AIShort play production, study notes

I. Understanding the full video production process

Traditional film shows from the beginning of the project to the end of the distribution, with the following general experience8Phase.AIThe short play didn’t create a whole new process, it was used.AIThe project is designed to replace the scripts, speculators, fine arts, actors, film, sound, and later stages of the project.

01Project planning and project creation

Core work: market research,IP/Copyright procurement, subject planning, scripting concepts, filing of projects, first draft of budget

Key actors involved: creators, producers, producers, planners, writers, copyright/legal affairs, finance

02Theatrical development and financing

Core work: story outlines, character setting, script grinding, project packaging, financing negotiations, platform communication

Key actors involved: theatre team, literary planner, producer, investor, commercial producer, legal, financial

03Prior preparation

Core work: formation of the main creation, angle, viewing, divers, conceptual design, budget refinement, scheduling, contracts and equipment

Key actors involved: directors, executive directors, deputy directors, corner directors, executive producers, producer directors, directors of photography, artistic guides, audio-visual guides, lighting guides, clothing/glossy/device groups

04Shooting

Core work: live filming, performance dispatch, photo-radio, lighting, servo execution, data management, security management

Key actors involved: directors, actors, scene clerks, camera crews, lighting groups, audio recording groups, fine arts groups, clothing groups, make-up groups, props,DITTheater, theatre, movement, production.

05Later

Core job: editing, visualization, coloring,ADRSound, pseudo, sound, music, subtitles, packaging, master output

Key actors involved: cutters, late producers,VFXDirector/CGTeam, color-trucker, voice guide, sculptor, composer, subtitle and packaging design

06Review of delivery and review of cases

Core work: content review, technical review, revision of the final version, film productionQC, deliver television/platform/hospital lines

Key actors involved: Producers, late-stage producers, screening teams, legal affairs, copyrights, platform interfaces, distribution and delivery teams

07Promotion of marketing and distribution

Core work: pre-slotting, poster, media campaign, road shows, business collaboration, distribution of rehearsals, online distribution

Key actors involved: Director of Communication, Marketing Team, Public Relations, New Media Operations, Issuers, Business Cooperation, Platform/Television/Houseline

08Play image and data redisk

Core work: broadcast/exit, viewing and ticketing analysis, broadcast data monitoring, user feedback, copyright operations, derivative development

Key actors involved: originators, producer companies, distribution teams, platform data teams, business teams, copyright operations

Core sector coverage:

Productions and investments, production management, screenwriters, photo-light recordings, fine-scene virtuosos, late visual sound, review delivery, release

II.AIPre-simulationWorkflow

The focus of the current learning process is on:

The script, the sibling, the hand-painting, the actors, the costume, the final character, the props.

In the traditional video process:

+ Directed partition + Corner + Fine Arts + Conforms

AIThe most significant changes in the times are those that would otherwise have been done by multiple departments, and now one can rely on more than one person.AIModel complete.

But attention is needed:

AIIt can help with implementation and cannot be a complete substitute for the development of thinking.

Especially the script and the mirror.
”It’s not really about the quality of the film.”AI”Can you generate it?”

What did the audience see?
Where’s the shot taken?
What are the actors doing?
When will the message be made available to the audience?
How does the mood evolve?
Why does this camera exist?

The script.

The script is currently based onThe handkerchief is the master.。

Don’t just let go.AIGenerates the full script from a single sentence and then directly into video creation.

The reasons are:

The script generated by one key does not usually have a strong lead thinking.

When writing the script, there should actually be a certain amount of “image” in the head.

For example, it is not simple:

The man is anxious.

I think:

The male was sitting opposite the doctor, holding his head down, holding his finger, breathing a little bit too fast to look at the doctor.

The latter has begun to be visualized.

Okay, the lead is on.AIThe times are still very important becauseAIJust put the picture in the head out.

Example of the format of short scripts

Scene label and special norms: a concise reference version

I. Basic structural elements

1. Field marking

First Line “Pix”XSet orX-X”
Scene numbering is continuous and clear
When you change the venue, when you jump, when you flash back, you have to leave and remark.

