DiSCO: Defending text-to-image generation through distribution-guided contrastive prompt optimization

  • 类型:arxiv
  • 标识:2608.17067
  • 链接:https://arxiv.org/abs/2608.17067
  • 主分类:multimodal
  • 形态:method
  • 被引:0
  • 被引来源:Semantic Scholar
  • S2被引:0
  • 影响力被引:0
  • TLDR:DiSCO is proposed, a zero-shot, strictly black-box defense that operates entirely at the prompt level as a plug-and-play module, requiring no model retraining, fine-tuning, or access to model internals, and can be readily applied to any text-to-image system without necessitating any changes to the model itself.
  • 副分类:risk
  • 待LLM分类:否
  • 标题中文:DiSCO:通过分布引导的对比提示优化保护文本到图像生成
  • TLDR中文:提出 DiSCO,一种零样本、严格黑盒的防御方法,完全在提示层面以即插即用模块的形式运行,无需模型重训练、微调或访问模型内部,可直接应用于任何文本到图像系统,无需对模型本身进行任何修改。
  • 来源文件
  • /inbox/tom/_candidates/2026-08-19-agent-rag-longcontext-candidates.json
  • [S2 enrich]
  • /inbox/tom/_candidates/2026-08-20-agent-rag-longcontext-candidates.json