When to open:When you want to make AI comic dramas / AI outfit videos / digital humans / SillyTavern character videos / supporting videos for AI web pages. This wiki extracts and compares the SOPs from 4 schools.
One-sentence core:Consensus across the 4 SOP schools = "Script/Storyboard → Asset Library → Image Generation → Video Generation → Stitching";Differences lie in tool stack selection + character consistency strategy + granularity of engineering layering. AI video generation has evolved from a "personal toy" to "agency small-team production (5x revenue share for comic dramas) + Skill-based batch production (Vibe Marketing) + platform 400 million yuan financing (Astro Kid)".
0. "AI Video" from the 4 Schools
| Perspective | Source | Core Model |
|---|---|---|
| Agency Small-Team School | Caviar AI Comic Drama Production Course / Complete methodology from Section 1 live teaching | 5-step AI comic drama process + 5x revenue share on Hongguo Theater + 100 minutes of content with a 1-month production cycle |
| Canvas-Based Creation School | Xiaohongshu / Complete workflow for 1-to-1 replication of AI videos / Xiaohongshu / AI web pages + 3D interaction + Pretext design collection | 5-step 1-to-1 replication for AI outfits / 6-step cinematic web pages / reverse-engineered prompts + image-to-image fusion |
| Engineering Layering School | Wang Kai SillyTavern Video Production SOP (Skill implementation example) | Phase 1-7 / Components + Compositions layering / Jimeng CLI premium membership + Skill system |
| Digital Human Translation School | dontbesilent AI self-media course / 13-section systematic methodology §L05 + §L10 | 5-tool combo + one-click translation into 30 languages / 2 million views on Douyin |
1. Cross-Author Consensus (5 Points)
1.1 Core workflow is consistent:Script/Storyboard → Asset Library → Image Generation → Video Generation → Stitching
| School | Workflow |
|---|---|
| Caviar Comic Drama | Novel → AI dynamic comic script → Storyboard sheet → Asset library (characters/scenes) → Jimeng image fusion storyboard frames → Jimeng/Vidu image-to-video → CapCut stitching |
| Xiaohongshu Outfits | Original video → Reverse-engineered prompts → Image-to-image fusion (clothing × model) → Motion transfer → Batch nodes → Editing |
| Wang Kai SillyTavern | Directory skeleton → Reference analysis → Asset scraping → Skill system → First video → Three types of videos → Scaling |
| Lao D Digital Human | Doubao white-background image → Live-action recorded video → wan2.2 animate face swap → ElevenLabs voice swap → CapCut compositing → HeyGen translation |
1.2 "Director mindset" is the core (exclusive to Caviar but agreed by all schools)
Caviar AI Comic Drama Production Course / Complete methodology from Section 1 live teaching:
"Essentially, comic dramas are pieced together from different shots,as a drama maker you are essentially the director, AI is only an assistant. "
Key mindset — Image first:
"Don't write segment by segment; first carefully list and finish all storyboards, then generate images with a sense of the visuals. "
Why:AI is a material library,the more detailed the description, the fewer choices left for AI, the higher the hit rate of card draws.
1.3 Character consistency is the biggest challenge
| School | Consistency Strategy |
|---|---|
| Caviar | Three-view character locking(hairstyle is the key distinguishing dimension)/ Store characters in asset library / Jimeng image fusion |
| Xiaohongshu Outfits | Image-to-image fusion(clothing × specified model image)/ Motion transfer |
| Wang Kai SillyTavern | Components (reusable units)/ Jimeng image2image anchoring |
| Lao D Digital Human | White-background image ★ Key(avoids background recognition errors)/ wan2.2 animate face swap |
See wiki-content-replication(memory) and jimeng-multishot-replicate skill —— "Jimeng multi-storyboard must use image2image anchoring".
