https://github.com/MeiGen-AI/InfiniteTalk.git
+
+# 进入仓库目录
+WORKDIR /app/InfiniteTalk
+
+# 4. 安装 Python 依赖 (严格按照 README 顺序)
+# 先升级 pip
+RUN pip install --no-cache-dir --upgrade pip
+
+# 安装 PyTorch 相关 (基础镜像已有，但为了确保 xformers 匹配，我们可以再次确认或跳过)
+# README: pip install -U xformers==0.0.28 --index-url https://download.pytorch.org/whl/cu121
+RUN pip install -U xformers==0.0.28 --index-url https://download.pytorch.org/whl/cu121
+
+# 安装 Flash Attention 及其依赖 (README 步骤 2)
+# misaki[en], ninja, psutil, packaging, wheel
+RUN pip install --no-cache-dir misaki[en] ninja psutil packaging wheel
+# 安装 flash-attn
+RUN pip install flash_attn==2.7.4.post1
+
+# 安装其他依赖 (README 步骤 3)
+RUN pip install --no-cache-dir -r requirements.txt
+# 补充一些 Space 运行需要的库
+RUN pip install --no-cache-dir gradio huggingface_hub
+
+# 5. 设置权限 (Hugging Face Space 安全要求)
+RUN useradd -m -u 1000 user
+RUN chown -R user:user /app
+USER user
+
+# 6. 设置 Gradio 环境变量，确保可以通过公网访问
+ENV GRADIO_SERVER_NAME="0.0.0.0"
+
+# 7. 复制我们的启动脚本 (下一步创建)
+COPY --chown=user:user run.py /app/InfiniteTalk/run.py
+
+# 8. 启动
+CMD ["python", "run.py"]

run.py

@@ -0,0 +1,76 @@
+import os
+import subprocess
+from huggingface_hub import snapshot_download, hf_hub_download
+
+# 权重保存的基础目录 (对应 README 中的 ./weights)
+WEIGHTS_BASE = "weights"
+
+def download_models():
+    print("🚀 [Step 1/2] 开始下载模型权重，请耐心等待...")
+    
+    # 1. 下载 Wan2.1 底模
+    # README: huggingface-cli download Wan-AI/Wan2.1-I2V-14B-480P --local-dir ./weights/Wan2.1-I2V-14B-480P
+    wan_path = os.path.join(WEIGHTS_BASE, "Wan2.1-I2V-14B-480P")
+    if not os.path.exists(wan_path):
+        print(f"正在下载 Wan2.1-I2V-14B-480P 到 {wan_path} ...")
+        snapshot_download(repo_id="Wan-AI/Wan2.1-I2V-14B-480P", local_dir=wan_path)
+
+    # 2. 下载 InfiniteTalk 权重
+    # README: huggingface-cli download MeiGen-AI/InfiniteTalk --local-dir ./weights/InfiniteTalk
+    it_path = os.path.join(WEIGHTS_BASE, "InfiniteTalk")
+    if not os.path.exists(it_path):
+        print(f"正在下载 InfiniteTalk 到 {it_path} ...")
+        snapshot_download(repo_id="MeiGen-AI/InfiniteTalk", local_dir=it_path)
+
+    # 3. 下载 Audio Encoder (wav2vec2)
+    # README 提到这个比较特殊，分为两步：主仓库文件 + 特殊分支的 model.safetensors
+    w2v_path = os.path.join(WEIGHTS_BASE, "chinese-wav2vec2-base")
+    
+    # 3.1 下载主仓库
+    if not os.path.exists(w2v_path):
+        print(f"正在下载 chinese-wav2vec2-base 主体到 {w2v_path} ...")
+        snapshot_download(repo_id="TencentGameMate/chinese-wav2vec2-base", local_dir=w2v_path)
+    
+    # 3.2 下载特定 Revision 的 model.safetensors
+    # README: ... model.safetensors --revision refs/pr/1 ...
+    safetensor_file = os.path.join(w2v_path, "model.safetensors")
+    # 简单的检查：如果文件不存在或文件太小(可能没下载对)，则下载
+    if not os.path.exists(safetensor_file) or os.path.getsize(safetensor_file) < 1024:
+        print("正在下载 wav2vec2 的 model.safetensors (revision refs/pr/1) ...")
+        hf_hub_download(
+            repo_id="TencentGameMate/chinese-wav2vec2-base",
+            filename="model.safetensors",
+            revision="refs/pr/1",
+            local_dir=w2v_path,
+            local_dir_use_symlinks=False
+        )
+
+    print("✅ 所有模型下载完成！")
+
+def start_app():
+    print("🚀 [Step 2/2] 启动 Gradio 应用...")
+    
+    # 构建启动命令
+    # 参考 README "Run with Gradio" 部分
+    # python app.py --ckpt_dir ... --wav2vec_dir ...
+    
+    cmd = [
+        "python", "app.py",
+        "--ckpt_dir", "weights/Wan2.1-I2V-14B-480P",
+        "--wav2vec_dir", "weights/chinese-wav2vec2-base",
+        # README 示例中单人模式指向 single/infinitetalk.safetensors
+        "--infinitetalk_dir", "weights/InfiniteTalk/single/infinitetalk.safetensors",
+        # 添加低显存参数，防止 Space OOM (参考 README: Run with very low VRAM)
+        "--num_persistent_param_in_dit", "0",
+        "--motion_frame", "9"
+    ]
+    
+    # 打印命令方便调试
+    print(f"执行命令: {' '.join(cmd)}")
+    
+    # 运行
+    subprocess.run(cmd, check=True)
+
+if __name__ == "__main__":
+    download_models()
+    start_app()