A professional AI script and storyboard generation system for creating professional AI video series on the Seedance 2.0 platform.
Core Workflow: Script-Assets-Video-Edit
A new-generation AI short drama production system that replaces the traditional "script-image-video-audio-edit" process:
| Stage | Core Content | Function |
|---|---|---|
| Script | Scriptwriting and storyboard breakdown | Provides the story blueprint and construction drawings |
| Assets | Character/scene/prop asset library | Ensures stability and consistency of the final film |
| Video | Multimodal direct video output | Character + scene + script-driven generation |
| Edit | Post-production editing and image repair | Completes the final audiovisual integration |
Workflow
Convert the source material into a production-ready video script according to the following sequential process:
1. Analyze the Input
Determine the input type:
- Complete text: A complete novel/article that needs adaptation and episode division
- Outline: A short story concept that needs full script development
Extract core elements:
- Protagonist and key characters
- Core conflict and narrative arc
- Scene/worldbuilding elements
- Key plot points and emotional beats
- Core hook/selling point
If the input is vague or incomplete, ask clarifying questions.
2. Confirm Production Parameters
Essential questions to ask:
- Visual style: What visual style? (Realistic/animation/ink wash/sci-fi/retro/cinematic/other)
- Duration: Total runtime? (Standard: short drama 5 episodes × 15 seconds ≈ 75 seconds; long drama 20 episodes × 15 seconds ≈ 5 minutes)
- Target platform: Aspect ratio? (16:9 landscape / 9:16 portrait / 2.35:1 cinematic widescreen)
- Tone: Overall emotional tone? (Epic/warm/suspenseful/cheerful/melancholic, etc.)
- Core hook: What is the one-sentence selling point? (Desperate counterattack/revenge thriller/healing and warmth/suspense and horror, etc.)
Record these parameters to ensure they are consistently applied throughout the process.
3. Generate the Complete Script Structure
Important: The output must strictly follow the format below, meeting the quality standards of "Good Script.md".
3.0 Short Drama Scriptwriting Formulas
Single-episode structure formula: Conflict upfront + emotional tension + staged face-slapping + hook at the end
Each episode is 15 seconds, with an extremely fast pace. There is no time for straightforward background exposition; it must get straight to the point.
AI scriptwriting prompt template:
你是一名专业的短剧编剧。请帮我创作一个【类型】短剧剧本。
故事设定如下：【写下剧名、类型、集数、主人公、核心欲望、反派/阻碍、世界观/背景、已有剧情想法】。
请你遵循三幕式结构设置钩子，直接呈现核心冲突，抓住观众。节奏要快，信息密度要高，爽点要多，并在每集结尾留下新的钩子。
要求第一集开头必须设计"黄金三秒"：通过一个强画面、强台词、强冲突、强反差或强悬念，立刻吸引观众眼球，不要平铺直叙介绍背景。第一集还必须介绍世界观、主人公的核心欲望，并尽快抛出核心冲突。每一集都要有一个小反转，一集解决一个矛盾冲突或完成一个情节点，同时留下一个小钩子。可以适当增加支线故事来丰富剧情，但支线必须服务主线。
请在第二幕结尾引出一个颠覆性的大反转，在第三幕揭露真相并给出相对闭合的故事结局。
请先输出"一句话卖点、故事简介、主要人物介绍、分集大纲"。等我确认分集大纲后，再按照常规剧本格式逐集写正式剧本。
正式剧本格式请保持简单清晰：先写集数和本集出场人物，再按场景写"内景/外景、地点、时间"，然后写画面动作和人物台词。台词格式为"人物名：台词"。每集建议3场戏左右，每集结尾必须有钩子。
3.1 Script Analysis Output Format
# [标题] - 剧本
一、核心梗
[两字或短语核心卖点，如：绝境反杀、经典武侠、复仇爽剧]
