diff --git a/docs/FEISHU_MODEL_API.md b/docs/FEISHU_MODEL_API.md
new file mode 100644
index 0000000..a31474a
--- /dev/null
+++ b/docs/FEISHU_MODEL_API.md
@@ -0,0 +1,255 @@
+# 飞书版模型服务 API 文档
+
+面向“故事到视频”链路，整理了可直接在飞书文档发布的接口说明，覆盖语言模型分镜生成、文生图、图生视频以及任务查询与 WebSocket 推送。
+
+## 基础信息
+- **Base URL**：`https://api.example.com`（按部署替换）
+- **版本**：`v1`
+- **认证**：`Authorization: Bearer <token>`，可选 `X-Request-Id` 便于链路追踪
+- **编码**：`Content-Type: application/json; charset=utf-8`
+- **幂等**：推荐传 `Idempotency-Key` 防重复提交
+- **默认模型**：文生图 `sd-turbo`，图生视频 `svd-img2vid`
+
+### 通用返回格式
+```json
+{
+  "code": 0,
+  "message": "ok",
+  "data": {}
+}
+```
+- `code`：0 表示成功；常见错误码：400 参数错误，401 未授权，404 不存在，429 频控，500 内部错误。
+
+### 任务模型
+- 耗时任务返回 `job_id`，通过查询接口或 WebSocket 订阅进度。
+- 状态枚举：`pending` / `running` / `succeeded` / `failed` / `canceled`。
+
+#### 通用任务查询
+- `GET /v1/jobs/{job_id}`
+
+响应示例：
+```json
+{
+  "code": 0,
+  "message": "ok",
+  "data": {
+    "job_id": "job_xxx",
+    "type": "llm|t2i|i2v",
+    "status": "running",
+    "progress": 42,
+    "result": {},
+    "error": {}
+  }
+}
+```
+
+## 接口一览
+1. 语言模型：生成分镜 `POST /v1/llm/storyboard`
+2. 文生图：生成关键帧 `POST /v1/image/generate`
+3. 图生视频：生成短视频片段 `POST /v1/video/generate`
+4. 任务查询：`GET /v1/jobs/{job_id}`
+5. WebSocket 进度推送（可选）：`GET wss://api.example.com/v1/ws`
+
+---
+
+## 1) 语言模型：生成分镜
+- **URL**：`POST /v1/llm/storyboard`
+- **功能**：将故事文本转为分镜 JSON，含标题、Prompt、旁白、BGM 建议等。
+
+请求体：
+```json
+{
+  "story_id": "story_001",
+  "story_text": "很久以前……",
+  "style": "movie",
+  "target_shots": 8,
+  "lang": "zh",
+  "extras": {
+    "tone": "warm",
+    "duration_hint_sec": 60
+  }
+}
+```
+
+成功响应（任务创建）：
+```json
+{
+  "code": 0,
+  "message": "ok",
+  "data": {
+    "job_id": "job_llm_123",
+    "status": "running"
+  }
+}
+```
+
+任务完成结果（经 `GET /v1/jobs/{job_id}` 获取）：
+```json
+{
+  "code": 0,
+  "message": "ok",
+  "data": {
+    "job_id": "job_llm_123",
+    "status": "succeeded",
+    "result": {
+      "story_id": "story_001",
+      "shots": [
+        {
+          "shot_id": "shot_001",
+          "title": "清晨的街道",
+          "prompt": "cinematic morning street, soft light, ...",
+          "narration": "清晨，主角踏上旅程……",
+          "bgm_hint": "lofi calm",
+          "duration_sec": 5
+        }
+      ]
+    }
+  }
+}
+```
+
+---
+
+## 2) 文生图：生成关键帧
+- **URL**：`POST /v1/image/generate`
+- **功能**：基于分镜 Prompt 生成关键帧图片（默认 SD Turbo）。
+
+请求体：
+```json
+{
+  "shot_id": "shot_001",
+  "prompt": "cinematic morning street, soft light, ...",
+  "negative_prompt": "blurry, lowres",
+  "style": "movie",
+  "size": "1024x576",
+  "model": "sd-turbo",
+  "seed": 42,
+  "steps": 15,
+  "guidance_scale": 3.5,
+  "scheduler": "euler",
+  "safety_check": true
+}
+```
+
+成功响应：
+```json
+{
+  "code": 0,
+  "message": "ok",