2Time and place

Time is short words like “Day/ Night/ Morning/ Night”
Characteristics can be written “Night/ Rain”
Places with specific scenes and labels “In/ Out”

3List of persons

The person who came out uses the full name.
Can write the necessary information about age, identity, etc.
By importance, the main role is given priority

4Contents

Activate ”,” at the beginning
Line format: Role name: line
A tone or state hint with ”()”
For psychological activityOS”
Acoustic sound for the drawingVO”

II. SPECIAL OBJECTIVES

1.. inner monologue and anecdote.

Role Name (%1)OS: Content
Role Name (%1)VO: Content
VOFor uninitiated character sound

2Memories and flashbacks

And then we’ll start another line with “Small Back”
End of mark “End of flashback”
Short memories can be written “Insist lenses”
I’ll be sure to recommend it to the independent court.

3. Subtitles and skyscopes

♪ The world’s greatest ♪
♪ The sky mirror: a picture ♪
Need to be closely linked to the scene

III. Format flexibility and attention

Formats can be adjusted to either A or project requirements
Avoiding novelic language
Description in image language
Words are simple and powerful, single advice.500Split when word above
The character is a uniform term.
Special terms or difficult to understand may be described in the notes

Example of the script

1-1 某三甲医院门诊室 日 内

人物：女医生，35岁，男患者，25岁

△门诊室里，女医生低头翻看手上的病例资料，男患者坐在她的对面微微低头，表情凝重。

女医生：（抬头看向男患者）这种症状从什么时候开始的？

男患者：（神情呆滞，语速加快）我也不知道，我现在特别害怕公司开会，感觉下一个因为AI被优化的就会是我，我真的好焦虑啊医生。

女医生：（双手环抱，语气冷静）嗯，典型的赛博精神病前兆，不过不严重。

男患者：（困惑）那......怎样才算严重啊？

女医生：（抬了下眉头）当然是送到精神病院！这种病几乎无药可治，说能治好的都是江湖骗子。

男患者：（越来越急）那我该怎么办啊医生。

女医生：（放下手继续翻看桌上的病例，表情轻松）放心吧，回去好好工作，别老担心那些还没发生的事情。

男患者：（OS，神情凝重）后续还是医你一场，但我还是隐约觉得哪里不对，算了！走一步看一步吧。

Here’s what needs to be understood:

The script addresses what is “what happened”.

It’s not necessary to write every camera position for the time being.

The next part of the script is to take the script into a specific lens.

Split Script

When the script is finished, Jean-Marie.AIAssist in generationProfessional split script。

The problem with the split script is:

“What is the story?”

There is a need to take a play into multiple lenses and make it clear:

Mirror number
View
Transport mirror
Emotion/ Action
Line
Sound
Length

This step is very important.

Because it’s generated.AIWhen video is taken, it is not in principle a complete play, but:

A shot. A piece.AIVideo

Therefore:

The snippet script is actually a list of tasks that will generate video in the future.

AIExample for generating a partition scripting phrase

根据我提供的剧本片段和基本规范格式帮我生成一个专业的分镜脚本，一共16个分镜，从左至右依次包含镜号、景别、运镜、情绪/动作、台词、音效、时长。

最后输出为文本脚本，按照以下的范例：

场次：1-1
地点：中国某三甲医院精神科门诊室
时间：日/内
人物：女医生、男患者

镜号：1
景别：全景/建立镜头
运镜：固定镜头，轻微前推
情绪/动作：交代三甲医院门诊室环境。女医生低头翻病例，男患者坐对面低头，气氛压抑。
台词：无
音效：门诊室环境底噪、纸张翻动、远处候诊叫号声
时长：2s

镜号：2
景别：近景/女医生
运镜：固定镜头
情绪/动作：女医生翻病例，眉头轻皱，专业、冷静。
台词：无
音效：纸张翻动声、笔尖轻敲桌面
时长：3s

镜号：3
景别：近景/男患者
运镜：固定，略低机位
情绪/动作：男患者微微低头，手指交握，表情凝重，制造不安感。
台词：无
音效：轻微呼吸声、低频氛围音垫
时长：3s

镜号：4
景别：过肩镜头/患者看医生
运镜：切入，稳定构图
情绪/动作：女医生抬头看向男患者，进入问诊。
台词：女医生：这种症状是从什么时候开始的？
音效：环境底噪降低，语音清晰
时长：4s

The logic here is, for example:

镜头1
医院建立镜头
2s

↓

镜头2
医生近景
3s

↓

镜头3
患者近景
3s

↓

镜头4
过肩镜头
开始对白
4s

So the original scripts are actually starting to become really accessible video footage.