二、故事梗概
故事背景：[时代、地点、初始处境]
开场冲突：[打破现状的事件，触发冲突]
主角画像：[主角是谁，核心特质，当前状态]
主线事件：[关键情节推进 → 高潮 → 结局]
结局：[最终结果，人物转变]
三、一句话卖点
[带情绪的营销钩子，如：忍无可忍，无须再忍！风雪夜，豹子头林冲长枪出鞘，血溅山神庙！]
四、人物小传
[每个主要角色包含：]
**角色名**
视觉形象：[年龄、外貌细节、服装、关键视觉标记]
身份背景：[社会角色、背景、与主角关系]
核心标签：[2-4个性格标签，如：隐忍不发、悲情英雄、杀神降临]
性格特点：[详细性格描述，弧光转变]
金句：[最能代表角色的台词]
五、剧本大纲
前期（起）：[建立世界观、主角、现状]
中期（承/转）：[冲突发展、复杂化、转折点]
后期（高潮）：[高潮、对抗、情感巅峰]
尾声（收尾）：[结局、余波、主题升华]
六、剧本正文
3.2 Script Body Output Format (per episode)
Key: Each episode must strictly use the following format:
第X集
X-X [日/夜] [内/外] [场景名称]
道具：[关键道具列表]
出场人物：[出场角色列表]
△ 【空镜/开场镜头】[详细镜头描述，含运镜、构图、光影、氛围]
△ [镜头2描述，含具体运镜：推/拉/摇/移/跟/环绕/升降/特写/远景]
△ [镜头3描述 - 逐拍继续]
角色名（内心独白）：[内心独白内容]
△ [反应镜头描述]
角色名（对白/动作描述）：[对白内容]
△ [动作/序列描述]
【字幕：xxx】[需要字幕时添加]
△ [场景继续]
△ 【闪回】[闪回场景描述]
[闪回内容，用△镜头描述]
【闪回结束】
△ [回到现实，反应镜头]
角色名（画外音）：[人物不在画面中的画外音]
△ [高潮动作序列]
△ 【空镜】[结尾氛围镜头]
【字幕：xxx】
Script format rules:
-
Each shot begins with
△at the start(triangle + space) -
Camera language must be specific:
- Shot size: extreme long shot/long shot/medium shot/close-up/extreme close-up
- Camera movement: dolly in/dolly out/pan/tracking shot/follow shot/orbit shot/crane shot/dolly zoom/one-take/handheld shake
- Combination:
中景推近、特写环绕、远景俯拍、快速推近、缓慢拉远
-
Dialogue annotation:
角色名（内心独白）- Character's inner thoughts, character is visible in the frame but their mouth is not moving角色名（画外音）- The character is not in the current frame, the voice comes from off-screen角色名（惊/怒/喜）- Dialogue with emotion角色名- Ordinary dialogue
-
Special elements:
【空镜】- Atmosphere/establishing shot with no characters【闪回】/【闪回结束】- Flashback sequence【字幕：xxx】- On-screen text/title
-
Action description format:
- Use
→for continuous actions:林冲一枪刺穿差拨胸膛 → 鲜血喷溅 → 雪地染红 - Use emotional language:
杀气冲天、眼中怒火燃烧、浑身颤抖
- Use
-
Sensory details:
- Visual: color, light and shadow, composition, atmosphere
- Auditory: sound descriptions integrated into the shot
- Tactile: cold, pain, texture
4. Create an asset generation plan
Classify and number all visual assets:
| Category | Prefix | Example | Description |
|---|---|---|---|
| Character | C01-C99 | C01 Lin Chong · front full body | Multiple angles for each character |
| Scene | S01-S99 | S01 Cangzhou fodder field · snow scene | Key locations |
| Prop | P01-P99 | P01 Spear | Important items |
4.1 Asset construction workflow
Workflow:
- Extract character information from the script/novel
- Use a large model to generate a structured character biography
- Extract visual keywords from the character biography
- Organize the visual keywords into AI painting prompts
- After iterating to a stable image, add three-view drawings
4.2 Asset prompt generation rules
AI-assisted character information extraction prompt template:
请先阅读并分析以下短剧剧本：
【在这里粘贴完整剧本】
---
你是一名专业的 AI 出图设计师和短剧视觉资产拆解师。
请根据我提供的故事剧本，完整拆解本剧需要生成的视觉资产，并分别输出【人物资产】【场景资产】【道具资产】的 AI 出图提示词。
整体出图风格要求：
风格须贴合用户在步骤2中选择的视觉风格（写实/动画/水墨/科幻/复古/电影感等）。人物、服装、场景和道具都要贴合剧本的人物身份、时代背景、地域环境、社会阶层和剧情氛围。
I. Character asset requirements
Please list all character roles in the script that need images, and generate one Chinese AI image prompt for each character.