+  "data": {
+    "job_id": "job_t2i_456",
+    "status": "running"
+  }
+}
+```
+
+任务完成结果：
+```json
+{
+  "code": 0,
+  "message": "ok",
+  "data": {
+    "job_id": "job_t2i_456",
+    "status": "succeeded",
+    "result": {
+      "shot_id": "shot_001",
+      "image_url": "https://cdn.example.com/shot_001.png",
+      "meta": {
+        "seed": 42,
+        "size": "1024x576",
+        "steps": 15,
+        "model": "sd-turbo"
+      }
+    }
+  }
+}
+```
+
+---
+
+## 3) 图生视频：生成短视频片段
+- **URL**：`POST /v1/video/generate`
+- **功能**：以关键帧生成短视频片段，默认 Stable Video Diffusion（Img2Vid）。
+
+请求体：
+```json
+{
+  "shot_id": "shot_001",
+  "image_url": "https://cdn.example.com/shot_001.png",
+  "duration_sec": 4,
+  "fps": 24,
+  "resolution": "1280x720",
+  "model": "svd-img2vid",
+  "transition": "kenburns",
+  "motion_strength": 0.7,
+  "seed": 123,
+  "audio": {
+    "voiceover_url": "https://cdn.example.com/vo_shot001.wav",
+    "bgm_url": "https://cdn.example.com/bgm_lofi.mp3",
+    "ducking": true
+  }
+}
+```
+
+成功响应：
+```json
+{
+  "code": 0,
+  "message": "ok",
+  "data": {
+    "job_id": "job_i2v_789",
+    "status": "running"
+  }
+}
+```
+
+任务完成结果：
+```json
+{
+  "code": 0,
+  "message": "ok",
+  "data": {
+    "job_id": "job_i2v_789",
+    "status": "succeeded",
+    "result": {
+      "shot_id": "shot_001",
+      "video_url": "https://cdn.example.com/shot_001.mp4",
+      "meta": {
+        "duration_sec": 4,
+        "fps": 24,
+        "resolution": "1280x720",
+        "model": "svd-img2vid"
+      }
+    }
+  }
+}
+```
+
+---
+
+## 4) WebSocket 进度推送（可选）
+- **URL**：`GET wss://api.example.com/v1/ws`
+- **鉴权**：沿用 HTTP Header。
+- **订阅**：传入 `job_id` 列表。
+
+订阅示例：
+```json
+{ "action": "subscribe", "job_ids": ["job_llm_123","job_t2i_456","job_i2v_789"] }
+```
+
+推送示例：
+```json
+{
+  "job_id": "job_i2v_789",
+  "status": "running",
+  "progress": 65,
+  "message": "rendering frames"
+}
+```
+
+---
+
+## FAQ
+- **并行与合成**：每个镜头独立 job，可并行提交；最终合成由客户端/服务端 FFmpeg 处理。
+- **安全与审计**：可在文生图/图生视频前置安全审核（暴恐涉政等）。
+- **失败重试**：结合 `Idempotency-Key` 识别重复请求，失败可重提或按 `job_id` 重拉结果。
diff --git a/model/.dockerignore b/model/.dockerignore
new file mode 100644
index 0000000..ff36840
--- /dev/null
+++ b/model/.dockerignore
@@ -0,0 +1,13 @@
+__pycache__
+*.pyc
+*.pyo
+*.pyd
+*.swp
+.env
+venv
+build
+dist
+.cache
+.huggingface
+models
+weights
diff --git a/model/API.md b/model/API.md
new file mode 100644
index 0000000..c811280
--- /dev/null
+++ b/model/API.md
@@ -0,0 +1,136 @@
+# Model Node API Documentation
+
+This document specifies the HTTP API exposed by the model node FastAPI service. All endpoints are JSON-based and designed to be wired to real pipelines (Qwen/Ollama, Stable Diffusion Turbo, Stable-Video-Diffusion-Img2Vid, CosyVoice-mini).
