Tuzi Short Video

Short Video Creator

Creates short videos for social media platforms. Analyzes content, adapts to platform specs, generates prompts, and calls tuzi-video-gen backend.

Script Directory

This is a workflow-only skill (no scripts). It delegates video generation to tuzi-video-gen.

VIDEO_GEN_DIR: Resolve from skills/tuzi-video-gen/SKILL.md relative to project root.

Step 0: Load Preferences ⛔ BLOCKING

CRITICAL: This step MUST complete BEFORE any video generation.

0.1 Check API Key

echo "${TUZI_API_KEY:-not_set}"
grep -s TUZI_API_KEY .tuzi-skills/.env "$HOME/.tuzi-skills/.env"

0.2 Check EXTEND.md

test -f .tuzi-skills/tuzi-short-video/EXTEND.md && echo "project"
test -f "$HOME/.tuzi-skills/tuzi-short-video/EXTEND.md" && echo "user"

EXTEND.md Supports: Default platform | Default model

Schema: references/config/preferences-schema.md

Step 1: Analyze User Input

Analyze the user's content (text, article, script, topic description):

• Identify main topic and key points
• Determine content type (tutorial, story, showcase, explainer)
• Estimate ideal video length and segment count

Step 2: Confirm Platform and Parameters

Use AskUserQuestion to let user choose target platform (unless specified or saved in EXTEND.md):

header: "Target Platform"
question: "目标发布平台？"
options:
  - label: "小红书"
    description: "9:16 竖屏, 720x1280"
  - label: "抖音"
    description: "9:16 竖屏, 1080x1920"
  - label: "X/Twitter"
    description: "16:9 横屏, 1280x720"
  - label: "视频号"
    description: "9:16/16:9, 1080x1920"

Platform Presets

Platform details: See references/platforms/

Step 3: Generate Video Script

Based on content analysis and platform:

Single Video (content fits one clip)

1. Write a detailed video generation prompt in English 2. Include: visual scenes, camera movements, style, mood, lighting 3. Save prompt to prompts/01-video-prompt.md

Long Video (content needs multiple segments)

1. Split content into logical segments (opening + body + ending) 2. Write per-segment prompts with continuity notes 3. Save to prompts/01-segment-opening.md, 02-segment-main.md, etc.

Workflow details: See references/workflows/

Step 4: Generate Video

Single Video

npx -y bun ${VIDEO_GEN_DIR}/scripts/main.ts \
  --promptfiles <output-dir>/prompts/01-video-prompt.md \
  --video <output-dir>/01-video-<slug>.mp4 \
  --size <platform-size> \
  --seconds <duration>

Long Video

npx -y bun ${VIDEO_GEN_DIR}/scripts/main.ts \
  --video <output-dir>/01-video-<slug>.mp4 \
  --segments <N> \
  --segment-prompts <output-dir>/prompts/01-segment-opening.md <output-dir>/prompts/02-segment-main.md ... \
  --size <platform-size> \
  --seconds <per-segment-seconds>

Step 5: Completion Report

Display:

• Output video path
• Platform and specs used
• Model used
• Duration

Output Directory

short-video/{topic-slug}/
├── source-{slug}.md
├── prompts/
│   ├── 01-video-prompt.md          (single video)
│   ├── 01-segment-opening.md       (long video)
│   ├── 02-segment-main.md
│   └── 03-segment-ending.md
├── segments/                        (long video temp, cleaned up)
└── 01-video-{slug}.mp4             (final output)

Extension Support

Custom configurations via EXTEND.md. See Step 0 for paths and supported options.

First-Time Setup

Same API key setup as tuzi-video-gen. See skills/tuzi-video-gen/references/config/first-time-setup.md for API key flow.

EXTEND.md Setup

Use AskUserQuestion:

Question 1: Default Platform

header: "Platform"
question: "默认发布平台？"
options:
  - label: "小红书"
    description: "9:16 竖屏, 720x1280, 15-60s"
  - label: "抖音"
    description: "9:16 竖屏, 1080x1920, 15-60s"
  - label: "X/Twitter"
    description: "16:9 横屏, 1280x720, 5-140s"
  - label: "视频号"
    description: "9:16/16:9, 1080x1920, 15-60s"

Question 2: Save Location

header: "Save"
question: "偏好保存位置？"
options:
  - label: "Project (Recommended)"
    description: ".tuzi-skills/ (仅当前项目)"
  - label: "User"
    description: "~/.tuzi-skills/ (所有项目)"

EXTEND.md Template

---
version: 1
default_platform: null
default_model: null
---

Preferences Schema

Full Schema

---
version: 1

default_platform: null     # xiaohongshu|douyin|x-twitter|weixin-video|null

default_model: null         # veo3|veo3.1|sora-2|null (null = veo3.1)
---

Field Reference

抖音 (Douyin / TikTok)

Video Specs

Content Tips

• 竖屏 9:16 是标准格式
• 支持更高分辨率（1080p）
• 前 1-3 秒决定用户是否继续观看
• 15s 短视频适合快节奏内容
• 60s 适合教程/故事类内容

微信视频号 (WeChat Channels)

Video Specs

Content Tips

• 竖屏 9:16 优先，信息流展示更大
• 也支持 16:9 横屏
• 15-30s 短视频完播率高
• 内容偏向知识分享、生活记录
• 封面和标题影响点击率

X / Twitter

Video Specs

Content Tips

• 横屏 16:9 是主流格式
• 短视频（15-30s）传播效果好
• 支持较长视频（最长 140s）
• 自动播放无声，前几秒需视觉吸引力
• 建议添加字幕

小红书 (Xiaohongshu / RedNote)

Video Specs

Content Tips

• 竖屏优先，9:16 是最佳比例
• 前 3 秒抓住注意力
• 15-30s 短视频效果最好
• 字幕/文字叠加提升完播率
• 封面图很重要，建议单独生成

Long Video Workflow

Flow

1. Analyze content, determine total segments needed 2. Split content into logical segments (opening, body sections, ending) 3. Generate per-segment prompts with continuity notes 4. Call tuzi-video-gen with --segments N --segment-prompts 5. Report result

Segment Planning

• Opening (1 segment): Hook, introduce topic
• Body (1-N segments): Main content, each segment covers one key point
• Ending (1 segment): Summary, call to action

Continuity Tips

• tuzi-video-gen auto-extracts last frame for next segment reference
• Keep visual style consistent across prompts
• Mention recurring elements (characters, settings) in each prompt
• Transition hints: "continuing from previous scene..."

tuzi-video-gen Call

npx -y bun ${VIDEO_GEN_DIR}/scripts/main.ts \
  --video <output-path> \
  --segments <N> \
  --segment-prompts seg1.md seg2.md seg3.md \
  --model <model> \
  --size <platform-size> \
  --seconds <per-segment-duration>

Single Video Workflow

Flow

1. Analyze user input (text, article, script) 2. Determine platform preset (size, duration, aspect ratio) 3. Generate video prompt based on content 4. Call tuzi-video-gen to generate video 5. Report result

Prompt Generation Guidelines

• Describe visual scenes, camera movements, lighting
• Include style keywords (cinematic, documentary, vlog, etc.)
• Specify mood and atmosphere
• Keep prompt under 500 words for best results
• For Chinese content, write prompts in English for better model understanding

tuzi-video-gen Call

npx -y bun ${VIDEO_GEN_DIR}/scripts/main.ts \
  --promptfiles <prompt-file> \
  --video <output-path> \
  --model <model> \
  --size <platform-size> \
  --seconds <duration>