1.4 Jimeng + Vidu are the mainstream tool stack
| Tool | Use |
|---|---|
| Jimeng | Text-to-image + image-to-video + character fusion / Premium membership offers high efficiency |
| Vidu | Video generation supplement |
| HeyGen | Lip sync / Digital human / 30-language translation |
| Tongyi Wanxiang wan2.2 animate | Character replacement face swap |
| ElevenLabs | Voiceover / Voice Changer for 29 languages |
| CapCut | Keyframe compositing / Subtitles / Loudness normalization |
1.5 Prompt details determine success or failure
Caviar AI Comic Drama Production Course / Complete methodology from Section 1 live teaching Prompt quick reference checklist:
- Body type (height / weight / body description)
- Hairstyle (key distinguishing dimension)
- Clothing (color / style / material)
- Accessories (glasses / necklace / bag)
- Art style (anime / realistic / retro)
2. Explicit Differences (3 Points — Not Smoothed Over)
2.1 Difference 1: Personal account building vs agency small-team system?
| Perspective | Position |
|---|---|
| Caviar(Agency small-team system) | "Do not recommend personal account building" —— Account building cycle is too long, drama production takes too long, hard to manage both ends;Join the agency small-team system:several people co-produce one drama, with revenue share + production fees |
| Xiaohongshu vibecoding | Led by individual creators —— 1-to-1 replication / 30-day ARPG challenge / buildinpublic |
| Wang Kai SillyTavern | Personal engineering = transferable to any product —— Phase 1-7 / Skill system / Mass production of SillyTavern videos by one person |
| Lao D | Personal digital human + multilingual —— 2 million views on a single Douyin video |
Real difference:Whether a team is needed:
- 100-minute comic drama × 1 month → Agency small-team system is mandatory
- AI outfits / digital humans → Fully feasible for individuals
- SillyTavern mass production → Personal engineering (Skill) is sufficient
2.2 Difference 2: Manual workflow vs Skill-based batch production?
| Perspective | Position | |---|---| | Caviar(Manual workflow) | 9-column storyboard sheet + 6 major camera movements filled manually | Comic dramas require the director's sense of visuals | | Xiaohongshu 1-to-1 replication | Canvas nodes + batch nodes | Visualized workflow, batch nodes run multiple groups | | Wang Kai SillyTavern | Skill system + Components/Compositions | Transferable to mass production of any product videos | | Xiaohongshu Vibe Marketing | Claude Code Skill + FFmpeg local stitching | "Batch produce only when the formula for one type of video is very clear" | | Lao D Digital Human | 5-tool combo manually chained | 2 million Douyin views but still manual |
Real difference:Boundary between Skill-based and manual:
- Clear formula + repeated many times → Skill-based
- Creativity-led + one-off → Manual
- Creativity + replication → Components (reusable units) + Compositions (assemblies)
2.3 Difference 3: Jimeng, Sora, or others?
| Perspective | Choice |
|---|---|
| Caviar(Comic drama) | Jimeng(three-view) + Vidu(supplement) + start and end frames can be added |
| Xiaohongshu 1-to-1 replication | Canvas-based operation + Jimeng workflow |
| Xiaohongshu AI web pages(Yanyan) | Higgsfield + Seedance(cinematic web page character video) + Google AI Studio |
| Wang Kai(Jimeng CLI) | Jimeng premium membership + CC-installed Jimeng CLI running automatically across multiple windows "Each YouTube account needs at least 10 pieces of material per day" |
| Lao D Digital Human | Doubao image generation + wan2.2 animate(character replacement) + ElevenLabs(voice swap) |
| Sora(OpenAI) | (lenny mentioned Sora shipped Android in 18 days, but it has not become a mainstream tool) |
Real difference:Niche determines tools:
- Comic drama → Jimeng + Vidu (available in China / friendly to Hongguo Theater)
- Outfits → Canvas-based + reverse-engineered prompts
- Cinematic web pages → Higgsfield + Seedance + Google AI Studio
- Digital human → Doubao + wan2.2 + HeyGen translation
3. Caviar AI Comic Drama 5-Step Process (Exclusive Full Version)
Caviar AI Comic Drama Production Course / Complete methodology from Section 1 live teaching:
