AI 生成视频：它改了哪一格AI-Generated Video: Which Cell It Changed

难度LEVELL1 · 入门L1 · Beginner 领域TOPIC调色

学习目标LEARNING OBJECTIVES

学完你能在五维图上指出 AI 生成视频改了哪一格、没改哪一格，以及它带来的新问题。

概念说明CONCEPTS

它改的是光的行为：从传感器记录变成模型输出——动态范围与色彩不再「拍到了多少」，而是「模型给了多少」。交付链路不变：仍走 H.264/H.265；主要按 sRGB、8bit 交付，比 709 素材短一格。好看 ≠ 五个维度都站得住。

对判断意味着什么：物理约束被绕开——没拍到的光可以被发明——但多个镜头、连续运动、同一角色的稳定性，落在模型身上，而不是片场。一致性与可复现性成了新问题。

操作步骤STEPS

  1. 打开五维图页，按 4 看 AI 生成素材的形状
  2. 短板落在 Dynamic Range
  3. 对照 709 实拍素材的形状，看差在哪几格

练习PRACTICE

动手任务

找一条 AI 生成片段与一条实拍素材，按五维图逐格对比，列出各自的短板。

自检清单

做完任务后逐条自查，全部能打勾才算过。

常见问题FAQ

AI 视频的动态范围真的更高吗？

那是模型给的——不是传感器拍到的，能不能用要看场景。

好看就等于画质好吗？

好看 ≠ 五个维度都站得住。

Learning Objectives学习目标

After this card you can point out, on the five-axis chart, which cell AI-generated video changed and which it didn't — and the new question it adds.

Concepts概念说明

What it changes is the behaviour of light: from a sensor recording it to a model outputting it — dynamic range and colour are no longer "how much was captured" but "how much the model gave you". The delivery chain is unchanged: still H.264/H.265; mostly delivered as sRGB, 8-bit — one notch shorter than 709 footage. Looking good ≠ all five dimensions hold up.

For judgement it adds a new question: the physical constraints are bypassed — light that was never captured can be invented — but keeping several shots, continuous motion and the same character stable now rests on the model, not on set. Consistency and reproducibility become the new problem.

Steps操作步骤

  1. Open the five-axis page, press 4 for the AI-generated shape
  2. The weak vertex lands on Dynamic Range
  3. Compare against a 709 live-action shape to see the gap

Practice练习

Hands-on task

Find an AI-generated clip and a live-action clip, compare them cell by cell on the five-axis chart, and list each one's weak side.

Self-check

Work through the task, then tick every line; all checked means you own this card.

FAQ常见问题

Is the dynamic range of AI video really higher?

That is what the model gives you — not what a sensor captured; whether it works depends on the scene.

Does looking good mean good quality?

Looking good ≠ all five dimensions hold up.

本卡术语Terms in this card

H.264H.264 静帧Still AI 生成视频AI-Generated Video 动态范围Dynamic Range