Group 1 - The core achievement of the company is the introduction of two innovative results selected for ICCV 2025, focusing on image generation and video enhancement, showcasing breakthroughs in generative AI technology [1][2] - In image generation, the company developed a new denoising masked autoregressive generation paradigm called De-MAR, which addresses key bottlenecks in autoregressive models for visual generation, improving detail representation and inference speed [1] - The De-MAR framework utilizes a dual-token optimization mechanism, incorporating diffusion and denoising heads, achieving top-tier FID scores of 1.47 and 5.27 on ImageNet and MS-COCO datasets, respectively, while generating images 45% faster than DiT-XL/2 [1] Group 2 - In video enhancement, the company introduced the generative video quality enhancement framework GenVE, which overcomes detail loss issues in traditional methods through a dual alignment mechanism [2] - GenVE employs an image diffusion model for semantic reference generation and a local perception cross-attention module for precise texture detail transfer to videos, enhancing robustness through multiple strategies [2] - The framework has shown superior performance on datasets like YouHQ40 and VideoLQ, effectively restoring details such as hair and clothing folds, resulting in more natural and fluid video visuals [2]