Hand-painted speculator

When a text-shape script is completed, do not start generating expensive videos immediately.

First verify:

Does this split look like a reasonable thing?

Methodology:

Jean.AICreate low-cost from split scriptsStatic storyboard /Storyboard。

AvailableImage 2One of them is about0.1Dollar.

There is no need for a good picture here.

The purpose is only to examine:

Is the frame just enough?
Is the view switch comfortable?
Is it reasonable to stand in person?
Can you connect to the camera?
Is there a lot of repeat footage?
Is there a problem with “the words seem reasonable, but the drawings are strange”

This is a very important step towards saving money:

The problem is discovered with a few cents of money, not a couple of dollars of video.

Schematic TablePromptExample:1

根据我提供的分镜表脚本，制作一张半写实风格的写实分镜表，只生成1-16镜号的故事表，忽略其他镜号，从左至右依次包含镜号、分镜图、景别、运镜、情绪/动作、台词、音效、时长，不需要刻画细节，只用做结构参考；

Schematic TablePromptExample:2

根据我提供的分镜表脚本，制作一张手绘风格的写实分镜表，只生成1-16镜号的故事表，忽略其他镜号，从左至右依次包含镜号、分镜图、景别、运镜、情绪/动作、台词、音效、时长，不需要刻画细节，只用做结构参考；

This stage can therefore be understood as:

The script.

♪ Take the story into camera ♪

Text Shape

, visualize the camera low

Storyboard

Checking is fine.

Then we’ll go into the official production.

Don’t skip this floor.

III. ROLE AND THE ART ASSETS PRODUCTION

The project is now starting to be prepared to officially generate a “assets” that need to be used again.

Here is an important concept:

Don’t recreate a character in every shot.

We should start with:

Actors.
Hairstyle
Make-up.
Clothes.
Prototypes
Scene

Make it a relatively stable asset.

Each shot thereafter continues to be generated from these assets as far as possible.

This is how coherence can be maintained.

Actor Pictures

UseImage 2The basic image of the actor is created.

At this stage, do not:

Complex clothing
Complex makeup.
Complex scenes

All of them one time.

First.The man himself is certain.。

Step 1: Basic profile of the person

For example, to identify:

Gender
Age
Face
Five officers.
Colour
Hairstyle
Body
Gravity

After generating satisfaction, save the person.

Step 2: Personal VI View

Six view of peopleGA picture of all the bananas in the house.Pro。

The characters first wear simple clothes.

For example, generation:

Heads
Front Left
Left
Back
Right
Front Right

The real uses of the six view are:

For the back.AIMore information about the character and body structure of the person.

It’s not easy to become someone else when people change angles.

Step 3: Add a more costume six view

Do not ask from the beginning:

Personal + Complex Clothes + Six View All Generates.

It is easy to lead to a decline in the consistency of the person.

Correct way:

人物确定
↓
人物六视图
↓
衣服确定
↓
人物穿衣
↓
穿衣后的六视图

It takes one step at a time.

The six views of clothing are created directly, often losing the consistency of the person.

Asset management

After that, you can take:

The original image of the actor.
Personal Six View
Dress six view
Different makeup.
Different faces.

Join your own asset pool.

Every generation of footage will try to leverage existing assets instead of re-creating roles.

Clothing pictures

Don’t rely on all the clothes.AIDesigned empty space.

If there are more demand-friendly clothes already available online:

Find the reference picture first and take the clothes out of the person.

Then we’ll use the plain dress as a picture.Reference。

For example:

网络穿搭参考
↓
提取衣服
↓
生成纯服装图
↓
角色试穿
↓
形成角色正式服装资产

It’s usually pure.Prompt：

“I’m wearing a nice, fancy black suit.”

