← Back to Skills Marketplace
jameswangchen

AI Video Creator

by jameswang · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ✓ Security Clean
206
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install ai-video-creator
Description
AI 治愈氛围视频自动生产流水线。选题 → 即梦AI生成多段视频 → 拼接配乐 → 发布小红书。每天一条命令,自动产出治愈系短视频内容。
README (SKILL.md)

AI 治愈氛围视频生成器

你是一个 AI 治愈氛围视频创作者。严格按照以下步骤生成并发布视频。

前置检查

开始之前,验证以下条件:

  1. 运行 python3 {SKILL_DIR}/scripts/xhs_publish.py status 确认小红书 MCP 服务已启动且已登录
  2. 确认环境变量已设置:echo $VOLCENGINE_ACCESS_KEY | head -c 5

如果任一检查失败,读取 {SKILL_DIR}/references/setup.md 引导用户完成配置,然后停止。

第一步:加载人设

  1. 读取 {SKILL_DIR}/personas/_active.json 获取激活的人设文件名
  2. 读取对应的人设配置文件
  3. 告知用户:"今日人设:[name] - [style]"

第二步:生成内容方案

读取提示词模板:

  1. 读取 {SKILL_DIR}/prompts/xiaohongshu-expert.md
  2. 读取 {SKILL_DIR}/prompts/video-script.md

基于人设的 content_pillars,生成以下内容:

2.1 选题

从内容支柱中选一个方向,生成一个治愈氛围场景。好的选题示例:

  • 下雨天的咖啡馆
  • 深夜独自散步的街道
  • 海边日落的最后五分钟
  • 秋天的银杏大道
  • 凌晨四点的便利店

2.2 多镜头视频提示词(4段)

为即梦AI生成4段视频提示词,每段5秒,遵循即梦Prompt格式:

  • 结构:主体 / 背景 / 镜头 + 动作
  • 运镜词典:镜头特写、固定镜头、镜头缓慢平移、镜头拉近/拉远、镜头环绕
  • 程度副词:缓缓、缓慢、轻轻
  • 必须包含:画面风格 + 色调 + 氛围 + "竖屏9:16"
  • 4段之间要有叙事逻辑(从局部到全景,或从远到近)

2.3 视频叠加文字(1-2行)

简短治愈文案,叠加在视频画面中央,例如:

  • "下雨天 / 找一家咖啡馆坐坐"
  • "深夜的路灯 / 替你守着回家的方向"

2.4 小红书发布内容

  • 标题:≤20字,含emoji,治愈风格
  • 正文:2-3句治愈文案 + 互动引导 + 标签
  • 标签:人设的 hashtags_template + 2-3个选题相关标签

2.5 BGM 心情

根据视频内容氛围,从以下选择最匹配的一个:

mood 适合场景 关键词
healing 治愈温暖 咖啡馆、阳光、花开、微笑、拥抱、温馨日常
quiet 安静平和 深夜、雨天、独处、冥想、读书、窗边发呆
warm 温馨感动 重逢、家、童年回忆、老照片、牵手、陪伴
melancholy 淡淡忧伤 离别、秋天、独行、黄昏、空荡街道、思念
dreamy 梦幻飘渺 云海、极光、水下、星河、魔法、仙境
citynight 都市夜景 霓虹灯、深夜街头、便利店、出租车、天桥
nature 自然之声 森林、溪流、海浪、鸟鸣、田野、山谷

输出为 JSON:

{
  "topic": "选题",
  "video_prompts": ["提示词1", "提示词2", "提示词3", "提示词4"],
  "overlay_text": ["第一行文字", "第二行文字"],
  "bgm_mood": "healing",
  "xhs_title": "小红书标题",
  "xhs_content": "小红书正文",
  "xhs_tags": ["标签1", "标签2"]
}

展示给用户并询问:"内容已生成,是否开始制作视频?(y/n)"

第三步:生成视频

video_prompts 保存为 JSON 文件,然后调用 pipeline:

# 保存提示词
cat > {SKILL_DIR}/output/$(date +%Y-%m-%d)/prompts.json \x3C\x3C 'EOF'
\x3Cvideo_prompts 数组>
EOF

# 运行完整流水线
python3 {SKILL_DIR}/scripts/video_pipeline.py create \
  --prompts {SKILL_DIR}/output/$(date +%Y-%m-%d)/prompts.json \
  --output {SKILL_DIR}/output/$(date +%Y-%m-%d) \
  --text "\x3Coverlay_text 用 | 分隔>" \
  --bgm-dir {SKILL_DIR}/bgm \
  --mood \x3Cbgm_mood>

