shot-by-shot-analysis

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Shot-by-Shot Analysis

逐镜头分析

Purpose

目的

Turn multiple reference videos into one evidence-backed Markdown report that a human can review and downstream generation skills can use. Analyze first; synthesize only after every source has been inspected.

将多个参考视频转化为一份基于证据的Markdown报告，供人工审核及下游生成工具使用。先进行分析，仅在所有来源视频都完成检查后再进行整合。

Execution Boundary

执行边界

Use
```
video-analysis
```
for actual video understanding. Analyze each source in a separate request with the same shot-table contract.
Do not generate images, videos, prompts, workflows, or provider payloads.
Do not infer video content from metadata, filenames, transcripts, thumbnails, or general knowledge when the footage itself has not been inspected.
Keep raw hosted responses and temporary request records internal. Deliver one Markdown report named
```
<case>-shot-by-shot-analysis.md
```
.

使用
```
video-analysis
```
进行实际视频理解。针对每个来源发起单独请求，并遵循相同的镜头表格规范。
不得生成图像、视频、提示词、工作流或服务商负载。
当未检查视频素材本身时，不得从元数据、文件名、转录文本、缩略图或常识推断视频内容。
原始托管响应和临时请求记录需内部留存。交付一份名为
```
<case>-shot-by-shot-analysis.md
```
的Markdown报告。

Workflow

工作流程

Inventory every input video. Record a stable source label, path or URL, duration when known, and source basis.
Inspect each video independently. Do not expose earlier analyses or a draft shared style theory while analyzing later videos.
Segment by observable shot or meaningful continuous beat. Use start-end timecodes. Mark approximate boundaries when exact timing is uncertain.
Complete the shot table for the current video before interpreting its broader purpose. Describe visible and audible evidence before inference.
Verify that the table covers the full usable duration, preserves shot order, and does not invent speech, sound, motion, products, or production intent.
Repeat until every video has a complete table.
Compare all completed tables. Separate recurring grammar from one-off details and unique recognizable elements.
Write the Style Grammar Report with concrete rules, supporting source and timecode references, recurrence, and confidence.
Add a compact downstream handoff section. Preserve the style logic without turning it into final prompts or execution requests.

盘点所有输入视频。记录稳定的来源标签、路径或URL、已知时长以及来源依据。
独立检查每个视频。分析后续视频时，不得暴露之前的分析结果或初步的共享风格理论。
按可观测的镜头或有意义的连续节拍进行分段。使用开始-结束时间码。当无法确定精确时间时，标记近似边界。
在解读当前视频的更广泛用途前，先完成该视频的镜头表格。先描述可见和可听的证据，再进行推断。
验证表格覆盖了所有可用时长、保留了镜头顺序，且未编造语音、声音、动作、产品或制作意图。
重复上述步骤，直到所有视频都有完整的表格。
对比所有已完成的表格。区分反复出现的规则、一次性细节和独特可识别元素。
撰写风格规则报告，包含具体规则、支持的来源和时间码引用、重复情况及置信度。
添加一个简洁的下游交接部分。保留风格逻辑，但不要将其转化为最终提示词或执行请求。

Shot Table Contract

镜头表格规范

Use this exact column order for every video:

timecode | visual | framing/composition | camera motion | subject/action | product/prop relationship | lighting/color | edit rhythm | sound/music | speech/text | emotional state | production purpose

Apply these rules:

Describe what is visible and audible before interpreting it.
Keep
```
visual
```
factual. Put inferred narrative or commercial function only in
```
production purpose
```
.
Describe framing with production terms such as shot size, angle, headroom, subject placement, depth, foreground, and negative space.
Distinguish camera movement from subject movement.
Quote speech or visible text only when legible or audible. Otherwise write
```
unclear
```
,
```
partially audible
```
, or another precise uncertainty marker.
Describe edit rhythm using observable cut length, cut type, pauses, speed changes, montage density, or continuity behavior.
Mark inferred emotional state or production purpose with confidence when the evidence is ambiguous.
Use
```
not present
```
when a field truly does not occur. Do not fill absence with speculation.

每个视频都必须使用以下精确的列顺序：

timecode | visual | framing/composition | camera motion | subject/action | product/prop relationship | lighting/color | edit rhythm | sound/music | speech/text | emotional state | production purpose

遵循以下规则：

先描述可见和可听的内容，再进行解读。
```
visual
```
字段需保持事实性。仅将推断的叙事或商业功能放入
```
production purpose
```
字段。
使用专业制作术语描述构图，如镜头尺寸、角度、头部空间、主体位置、景深、前景和负空间。
区分摄像机运动和主体运动。
仅在清晰可辨或可听时引用语音或可见文本。否则请填写
```
unclear
```
、
```
partially audible
```
或其他精确的不确定性标记。
使用可观测的剪辑时长、剪辑类型、停顿、速度变化、蒙太奇密度或连续性行为描述剪辑节奏。
当证据模糊时，标记推断的情绪状态或制作目的及其置信度。
当某个字段确实不存在时，填写
```
not present
```
。不得用猜测填补空缺。

Synthesis Rules

整合规则

Do not average the videos early. Synthesize only from completed shot tables.
Treat a pattern as recurring only when supported by more than one source, or by repeated independent examples within a source. State the narrower basis.
Cite evidence as
```
<source-label> @ <timecode>
```
.
For every grammar dimension, distinguish:
- ```
shared rule
```
  : the reusable production behavior
- ```
evidence
```
  : supporting source and timecode references
- ```
recurrence
```
  : recurring, source-specific, or uncertain
- ```
confidence
```
  : high, medium, or low
Prefer concrete language over labels such as
```
premium
```
,
```
cinematic
```
,
```
authentic
```
, or
```
native
```
. Use those labels only after describing the visible and audible evidence that supports them.
Preserve camera grammar, timing, object logic, relationship logic, sound behavior, and emotional progression without preserving creator identity.

