AI修图翻车四成，换个思路让AI自己挑
让AI按一句话改图，现在最通用的模型有超过四成情况直接交不出能用的结果——不是改得不好，是它卡在自己的小工具箱里：每一步失败就换着法子重试，而负责验收的“裁判”也只盯着像素看，看不出改没改对。这篇把两个空间同时撑开：每一步不再只调一个通用模型，而是把任务拆给擅长不同活儿的专用工具；每个候选结果附上一份“验证记录”，记录它改了什么、怎么改的，让裁判不再只看图，而是看图加记录来打分。结果，现成的视觉语言模型当裁判，比专门训练的打分模型还准；候选结果并行生成、一次性排序，专家在78.5%的对比里更喜欢这种方案，而且裁判调用次数减半。它不是你明天就能用的插件，但指向一个更靠谱的方向：与其让AI闭门造车，不如让它把过程摊开给人看。
📄 原文摘要(英文)
Agentic image editors decompose complex requests into steps, but when a step fails, they mostly retry with the same narrow set of tools and a judge that scores each output by eye directly. We argue that such agents are bounded by two spaces: the action space of what the agent can do, and the observation space of what its judge can see. On our expert-annotated benchmark, a general-purpose editor yields no usable result in over 40% of the cases, and no judge is reliable across edit types. We present EditAgent, which expands both spaces. It routes each step to general editors and specialized methods, and attaches to each candidate a verification record. With these records, off-the-shelf VLM judges outperform finetuned reward models. Candidates are generated in parallel and judged in one listwise call: experts prefer this to sequential refinement in 78.5% of decided comparisons, at no more than half the judge calls. EditAgent improves two open-weight editors on three benchmarks and serves as a data engine for training stronger editors.