3.1 5-step production process
小说 / 文章
↓ 1. AI 生成 → 动态漫剧本(标准格式)
↓ 2. AI 转换 → 分镜表(9 列)+ 资产库提炼
↓ 3. 即梦生图(三视图)→ 角色 / 场景 / 物品资产库
↓ 4. 即梦融图 + 文生图 → 分镜图(单帧)
↓ 5. 即梦/Vidu 图生视频(可加首尾帧)→ 分镜视频
↓ 6. 剪映拼接 → 成片
3.2 9-column storyboard sheet structure (exclusive)
| Column | Content |
|---|---|
| 1 | Shot number (e.g. 2-1 / 2-2 / 2-3, close-ups within the same storyboard use sub-numbers) |
| 2 | Scene (day/night + interior/exterior + location) |
| 3 | Characters (list of appearing roles) |
| 4 | Action description (script actions → visual narration for AI) |
| 5 | AI text-to-image logic prompt(directly copy to Jimeng) |
| 6 | Camera movement(push in / pull out / pan / track / follow / orbit + combinations) |
| 7 | v.0.1 Dialogue |
| 8 | Dialogue corresponding to camera movement (detailed mapping to extreme long shot / full shot / medium shot / close-up / extreme close-up) |
| 9 | Notes (director's hand-drawn sketches / links to drawn images / links to generated videos) |
3.3 6 basic camera movements + 1 subjective perspective
| Camera Movement | Applicable Scenario |
|---|---|
| Push in | Zoom in to focus on details (protagonist's facial expression / close-up of objects) |
| Pull out | Zoom out to show the environment (cut from close-up to long shot, often used at the end) |
| Pan | Sweep across the scene horizontally (show spatial sense) |
| Track | Lens moves horizontally/vertically |
| Follow | Follow the character |
| Orbit | 360-degree display |
| Subjective perspective | First-person view |
3.4 Hongguo Theater monetization
- 2026 Hongguo heavily promotes AI comic dramas
- New revenue share model released in 2026-12:5x revenue share coefficient
- A full episode requires ~100 minutes / Solo production cycle ~1 month
- Do not recommend personal account building, join the agency small-team system
4. Xiaohongshu AI Outfit 1-to-1 Replication SOP (Exclusive)
Xiaohongshu / Complete workflow for 1-to-1 replication of AI videos:
4.1 Complete workflow (canvas-based operation)
[原视频] → [文生图反推] → [图生图融合(服装×模特)] → [动作迁移裁剪]
→ [批量节点生成视频] → [剪辑打关键帧] → [成品]
4.2 Detailed 5 steps
Step 1: Text-to-image reverse engineering — generate the same scene
1. 将需要复刻的同款视频拖入画布
2. 在视频节点选择一帧满意的图片
3. 点击上方"截帧"
4. 在文本生成节点贴入提示词 — 通过"反推提示词"得到同样风格的全新图片
Step 2: Image-to-image fusion — specify model to wear the outfit
1. 上传服装图片
2. 上传指定模特图片
3. 通过图生图融合 → 得到指定模特的服装上身图片
4. 重复 1-3 → 各种风格穿搭图片
Step 3: Motion transfer cropping
- Perform motion transfer referencing the original video
- Crop a certain motion segment (crop once for each outfit)
Step 4: Batch nodes generate videos
- Pass the cropped video clips + outfit-on-model images into batch nodes respectively
- Select "motion and expression, pose transfer" for the workflow
- Consistency prompt simplified: "woman posing"
- Click batch run → background batch processing → 4 sets of different dynamic videos
Step 5: Editing
- Import into editing software → Add keyframes → Done
5. Wang Kai SillyTavern Video Engineering (Phase 1-7 Exclusive Template)
Wang Kai SillyTavern Video Production SOP (Skill implementation example) —— Transferable to mass production of any product videos:
| Phase | Content |
|---|---|
| 1. Directory skeleton | Video project structure / Materials / Components / Compositions layering |
| 2. Reference video analysis 8 dimensions | Topic / Shot / Rhythm / Emotion / Theme / Style / BGM / Transition |
| 3. Real asset scraping 5 categories | Screenshots / Characters / Scenes / Objects / Voice |
| 4. Skill system / Components / Compositions layering | Reusable units → Assemblies |
| 5. First video | MVP validation |
| 6. Three types of videos + Jimeng mixed editing | Content matrixization |
| 7. Scaling | Copy + Skill Optimization |
5.1 Jimeng CLI Multi-Window Auto-Run (Wang Kai Exclusive)
"Combining Jimeng Premium membership + installing Jimeng CLI in CC, running automatically across multiple windows,each YouTube account produces at least 10+ pieces of footage per day. "
Cost Observation:"The cost of making videos with Claude Code is much lower than with Jimeng and similar tools" — Remotion AI Skill programmatic video production.
