Group 1 - Alibaba Cloud's Lingyang launched the AgentOne platform to help enterprises transition from passive responses to proactive intelligence, aiming to create "super companies" in the AI era. The platform offers lifecycle management for intelligent agents and has over 20 ready-to-use enterprise-level agents covering marketing, customer service, analysis, and operations [2] - Alibaba introduced the upgraded Qwen3-VL series, featuring the flagship model Qwen3-VL-235B-A22B, which includes Instruct and Thinking versions. The Instruct version performs at or above Gemini 2.5 Pro in several visual perception evaluations, while the Thinking version achieved state-of-the-art performance in multi-modal reasoning benchmarks [2] - Shengshu Technology released the new generation video model Vidu Q2, focusing on "subtle expression generation." The model has made advancements in expression variation, camera movement, generation speed, and semantic understanding, offering various modes including image-to-video and customizable duration [2] - JD.com launched a digital assistant named "He She It" during the JD Discovery 2025 conference. The CEO demonstrated its capabilities by ordering coffee through voice commands, showcasing its ability to recommend products and complete transactions seamlessly [2]

锦秋基金于2023年年中投资了生数科技，是生数科技的早期机构投资人。 锦秋基金，作为12 年期的 AI Fund，始终以长期主义为核心投资理念，积极寻找那些具有突破性技术和创新商业模式的通用人工智能初创企业。 9月25日，锦秋基金被投公司生数科技正式发布新一代图生视频大模型Vidu Q2。新模型以" Vidu Q2 看AI演戏 "为主题，"细微表情生成"为核心提升场景，在极致表 情变化、推拉运镜、生成速度及语义理解方面取得突破性进展，实现从"生成视频"到"生成演技"，从"动态流畅"到"情感表达"的革命性跨越，标志着AI视频生成技 术正式从追求"形似"进入追求"神似"的新阶段，将为内容创作、影视产业、广告营销等领域带来全新升级。 以下为此次新闻的相关内容。 生数科技全球发布Vidu Q2，推动"视频生成"走向"演技生成"时代 9月25日，生数科技正式发布新一代图生视频大模型Vidu Q2。新模型以" Vidu Q2 看AI演戏 "为主题，"细微表情生成"为核心提升场景，在极致表情变化、推拉运 镜、生成速度及语义理解方面取得突破性进展，实现从"生成视频"到"生成演技"，从"动态流畅"到"情感表达"的革命性跨越，标 ...

据悉，本次更新的Vidu Q2主要包括图生视频、首尾帧视频、时长可选（2-8s）、电影大片及闪电出片 两种模式，在复杂表情变化的文戏，常见的多人打斗场景的武戏，及影视剧情中令人震撼的炫酷特效中 表现出彩，能够通过媲美人类的AI真实表演，将AI演技强力渗透影视、短视频剧情创作、广告创意等 生产力场景，推动AI视频生成主角从原本僵硬、机械、无感情的"AI木头"提升为灵动、有情感、会演 戏"AI人"。 责任编辑：何俊熹 9月25日消息，生数科技今天正式发布新一代图生视频大模型Vidu Q2。新模型以"Vidu Q2 看AI演戏"为 主题，"细微表情生成"为核心提升场景，在极致表情变化、推拉运镜、生成速度及语义理解方面取得的 突破性进展，实现从"生成视频"到"生成演技"，从"动态流畅"到"情感表达"的革命性跨越，标志着AI视 频生成技术正式从追求"形似"进入追求"神似"的新阶段，将为内容创作、影视产业、广告营销等领域带 来全新升级。 ...

Core Viewpoint - The rapid iteration of AI video models has created a landscape where users can easily generate videos, but practical application remains a challenge for ordinary users [2][3][4]. Group 1: User Needs and Model Evaluation - Many users require clear narratives, reasonable actions, and smooth visuals rather than complex effects [4][6]. - The evaluation focuses on whether these models can solve real problems in practical applications, particularly for novice content creators [5][7]. - A series of assessments were designed to test the models' capabilities in real-world scenarios, emphasizing practical video content creation [8][9]. Group 2: Model Selection and Testing - Eight popular video generation models were selected for testing, including Veo3, Hailuo02, and Jimeng3.0, which represent the core capabilities in the current video generation landscape [11]. - The testing period was set for July 2025, with specific attention to the models' performance in generating videos from text prompts [11]. Group 3: Evaluation Criteria - Five core evaluation dimensions were established: semantic adherence, physical laws, action amplitude, camera language, and overall expressiveness [20][25]. - The models were assessed on their ability to understand prompts, maintain physical logic, and produce coherent and stable video outputs [21][22][23][24][25]. Group 4: Practical Application and Limitations - The models can generate usable visual materials but are not yet capable of producing fully deliverable commercial videos [57]. - Current models are better suited for creative sketch generation and visual exploration rather than high-precision commercial content [65]. Group 5: Future Directions - Future improvements may focus on enhancing structural integrity, semantic understanding, and detail stability in video generation [60][61][62]. - The rise of image-to-video models may provide a more practical solution for commercial applications, bypassing some of the challenges faced by text-to-video models [62].