JarvisEvo：面向协同编辑器-评估器优化的自进化照片编辑智能体
JarvisEvo: Towards a Self-Evolving Photo Editing Agent with Synergistic Editor-Evaluator Optimization
把交错多模态思维链（iMCoT）与编辑器-评估器协同策略优化（SEPO）引入图像编辑智能体，为视觉研究者提供一种无需外部奖励模型即可自我改进、并缓解奖励黑客的可复用训练范式；同时其保真型/生成型编辑双工具集成与 ArtEdit-Bench 评测协议可直接用于评估编辑中的像素级保真度。
原文依据 a synergistic editor-evaluator policy optimization (SEPO) framework that enables self-improvement without external rewards
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论基于智能体的编辑模型已在交互体验、处理质量和创作灵活性方面取得显著进展。然而，仍存在两个关键挑战：（1）指令幻觉——纯文本的思维链（CoT）推理由于固有的信息瓶颈，无法完全避免事实性错误；（2）奖励黑客——针对静态奖励模型的动态策略优化使智能体得以利用奖励函数中的缺陷。为解决这些问题，我们提出 JarvisEvo，一个统一的图像编辑智能体，它模仿人类专家设计师，通过迭代编辑、选择合适的工具、评估结果并反思自身决策来改进输出。JarvisEvo 具有三项关键优势：（1）交错多模态思维链（iMCoT）推理机制，可增强指令遵循能力与编辑质量；（2）编辑器-评估器协同策略优化（SEPO）框架，无需外部奖励即可实现自我改进，有效缓解奖励黑客问题；（3）通过无缝集成 Adobe Lightroom 与 Qwen-Image-Edit 工具，同时支持保真型编辑与生成型编辑。在 ArtEdit-Bench 上，JarvisEvo 在保真型编辑指标上平均超越 Nano-Banana 18.95%，其中像素级内容保真度提升达 44.96%，同时在生成型编辑任务上保持有竞争力的表现。
英文原文摘要 1422 字符
Agent-based editing models have substantially advanced interactive experiences, processing quality, and creative flexibility. However, two critical challenges persist: (1) instruction hallucination--text-only chain-of-thought (CoT) reasoning cannot fully prevent factual errors due to inherent information bottlenecks; (2) reward hacking--dynamic policy optimization against static reward models allows agents to exploit flaws in reward functions. To address these issues, we propose JarvisEvo, a unified image editing agent that emulates an expert human designer by iteratively editing, selecting appropriate tools, evaluating results, and reflecting on its own decisions to refine outcomes. JarvisEvo offers three key advantages: (1) an interleaved multimodal chain-of-thought (iMCoT) reasoning mechanism that enhances instruction following and editing quality; (2) a synergistic editor-evaluator policy optimization (SEPO) framework that enables self-improvement without external rewards, effectively mitigating reward hacking; and (3) support for both preservative and generative editing through seamless integration of Adobe Lightroom and Qwen-Image-Edit tools. On ArtEdit-Bench, JarvisEvo outperforms Nano-Banana by an average of 18.95% on preservative editing metrics, including a substantial 44.96% improvement in pixel-level content fidelity, while maintaining competitive performance in generative editing tasks.