6. Lao D Digital Human 5-Tool Combo (Exclusive)
dontbesilent AI Self-Media Course / 13 Systematic Methodology Lessons §L10:
6.1 Complete Workflow
1. 豆包生图(白底图 ★ 关键)
→ 提示词:让豆包反推一段 prompt 给你
→ 生成数字人图片(白底,方便后续抠人)
2. 真人录视频(你自己拍)
3. 阿里通义万相 wan2.2 animate(角色替换)
→ 视频 → 数字人替代脸
→ 但声音还是你的男声 ← 问题
4. ElevenLabs Voice Changer
→ 用女声换音色(支持 29 国语言)
→ 选中文女声(如安娜苏)
→ 输出 MP3
5. 剪映合成
→ 音频替换 + 响度统一 + 字幕
6. (可选)HeyGen 翻译视频
→ 一键翻译成日 / 法 / 韩 / 西班牙 / 俄等 30 国语言
→ 抖音覆盖海外市场
6.2 Key Conventions
| Pitfall | Solution |
|---|---|
| Video images must be white-background images | otherwise background recognition will be garbled (wall surface distortion) |
| Tongyi Wanxiang requires a paid membership | The standard version has queue lag and stuttering |
| Inconsistent durations across different tools | Align and cut to the shortest clip in CapCut |
| Select the target language for HeyGen translation | If not selected, it will auto-detect but is prone to errors |
6.3 Practical Returns
"A video Lao D posted on Douyin yesterday is approaching 2 million views,after the digital human covers 30 languages, single-video traffic doubles. "
7. Tool Stack Comparison Table (4 Schools)
| Task | Caviar Comic Drama | Xiaohongshu Outfit | Wang Kai SillyTavern | Lao D Digital Human |
|---|---|---|---|---|
| Script | Doubao / DeepSeek / Gemini | (using viral videos as blueprints) | 8-dimension reference analysis | AI writes voiceover scripts |
| Storyboard | 9-column storyboard sheet + 6 major camera movements | Canvas nodes | Components / Compositions | Not a strong requirement |
| Image Generation | Jimeng three-view drawing | Reverse prompt engineering + image-to-image fusion | Jimeng image2image anchoring | Doubao white-background images |
| Video Generation | Jimeng / Vidu + first and last frames | Jimeng batch nodes + motion transfer | Jimeng CLI Premium / Remotion | Real-person recording + wan2.2 face swap |
| Voiceover | Not a strong requirement | Not a strong requirement | ElevenLabs | ElevenLabs 29 countries |
| Multilingual | Not a strong requirement | Not a strong requirement | Not a strong requirement | HeyGen 30-country one-click |
| Editing/Stitching | CapCut | CapCut + keyframes | FFmpeg + Skill | CapCut |
| Monetization | Hongguo 5x revenue share | Xiaohongshu / Personal IP | YouTube multi-account | Douyin 2 million views |
| Usable in mainland China | ✅ | ✅ | ❌(YouTube) | ✅ |
8. How to Choose — AI Video Path by Identity
| Who you are | Primary method | Starting point |
|---|---|---|
| Want to join an agency to make manga dramas | Caviar 5-step process | 9-column storyboard sheet + 6 major camera movements + Hongguo 5x revenue share |
| Make AI outfits / 1-to-1 replication | Xiaohongshu canvas-based workflow | Reverse prompt engineering + image-to-image fusion + motion transfer + batch nodes |
| Make product promotional videos | Wang Kai SillyTavern Phase 1-7 | Directory skeleton + Components + Compositions + Skill system |
| Make digital humans + multilingual content | Lao D 5-tool combo | Doubao white background + wan2.2 + ElevenLabs + CapCut + HeyGen translation |
| Make cinematic web companion videos | Xiaohongshu AI Web 6-step | Higgsfield + Seedance + Google AI Studio + Landingbook |
| Batch produce ad creatives | Vibe Marketing Skill | Claude Code Skill + FFmpeg local stitching |
General Disciplines:
- Essentially you are the director, AI is the assistant(Caviar)
- Visuals first(list all storyboards before generating images)
