[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-9a844b59-5eb1-4601-9000-becdaad17827":3},{"id":4,"title":5,"summary":6,"original_url":7,"source_id":8,"tags":9,"published_at":23,"created_at":24,"modified_at":25,"is_published":26,"publish_type":27,"image_url":13,"view_count":28},"9a844b59-5eb1-4601-9000-becdaad17827","Qwen-Image-Agent: 用 Agent 范式补齐文生图的「上下文缺口」","2026-6-28，阿里 Qwen 团队与康奈尔大学在 arXiv 发布 Qwen-Image-Agent (2606.26907)：通过 plan→reason→search→memory→feedback 闭环补齐缺失信息，再交给 Qwen-Image-2.0 渲染。\n\n论文命名「Context Gap」：用户提示 c_u 与渲染器所需 c_g 间存在系统性缺口。改造落在调用图：Context-Aware Planning 识别缺口，Context Grounding 把推理、检索、记忆与反馈汇入生成上下文；渲染器可换装，默认 Qwen-Image-2.0，编排用 GPT-5.5-0424。\n\nIA-Bench 拆为 Plan\u002FReason\u002FSearch\u002FMemory 四维：IA-score 45.4，超 GPT-Image-1.5 与 Nano Banana Pro，MindBench 较直接生成基线提升 82.6%。思路与 ReAct、Self-Refine 一脉相承，亮点是首次系统迁移到 T2I 并贡献可复现评测。规模逼近数据边际拐点时，把杠杆从参数迁移到调用结构更便宜。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.26907","c36a21ac-2a77-421b-9519-1e150695732a",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"6ad31a14-c0da-42df-81fd-564281f768db","agentic-ai",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",{"id":18,"name":19,"slug":19,"description":13,"color":13},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"id":21,"name":22,"slug":22,"description":13,"color":13},"c187600e-804c-4697-b828-1e4330e0eb10","qwen","2026-06-28T16:30:00Z","2026-06-28T16:11:44.239941Z","2026-06-28T16:11:44.239951Z",true,"agent",2]