+
+## General
+- **Base URL (default docker compose)**: `http://localhost:8000`
+- **Content type**: `application/json`
+- **Authentication**: not required in the sample; add gateway/token in production.
+- **Swagger UI**: `GET /docs`
+- **OpenAPI JSON**: `GET /openapi.json`
+
+## Health
+- **Endpoint**: `GET /health`
+- **Purpose**: Liveness/readiness signal.
+- **Response**
+  ```json
+  {
+    "status": "ok",
+    "ts": "2024-07-01T10:00:00.000000"
+  }
+  ```
+
+## Storyboard (LLM)
+- **Endpoint**: `POST /llm/storyboard`
+- **Description**: Convert a free-form story into structured shots.
+- **Request body**
+  ```json
+  {
+    "story": "夕阳下的海边散步",
+    "style": "pixar"
+  }
+  ```
+  - `story` (string, required): user story text.
+  - `style` (string, optional): tone/visual hint.
+- **Response 200**
+  ```json
+  {
+    "shots": [
+      {
+        "title": "自动生成分镜",
+        "prompt": "pixar | 夕阳下的海边散步",
+        "narration": "夕阳下的海边散步",
+        "bgm": "lofi-chill"
+      }
+    ],
+    "generated_at": "2024-07-01T10:00:00.123456"
+  }
+  ```
+
+## Text-to-Image
+- **Endpoint**: `POST /sd_generate`
+- **Description**: Generate a keyframe image from a prompt.
+- **Request body**
+  ```json
+  {
+    "prompt": "sunset beach cinematic",
+    "style": "anime",
+    "width": 1024,
+    "height": 576
+  }
+  ```
+  - `prompt` (string, required): image description.
+  - `style` (string, optional): style preset/tag.
+  - `width` (int, optional, default 1024)
+  - `height` (int, optional, default 576)
+- **Response 200**
+  ```json
+  {
+    "url": "https://example.com/keyframe.png",
+    "note": "Requested 1024x576 image in style=anime"
+  }
+  ```
+
+## Image-to-Video (optional)
+- **Endpoint**: `POST /img2vid`
+- **Description**: Turn a keyframe into a short clip.
+- **Request body**
+  ```json
+  {
+    "image_url": "https://example.com/keyframe.png",
+    "duration_seconds": 3.0,
+    "transition": "dissolve"
+  }
+  ```
+  - `image_url` (string, required): source frame.
+  - `duration_seconds` (float, optional, default 3.0): clip length.
+  - `transition` (string, optional): e.g., `dissolve`, `zoom`, `cut`.
+- **Response 200**
+  ```json
+  {
+    "url": "https://example.com/clip.mp4",
+    "note": "duration=3.0s transition=dissolve"
+  }
+  ```
+
+## Text-to-Speech
+- **Endpoint**: `POST /tts`
+- **Description**: Generate narration audio for a shot.
+- **Request body**
+  ```json
+  {
+    "text": "欢迎使用 StoryToVideo",
+    "voice": "female"
+  }
+  ```
+  - `text` (string, required): narration text.
+  - `voice` (string, optional): speaker/voice style.
+- **Response 200**
+  ```json
+  {
+    "url": "https://example.com/narration.wav",
+    "note": "voice=female"
+  }
+  ```
+
+## Error model
+- **Status codes**: `200` on success. FastAPI will emit `422` for validation errors.
