Full text-to-video pipeline (MoneyPrinterTurbo style). Accepts a topic or script, generates narration via SiliconFlow TTS, assembles video from local footage, local images, or AI-generated clips, and outputs an MP4. Designed to be called directly by users or spawned by pro-selfmedia-operator.
Use this skill when:
shorts-compose provides (local files, AI clips)Pipeline:
Topic / Script
→ LLM Script (if topic given)
→ SiliconFlow TTS (audio narration)
→ Video Materials (local video / local images / AI-generated via SiliconFlow)
→ MoviePy Composition
→ MP4 + metadata.json
t2video vs shorts-compose
t2videouses SiliconFlow TTS (Chinese-optimized, no external process needed)t2videonatively handles local video/image files as material input
shorts-compose uses edge-tts and focuses on Pexels/AI-image modes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 |
| 中段讲卖点 | 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 openai moviepy requests pillow
# default: SiliconFlow generates images as background frames
python3 {baseDir}/scripts/t2video.py \
--topic "为什么AI正在改变每一个行业" \
--language zh \
--output-dir ./output_videos
python3 {baseDir}/scripts/t2video.py \
--script "人工智能正以前所未有的速度渗透到每一个行业..." \
--language zh \
--output-dir ./output_videos
# Use pre-downloaded .mp4 / .mov files from a directory
python3 {baseDir}/scripts/t2video.py \
--topic "城市夜景之美" \
--footage-dir ./assets/city_night \
--aspect 9:16 \
--output-dir ./output_videos
# Slideshow from a directory of .jpg / .png files
python3 {baseDir}/scripts/t2video.py \
--topic "春天的花朵" \
--images-dir ./assets/spring_photos \
--aspect 16:9 \
--output-dir ./output_videos
# Generates video clips via SiliconFlow Wan2.2 API for each scene
# WARNING: each clip takes 1-5 minutes — total may be 10-30 min for full video
python3 {baseDir}/scripts/t2video.py \
--topic "深海生物的神秘世界" \
--auto-gen-video \
--aspect 9:16 \
--output-dir ./output_videos
python3 {baseDir}/scripts/t2video.py \
--topic "10个提升效率的方法" \
--batch 3 \
--output-dir ./output_videos
--topic <text> Video topic (LLM generates script from this)
--script <text> Direct script text (skips LLM)
--language zh|en Language for TTS (default: zh)
--aspect 9:16|16:9|1:1 Output aspect ratio (default: 9:16)
Material source (choose one; default = SiliconFlow image generation):
--footage-dir <dir> Directory of local video clips (.mp4/.mov/.avi)
--images-dir <dir> Directory of local image files (.jpg/.png)
--auto-gen-video Generate video clips via SiliconFlow Video API (slow)
TTS options:
--voice <name> SiliconFlow TTS voice name (default: FishAudio/speech-01-hd)
Leave empty string "" for dynamic voice
Composition:
--clip-duration <sec> Max seconds per footage/image clip (default: 5)
--transition none|fade Clip transitions in footage/images mode (default: none)
--n-images <N> Number of AI images in default image mode (default: 5)
--batch <N> Generate N versions (default: 1)
--output-dir <dir> Output directory (required)
output_videos/
2026-04-05-AI行业-1.mp4 ← final video
2026-04-05-AI行业-1.json ← metadata (script, terms, duration, voice, aspect)
assets/
audio.mp3 ← TTS narration
scene_*.png / clip_*.mp4 ← intermediate frames/clips
| Situation | Recommended mode |
|---|---|
| Have real-life footage for topic | --footage-dir |
| Have themed photos | --images-dir |
| Want highest quality AI visuals (slow) | --auto-gen-video |
| Quick draft / no materials | default (AI images, fast) |
| Voice ID | Style | Best for |
|---|---|---|
FishAudio/speech-01-hd | Natural, HD | General Chinese narration (default) |
FishAudio/speech-01-turbo | Fast | High-volume batch generation |
"" (empty string) | Dynamic | Auto-selected by platform |
Note on TTS API: SiliconFlow TTS uses
/v1/audio/speech(OpenAI-compatible). The user-facing docs at/audio/create-audio-transcriptionscover speech-to-text, not TTS.
| Variable | Description |
|---|---|
SILICONFLOW_API_KEY | API key for LLM + TTS + image/video generation (required) |
LLM_API_BASE | Override LLM endpoint (default: https://api.siliconflow.cn/v1) |
LLM_API_KEY | Override LLM key (defaults to SILICONFLOW_API_KEY) |
LLM_MODEL | Model name (default: Qwen/Qwen2.5-7B-Instruct) |
| Error | Action |
|---|---|
SILICONFLOW_API_KEY not set | Check environment variables |
| TTS request fails | Verify API key; try --voice "" for dynamic voice |
moviepy not installed | pip install moviepy |
--auto-gen-video timeout | Each clip may take up to 10 min; consider reducing scene count or using --footage-dir |
| No material found in directory | Check path; ensure files have correct extensions |