发表机构
Kuaishou Group（快手集团）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
KwaiMind 结合通用能力与电商专用优化，通过数据引擎、多模态扩散 Transformer 及 CTR 等奖励对齐，在多项基准上取得最优，并显著提升实际点击率。
AI 中文摘要
商业图像编辑需要在保证通用编辑质量的同时，实现产品身份保持、准确的文本渲染和用户吸引力。我们提出了 KwaiMind，一个将通用能力与电子商务专业化相结合的图像编辑系统。一个基于智能体的数据引擎维护了约 180 万对高质量编辑数据对。KwaiMind 构建于多模态扩散 Transformer 之上，经历了持续预训练和有监督微调，随后进行偏好优化和在线强化学习。一个通用的视觉-语言评判器以及针对点击率（CTR）、文本渲染和产品一致性的专门奖励引导专门策略，这些策略通过在线策略蒸馏进行整合。我们引入了 Ecom-Bench，涵盖 11 项商业编辑任务，并包含任务特定的视觉评估和基于 CTR 的排名。在评估的开源编辑器中，KwaiMind 在 ImgEdit、GEdit、REDEdit 的两种语言子集以及 Ecom-Bench 视觉质量上取得了最强的总体得分，并在比较系统中获得了最高的综合 CTR 排名得分。离线情况下，CTR 引导的优化将生成图像中预测 CTR 超过原始产品图像的比例从 12.16% 提升至 37.41%。在在线 A/B 实验中，基于 CTR 选择产品主图使得实际 CTR 相对提升了约 2.44%。这些结果证明了领域特定数据和奖励驱动的对齐对于商业图像编辑的价值。
英文摘要
Commercial image editing requires product identity preservation, accurate text rendering, and user appeal alongside general editing quality. We present KwaiMind, an image editing system combining general capabilities with e-commerce specialization. An agent-based data engine maintains approximately 1.8 million high-quality editing pairs. Built on a multimodal diffusion transformer, KwaiMind undergoes continued pre-training and supervised fine-tuning, followed by preference optimization and online reinforcement learning. A general-purpose vision-language judge and specialized rewards for click-through rate (CTR), text rendering, and product consistency guide specialized policies, which are consolidated through on-policy distillation. We introduce Ecom-Bench, covering 11 commercial editing tasks with task-specific visual evaluation and CTR-based ranking. KwaiMind achieves the strongest overall scores among evaluated open-source editors on ImgEdit, GEdit, both language splits of REDEdit, and Ecom-Bench visual quality, and the highest aggregate CTR ranking score among compared systems. Offline, CTR-guided optimization increases the proportion of generated images whose predicted CTR exceeds that of the original product image from 12.16% to 37.41%. In an online A/B experiment, CTR-based selection of product main images yields an approximately 2.44% relative increase in actual CTR. These results demonstrate the value of domain-specific data and reward-driven alignment for commercial image editing.
CommentsKwaiMind Team, Kuaishou Group