+- **Example 422 response**
+  ```json
+  {
+    "detail": [
+      {
+        "loc": ["body", "story"],
+        "msg": "Field required",
+        "type": "value_error.missing"
+      }
+    ]
+  }
+  ```
+
+## Quick smoke tests
+- Health: `curl http://localhost:8000/health`
+- Storyboard: `curl -X POST http://localhost:8000/llm/storyboard -H "Content-Type: application/json" -d '{"story":"夕阳下的海边散步","style":"pixar"}'`
+- Text-to-Image: `curl -X POST http://localhost:8000/sd_generate -H "Content-Type: application/json" -d '{"prompt":"sunset beach cinematic","style":"anime"}'`
+- Img2Vid: `curl -X POST http://localhost:8000/img2vid -H "Content-Type: application/json" -d '{"image_url":"https://example.com/keyframe.png"}'`
+- TTS: `curl -X POST http://localhost:8000/tts -H "Content-Type: application/json" -d '{"text":"欢迎使用 StoryToVideo"}'`
diff --git a/model/Dockerfile b/model/Dockerfile
new file mode 100644
index 0000000..46f1658
--- /dev/null
+++ b/model/Dockerfile
@@ -0,0 +1,30 @@
+# CUDA 12.4 + cuDNN + Python 3.10 for GPU model serving
+FROM nvidia/cuda:12.4.1-cudnn-devel-ubuntu22.04
+
+ENV DEBIAN_FRONTEND=noninteractive \
+    PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1 \
+    PIP_NO_CACHE_DIR=1
+
+WORKDIR /workspace
+
+# System deps
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends \
+       python3.10 python3.10-venv python3-pip git curl ca-certificates \
+    && rm -rf /var/lib/apt/lists/*
+
+# Upgrade pip and install torch with CUDA 12.4 wheels
+RUN python3 -m pip install --upgrade pip \
+    && python3 -m pip install \
+       torch==2.4.0 torchvision==0.19.0 torchaudio==2.4.0 \
+       --index-url https://download.pytorch.org/whl/cu124
+
+COPY requirements.txt ./
+RUN python3 -m pip install -r requirements.txt
+
+COPY main.py ./
+
+EXPOSE 8000
+
+CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000"]
diff --git a/model/README.md b/model/README.md
index d8df77d..e055b8d 100644
--- a/model/README.md
+++ b/model/README.md
@@ -1,11 +1,10 @@
 # Model Node (Local GPU) / 模型节点
 
-Purpose: host generation capabilities decoupled from server; accessed via HTTP (FastAPI) and optionally exposed through FRP.  
+Purpose: host generation capabilities decoupled from server; accessed via HTTP (FastAPI) and optionally exposed through FRP.
 目的：承载生成能力，与服务端解耦，通过 FastAPI/HTTP 暴露，必要时用 FRP 打通。
 
 ## Suggested components / 推荐组件
-- LLM: Qwen2.5-0.5B via Ollama → story structure / storyboard JSON / narration draft.  
-  文本生成分镜 JSON/旁白。
+- LLM: Qwen2.5-0.5B via Ollama → story structure / storyboard JSON / narration draft. 文本生成分镜 JSON/旁白。
 - T2I: Stable Diffusion Turbo (diffusers) → keyframes. / 关键帧生图
 - I2V (optional): Stable-Video-Diffusion-Img2Vid → short clips. / 图生视频（可选）
 - TTS: CosyVoice-mini → narration audio. / 旁白语音
@@ -24,6 +23,128 @@ async def sd_generate(req: dict):
     return {"url": "https://.../image.png"}
 ```
 
+## GPU Dockerized model node (RTX 4060 Laptop, CUDA 12.4)
+针对截图中的 Windows 11 + RTX 4060 Laptop + 最新 NVIDIA 驱动（550+），新增了一个可直接构建的 GPU 模型容器。容器默认暴露 FastAPI 模型桩服务，并附带一个 GPU 版 Ollama 服务用来拉取 Qwen2.5-0.5B。
+
+### 1) Host prerequisites / 主机前置
+- Windows 11 + WSL2 + Docker Desktop；NVIDIA 550+ 驱动（支持 CUDA 12.4）。
+- 安装 `nvidia-container-toolkit`，确保 `nvidia-smi` 在 WSL 中可用：
+  ```bash
+  # inside WSL
+  distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
+  curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
+  curl -s -L https://nvidia.github.io/libnvidia-container/$distribution/libnvidia-container.list | \
+    sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list
+  sudo apt-get update && sudo apt-get install -y nvidia-container-toolkit
+  sudo nvidia-ctk runtime configure --runtime=docker
+  sudo systemctl restart docker
+  ```
+
