Viral Video Studio
End-to-end workflow for cloning viral TikTok content structure and building a monetizable channel.
Workflow Overview
CODEBLOCK0
Step 0: Define Character (Do This Once)
Before making any video, create a Character Bible. Read references/character-bible.md.
Key decisions:
- 1. Pick archetype: Overconfident Underdog / Dignified Aristocrat / Innocent Explorer / Unlikely Professional
- Define: species, color, signature outfit, personality, fatal flaw, motivation
- Write a "consistency prompt prefix" — prepend to EVERY AI image prompt
- Choose series format: single gags OR running series (series = faster follower growth)
Output: A completed character sheet saved to assets/character-sheet.md
See assets/character-sheet-template.md for the fill-in-the-blank template.
Step 1: Analyze Reference Video
When user provides a TikTok URL:
CODEBLOCK1
Then use the image tool to analyze /tmp/tiktok-thumb.jpg and describe:
- - Content type: Comedy / Educational / Story / Trend
- Format: AI-generated images / Live action / Screen recording / Animation
- Hook style: Text hook / Visual hook / Sound hook
- Pacing: Fast cuts / Slow / Building
- Emotion trigger: Humor / Surprise / Curiosity / Relatability
Step 2: Deep Viral Analysis
This is the most important step. Read references/viral-analysis.md for full protocol.
Quick analysis (run for every reference video):
- 1. 7 Triggers check — which psychological triggers fire? (surprise, relatability, curiosity, social currency, empathy, awe, identity)
- Frame-by-frame breakdown — what emotion does each moment create?
- The Real Why — what universal human experience is at the core?
- Share mechanic — who sends this to whom, and why?
- Red flags check — does the reference video have any fail patterns to avoid?
Output a Viral Analysis Report (template in viral-analysis.md) before writing any script.
Step 3: Reverse-Engineer Viral Structure
Decode the video into this framework:
CODEBLOCK2
Identify:
- - Character archetype: Small vs Big / Underdog / Relatable everyman
- Emotion arc: Anticipation → Tension → Release → Satisfaction
- Loop factor: Does it make viewers rewatch? How?
Step 4: Generate New Video Concept
Create a new video using the SAME structure but DIFFERENT content:
Concept generation prompt:
CODEBLOCK3
Pick the best concept and proceed.
Step 5: Scene-by-Scene AI Image Prompts
For each of the 6 frames, write Midjourney/DALL-E prompts following this formula:
CODEBLOCK4
Quality modifiers to always include:
- - INLINECODE6
- INLINECODE7 (TikTok native)
- INLINECODE8 or INLINECODE9
- INLINECODE10 /
shocked expression / INLINECODE12
Consistency tip: Start each prompt with "Same [character description]" to maintain visual consistency across frames.
🎀 Cute Factor: Cute characters — chubby body, round shape, big eyes — create instant emotional investment. The cuter the character, the faster viewers bond and the harder they laugh at failures. Lean into exaggerated cuteness. See references/character-bible.md for the Cute Factor framework.
Video generation options (choose by budget):
- - Budget: DALL-E images → CapCut slideshow (~$0.50/video)
- Mid: Hailuo AI video clips → CapCut (~$0.75/video)
- Best quality: Kling AI with character reference image (~$1-3/video)
- See
references/video-generation-tools.md for full comparison, pricing, and all 15 Kling AI prompts for Hammy video.
See references/prompt-templates.md for 10 ready-to-use prompt templates.
See references/script-templates.md for 4 full script templates + 30 content ideas.
See references/character-bible.md for character archetypes, quality checklist, and emotional arc framework.
See references/viral-analysis.md for the 7 viral triggers, deep analysis protocol, and red flags checklist.
Step 6: Full Scene-by-Scene Script
For EVERY scene (frame), specify ALL of the following in detail:
CODEBLOCK5
Ví dụ cảnh chi tiết:
CODEBLOCK6
Quy tắc âm thanh theo từng loại cảnh:
| Loại cảnh | Nhạc nền | SFX gợi ý |
|---|
| HOOK | Dramatic sting / Silence → bang | Single piano chord |
| SETUP |
Sneaky/spy theme | Footsteps, paper rustle |
| CONFLICT | Tension buildup | Machine hum, creaking |
| PEAK | Music STOPS → chaos | Explosion, scream |
| TWIST | Sad trombone / Magic sparkle | Record scratch, fail horn |
| CTA | Upbeat happy | Victory fanfare (mini) |
Quy tắc hình ảnh đa frame → video:
Mỗi video = 6-15 frames ảnh AI ghép thành video trong CapCut:
- - Frame 1-2: HOOK (3-6 giây)
- Frame 3-5: SETUP + CONFLICT (10-15 giây)
- Frame 6-8: PEAK (8-10 giây)
- Frame 9-10: TWIST (5-7 giây)
- Frame 11-12: RESOLUTION + CTA (5-7 giây)
Ken Burns effect trên mỗi frame tạo cảm giác chuyển động dù ảnh tĩnh:
- - Zoom in: tạo tension, dramatic
- Zoom out: reveal surprise
- Pan left/right: following character movement
Step 7: Algorithm Optimization
TikTok ranks videos by priority order:
- 1. Completion rate — watch to the end? (most important)
- Rewatch rate — watch again?