- Character consistency: three-view locking / image2image anchoring / white-background images
- The more detailed the description, the higher the AI "gacha" hit rate
- Batch only when the formula is clear; creative-led work requires manual handling
- Multilingual translation = doubled traffic(Lao D empirical evidence)
9. Reference List
Main original author wikis:
- Caviar AI Comic Drama Production Course / Complete methodology from Section 1 live teaching — Caviar AI Comic Drama Section 1 / 5-step process / 9-column storyboard sheet / 6 major camera movements / Hongguo 5x revenue share / prompt quick reference
- Xiaohongshu / AI Video 1-to-1 Replication Complete Workflow — AI Outfit 1-to-1 Replication SOP / 5-step canvas workflow / reverse prompt engineering + image-to-image fusion
- Xiaohongshu / AI Web + 3D Interaction + Pretext Design Collection — Pretext / cinematic web 6 steps / Higgsfield + Seedance / 3 libraries + 3 libraries + Hitem3D
- Xiaohongshu / AI Tools + Skill Tutorial Collection — Vibe Marketing Skill batch videos (FFmpeg) / Pixel Agent
- Wang Kai SillyTavern Video Production SOP (Skill Implementation Example) — SillyTavern Phase 1-7 / Components + Compositions / Jimeng CLI multi-window
- dontbesilent AI Self-Media Course / 13 Systematic Methodology Lessons §L05 + §L10 — AI narrative short videos / HeyGen digital human 5-tool combo / 30-language translation
Related topic wikis (this series):
- AI Self-Media and Content Engineering (Cross-Author Topic Synthesis) — AI Self-Media and Content Engineering (Batch 1)
- Vibe Coding and Independent Development (Cross-Author Topic Synthesis) — Vibe Coding and Independent Development (this batch Batch 5)
- Claude Skill Engineering (Cross-Author Topic Synthesis) — Skill Engineering (Batch 3)
- AI Video Post-Production Head Floating (Deterministic Solution for POV Anthropomorphism) — Post-production head floating (first-hand testing at work/Storeel): deterministic solution for "dead still" POV videos + Veo/Seedance behavior comparison + five core POV prompt principles
Related skills (already operationalized for users):
jimeng-multishot-replicate— Jimeng multi-storyboard replication / image2image anchoring / 9 types of storyboard templates / 8 major pitfallsai-mandrama-pipeline— AI manga drama pipelineffmpeg-reels-pipeline— FFmpeg video stitchingins-reels-pick-and-score— ins Reels clip selectionins-reels-launch-sop— ins Reels launch SOPins-content-paraphrase-strategy— Subtitle paraphrase strategyins-account-highlights— ins account Highlights
History
- 2026-05-18: Topic-dimension wiki first release. Synthesized and rewritten from 6 author wikis. 3 divergences explicitly marked: individual vs agency / manual vs Skill-based / Jimeng vs Sora vs Higgsfield
Sources and Related Materials
- Caviar AI Comic Drama Production Course / Complete methodology from Section 1 live teaching
- Xiaohongshu / AI Video 1-to-1 Replication Complete Workflow
- Xiaohongshu / AI Web + 3D Interaction + Pretext Design Collection
- Xiaohongshu / AI Tools + Skill Tutorial Collection
- Wang Kai SillyTavern Video Production SOP (Skill Implementation Example)
- dontbesilent AI Self-Media Course / 13 Systematic Methodology Lessons
- AI Self-Media and Content Engineering (Cross-Author Topic Synthesis)
- Vibe Coding and Independent Development (Cross-Author Topic Synthesis)
- Claude Skill Engineering (Cross-Author Topic Synthesis)
- AI Video Post-Production Head Floating (Deterministic Solution for POV Anthropomorphism)