+### 2) Build & run / 构建与运行
+```bash
+cd model
+# 构建 CUDA12.4 + PyTorch 2.4.0 的模型服务镜像
+docker compose -f docker-compose.gpu.yml build
+# 启动模型节点（FastAPI）和 GPU 版 Ollama
+docker compose -f docker-compose.gpu.yml up -d
+```
+- FastAPI 模型桩：`http://localhost:8000`（健康检查 `/health`，业务接口 `/llm/storyboard`、`/sd_generate`、`/img2vid`、`/tts`）。详见 [完整 API 文档](./API.md)，或查看便于飞书分享的精简版接口说明 [docs/FEISHU_MODEL_API.md](../docs/FEISHU_MODEL_API.md)。
+- Ollama：`http://localhost:11434`。
+
+### 3) Pull Qwen 模型（在容器内执行）
+```bash
+# 进入 ollama 容器，拉取 Qwen2.5 0.5B
+docker compose -f docker-compose.gpu.yml exec ollama ollama pull qwen2.5:0.5b
+```
+
+### 4) 挂载与缓存
+- `./weights` 挂载到容器 `/models`，用于 HF/SD/SVD/CosyVoice 等权重缓存。
+- Ollama 权重持久化到 compose 中的 `ollama` 卷，可跨重启保留。
+
+### 5) 接入提示
+- 将真实推理逻辑接到 `model/main.py` 中的 TODO（Qwen/Ollama、SD Turbo、SVD、CosyVoice）。
+- 如果需要对外暴露端口到公网，复用仓库 `frp/` 下的示例配置。
+
+### 6) 本地自动化测试
+- 安装依赖（含 pytest）：`pip install -r requirements.txt`
+- 运行测试：
+  ```bash
+  cd model
+  pytest
+  ```
+
+### 7) 镜像构建后如何测试 API
+容器启动后，可直接在宿主机用 `curl`/Postman/浏览器验证：
+
+- 健康检查：
+  ```bash
+  curl http://localhost:8000/health
+  ```
+- 分镜/LLM：
+  ```bash
+  curl -X POST http://localhost:8000/llm/storyboard \
+    -H "Content-Type: application/json" \
+    -d '{"story": "夕阳下的海边散步", "style": "pixar"}'
+  ```
+- 文生图（Stable Diffusion Turbo 桩）：
+  ```bash
+  curl -X POST http://localhost:8000/sd_generate \
+    -H "Content-Type: application/json" \
+    -d '{"prompt": "sunset beach cinematic", "style": "anime", "width": 1024, "height": 576}'
+  ```
+- 图生视频（Stable Video Diffusion 桩）：
+  ```bash
+  curl -X POST http://localhost:8000/img2vid \
+    -H "Content-Type: application/json" \
+    -d '{"image_url": "https://example.com/keyframe.png", "duration_seconds": 3.0, "transition": "dissolve"}'
+  ```
+- 旁白 TTS（CosyVoice 桩）：
+  ```bash
+  curl -X POST http://localhost:8000/tts \
+    -H "Content-Type: application/json" \
+    -d '{"text": "欢迎使用 StoryToVideo", "voice": "female"}'
+  ```
+
+> 也可以打开 `http://localhost:8000/docs` 使用 FastAPI 提供的交互式 Swagger UI 进行可视化调试；出现错误时用 `docker compose -f docker-compose.gpu.yml logs -f model` 查看容器日志。
+
+### 8) 将镜像推送到仓库（供前后端同学复用）
+构建完成后，可将镜像推送到 Docker Hub 或 GHCR，便于前后端直接 `pull` 部署。
+
+- **Docker Hub 示例**
+  ```bash
+  # 1) 构建（如果尚未构建）
+  docker compose -f docker-compose.gpu.yml build
+
+  # 2) 登录 Docker Hub
+  docker login
+
+  # 3) 打 tag 并推送（替换 yourname 为你的仓库命名空间）
+  docker tag storytovideo-model-node:cuda12.4 yourname/storytovideo-model-node:cuda12.4
+  docker push yourname/storytovideo-model-node:cuda12.4
+  ```
+
+- **GitHub Container Registry (GHCR) 示例**
+  ```bash
+  export GH_USER="your-github-username"
+  echo "$GH_PAT" | docker login ghcr.io -u "$GH_USER" --password-stdin
+
+  docker tag storytovideo-model-node:cuda12.4 ghcr.io/$GH_USER/storytovideo-model-node:cuda12.4
+  docker push ghcr.io/$GH_USER/storytovideo-model-node:cuda12.4
+  ```
+
+- **前后端同学使用方法**：直接 `docker pull <registry-tag>`，或在部署 compose 中用远端镜像替换本地构建：
+  ```yaml
+  services:
+    model-node:
+      image: ghcr.io/your-namespace/storytovideo-model-node:cuda12.4
+      runtime: nvidia
+      ports:
+        - "8000:8000"
+      volumes:
+        - ./weights:/models
+  ```
+  若需要从私有仓库拉取，请提前配置 `docker login` 或在 CI/CD 中通过密钥注入。
+
 ## Deployment / 部署
 - Run on local GPU; package models separately from server. / 本地 GPU 运行，独立包模型。
 - Expose ports via `frpc` to cloud `frps`. / 用 frpc 将端口暴露给公网 frps。
diff --git a/model/__init__.py b/model/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/model/docker-compose.gpu.yml b/model/docker-compose.gpu.yml
new file mode 100644
index 0000000..5a2df90
--- /dev/null
+++ b/model/docker-compose.gpu.yml
@@ -0,0 +1,47 @@
+version: "3.8"
+
+services:
+  model-node:
+    build:
+      context: .