不得过早对视频进行平均处理。仅在所有镜头表格完成后再进行整合。
只有当模式得到多个来源的支持，或同一来源内有多个独立重复示例时，才将其视为反复出现的模式。说明其适用范围。
以
```
<source-label> @ <timecode>
```
格式引用证据。
对于每个规则维度，需区分：
- ```
shared rule
```
  ：可复用的制作行为
- ```
evidence
```
  ：支持的来源和时间码引用
- ```
recurrence
```
  ：反复出现、特定来源或不确定
- ```
confidence
```
  ：高、中或低
优先使用具体语言，而非
```
premium
```
、
```
cinematic
```
、
```
authentic
```
或
```
native
```
这类标签。仅在描述了支持这些标签的可见和可听证据后，才可使用它们。
保留摄像机规则、时间安排、对象逻辑、关系逻辑、声音行为和情绪递进，但不保留创作者身份。

Imitation Boundary

模仿边界

Do not copy or recommend copying:

creator identity, face, voice identity, or distinctive personal mannerisms
exact scenes, locations, wardrobe, dialogue, jokes, or shot sequences
brand marks, packaging identity, private details, or recognizable background information
unique overlays, signature graphics, or other source-identifying elements

Translate useful reference evidence into general production rules. Put recognizable or source-specific elements under

avoid

, not

must_keep

不得复制或推荐复制：

创作者身份、面部、声音特征或独特的个人举止
精确场景、地点、服装、对话、笑话或镜头序列
品牌标识、包装特征、私人细节或可识别的背景信息
独特的叠加层、标志性图形或其他可识别来源的元素

将有用的参考证据转化为通用制作规则。将可识别或特定来源的元素归入

avoid

（需避免）类别，而非

must_keep

（需保留）类别。

Markdown Deliverable

Markdown交付物

Read

references/report-template.md

. Produce one report with complete source tables, the ordered

Style Grammar Report

, and a concise downstream handoff.

阅读

references/report-template.md

。生成一份包含完整来源表格、有序的《风格规则报告》以及简洁下游交接部分的报告。

Downstream Handoff

下游交接

Keep the handoff concise and derived from the report:

Image Generation: Extract static framing, subject state, scene, lighting, palette, product placement,
```
must_keep
```
,
```
can_vary
```
, and
```
avoid
```
. Hand this to creative planning before
```
image-batch-runner
```
.
Video Prompt Architecture: Extract camera language, movement, edit rhythm, sound, speech posture, emotional progression, product behavior, continuity, and generator risks before
```
video-batch-runner
```
.
Workflow Creation: Provide the full report as source evidence. Compile its rules into positive, executable scene carriers and explicit reference jobs; do not paste analysis language or imitation warnings into model prompts.

Do not create separate handoff files unless the user explicitly requests them.

交接内容需简洁且基于报告：

图像生成：提取静态构图、主体状态、场景、灯光、调色板、产品摆放、
```
must_keep
```
、
```
can_vary
```
和
```
avoid
```
。在运行
```
image-batch-runner
```
前，将此内容提交给创意规划团队。
视频提示架构：在运行
```
video-batch-runner
```
前，提取摄像机语言、运动、剪辑节奏、声音、语音姿态、情绪递进、产品行为、连续性和生成器风险。
工作流创建：提供完整报告作为来源证据。将其规则整理为积极可执行的场景载体和明确的参考任务；不得将分析语言或模仿警告粘贴到模型提示词中。

除非用户明确要求，否则不得创建单独的交接文件。

Quality Check

质量检查

Before handoff, confirm:

every input video has a complete, ordered shot table
synthesis begins only after the final table
shared claims cite supporting sources and timecodes
recurring and one-off details are separated
uncertainty is visible rather than silently resolved
production-purpose inference is not presented as observation
unique identity, brand, private, and recognizable elements are excluded from reusable rules
the report is readable by a human and specific enough for downstream prompt construction

交接前，确认：

每个输入视频都有完整、有序的镜头表格
整合仅在最后一份表格完成后才开始
共享结论都引用了支持的来源和时间码
区分了反复出现的细节和一次性细节
不确定性已明确标注，而非默默解决
制作目的推断未被当作观测结果呈现
独特身份、品牌、私人和可识别元素已从可复用规则中排除
报告易于人类阅读，且足够具体，可用于下游提示词构建

Stop Conditions

停止条件

Stop when a source video cannot be accessed or actually inspected.
Stop when missing sources would materially change a requested cross-video synthesis.
If the hosted analysis command fails, report the exact error and do not replace video evidence with guesses.

当无法访问或实际检查某一来源视频时，停止操作。
当缺失的来源会实质性改变所需的跨视频整合结果时，停止操作。
如果托管分析命令失败，报告确切错误，不得用猜测替代视频证据。