Video Prompt Reverse

English · 简体中文

English

video-prompt-reverse is a Codex skill for reverse-engineering a supplied video into a shot-by-shot generation specification, evidence frames, and paired Chinese-English prompt packages.

It supports generic prompt adaptation and model-oriented workflows for Veo, Sora, Kling, and Wan. It focuses on semantic and cinematic similarity; text prompts alone cannot guarantee frame-identical reconstruction.

Install in Codex

Copy this repository into:

%USERPROFILE%\.codex\skills\video-prompt-reverse

The skill is activated by requests such as “视频反推提示词”, “视频转提示词”, and “video-to-prompt”.

Contents

• SKILL.md — skill instructions and invocation rules
• agents/openai.yaml — Codex agent metadata
• references/ — analysis contract, model adapters, and optimization guidance
• scripts/ — video probing, prompt compilation, validation, and bounded-loop utilities

Community link

For more AI and developer discussions, visit Linux Do.

简体中文

video-prompt-reverse 是一个 Codex skill，用于将视频反向拆解为逐镜头生成规格、带证据的视频帧，以及中英文成对的提示词包。

它支持通用提示词适配，并可面向 Veo、Sora、Kling 和 Wan 等模型工作。它侧重语义与电影感相似度；仅凭文字提示词无法保证逐帧一致的重建结果。

在 Codex 中安装

将本仓库复制到：

%USERPROFILE%\.codex\skills\video-prompt-reverse

当用户提出“视频反推提示词”“视频转提示词”“从视频生成提示词”或 “video-to-prompt” 等请求时即可触发此 skill。

目录内容

• SKILL.md：skill 说明与调用规则
• agents/openai.yaml：Codex agent 元数据
• references/：分析契约、模型适配器与优化流程说明
• scripts/：视频探测、提示词编译、校验与有限闭环工具

社区友链

欢迎访问 Linux Do，参与 AI 与开发者社区交流。

返回 English

This project produces reverse-engineered generation specifications. It does not claim to recover a source video's original prompt.