Files
nexus/wiki/concepts/NegativePromptingLibrary.md
2026-04-25 00:02:50 +08:00

1.4 KiB
Raw Blame History

title, type, tags, sources, last_updated
title type tags sources last_updated
Negative Prompting Library concept
prompt-engineering
generative-ai
bias-mitigation
image-generation
video-generation
design-inclusive-visuals-specialist
2026-04-24

Definition

AI 图像/视频生成中显式列举"应避免生成的内容"的约束库——通过负向提示词告知模型不要生成特定元素,是对抗 AI 幻觉(克隆脸、乱码符号、超现实瑕疵等)的核心技术手段。

Use Cases

Image Generation (Midjourney/DALL-E/Flux)

  • No cloned faces — 强制面部结构差异
  • No gibberish text or symbols — 避免非英语文字幻觉
  • No generic "stock photo" smiles — 避免库存照片套路
  • No oversized cultural symbols dominating composition — 避免英雄符号构图
  • No extra fingers, unnatural lighting on dark skin — 避免物理失真

Video Generation (Sora/Runway)

  • No cloned background actors — 背景人物必须展现交叉性差异
  • No text or symbols on whiteboards/screens — 避免文字幻觉
  • No futuristic/sci-fi tropes in real-world contexts — 避免超现实元素混入写实场景
  • No physics errors on mobility aids — 轮椅/拐杖/假肢运动一致性