- Shares — send to someone?
- Comments — feel compelled to respond?
- Follows — earn new followers?
Every creative decision must serve one of these metrics.
Hook Formulas (First 3 Seconds — Most Critical)
CODEBLOCK7
Video Length Strategy
- - 7-15s: Memes, single-gag AI images → highest completion rate
- 15-30s: Story arc (6 frames) → best balance of value + retention
- 30-60s: Extended story, series episode → loyal audience
- 1-3 min: Deep dives → lower reach but higher-quality followers
Engagement Tactics
- - Reply to every comment in first hour → algorithm boost
- Pin the funniest comment on each video
- Create follow-up videos responding to comments
- Ask a specific question in CTA (not generic "comment below")
- Use "Part 2?" as CTA to tease continuation
Step 8: CapCut Assembly Instructions
CODEBLOCK8
Step 9: 30-Day Posting Calendar + Monetization Roadmap
Posting Calendar
See references/posting-calendar.md for the full template.
Quick schedule:
- - Week 1-2: Post 2x/day, test 4 different content styles
- Week 3-4: Double down on best-performing style, post 3x/day
- Best times (Vietnam timezone GMT+7): 7-9 AM, 12-1 PM, 8-10 PM
Content mix per week:
CODEBLOCK9
See references/trend-riding.md for the full trend-riding playbook.
See references/channel-setup.md for new channel setup guide.
Monetization Roadmap
See references/monetization.md for full details.
Quick milestones:
CODEBLOCK10
Fastest monetization paths for AI comedy content:
- 1. Midjourney affiliate: 20% recurring commission — promote in bio
- CapCut Pro affiliate: Promote editing workflow
- Merchandise: Print-on-demand with your character designs
- Course: "How I make viral AI videos" — sell for $29-99
Viral Video Studio
克隆病毒式TikTok内容结构并建立可盈利频道的端到端工作流程。
工作流程概览
第0步:定义角色(一次性完成)— 创建角色圣经,填写角色表
第1步:分析参考视频 — 下载、提取帧、解码内容类型
第2步:深度病毒式分析 — 7个触发点、逐帧分解、分享机制
第3步:逆向工程病毒式结构 — 钩子/铺垫/冲突/高潮/反转/行动号召
第4步:生成新视频概念 — 相同结构,新角色/场景
第5步:逐场景AI图像提示 — 6帧,Midjourney/DALL-E
第6步:完整脚本含文字叠加 — 每帧规格含音效提示
第7步:算法优化 — 完播率、钩子公式、时长策略
第8步:CapCut剪辑指南 — 导入、节奏、特效、导出
第9步:30天发布日历 + 盈利路线图
第0步:定义角色(一次性完成)
在制作任何视频之前,创建角色圣经。阅读 references/character-bible.md。
关键决策:
- 1. 选择原型:过度自信的弱者 / 高贵的贵族 / 天真的探索者 / 不靠谱的专业人士
- 定义:物种、颜色、标志性服装、个性、致命缺陷、动机
- 编写一致性提示前缀 — 在每个AI图像提示前添加
- 选择系列格式:单次搞笑或连载系列(系列 = 更快的粉丝增长)
输出: 完成的角色表保存至 assets/character-sheet.md
参见 assets/character-sheet-template.md 获取填空模板。
第1步:分析参考视频
当用户提供TikTok链接时:
bash
下载视频
yt-dlp -o /tmp/tiktok-ref.%(ext)s
提取第一帧用于视觉分析
ffmpeg -i /tmp/tiktok-ref.mp4 -ss 00:00:01 -frames:v 1 /tmp/tiktok-thumb.jpg -update 1
尝试获取字幕(可能不可用)
yt-dlp --write-auto-sub --sub-lang vi,en,zh --skip-download -o /tmp/tiktok-sub
然后使用 image 工具分析 /tmp/tiktok-thumb.jpg 并描述:
- - 内容类型:喜剧 / 教育 / 故事 / 趋势
- 格式:AI生成图像 / 实拍 / 屏幕录制 / 动画
- 钩子风格:文字钩子 / 视觉钩子 / 音效钩子
- 节奏:快速剪辑 / 缓慢 / 递进
- 情感触发:幽默 / 惊喜 / 好奇 / 共鸣
第2步:深度病毒式分析
这是最重要的一步。阅读 references/viral-analysis.md 获取完整协议。
快速分析(对每个参考视频执行):
- 1. 7个触发点检查 — 哪些心理触发点被激活?(惊喜、共鸣、好奇、社交货币、共情、敬畏、身份认同)
- 逐帧分解 — 每个时刻创造什么情感?