+      dockerfile: Dockerfile
+    image: storytovideo-model-node:cuda12.4
+    runtime: nvidia
+    environment:
+      - NVIDIA_VISIBLE_DEVICES=all
+      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
+      - HF_HOME=/models/hf
+      - TRANSFORMERS_CACHE=/models/hf
+      - TORCH_CUDNN_V8_API_ENABLED=1
+    volumes:
+      - ./weights:/models
+    ports:
+      - "8000:8000"
+    command: ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000"]
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: 1
+              capabilities: [gpu]
+
+  ollama:
+    image: ollama/ollama:latest
+    runtime: nvidia
+    environment:
+      - OLLAMA_HOST=0.0.0.0
+    ports:
+      - "11434:11434"
+    volumes:
+      - ollama:/root/.ollama
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: 1
+              capabilities: [gpu]
+
+volumes:
+  ollama:
diff --git a/model/main.py b/model/main.py
new file mode 100644
index 0000000..1a0b34b
--- /dev/null
+++ b/model/main.py
@@ -0,0 +1,103 @@
+"""Minimal FastAPI entrypoint for model node.
+
+This keeps the same endpoint names used in the docs and can be wired to
+real pipelines (Qwen via Ollama, Stable Diffusion, Stable Video Diffusion,
+CosyVoice, etc.). For now, it echoes the request to keep the container
+lightweight while providing a health surface for integration.
+"""
+from datetime import datetime
+from typing import List, Optional
+
+from fastapi import FastAPI
+from pydantic import BaseModel, Field
+
+
+class StoryboardRequest(BaseModel):
+    story: str = Field(..., description="User story text")
+    style: Optional[str] = Field(None, description="Tone or visual style hint")
+
+
+class Shot(BaseModel):
+    title: str
+    prompt: str
+    narration: str
+    bgm: Optional[str] = None
+
+
+class StoryboardResponse(BaseModel):
+    shots: List[Shot]
+    generated_at: datetime
+
+
+class SDRequest(BaseModel):
+    prompt: str
+    style: Optional[str] = None
+    width: int = 1024
+    height: int = 576
+
+
+class SDResponse(BaseModel):
+    url: str
+    note: Optional[str] = None
+
+
+class Img2VidRequest(BaseModel):
+    image_url: str
+    duration_seconds: float = 3.0
+    transition: Optional[str] = Field(None, description="e.g. dissolve, zoom")
+
+
+class Img2VidResponse(BaseModel):
+    url: str
+    note: Optional[str] = None
+
+
+class TTSRequest(BaseModel):
+    text: str
+    voice: Optional[str] = Field(None, description="voice name or speaker id")
+
+
+class TTSResponse(BaseModel):
+    url: str
+    note: Optional[str] = None
+
+
+app = FastAPI(title="StoryToVideo Model Node", version="0.1.0")
+
+
+@app.get("/health")
+def health() -> dict:
+    return {"status": "ok", "ts": datetime.utcnow().isoformat()}
+
+
+@app.post("/llm/storyboard", response_model=StoryboardResponse)
+def storyboard(req: StoryboardRequest) -> StoryboardResponse:
+    # TODO: wire to Qwen/Ollama
+    shot = Shot(
+        title="自动生成分镜",
+        prompt=f"{req.style or '默认风格'} | {req.story[:80]}",
+        narration=req.story,
+        bgm="lofi-chill"
+    )
+    return StoryboardResponse(shots=[shot], generated_at=datetime.utcnow())
+
+
+@app.post("/sd_generate", response_model=SDResponse)