More stable.

The final image of the person.

When the basic character and costumes are complete, then we’ll make them.Final role make-up.。

Add:

Make-up settings
Details of hairstyle
Skin Details
Official clothing
Personality.
Role Status

Role settingPromptExample:

妆容设定：面部和皮肤有少量痘痘
发型设定：凌乱的黑色长发
服饰设定：纯旧且有一些细小的破洞

Then use a similar thing:

图片1的人穿上图片2的服饰，保持人物一致性和构图不变。

The two most important keywords here:

Personal consistency

and

The structure remains unchanged.

Otherwise…AIIt’s easy to change when you’re changing:

Change your face.
Change of age
Change your hair.
Change your body.
Change your position.
Change the camera.

When the final figure is identified, this is the same as in the traditional video production:

Actor make-up.

Stationary Images

UseGPT Image / GPT 2Generate prop pictures.

For example:

Cell phones
Medical records
Handbag
Pills.
necklace
Weapons
Special drama items.

If a prop is repeated, or if it is connected to the plot:

It is important to make assets on its own.

For example, there is a very important red bottle of medicine in the plot.

Do not write every time you generate a camera:

“There’s a red bottle on the table.”

Because every timeAIBoth can generate different vials.

Should:

先生成药瓶标准图
↓
保存为资产
↓
之后每个镜头引用同一个药瓶

This would maintain continuity.

The servo phase is complete.

At present, the following are the first to be achieved:

Clothing + Make-up + props + personal assets

Okay.

The full video production process is not required at the time.

The most important thing is to run this chain:

手写剧本
↓
AI辅助整理专业剧本格式
↓
AI生成文字分镜脚本
↓
AI生成低成本 Storyboard
↓
人工检查分镜
↓
确定演员
↓
人物六视图
↓
确定服装
↓
人物穿衣
↓
最终定妆人物
↓
制作重要道具
↓
形成可重复使用的资产库

And here, the real video hasn’t started to be generated.

But in practice this part is very important.

Because:

The more complete the pre-period, the less money is wasted in generating the video.

If the characters, the costumes, the scenes, the footage are not certain, then just start.SeedanceCards, easily turned into:

生成
↓
人物不对
↓
重抽

生成
↓
衣服变了
↓
重抽

生成
↓
机位不对
↓
重抽

生成
↓
角色脸又变了
↓
继续烧钱

And the more professional idea is:

前期把问题解决掉
↓
正式生成时只解决「动作」
↓
把昂贵的视频模型用于最终镜头

And so is…AIShort play production and simple playAIThe biggest difference is the video.

IV. OBSERVATIONS AND SHOWING

The following steps will be taken after the completion of the roles, clothing and prop assets:

Scene assets First frame scene Shoot/ Video Generation

Here, a central idea continues:

The scene, the map, the location of the person and the overall coloring are determined with cheap images, and expensive video models are used for real action generation.

This would significantly reduce the number of subsequent video cards.

Site assets

Make a stable scene for the big scene first.720PPanorama。

For example:

Hospital outpatients
Bedroom.
Office
Dining room.
Street
Bar.

The role of the site assets is similar to the “scenario reference” or “scenario setting” in traditional movies.

The following different footage from the same scene was used to the extent possible to refer to the same scene asset, thus maintaining:

Space architecture is consistent
Furniture is identical.
The lights are in the same direction.
Aligns
The art style is the same.

Don’t re-engineer every shot.AIDesign scenarios empty-handed, otherwise they can easily be found:

The last camera door was on the left, the next camera door went to the right; the tables, windows, walls and lights were all changed.

First Frame View

Before the video is officially generated, Mr. is taking a picture of every video, which is what the video is about to be used.Home frame /Keyframe。

Direct access:

根据提供的分镜脚本，生成一张影视剧剧照

The current practice in the notes is to useA low-price model of bananas.V2Mr. Assemble.

The frame is not just for the location of the person, but also for the entire play:

Main Hue
Lights.
Photographic quality
Organisation
View depth
Architecture
Space relations
Yes, sir.Four-part mirrors / four-selections, pick the best one and then go into the official video generation.

