Full pipeline to generate short-form videos from topic or script. Supports AI image mode (SiliconFlow) and real footage mode (Pexels/Pixabay clips). Multiple aspect ratios, transitions, and batch generation.
Use this skill when generating any short-form video. Two operating modes:
| Mode | When to use | Visual material |
|---|---|---|
| Image mode (default) | No footage clips available | AI-generated images via SiliconFlow |
Footage mode (--footage-dir) | After running pexels-footage or pixabay-footage | Real copyright-free video clips |
用 --topic 触发 LLM 生成脚本时，prompt 中须明确要求以下三段式结构：
| 段落 | 时长占比 | 目标 | 示例套路 |
|---|---|---|---|
| 开篇抓眼球 | 前 15–20% | 3 秒内让人停止划走 | "99% 的人都不知道…" / "我花了 XX 才搞明白" / 强反差开场 |
| 60–70% |
| 展示核心价值，每个卖点一句 |
| 场景化痛点 → 产品/方法解决 → 数据/案例佐证 |
| 结尾促下单 | 后 15–20% | 明确 CTA，降低决策门槛 | "链接在简介" / "点击立即领取" / "限时优惠只剩 XX 件" |
若调用方已提供完整剧本（
--script），跳过此结构，直接使用传入内容。
pip install edge-tts moviepy openai requests pillow
python3 {baseDir}/scripts/compose.py \
--topic "AI工具如何改变工作方式" \
--language zh \
--output-dir ./output_videos
# Step 1: Download footage
python3 {pexels_baseDir}/scripts/pexels_search.py \
--terms "technology,office,innovation" \
--aspect 9:16 \
--output-dir ./video_assets/footage
# Step 2: Compose with footage
python3 {baseDir}/scripts/compose.py \
--topic "AI工具如何改变工作方式" \
--footage-dir ./video_assets/footage \
--aspect 9:16 \
--transition fade \
--output-dir ./output_videos
python3 {baseDir}/scripts/compose.py \
--topic "大自然风光" \
--footage-dir ./video_assets/footage \
--aspect 16:9 \
--output-dir ./output_videos
python3 {baseDir}/scripts/compose.py \
--topic "AI工具推荐" \
--batch 3 \
--output-dir ./output_videos
--topic <text> Video topic (LLM generates script)
--script <text> Direct script (skip LLM step)
--language zh|en Language (default: zh)
--aspect 9:16|16:9|1:1 Aspect ratio (default: 9:16)
--footage-dir <dir> Directory of pre-downloaded video clips (enables footage mode)
--clip-duration <sec> Max seconds per clip in footage mode (default: 5)
--transition none|fade Clip transitions (default: none)
--n-images <N> Number of AI images in image mode (default: 5)
--duration-hint <sec> Target video duration hint (default: 45)
--batch <N> Generate N videos (default: 1)
--output-dir <dir> Output directory (required)
--assets-dir <dir> Intermediate files dir (default: output-dir/assets)
output_videos/
2026-04-05-AI工具.mp4 ← final video
2026-04-05-AI工具.json ← metadata
# Batch mode:
2026-04-05-AI工具-1.mp4
2026-04-05-AI工具-2.mp4
2026-04-05-AI工具-3.mp4
Metadata JSON includes: title, description, tags, duration, aspect, script, footage_mode.
1. Choose aspect ratio based on target platform:
- 9:16 → YouTube Shorts, TikTok, Instagram Reels, Douyin
- 16:9 → YouTube standard, Bilibili, WeChat Video
- 1:1 → Instagram Feed, WeChat Moments
2. Generate footage search terms from topic keywords (in English, simple nouns work best)
e.g., topic "AI工具推荐" → terms "technology,computer,work,office"
3. Run pexels-footage (or pixabay-footage if Pexels has no results)
4. Run shorts-compose with --footage-dir
5. If batch mode: show user all N videos, ask which to keep/publish
| Variable | Description | Default |
|---|---|---|
SILICONFLOW_API_KEY | API key for LLM + image gen | Required |
LLM_API_BASE | Override LLM endpoint | https://api.siliconflow.cn/v1 |
LLM_API_KEY | Override LLM key | SILICONFLOW_API_KEY |
LLM_MODEL | Model name | Qwen/Qwen2.5-7B-Instruct |
EDGE_TTS_VOICE | TTS voice | auto by language |
Default TTS voices: zh-CN-YunxiNeural (Chinese) · en-US-GuyNeural (English)
| Error | Action |
|---|---|
SILICONFLOW_API_KEY not set | Check environment |
edge-tts / moviepy not found | pip install edge-tts moviepy |
| No clips in footage-dir | Falls back to placeholder image, continues |
| Image generation fails | Uses placeholder image, continues |
| Batch: some videos fail | Reports per-video status, exits 1 only if all fail |