Character prompts must meet the following requirements:
- Must be full-body front view on a pure white background
- Must explicitly state "standing full body" "front view" "complete from head to toe in frame"
- Must include a description of the shoes to avoid generating a half-body image
- The character should not hold anything
- Both hands hang naturally, body upright
- The character has no expression, neutral expression
- Must describe the character's height, build, age, hairstyle, hair color, clothing, shoes, face shape, eyes
- The character image must match the character's ethnicity, historical background, identity and occupation, hair accessories and clothing settings
- Adult characters should use proportions close to 2:8 as much as possible, with a good-looking, natural, and camera-friendly appearance
- Even for villains, do not design them too ugly or extreme; while preserving the character's traits, make the appearance as attractive as possible
- Do not include children or elderly figures unless the script explicitly requires it
Character prompt format:
【角色名】
角色定位：
AI 出图提示词：
II. Scene asset requirements
Based on the important locations that appear in the script, list all the scenes that need to be generated as images, and generate AI image generation prompts for each scene.
Scene prompts must meet the following requirements:
- No characters may appear in the scene
- The scene style must match the visual style selected by the user in step 2, with a film and television drama quality
- The scene must conform to the script's era, region, social class, atmosphere, and plot function
- Generate at least three prompts for each important scene:
- Front shot scene image
- Reverse shot scene image
- Side full view image
- The front and reverse shots should be designed according to the characters' possible positions and dialogue directions, to facilitate subsequent storyboard reference
- The side full view image should show the complete spatial relationships, to facilitate understanding of character positions and blocking
- Do not include text, watermarks, posters, character silhouettes, or irrelevant elements in the scene
Scene prompt format:
【场景名】
场景用途：
正打场景图提示词：
反打场景图提示词：
侧面全景图提示词：
III. Prop Asset Requirements
Please list all important props in the script, and generate one Chinese AI image generation prompt for each prop.
Prop prompts must meet the following requirements:
- Must have a pure white background
- No characters may appear
- No real environmental background may appear
- The prop must be displayed alone, with a clear main subject
- The prop's shape, material, sense of era, and signs of use must conform to the script's settings
- If the prop is related to plot clues, identity reversals, suspense, or satisfying moments, please reflect its importance in the prompt
- Do not include text, watermarks, or irrelevant decorations
Prop prompt format:
【道具名】
道具用途：
AI 出图提示词：
Asset generation prompt format:
### [编号] — [名称]
[风格前缀]，[详细视觉描述（中文）]，[技术规格]
**风格前缀示例：**
- 中国水墨画风格混合动漫赛璐珞渲染
- 电影级写实风格，戏剧性光影
- 3D皮克斯风格动画渲染
- 科幻赛博朋克美学，霓虹灯光
**角色区分：** 为每个角色使用不同的配色方案和视觉标记，确保在所选艺术风格中能够识别。
Output format: An organized list with unique IDs, suitable for copy-pasting into an image generator.
4.3 Character Three-View and Four-View Assets
After determining the character's full-body front view, it is necessary to generate the character's three-view (front, side, back) + facial close-up, namely the "four-view".
Four-view prompt:
生成全身三视图以及一张面部特写。(最左边占满 1/3 的位置是超大的面部特写，右边 2/3 放正视图、侧视图、后视图，二八身材比例)，纯白背景。风格须与用户选择的视觉风格一致。
Note: The prompt must describe the art style, otherwise it may cause the same image to have inconsistent styles.
