Kuaishou's technical team has unveiled KwaiMind, a cutting-edge image editing foundation model tailored explicitly for real-world e-commerce content creation. This innovative model seamlessly blends general image editing prowess with specialized e-commerce functionalities, addressing a wide array of requirements including adding, deleting, replacing elements, adjusting composition, and rectifying local issues inherent in e-commerce material production.

KwaiMind constructs and upholds top-tier training samples through a sophisticated multi-agent data engine, adopting a training methodology that harmonizes specialized capability optimization with multi-teacher distillation techniques. It amalgamates multiple capabilities and has pioneered an exclusive e-commerce evaluation system, Ecom-Bench. According to related tests, KwaiMind has emerged as the frontrunner among competing open-source models, securing the top spot in comprehensive scores across various general image editing benchmarks and self-constructed e-commerce evaluations.

In online A/B experiments aimed at optimizing main product image materials, the model has demonstrated a notable relative increase in click-through rates, approximately 2.44%, thereby offering an efficient and effective image editing solution for e-commerce content production. Looking ahead, the Kuaishou team is committed to continuously refining and expanding the model's capabilities to further enhance its utility and performance in the e-commerce landscape.