2025-11 live teaching (Caviar Agency, lead instructor + 2 students). Covers the complete 0-to-1 production process for AI comic dramas, suitable for zero-foundation beginners to get started with the Hongguo Theater monetization chain.
1. AI Comic Drama Monetization Path (Why Do It)
Hongguo Theater — The Most Direct Monetization Channel
- In 2025, Hongguo will heavily promote AI comic dramas, a new revenue-sharing model will launch in 2025-12:5x revenue-sharing coefficient
- A full episode requires ~100 minutes / solo production cycle is ~1 month
- Not recommended for individuals to start an account from scratch — the account startup cycle is too long, drama production takes a long time, making it difficult on both ends
- Join an agency team system: several people work on one drama together, with revenue sharing + production fees
One-sentence summary
Essentially, comic dramas are pieced together from different shots; as a drama maker, you are essentially the director, and AI is only an assistant.
2. 5-Step Comic Drama Production Process (Overall Diagram)
小说 / 文章
↓ 1. 通过 AI 生成
动态漫剧本(标准格式)
↓ 2. 通过 AI 转换
分镜表(表格化)+ 核心场景&角色介绍(资产库提炼)
↓ 3. 即梦生图(三视图)
资产库(角色 / 场景 / 物品)
↓ 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
Key conventions
| Step | Tool | Output |
|---|---|---|
| 1. Script generation | Doubao / DeepSeek / Google Gemini | Dynamic comic script (.docx) |
| 2. Storyboarding | Doubao / DeepSeek + spreadsheet template | Storyboard table (8-column structure) |
| 3. Asset library extraction | AI text-to-image description word generation | Character + scene prompt document |
| 4. Image generation | Jimeng | Three-view / character fusion images |
| 5. Video generation | Jimeng / Vidu | mp4 storyboard clips |
| 6. Editing | Jianying | Finished video |
3. Storyboard Table — The "Skeleton" of a Comic Drama
Standard 9-column storyboard table structure
| 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(copy directly to Jimeng) |
| 6 | Camera movement(push/pull/pan/track/follow/orbit + combinations) |
| 7 | v.0.1 Dialogue |
| 8 | Dialogue corresponding to camera movement (detailed matching for extreme long shot/full shot/medium shot/close-up/extreme close-up) |
| 9 | Notes (director's hand-drawn sketches / links to selected images / links to generated videos) |
Core mindset: visuals first
Do not write segment by segment while reading; first carefully list and write out all storyboards completely, then generate images with a clear sense of the visuals.
Why: AI is a material library; the more detailed the description, the fewer options left for AI, and the higher the hit rate when "drawing cards".
Dual role of the storyboard table
- Mental mind map — helps understand the direction of the entire episode
- AI operation input — prompts + camera movements are directly usable
4. Camera Movement — 6 Basic Types + 1 Subjective Perspective
Full camera movement list
| Camera movement | Applicable scenarios |
|---|---|
| Push in | Zoom in to focus on details (protagonist's facial expression / close-up of an object) |
| Pull out | Zoom out to show the environment (cut abruptly from close-up to long shot, often used at endings) |
| Pan | Sweep horizontally across the scene (to show sense of space) |
| Track | Lens moves horizontally/vertically |
| Follow | Follow the moving target (protagonist running / object flying) |
| Orbit | Circle around the target once (show the object in 360°) |
| Subjective | First-person perspective (the audience sees what the character sees) |
Combination shots
"For all anime / film and television works on the market, camera movements are essentially these 6 plus subjective. Everything else is a combination."
Common combinations:
- Orbit + push in: circle around while gradually zooming in
- Follow + push in: follow the object while gradually approaching (e.g. an eagle flying toward its target)
- Pull out + fade to black transition: pull out to the sky → fade to black → cut to the next shot
Transition design — first and last frame thinking
Core: the last frame of the first shot = the first frame of the second shot.
Case(transition from ruins to Luming Manor swimming pool):
- Shot 1: long shot of ruins → close-up of screaming passerby → pull out to the sky(last frame = sky)
- Shot 2:pull down from the sky(first frame = sky)→ Luming Manor swimming pool
- Result: two completely different scenes transition smoothly using "sky" as the shared visual anchor point
5. Asset Library — The Solution for Character Consistency
What is an asset library
A pre-generated image library of characters + scenes + important items that appear in each episode, create a folder to store all materials for the entire episode., create a folder to store all materials for the entire episode.
Why it must be done
The most common pitfall for beginners: feeding the script directly to Jimeng/Vidu to generate videos → characters look different before and after, transitions are incoherent.
Correct approach: first generate three-view drawings(front / side / back)→ Jimeng locks the character image during image fusion.