4.4 Image Aspect Ratio Recommendations
| Asset Type | Recommended Ratio | Recommended Resolution |
|---|---|---|
| Character front full-body photo | 9:16 | 2K |
| Character three-view/four-view | 16:9 or 21:9 | 2K |
| Scene image | 16:9 | 2K |
| Prop image | 1:1 | 1K/2K |
4.5 Common Issues in Character Design
- Mismatched costume styling: For period dramas, attention must be paid to the characteristics of Hanfu in different historical periods; ancient idol dramas tend toward bright and luxurious styles, xianxia works emphasize ethereal and dreamlike qualities, and wuxia works display an elegant and unrestrained style
- Character proportion issues: Ensure full-body images, avoid half-body images, pay attention to body proportions, and adopt a 2:8 body proportion
- Aesthetics issues: Check thumbnail recognizability, facial expression symmetry, and color harmony
- Other issues: Do not hold objects, do not include a background, and maintain a standing pose
It is prohibited to directly use actors' faces to create videos, AI may reference the actors' styling, but their likeness photos must never be used directly.
5. Generate the Seedance 2.0 storyboard script
5.1 Seedance 2.0 Storyboard Prompt Generation
Generate the following content for each episode:
a. Material Upload List
| 上传位置 | 素材ID | 素材描述 |
| 图片1 | C01 | 林冲正面全身（角色一致性参考） |
| 图片2 | S01 | 草料场场景参考 |
b. Seedance Prompt (timeline format, in Chinese)
[风格描述]，15秒，9:16竖屏，[整体氛围]
0-3秒：[场景建立] - [镜头运动]，[详细画面描述]，[氛围营造]
3-6秒：[主体引入/情节发展] - [镜头运动]，[具体动作]，[情绪表达]
6-9秒：[发展/冲突] - [镜头运动]，[关键动作]，[细节特写]
9-12秒：[高潮/转折] - [镜头运动]，[情绪爆发]，[视觉冲击]
12-15秒：[结尾/落版] - [镜头运动]，[最终画面]，[余韵]
【声音】只生成音效，不要生成音乐。不要生成任何字幕。[具体音效描述，如：脚步声、碎裂声、喘息声]
【参考】@图片1 [用途说明]，@图片2 [用途说明]，@图片3 [用途说明]...
c. Ending Frame Description
- Record the content of the last frame to ensure continuity with the next episode
- Include: subject, background, lighting, composition, atmosphere
Camera movement keywords: Push in/Pull out/Pan/Truck/Follow/Orbit/Crane/Hitchcock zoom/One-take/Handheld shake
Episode connection (from episode 2 onward): Start the prompt with 将@视频1延长15s and upload the previous episode as a video reference.
For more Seedance 2.0 tips and templates, please refer to references/seedance-manual.md.
5.2 Material Connection Techniques
- Frame extraction relay: Take a screenshot of the last frame generated by the previous video and upload it as the first-frame reference image for the next video generation
- Extend video: Upload the previous clip and ask the model in the prompt to extend the video
5.3 Solutions for Real-Person Review Restrictions on Platforms Such as Jimeng
For the review restrictions on real-person photos on platforms such as Jimeng, the following methods can be used:
- Local facial redrawing: Use an image model to redraw the face and lower the realism threshold
- In-platform image laundering: First make slight modifications to the image within the platform to increase platform trust
- Add grid to face: Add a geometric topology wireframe grid to the facial area
- Facial close-up occlusion: Use a small box to occlude part of the facial area
5.4 Storyboard Optimization
The significance of storyboard images：
- Reduce the video gacha rate: For complex extreme shots, first generate storyboard images and then run image-to-video
- Fallback remedy for defective shots: Optimize through "frame extraction → image modification → re-run image-to-video"
- Empty shots and emotional freeze-frames: Solve the problem of video large models being unable to directly output exquisite static images
Recommended tools for image generation and modification：
