archival-fragments

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Archival Fragments

从零做出一条完整短片。用户可能只给你一个想法，也可能给一份写好的文案——两种都从下面的流程走。

Create a complete short film from scratch. The user may give you only an idea or a finished script—both follow the workflow below.

三条硬约束（违反任何一条，成片必废）

Three hard constraints (violating any one of them will ruin the final film)

1. 每一镜的画面必须不同

1. Every shot must have a different image

N 个镜头需要 N+1 张各不相同的关键帧。

"相邻两镜共用同一张图"的意思是：第 1 镜的尾帧 = 第 2 镜的首帧，这一张图被这两镜共用。不是整片共用一张图。

绝对禁止：

拿同一张图当所有镜头的输入
拿一张结构图、示意图、信息图当所有镜头的参考
把
```
assets/style-reference.png
```
当成生成输入——它是给人看的风格参考板，任何一次生成都不要挂它

这样做出来的片子几十个镜头长得一模一样，等于没拆镜，观众三秒划走。

自检（必须做，并把结果报给用户）：

关键帧张数 == 镜头数 + 1      ✓/✗
任意两张关键帧的画面描述不重复  ✓/✗
每张关键帧都有自己独立的提示词  ✓/✗

N shots require N+1 distinct keyframes.

“S sharing the same image between two adjacent shots” means: the first shot’s end frame = the second shot’s start frame, so that one image is shared by both shots. It does not mean using one image for the entire film.

Absolutely prohibited:

Using the same image as the input for every shot
Using a single structural diagram, schematic, or infographic as the reference for every shot
Using
```
assets/style-reference.png
```
as a generation input—it is a style reference board for humans and must never be attached to any generation request

A film made this way has dozens of shots that all look exactly the same, as if the shots had never been broken down. Viewers will swipe away within three seconds.

Self-check (mandatory; report the results to the user):

Keyframe count == shot count + 1                 ✓/✗
No two keyframes have identical image descriptions  ✓/✗
Every keyframe has its own independent prompt       ✓/✗

2. 故事、画面、节奏必须由你亲自拆，且拆完要给用户过目

2. You must break down the story, visuals, and rhythm yourself, and present the breakdown for the user to review

不要拿到文案就直接开生成。必须先产出一张逐镜表，包含：每镜的旁白、画面内容、首尾帧、时长。拆完先给用户看，让他有机会否掉——改一张表 5 分钟，跑完再改是 4 小时。

详见 01-script.md。

Do not begin generating immediately after receiving the script. You must first produce a shot-by-shot table containing each shot’s narration, visual content, start and end frames, and duration. Present it to the user for review so they have a chance to reject or revise it—changing a table takes 5 minutes, while rerunning the entire production after changes takes 4 hours.

See 01-script.md for details.

3. 具象优先，不要抽象图形

3. Favor concrete imagery over abstract graphics

纸和红线是画面的地面，不是主体。每句话找一个观众不用想就懂的通用符号——世界地图+图钉、天平、台阶、一大堆实物。一条片子收敛到 2–3 个符号，各用几次。

用空白纸加一条线表达"这个位置是空的"、用三个方块表达"三家公司"——这类抽象图形观众看不懂，实测失败。

详见 style-guide.md 第零条。

The paper and red lines are the visual foundation, not the subject. For every sentence, find a universal symbol that viewers can understand without thinking—a world map with a pushpin, a set of scales, steps, a large pile of physical objects, and so on. A film should converge on 2–3 symbols, using each several times.

Using blank paper with a line to represent “this position is empty,” or using three squares to represent “three companies”—these abstract graphics are difficult for viewers to understand and have failed in real-world testing.

See Rule 0 in style-guide.md for details.

流程

Workflow

按顺序做，每一阶段有自己的文档：

阶段	文档	产出
0. 环境准备	00-setup.md	依赖装好、凭据可用、检查全绿
1. 想法 → 文案 → 拆故事/画面/节奏	01-script.md	`narration.json` + 逐镜表
2. 设计关键帧链	02-keyframes.md	`keyframes.json`
3. 写运动句	03-shots.md	`shots.json`
4. 排节奏	04-rhythm.md	镜长分配表
5. 生成与合成	05-produce.md	关键帧 → 镜头 → 配音 → 成片

视觉基线随时查 style-guide.md。

三个必须停下来问用户的节点：

阶段 0 检查完 —— 告诉用户环境就绪，请他给想法或文案
阶段 1 拆完 —— 把逐镜表给用户过目
阶段 5 跑完 2–4 个样片 —— 确认方向再跑全量

Proceed in order. Each stage has its own document:

Stage	Document	Output
0. Environment preparation	00-setup.md	Dependencies installed, credentials available, all checks passing
1. Idea → script → story/visual/rhythm breakdown	01-script.md	`narration.json` + shot-by-shot table
2. Design the keyframe chain	02-keyframes.md	`keyframes.json`
3. Write motion sentences	03-shots.md	`shots.json`
4. Arrange the rhythm	04-rhythm.md	Shot-duration allocation table
5. Generation and assembly	05-produce.md	Keyframes → shots → voiceover → final film

Consult style-guide.md at any time for the visual baseline.

Three checkpoints where you must stop and ask the user:

After completing the Stage 0 checks—tell the user the environment is ready and ask for an idea or script
After completing the Stage 1 breakdown—present the shot-by-shot table for the user to review
After Stage 5 has produced 2–4 samples—confirm the direction before running the full production

默认参数

Default parameters

4:3 / 2K（1920×1440）/ 视频

Minimax H3

frames2video

/ 图片

Seedream 5.0 Pro

count=1

/ 每镜生成 5 秒 / 并发上限 1，必须串行。

4:3 / 2K (1920×1440) / Video:

Minimax H3

frames2video

/ Images:

Seedream 5.0 Pro

count=1

/ 5 seconds per generated shot / Concurrency limit: 1; generation must run serially.

包里有什么

What’s included

```
references/
```
—— 六份规则文档，就是上表那几份
```
examples/
```
—— 一条已完成 67.3 秒成片的完整配置：35 张关键帧提示词原文、34 条运动句、镜长分配表、17 句旁白、文案。照着改结构，不要照抄内容——那些符号是为特定题材设计的
```
assets/style-reference.png
```
—— 风格参考板，只给人看
```
assets/example-01..08-*.jpg
```
—— 八张连续的关键帧，展示同一条片子里画面是怎么一张一张推进的

```
references/
```
— six rule documents corresponding to the stages in the table above
```
examples/
```
— the complete configuration for a finished 67.3-second film: 35 original keyframe prompts, 34 motion sentences, a shot-duration allocation table, 17 narration lines, and the script. Follow the structure, but do not copy the content—those symbols were designed for a specific subject
```
assets/style-reference.png
```
— a style reference board for human viewing only
```
assets/example-01..08-*.jpg
```
— eight consecutive keyframes showing how the visuals advance one frame at a time within the same film