- 真正的原因 — 核心是什么普遍的人类体验?
- 分享机制 — 谁发给谁,为什么?
- 红旗检查 — 参考视频是否有需要避免的失败模式?
在编写任何脚本之前,输出病毒式分析报告(模板见 viral-analysis.md)。
第3步:逆向工程病毒式结构
将视频解码为以下框架:
钩子(0-3秒): 什么阻止了滑动?
铺垫(3-10秒): 前提是什么?
冲突(10-25秒): 张力/挑战是什么?
高潮(25-35秒): 高潮是什么?
反转(35-45秒): 惊喜或笑点
行动号召(45-55秒):它要求什么行动?
识别:
- - 角色原型:小对大 / 弱者 / 有共鸣的普通人
- 情感弧线:期待 → 紧张 → 释放 → 满足
- 循环因素:是否让观众重看?如何做到的?
第4步:生成新视频概念
使用相同结构但不同内容创建新视频:
概念生成提示:
基于这个病毒式结构:[解码后的结构]
使用相同的情感弧线但不同的角色/场景,生成5个新视频概念。
每个概念:标题 + 角色 + 场景 + 反转
格式:AI生成图像系列(6帧)
选择最佳概念并继续。
第5步:逐场景AI图像提示
为每6帧编写Midjourney/DALL-E提示,遵循以下公式:
[主体] [动作] [场景] [情感] [风格:照片级真实/卡通/电影感] [宽高比:9:16竖屏]
始终包含的质量修饰词:
- - 照片级真实,超详细,专业摄影
- 9:16竖屏格式(TikTok原生)
- 戏剧性灯光 或 柔和自然光
- 搞笑表情 / 震惊表情 / 自信姿势
一致性技巧: 每个提示以相同[角色描述]开头,以保持跨帧视觉一致性。
🎀 可爱因素: 可爱的角色 — 圆润身体、圆形、大眼睛 — 创造即时的情感投入。角色越可爱,观众建立联系越快,对失败的反应越强烈。充分利用夸张的可爱。参见 references/character-bible.md 获取可爱因素框架。
视频生成选项(按预算选择):
- - 预算型:DALL-E图像 → CapCut幻灯片(约$0.50/视频)
- 中等:Hailuo AI视频片段 → CapCut(约$0.75/视频)
- 最佳质量:Kling AI带角色参考图像(约$1-3/视频)
- 参见 references/video-generation-tools.md 获取完整对比、定价以及Hammy视频的所有15个Kling AI提示。
参见 references/prompt-templates.md 获取10个即用提示模板。
参见 references/script-templates.md 获取4个完整脚本模板 + 30个内容创意。
参见 references/character-bible.md 获取角色原型、质量检查清单和情感弧线框架。
参见 references/viral-analysis.md 获取7个病毒式触发点、深度分析协议和红旗检查清单。
第6步:完整逐场景脚本
对每个场景(帧),详细指定以下所有内容:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
场景 [N] — [时间戳,例如 0:00-0:03]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
👁️ 视觉:
- 角色:[谁?在做什么?什么表情?什么姿势?]
- 背景:[地点,灯光,镜头角度]
- 运动:[静态 / Ken Burns放大 / Ken Burns缩小 / 平移]
- 重要细节:[什么物体突出?主色调?]
🎨 DALL-E / MIDJOURNEY提示:
[角色描述],[动作],[场景],[情感],[灯光],照片级真实,9:16竖屏
负面提示:模糊,变形,文字,水印
📝 文字叠加:
内容:[最多6个词 + emoji]
位置:[顶部 / 底部 / 居中]
字体样式:[粗体白色 + 黑色描边]
🔊 音效:
背景音乐:[音乐类型/情绪:狡猾/戏剧性/快乐/悲伤...]
音效:[具体音效名称:唱片刮擦 / 藤蔓爆裂 / 悲伤长号 / 无]
音乐音量:[100% / 70% / 50%]
⏱️ 时长:[X秒]
🎭 目标情绪:[观众在此场景应感受到什么?]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
详细场景示例:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
场景1 — 0:00-0:03(钩子)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
👁️ 视觉:
- 角色:Hammy(橙色胖仓鼠)直立,双手叉腰
表情:自信,眼中闪烁着决心
姿势:直视镜头,挺胸
- 背景:在巨大的银行门前,低角度仰拍
银行对Hammy来说像摩天大楼一样高
- 运动:Ken Burns慢速缩小以显示银行大小
- 细节:早晨阳光,Hammy的影子在人行道上拉长
🎨 DALL-E提示:
Hammy,一只小小的圆胖橙色仓鼠,黑色大眼睛,
穿着黄色樱桃印花短裤,自信