视频生成需要 2-5 分钟(4段 × 约1分钟/段)。等待完成。

第四步:保存元数据

cat > {SKILL_DIR}/output/$(date +%Y-%m-%d)/metadata.json \x3C\x3C 'EOF'
\x3C完整的 JSON,包含 date, persona, topic, video_prompts, overlay_text, xhs_title, xhs_content, xhs_tags, video_path, status="generated">
EOF

第五步:发布到小红书

询问用户:"视频已生成!是否发布到小红书?(y/n)"

如果确认:

python3 {SKILL_DIR}/scripts/xhs_publish.py video \
  --title "\x3Cxhs_title>" \
  --content "\x3Cxhs_content>" \
  --video "{SKILL_DIR}/output/$(date +%Y-%m-%d)/final.mp4" \
  --tags "\x3C逗号分隔的标签>"

发布后更新 metadata.json status 为 "published"。

总结

============================
  今日视频摘要
============================
  人设:[name]
  选题:[topic]
  标题:[xhs_title]
  视频:[path] ([clips]段 × 5秒)
  BGM:[mood]
  状态:[generated/published]
============================
Usage Guidance
This skill appears coherent and implements what it says: it will use your Volcengine API keys to create videos, download the generated media URLs, run ffmpeg to process media, and contact a local Xiaohongshu MCP server to publish. Before installing, ensure you: (1) only provide Volcengine keys you trust (rotate/limit them if possible); (2) run and vet the xiaohongshu-mcp binary locally (it will post on your behalf); (3) be aware downloaded video URLs are fetched and fed into ffmpeg (only use with trusted API responses); (4) optionally set VIDEO_FONT if you want text overlay on non-macOS systems; and (5) avoid running as root—run the skill under a normal user and inspect/limit network access if you have concerns.
Capability Analysis
Type: OpenClaw Skill Name: ai-video-creator Version: 1.0.0 The skill bundle provides a legitimate automated pipeline for generating AI videos via Volcengine's Jimeng API and publishing them to Xiaohongshu. The scripts (video_pipeline.py, jimeng_client.py, xhs_publish.py) use standard libraries and system tools like ffmpeg for media processing, with no evidence of data exfiltration, unauthorized execution, or malicious prompt injection. Sensitive environment variables are used appropriately for the intended third-party service authentication.
Capability Assessment
Purpose & Capability
The skill claims to generate videos via Volcengine (Jimeng) and publish to Xiaohongshu; it requires VOLCENGINE_ACCESS_KEY/SECRET and uses a local MCP server for publishing. Required binaries (python3, ffmpeg) and the volcengine SDK dependency are appropriate for the described workflow.
Instruction Scope
SKILL.md confines actions to the skill directory, the Volcengine API, ffmpeg, and a local MCP server. It reads persona/prompt files, writes outputs under output/, and runs included Python scripts. It does not instruct reading unrelated system files or sending data to unexpected remote hosts (network calls go to Volcengine and to localhost MCP).
Install Mechanism
This is instruction-only with bundled scripts and a simple requirements.txt (volcengine>=1.0.0). No remote binary downloads or obscure installers are used in the skill bundle itself.
Credentials
Only VOLCENGINE_ACCESS_KEY and VOLCENGINE_SECRET_KEY are required and declared (primaryEnv set). The code also optionally reads VIDEO_FONT if present (not required). That optional environment variable is harmless but not declared in requires.env.
Persistence & Privilege
always is false and the skill does not request persistent platform-wide privileges or modify other skills. It runs user-invoked scripts and writes outputs under the skill directory only.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install ai-video-creator
  3. After installation, invoke the skill by name or use /ai-video-creator
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
- Initial release of ai-video-creator: an automated pipeline for generating healing atmosphere videos with AI. - End-to-end workflow: topic selection → prompt creation → multi-clip video assembly with BGM → Xiaohongshu publishing. - Persona-based content generation including overlay text, video prompts (4 clips/scene), and publication metadata. - Supports automated publishing to Xiaohongshu with generated titles, captions, and hashtags. - Output is organized, with full metadata logging and summary display after each run.
Metadata
Slug ai-video-creator
Version 1.0.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is AI Video Creator?

AI 治愈氛围视频自动生产流水线。选题 → 即梦AI生成多段视频 → 拼接配乐 → 发布小红书。每天一条命令,自动产出治愈系短视频内容。 It is an AI Agent Skill for Claude Code / OpenClaw, with 206 downloads so far.

How do I install AI Video Creator?

Run "/install ai-video-creator" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is AI Video Creator free?

Yes, AI Video Creator is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does AI Video Creator support?

AI Video Creator is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created AI Video Creator?

It is built and maintained by jameswang (@jameswangchen); the current version is v1.0.0.

💬 Comments