TENCENT(HK:00700) Sou Hu Cai Jing·2025-05-16 09:15Core Insights - Tencent has launched its latest image generation technology, Hunyuan Image 2.0, which has garnered significant attention in the industry for its real-time image generation and hyper-realistic visual quality [1][10] - The model features a substantial increase in parameters compared to its predecessor, utilizing a high-compression image codec and a new diffusion architecture, resulting in image generation speeds that far exceed the industry average [1] - Hunyuan Image 2.0 achieves a response time in milliseconds, allowing users to see generated images instantly while typing or speaking, thus revolutionizing the traditional "wait-generate" model [1] - The quality of generated images has also improved significantly, employing advanced algorithms like reinforcement learning and incorporating extensive human aesthetic knowledge to produce images that are realistic and rich in detail, while avoiding common "AI flavor" seen in AIGC images [1] Performance Metrics - The accuracy of Tencent's Hunyuan Image 2.0 model exceeds 95% on the Geneval benchmark, outperforming other similar models and demonstrating its superior performance [2] Features and Innovations - The model includes a real-time painting board feature, allowing users to preview coloring effects while drawing sketches or adjusting parameters, thus breaking the traditional linear workflow of "draw-wait-modify" [1][8] - The real-time painting board supports multi-image fusion, enabling users to overlay multiple sketches on a single canvas and automatically coordinate perspective and lighting with AI, enhancing the interactive experience of AI image generation [1][8] Industry Impact - The release of Hunyuan Image 2.0 marks another significant milestone for Tencent in the image generation field, following its introduction of the first Chinese native DiT architecture model in 2014 [10] - Tencent continues to invest in image and video modalities, driving innovation and progress in technology, with plans to further explore multi-modal fields to deliver more surprises and breakthroughs to users [10]