Prompt Translator

This is a lookup table disguised as a translator. Instead of asking Claude to improvise how a Sora prompt should look in Kling, it references 110 hand-crafted examples across 10 scenarios (product shots, horror, wuxia fights, surfing dogs) and 11 models. The core trick is in-context learning from cross-model-matrix.json, so when you paste a Sora prompt and say "convert this to Veo," it finds the closest reference scenario, shows Claude the before-and-after pattern, and applies that structure to your input. Works best when your prompt resembles one of the 10 benchmark scenes. Won't predict actual video quality, but it will get the syntax right, which is half the battle when a model EOLs and you need to port 50 prompts by Tuesday.

GitHub stars: 269
last updated: Jun 1, 2026

View on GitHub

Install

$ npx -y skills add cclank/lanshu-awesome-ai-video-kit --skill prompt-translator --agent claude-code

Installs into .claude/skills of the current project.

Files

prompt-translator

跨模型提示词转换器。关键差异:不是凭 AI 直觉重写,而是查 110 条对照基准做 in-context learning。

何时不用此 skill

用户从零开始写一条新提示词(不是转换) → 用 seedance-prompter / kling-prompter / happyhorse-prompter 或 model-selector
用户问"用哪个模型好" → 用 model-selector
已有提示词出问题但不换模型 → 用 seedance-debugger

核心数据资产

prompts/data/cross-model-matrix.json — 这就是 translator 的"训练数据":

10 个核心场景:产品广告 / 情感重逢 / 滑板动作 / 图生视频 / 多人会议 / 恐怖悬疑 / 自然延时 / 抽象艺术 / 武侠决斗 / 萌宠爆款
每个场景 × 11 模型 = 110 条对照 prompt,每条严格遵循对应模型的官方公式
这构成"同一场景在 11 模型上的最佳写法对照",就是 translator 的查找表