Three-view + character locking method
Key insight: anime characters almost all look the same (pointed chins) when you remove their hairstyles. Therefore:
- Hairstyle is the most important character distinguishing feature(each character must be given a different hairstyle)
- Face shapes can be similar (it doesn't matter)
- Clothing can change per episode (just use Jimeng's "reference image generation" to change clothes)
Asset library workflow
1. 收到分镜表 + 核心场景&角色介绍文档
2. 对每个角色 → 让即梦生成 "三视图"
3. 对每个场景 → 生成场景图
4. 对重要物品 → 生成物品图(非重要可省)
5. 全部存入"第一集资产库"文件夹
6. 后续生图时用即梦"融图"功能 = 角色 + 场景 + 动作描述
Clothing change method
Changing clothes does not require regenerating the character — drop the generated character image into Jimeng's "reference image generation" and directly say "help me change this female lead's clothes to XXX", the image will remain locked.
6. Core Jimeng Usage Tips
1. Duration adjustment is key
Jimeng supports 3 / 5 / 8 / 10 seconds.Duration determines action completion:
- Too short duration → action cannot be completed → background jumps
- Too long duration → action is normal but very slow → speed up in post-editing
Recommendation: for a shot that theoretically needs 5 seconds, generate 8 seconds, then crop + speed up in post-production to get the best effect.
2. Three solutions for first and last frame flaws
| Problem | Solution |
|---|---|
| Action too slow | Adjust speed in post with Jianying (universal strategy, apply uniformly to all storyboards) |
| Image jump | Increase duration (give AI more time to complete the action) |
| No matter how you adjust it, it's still wrong | Change the last frame/first frame design (AI may not understand your splicing logic) |
Counterexample: female lead squats down to pick up a lighter → stands up and puts it in her bag (back view)→ AI cannot understand.Change to: female lead squats down → stands up facing front holding the lighter into the bag view → AI can understand.
3. Description word philosophy
| Task | The more detailed the better? |
|---|---|
| Image generation | ✅ The more detailed the better (locks down AI's options) |
| Video generation | ❌ Not the more detailed the better (over-constraining AI will instead cause image jumps) |
Core:
- Do not use colloquial language — use professional photography/film and television terminology
- Character descriptions run through the entire drama — write once and reuse repeatedly
- When generating videos, only describe the core action — characters / scenes are already locked
7. Prompt Description Word Quick Reference Checklist
Body type
Slim / plump / round and full / robust / muscular / broad shoulders narrow waist / short and stout / stocky / pear-shaped figure / pregnant / hunched / giant figure with oppressive presence
Hairstyle (★ most important character distinguishing dimension)
Short hair / ultra-short hair / buzz cut / medium-long hair / long hair / waist-length long hair / straight hair / curly hair / wavy curls / fluffy afro / straight bangs / side-swept bangs / no bangs / ponytail / twin tails / braided pigtails / bun / flower bud bun / updo / braided hair
Clothing — Traditional
Chinese Hanfu (can be subdivided: Tang style/Song style/Ming style)/ cheongsam / mandarin jacket / ruqun / long gown / Lolita dress (sweet/Gothic)/ Victorian (corset/bustle skirt/lace)/ ancient Greek robe / Roman toga / Japanese kimono / Korean traditional / ethnic Miao embroidery / Yi ethnic / Indigenous (feathers/animal hide)
Clothing — Modern
Casual T-shirt / jeans (straight leg/flared/ripped)/ hoodie / tracksuit / baseball jacket / suit (single-breasted/double-breasted)/ evening gown (strapless/mermaid/ballgown)/ women's business wear (shirt + half skirt/trousers)/ trendy hip-hop / punk (studded leather jacket/ripped jeans/chains)/ cyberpunk (reflective/techwear/neon)
Clothing details
Polka dots / stripes (horizontal/vertical)/ plaid (tartan/houndstooth)/ florals (rose/peony)/ geometric (triangles/squares)
Accessories
Belt (leather/metal buckle/wide/thin)/ necklace (pearl/diamond/statement design)/ earrings (long/round)/ hat (beret/top hat/fisherman)/ gloves (long/lace)
Painting style
Anime style / 2D animation / Japanese manga / 1990s Japanese manga / Korean manhwa / Lolita / comic art / otome game style / ACG / Dragon Ball style / game concept art / medieval / ancient Chinese style / Chinese wuxia / cartoon / picture book / blind box figurine/3D toy / cyberpunk / tech style / tech punk / metaverse / visual futurism / mecha / metal machinery / dark fantasy / dark horror
8. Complete Case: "Recruiting Team Members in the Apocalypse, I Only Pick the Most Useless"
Story outline
Lu Ming was an overworked grind king in his past life who died from overtime work; he transmigrates to an apocalyptic zombie world and awakens the "Apocalypse Slacker System": recruit useless team members and ensure they do not level up their abilities during the one-month assessment period, and he will receive system rewards (supplies/weapons/ability enhancements). However, the team members always accidentally improve their abilities, so Lu Ming can only drive them away and recruit new ones. As his fame grows, he attracts the attention of villains…
Core scenes and prompts for Episode 1
| Scene | AI text-to-image logic prompt |
|---|---|
| Ruins (daytime exterior) | Apocalyptic ruins style, dilapidated city buildings, zombies and human corpses covering the ground, black smoke curling up, motion-blurred running crowds, close-up of ferocious zombie faces, high-contrast lighting, 8K HD, cinematic texture, realistic style |