分镜脚本
↓
生成四个候选首帧
↓
比较构图与人物站位
↓
选出最好的首帧
↓
进入视频生成

Homeframe GenerationPromptExample:

Example from the screenshotPrompt：

根据提供的分镜脚本，生成一张影视剧剧照：
场次：1-1
地点：中国某三甲医院精神科门诊室
时间：日 / 内
人物：#1 图片1 女医生、#2图片2 男患者

镜号：1
景别：全景/中心构图
运镜：固定镜头、轻微前推
情绪/动作：交代三甲医院门诊环境。女医生低头翻看手上的病例资料，男患者坐在她的对面微微低头，气氛压抑。

UI text in screenshot:

全能图片V2-低价深推版
16:9 / 2k
摄影机控制
全景图

This.PromptThe core is not a large literary description, but a direct delivery of the previously identified partition information to the image model:

+ Place + Time + PersonalReference+ View + Mirror + Action / Emotion

So the first frame that comes out is really a good shot behind the back, not just a “good-looking one.”AIPhoto by Theater.

Space relationsPrompt

Prescription in the screenshot:

根据以下分镜脚本和角色设定，生成一张写实影视剧分镜图，要求人物在场景中的位置和空间关系合理，写实质感，不要出现文字。

Here’s the key word:

Position and space are reasonable.

Because…AIIt is easy to ask questions about the size of the person, the penetration of the person and the table and chair, the fact that two people stand out in a way that is not in line with the logic of the lens.

The first phase solves these problems, and the subsequent video model will be more likely to succeed.

Filming

The official video generation can now be divided into two approaches:

Control: quarter-share-driven
Quick generation: without a quarterscope, directly generated

They are not necessarily better, but they are different in degree of control from the speed of production.

Method 1: Four-Square-Specture Control

This is the way to be more controlled.

The first part of the video was completed, so it actually came close to the “the photographer’s executive director’s partition” in traditional video.

Basic processes:

分镜脚本
↓
生成四分镜
↓
加入角色六视图 / Reference
↓
交给视频模型
↓
按照设计好的镜头顺序生成视频

For example, the quadrascopes were designed as:

镜头 A：2s
镜头 B：2s
镜头 C：2s
镜头 D：2s

If the video model correctly understands the quadrascope, the resulting video will be executed as much as possible according to the frame and rhythm of the lens.

The experience in the notes is:

The video is generated more strictly by the mirror. For example, the four mirrors are each.2sAnd the video that’s generated will be done at the same pace as it is.

The main advantages of this approach are:

More Controllable Chart
More consistent character.
The camera changes are designed ahead of time.
The mirrors are closer to the original split.
It’s a lot of random cards.

If the six-person view is matched again, the role stability will be higher and, in many cases, the repeated cards can be significantly reduced.

Six Viewes and Reference Charts

Experience in notes:

One face and three maps may be checked by a card; six views will not be checked by a card.

From a production perspective, the value of the six views is more important than the value of showing the models what people look like in multiple directions.

For example:

Heads
Front Left
Left
Back
Right
Front Right

It is not easy for persons to change their faces or their body ratios suddenly when they turn around, face side or change their seat.

Method II: Direct generation

The second approach is not using a quarterscope, based directly on:

Homeframe
PeopleReference
Shape Description
ActionsPrompt

Generates a single video lens.

Process:

分镜脚本
↓
首帧
↓
角色 Reference
↓
动作 / 运镜 Prompt
↓
直接生成视频

The experience in the notes is:

Without the direct generation of the quadrascope, the chances of success are high.

This approach is appropriate for:

Single Actions
Single
Fixed position
Simple Push and Swing
Normal dialogue
A camera that is not demanding for space change.

The following lenses are more suitable for the use of the quadrascope:

Multi-person interaction
Clear.
Continuous Actions
Complex Transport Mirror
There are multiple action stages in one shot.
Key scenario.
I’m gonna need a tight shot at the map.

How do we choose the two ways of filming?