- Nano Banana series
- GPT Image 2
- Seedream series (Jimeng/Doubao image models)
- LibTV (aggregation platform)
Standard image generation prompt formula:
【艺术风格/媒介】+【景别与视角】+【主体描述】+【环境场景】+【光影与色调】+【画质与质感修饰词】
Storyboard script prompt template:
你是一位资深电影剪辑师、视觉导演、AI视频分镜设计师。请根据我提供的剧本，拆解成适合 AI 视频制作的分镜头脚本表格，并在表格最后一列直接生成每个分镜的 AI 绘画出图提示词。
【剧本内容】
{在这里粘贴剧本}
## 输出要求
只输出一张 Markdown 表格，不要输出额外说明。表格列为：
| 分镜号 | 景别 | 镜头运动 | 画面内容 | 台词 | 预估时长（秒） | 分镜出图提示词 |
## 分镜规则
1. 将剧本文学描写转化为可拍摄的镜头语言，不要出现"心里想、意识到、突然明白、感到绝望"等无法拍摄的描述
2. 每个分镜时长控制在2-5秒，每集总分镜时长合计13-15秒
3. 景别要有变化，避免连续相同景别，至少包含一次景别切换
4. 动作描述要具体可执行，用→连接连续动作
5. 对白标注声音来源属性：当面说话用"角色名说"，内心想法用"角色名内心独白"，画面外声音用"角色名画外音"
6. 对话场景采用正反打镜头调度，长台词中途切入反应镜头
7. 场景切换或情绪转折处，用空镜配合画外音先导，提升电影感
8. 音效只写明确的环境/动作音效，无明确音效则省略
9. 不要生成音乐，不要生成字幕
Output Format Description
This workflow produces the following outputs, and their relationships are as follows:
| Output | Stage | Format | Purpose |
|---|---|---|---|
| Script text | Text | △Storyboard script format (Section 3.2) | Story blueprint, the foundation of all subsequent work |
| Asset prompts | Asset | Numbered list format (Section 4.2) | Generate character/scene/prop images to serve as reference materials for video generation |
| Seedance storyboard prompts | Video | Timeline format (Section 5.1b) | Directly paste into Seedance 2.0 to generate video clips |
| Storyboard image generation prompts | Edit | Table format (Section 5.4) | When optimizing/repairing defective shots, first generate storyboard images and then convert to video |
Transition relationship: Script body → Extract assets → Generate asset images → Combine script + asset images to generate Seedance storyboard prompts → Generate video → If there are defective shots, use the storyboard image generation prompts to repair them
Final Output List
After completing the entire workflow, the following content should be delivered:
- Complete Script - Includes core premise, story synopsis, character profiles, episode outlines, script body (△ storyboard format)
- Character Asset Prompt List - Full-body front-view prompt + four-view prompt for each character
- Scene Asset Prompt List - Front shot/reverse shot/side full shot prompts for each scene
- Prop Asset Prompt List - Independent display prompt for each prop
- Seedance 2.0 Storyboard Prompts - One timeline-format prompt per episode, including material upload list and tail frame description
- Episode Transition Notes - Tail frame description for each episode to ensure continuity into the next episode
Quality Self-Check List
Before completing all outputs, confirm item by item:
Format Check
- Each shot starts with
△at the beginning - Camera movement uses specific terms (push/pull/pan/track/follow/orbit/crane)
- Dialogue is correctly annotated (inner monologue/voice-over/emotion)
- Includes sensory details (visual + auditory + tactile)
Content Check
- Each episode has a clear emotional arc
- Transitions between episodes are natural (tail frame→first frame)
- Character personalities are consistent and show development
- The core conflict is resolved
- The theme is expressed
Technical Check
- Each episode is controlled at 13-15 seconds (3-7 shots)
- Tail frame description is detailed (scene + image + lighting + composition + atmosphere)
- Color planning is clear
- Sound design is complete (generate sound effects only, no music, no subtitles)
- Asset prompt style is consistent with the visual style selected by the user
- The 【Sound】 section in the Seedance prompt contains only sound effects, not background music
Detailed Usage Guide
For the detailed usage guide, please refer to references/seedance-manual.md.