+def sd_generate(req: SDRequest) -> SDResponse:
+    # TODO: wire to Stable Diffusion Turbo pipeline
+    note = f"Requested {req.width}x{req.height} image in style={req.style or 'default'}"
+    return SDResponse(url="https://example.com/keyframe.png", note=note)
+
+
+@app.post("/img2vid", response_model=Img2VidResponse)
+def img2vid(req: Img2VidRequest) -> Img2VidResponse:
+    # TODO: wire to Stable-Video-Diffusion-Img2Vid
+    note = f"duration={req.duration_seconds}s transition={req.transition or 'cut'}"
+    return Img2VidResponse(url="https://example.com/clip.mp4", note=note)
+
+
+@app.post("/tts", response_model=TTSResponse)
+def tts(req: TTSRequest) -> TTSResponse:
+    # TODO: wire to CosyVoice-mini
+    note = f"voice={req.voice or 'default'}"
+    return TTSResponse(url="https://example.com/narration.wav", note=note)
diff --git a/model/requirements.txt b/model/requirements.txt
new file mode 100644
index 0000000..b467ed1
--- /dev/null
+++ b/model/requirements.txt
@@ -0,0 +1,13 @@
+fastapi==0.111.0
+uvicorn[standard]==0.30.1
+pydantic==2.7.3
+transformers==4.42.4
+diffusers==0.29.2
+accelerate==0.31.0
+safetensors==0.4.3
+sentencepiece==0.2.0
+numpy==1.26.4
+pillow==10.3.0
+soundfile==0.12.1
+pytest==8.2.2
+httpx==0.27.0
diff --git a/model/tests/test_api.py b/model/tests/test_api.py
new file mode 100644
index 0000000..303831c
--- /dev/null
+++ b/model/tests/test_api.py
@@ -0,0 +1,62 @@
+from fastapi.testclient import TestClient
+
+from model import main
+
+
+client = TestClient(main.app)
+
+
+def test_health():
+    resp = client.get("/health")
+    assert resp.status_code == 200
+    payload = resp.json()
+    assert payload.get("status") == "ok"
+    assert "ts" in payload
+
+
+def test_storyboard_endpoint():
+    payload = {"story": "夕阳下的海边散步", "style": "pixar"}
+    resp = client.post("/llm/storyboard", json=payload)
+    assert resp.status_code == 200
+    data = resp.json()
+    assert "shots" in data and isinstance(data["shots"], list)
+    assert data["shots"], "should return at least one shot"
+    first_shot = data["shots"][0]
+    assert first_shot["title"]
+    assert payload["story"] in first_shot["narration"]
+
+
+def test_sd_generate_endpoint():
+    payload = {
+        "prompt": "sunset beach cinematic",
+        "style": "anime",
+        "width": 1024,
+        "height": 576,
+    }
+    resp = client.post("/sd_generate", json=payload)
+    assert resp.status_code == 200
+    data = resp.json()
+    assert data["url"].startswith("http")
+    assert "1024x576" in data["note"]
+
+
+def test_img2vid_endpoint():
+    payload = {
+        "image_url": "https://example.com/keyframe.png",
+        "duration_seconds": 3.0,
+        "transition": "dissolve",
+    }
+    resp = client.post("/img2vid", json=payload)
+    assert resp.status_code == 200
+    data = resp.json()
+    assert data["url"].endswith(".mp4")
+    assert "duration=3.0" in data["note"]
+
+
+def test_tts_endpoint():
+    payload = {"text": "欢迎使用 StoryToVideo", "voice": "female"}
+    resp = client.post("/tts", json=payload)
+    assert resp.status_code == 200
+    data = resp.json()
+    assert data["url"].endswith(".wav")
+    assert "voice=female" in data["note"]