| Luming Manor swimming pool (daytime exterior) | Apocalyptic luxury manor swimming pool scene, azure pool water, exquisite lounge chairs, green plants and blue sky of the apocalyptic manor, warm tones/soft lighting, combination of cyberpunk and realism, 8K HD, cinematic texture, delicate character expression portrayal |
| Office (flashback) | Modern office scene, desk piled with documents, lit computer screen, cool-toned lighting, realistic style, 8K HD, portrayal of oppressive atmosphere |
| Unknown space (system awakening) | Floating sci-fi system panel, blue light effects, dark background highlighting the panel, cyberpunk style, 8K HD, special effect font displaying text |
| Apocalyptic market (daytime exterior) | Apocalyptic market scene, shabby stalls, crowded people, ability users emitting faint glows, warm yellow/cool/natural/messy lighting, realistic style, 8K HD, showing chaotic and lively atmosphere |
Core characters and prompts for Episode 1
Lu Ming (protagonist, male 25)
Short hair, black hair color, tall and robust build, wearing casual beachwear with black sunglasses in the manor scene, a simple bracelet on his wrist; wearing a black casual jacket over a white T-shirt in the market scene, cargo pants on the lower body, Martin boots on feet, facial expressions can switch between happy/excited/calm/determined, realistic style, 8K HD
Beautiful woman (supporting role, female 23)
Big wavy long hair, golden hair color, plump and sexy figure, initially wearing a beige trench coat over black sportswear; later changes into a red sexy bikini, holding a tall stemmed wine glass, facial expression with an awkward smile or relaxed expression, realistic style, 8K HD, delicate skin texture
Ning Xi (F-level healing ability, female 20)
Shoulder-length short hair, brown hair color, thin and frail figure, wearing a faded blue cloth shirt with gray trousers in the market scene, clothes have multiple patches, facial expression shifts from pleading/panicked/lost to surprised, eyes show timidity and expectation, realistic style, 8K HD, portrayal of subtle facial expressions
Captain 1 (villain NPC, male 35)
Buzz cut, black hair color, burly and sturdy build, wearing a black leather combat uniform, tattoos on arms, fierce and disgusted facial expression, contemptuous eyes, realistic style, 8K HD, clear portrayal of muscle lines
9. Quick Reference for Common Beginner Pitfalls
| Pitfall | Consequence | Solution |
|---|---|---|
| Feeding the entire script segment to Jimeng | Inconsistent characters before and after / incoherent transitions | First make the storyboard table + asset library |
| Description words too colloquial | AI cannot understand, generated images diverge | Use professional photography/film and television terminology |
| No three-view drawings made for characters | Face changes when generating images from a single angle and switching frames | Must make three-view drawings to lock the face |
| Video generation duration too short | Action not completed, image jumps | Set to 8-10 seconds and crop + speed up in post |
| First and last frame splicing logic is chaotic | AI cannot reason it out | Change the last frame angle / change subject pose |
| Not using first and last frames for shot transitions | Hard cuts feel awkward | Last frame of shot 1 = first frame of shot 2 |
| Not splitting into sub-shots | Long actions have poor completion | Split into 2-1 / 2-2 / 2-3 sub-storyboards (including close-ups) |
| Generating images directly without using image fusion | Characters / scenes deviate | First generate characters + scenes → fuse then generate actions |
10. Decision Checklist (Self-Check Before Starting)
- Is the Hongguo Theater account registered? Is the 5x revenue-sharing coefficient confirmed?
- Have you clearly assessed the 100 minutes per episode / 1 month production cycle?
- Working solo vs joining an agency team system — which to choose?
- Do you already have a script/novel source material?
- Do you know how to use Doubao/DeepSeek to have AI convert it into a dynamic comic script?
- Do you know how to use Jimeng's three-view drawing + image fusion features?
- Do you understand the selection of the 6+1 camera movements and the connection between start and end frames?
- Do you know how to use Jianying for splicing + speed adjustment processing?
- Is the folder structure of the asset library for the first episode prepared?
See also
- Wang Kai's Overseas Promotion SOP and Overseas Multi-Account Matrix Practical Combat — Wang Kai's AI video overseas distribution (English/Xiaohongshu matrix)
- newtype · Vibe Coding Independent Product Practical Combat — Lao Huang's newtype Profile orchestration (similar to the multi-Agent approach but applied to personal products)
- Personal memory
wiki-content-replication.md— Lao Huang's personal AI video replication rules (ethnicity hashtags / Jimeng multi-shot image2image anchoring), highly relevant to the methodology of this course
Secondary expansion suggestions
- The advanced course will teach:Specific usage scenario rules corresponding to push, pull, pan and track camera movements(This section is the foundational course)
- Vidu tool usage (this section mainly covers Jimeng, Vidu content to be supplemented)
- Hongguo Theater upload and publishing process
- Collaborative division of labor after joining an agency team system (specialized division of labor for characters / scenes / storyboards)