虚拟试衣、AI 试衣、服饰换装、扩散模型、图像生成
Virtual Try-On (VTON), AI TryOn, Outfit Swap, Diffusion Model, Diffusion-based Image Generation
面向 ToB 商家模特场景，通义虚拟试衣在复杂服饰版型、复杂服饰穿法、复杂模特姿态上持续提升，与绘蛙、可灵试衣相对比具有明显优势，达到行业领先水平。
Targeting ToB merchant and model scenarios, Tongyi Virtual Try-On continues to advance in handling complex garment patterns, intricate dressing configurations, and challenging model poses. Compared to competitors such as HuiWa and KeLing Virtual Try-On, it demonstrates clear advantages and has achieved industry-leading performance.
从左到右依次为：输入<平铺图，模特图>、通义虚拟试衣的换装结果、绘蛙试衣结果、可灵试衣结果 From left to right: input pair <garment image, model image>, result byTongyi Virtual Try-On, result by HuiWa Virtual Try-On, and result by KeLing Virtual Try-On.
在一阶段模型输出结果上进一步优化，构建二阶段模型，提升服饰整体质感和细节粒度，对毛衣、牛仔裤、针织衫等服饰细节质感提升明显，达到行业领先水平。
Building upon the output of the first-stage model, we developed a second-stage refinement model to enhance the overall fabric realism and fine-grained details of gerenated VTON results. This approach significantly improves the textural fidelity of challenging clothing items such as sweaters, denim jeans, and knitted tops, achieving industry-leading quality.
从左到右依次为：输入<平铺图，模特图>、基础版通义虚拟试衣结果（及局部放大图）、优化后的通义虚拟试衣结果（及局部放大图）、绘蛙试衣结果（及局部放大图） From left to right: input pair <garment image, model image>, Tongyi VTON Stage I result (with inset close-up), refined Tongyi VTON Stage II result (with inset close-up), and result from HuiWa (with inset close-up).
