duanjujingtoufantui.skill

短剧镜头反推工具：上传参考视频或图片，抽取关键帧，并调用你自己的视觉大模型生成镜头关键词、运镜脚本和视频生成提示词。

A lightweight video prompt workbench for turning reference videos or images into reusable AI video generation prompts.

duanjujingtoufantui.skill 是一个轻量级 Web 工作台，面向短剧、广告、电商 Listing、AI 视频制作团队。它把参考素材里的镜头效果、运镜方式、光线色彩、节奏氛围和负面关键词整理成可复制的生成提示词，方便复用到新人物、新产品或新场景的视频制作中。

项目本身不内置私有模型，也不包含 API Key。你可以配置 OpenAI 官方接口，或配置兼容 OpenAI Responses API 的中转接口。只要模型具备视觉理解能力，就可以用于素材拆解和关键词生成。

• 反推爆款短剧镜头的景别、机位、焦段感和节奏。
• 把广告、电商视频或参考图拆成可复用的视频生成关键词。
• 为 AI 视频团队沉淀内部镜头语言和提示词知识库。
• 用同一套参考镜头风格改写到新人物、新产品或新场景。
• 给客户项目按归档、项目和标签管理素材分析报告。

• 上传常见视频或图片格式。
• 自动抽取视频关键帧，生成关键帧联系表。
• 基于画面变化自动切分镜头段落。
• 调用 OpenAI Responses API 或兼容接口分析视觉内容。
• 输出镜头效果、运镜方式、光线色彩、节奏氛围、正向关键词、负面关键词和最终视频生成提示词。
• 支持内部工具模式：关闭登录后直接给可信团队使用。
• 支持账号模式：管理员可创建用户，并按项目隔离素材和报告。
• 自动沉淀 Markdown 报告和知识库文件，方便二次整理。
• 支持本地运行、VPS 部署、nginx 反向代理和 systemd 托管。

• Python 3.11+
• FastAPI
• Jinja2
• SQLite
• OpenCV
• Pillow
• NumPy
• OpenAI Responses API / OpenAI-compatible relay API

Clone the repository:

git clone https://github.com/Zzz-caomei/GitHub.git
cd GitHub

Create a virtual environment and install dependencies:

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env

Edit .env:

SHOT_PLATFORM_HOME=./runtime
OPENAI_API_KEY=your_api_key_here
OPENAI_BASE_URL=https://api.openai.com/v1
OPENAI_MODEL=gpt-5.5
AUTH_DISABLED=true
MAX_UPLOAD_MB=300
ADMIN_USER=admin
ADMIN_PASSWORD=change-me

Start the app:

set -a
source .env
set +a
uvicorn app.main:app --host 0.0.0.0 --port 8090

Open:

http://127.0.0.1:8090

1. Open the workbench in your browser.
2. Upload a reference video or image.
3. Choose the output type: all dimensions, shot style, camera movement, or 8-second script.
4. Add your analysis goal, such as "extract low-angle push-in, motion blur, cold/warm lighting, and negative prompts".
5. Wait for frame extraction and visual analysis.
6. Copy the generated video prompt or export the full report.

The app writes generated files under SHOT_PLATFORM_HOME:

data/uploads/          Uploaded source media
data/frames/           Extracted frames
data/contact_sheets/   Contact sheet previews
data/reports/          Markdown analysis reports
data/exports/          Exported zip files
data/database.sqlite   SQLite database
knowledge_base/        Reusable knowledge-base reports
logs/                  Runtime logs

Do not commit runtime data, uploaded media, databases, logs, reports, .env files, API keys, or customer materials to GitHub.

The application calls:

{OPENAI_BASE_URL}/responses

For an OpenAI-compatible relay, configure .env like this:

OPENAI_API_KEY=your_relay_key_here
OPENAI_BASE_URL=https://your-relay.example.com/v1
OPENAI_MODEL=your_vision_model

If the app returns an authentication error, check whether the API key is valid, the relay supports /responses, the account has enough quota, and the configured model supports image input.

For production or team usage:

• Put the app behind nginx or Caddy.
• Enable HTTPS on public networks.
• Set AUTH_DISABLED=false when the app is exposed outside a trusted network.
• Set a strong ADMIN_PASSWORD.
• Store runtime data outside the repository.
• Configure backups and cleanup rules for uploaded media and reports.

See docs/deployment.md for a systemd and nginx example.

Uploaded videos and images may contain sensitive people, products, customer data, or unpublished creative assets. Treat SHOT_PLATFORM_HOME as private application data.

• Never publish .env, uploaded files, generated reports, SQLite databases, logs, or customer examples.
• Use access control for public deployments.
• Review your model provider or relay provider's data policy before uploading confidential media.
• Report security concerns privately using the process in SECURITY.md.

• Automatic shot segmentation is based on visual change detection and should be reviewed by a human.
• Prompt quality depends on the configured vision model.
• Large videos can take longer to upload, sample, and analyze.
• The default SQLite setup is suitable for lightweight team usage, not high-volume multi-tenant SaaS workloads.

app/
  main.py              FastAPI application and processing pipeline
  static/style.css     Web UI styles
  templates/           Jinja2 pages
docs/
  deployment.md        Deployment notes
.env.example           Example environment configuration
requirements.txt       Python dependencies
LICENSE                MIT license

• Add screenshot examples for the workbench and report pages.
• Add Docker deployment files.
• Add model-provider configuration presets.
• Add background job queue support for heavier workloads.
• Add tests for media processing, auth, and export behavior.

Contributions are welcome. Please read CONTRIBUTING.md before opening an issue or pull request.

This project is released under the MIT License.