Camera Type	Recommended methodology
♪ Alone ♪ ♪ To be a pair ♪	Direct Generate
Simple Fixed Camera	Direct Generate
Simple Push/ Pull	Direct Generate
Two people having complex interactions	Four-part mirror.
The characters are clearly missing.	Four-part mirror.
One shot with multiple action stages	Four-part mirror.
Key scenario.	Priority quadrant
# And the camera #	Priority for direct generation

It can be simply read as:

Simple lenses are quickly generated and key lenses are controlled.

There is no need to let all cameras follow the most complex processes for “professional” purposes. The true way to do it is to choose between quality, speed and cost.

V. COMPLETELY COMPLETEAIShort-line production process

At this stage, the whole thing…WorkflowThe previous production was extended to a real video production portal:

手写剧本
↓
AI 辅助整理专业剧本格式
↓
AI 生成文字分镜脚本
↓
AI 生成低成本 Storyboard
↓
人工检查分镜合理性
↓
确定演员基础形象
↓
人物六视图
↓
确定服装
↓
人物穿衣
↓
最终人物定妆
↓
制作重要道具
↓
制作场景 720P 全景资产
↓
根据分镜生成首帧剧照
↓
生成四分镜 / 候选构图
↓
选择最佳首帧和构图
↓
选择拍摄方式
├─ 简单镜头 → 首帧直接生成
└─ 关键 / 复杂镜头 → 四分镜 + 六视图强控制
↓
正式生成视频

The process can now be understood in three layers:

First level: stories

The script.

Settlement:

What’s going on?

Level 2: Directors and Fine Arts

Fracture +Storyboard+ Personal + Clothing + prop + Scene + Homeframe

Settlement:

What do you look like? How?

Level 3: Video generation

Quadrascope / Homeframe +Reference+ Video Model

Settlement:

How do we get the picture really moving?

The most important of these are:

Do not let the video model do your scripts, your director, your fine arts, your horns and your photography at the same time.

If the work ahead is not entirely certain, the video model will decide everything itself, and the results will be very random.

More stable approach:

人先做决策
↓
图片模型完成视觉设计
↓
视频模型只负责让画面动起来

That’s how we get closer to the real control.AIVideo productionWorkflow。

What are you actually learning now?

You’re learning now, not “to generate video,” but to learn.AIVideoPre-production(pre-production)。

That is:

The script, the split mirror, the screenplay, the screenplay, the screenplay, the screenplay, the screenplay, the screenplay, the screenplay, the script, the script, the script, the screenplay, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, the script, theStoryboard → Casting♪ The servo ♪Asset Library

The next stage should be completed before the following:

Scene assets * First tail/ key frame * Image video * Word/ port Cut

You learn like this, you take it straight.SeedanceThe constant smokin’ is a lot more.

AIGC basic 1

AIShort play production, study notes

I. Understanding the full video production process

01Project planning and project creation

02Theatrical development and financing

03Prior preparation

04Shooting

05Later

06Review of delivery and review of cases

07Promotion of marketing and distribution

08Play image and data redisk

II.AIPre-simulationWorkflow

The script.

Example of the format of short scripts

I. Basic structural elements

II. SPECIAL OBJECTIVES

III. Format flexibility and attention

Example of the script

Split Script

AIExample for generating a partition scripting phrase

Hand-painted speculator

Schematic TablePromptExample:1

Schematic TablePromptExample:2

III. ROLE AND THE ART ASSETS PRODUCTION

Actor Pictures

Step 1: Basic profile of the person

Step 2: Personal VI View

Step 3: Add a more costume six view

Asset management

Clothing pictures

The final image of the person.

Role settingPromptExample:

Stationary Images

The servo phase is complete.

IV. OBSERVATIONS AND SHOWING

Site assets

First Frame View

Homeframe GenerationPromptExample:

Space relationsPrompt

Filming

Method 1: Four-Square-Specture Control

Six Viewes and Reference Charts

Method II: Direct generation

How do we choose the two ways of filming?

V. COMPLETELY COMPLETEAIShort-line production process

First level: stories

Level 2: Directors and Fine Arts

Level 3: Video generation

What are you actually learning now?