Core Viewpoint - The article discusses the rise of AI-generated video content, particularly focusing on a startup named Hedra, which has developed a technology for creating talking baby podcasts using AI-generated characters [1][2]. Group 1: Company Overview - Hedra was founded in 2023 and offers a web-based video generation and editing suite centered around its proprietary Character-3 model [1][5]. - The company completed a $32 million Series A funding round on May 15, led by Andreessen Horowitz, with existing investors participating [2][5]. - The CEO of Hedra, Michael Lingelbach, identified a market gap between companies like Synthesia and Runway, aiming to create longer dialogue scenes with greater control [2][5]. Group 2: Technology and Product Development - The Character-3 model, launched in March, has been a significant turning point for user growth and is expected to enable more customized AI character interactions [5][6]. - Hedra's technology allows users to integrate various models for video generation, including those for image and audio generation, enhancing the overall video production capabilities [7]. Group 3: Market Position and Competition - Hedra's competitors include Captions, Cheehoo, Synthesia, and HeyGen, with Hedra claiming its video characters are more expressive than those of its rivals [7]. - Andreessen Horowitz's Matt Bornstein noted that as AI-driven video generation evolves, more tools focusing on character, action, voice, and editing will emerge [7].

多模态 AI 技术正以前所未有的速度重塑内容创作领域。 从2024年 OpenAI Sora 点燃全球想象，到近期，吉卜力风图片席卷全网。这个一度被视为 AI 终极想象力边界 的领域，正以前所未有的速度冲破技术壁垒。 视频生成作为技术难度与应用潜力并存的关键环节，也吸引了全球范围内的广泛关注和投入。 在追求更长时长、更高分辨率、更惊艳视觉效果的同时，内容一致性难以保证、生成过程可控性不足、以及高 昂的计算成本等核心挑战，依然限制了其在专业领域、大众娱乐领域的规模化应用。 在此背景下，由生数科技研发的视频生成模型 Vidu，展现出一条差异化的发展路径。在多模态视频生成技术 的早期发展阶段，通过集中资源解决专业用户的核心痛点，如一致性、可控性、效率，建立起差异化优势和用 户基础，尤其是在动画等特定领域形成壁垒。 根据生数科技廖谦在近期访谈中的阐述，Vidu 的核心定位是"全球领先的AI内容生产平台 "，这也意味着 ，除 了追求基础生成能力的提升，也需要优先解决实际工作流中的关键痛点。 比如，生数科技敏锐的发现，纯粹的文生视频因为难以控制一致性，应用者并不多 。而 Vidu 推出的"参考 生"（Reference ...

Core Insights - The commercialization of the AI video generation industry is accelerating, with the Chinese AI company, Shengshu Technology, rapidly announcing partnerships with several leading enterprises in a short time frame [1] - Shengshu Technology's flagship product, Vidu, has achieved significant recognition in the multimodal generation field, indicating its advanced capabilities and leading position in the market [1] Group 1: B2B Commercialization Growth - Shengshu Technology has established deep partnerships across multiple sectors, including internet, hardware manufacturing, film and animation, and cultural media, indicating a robust and diversified commercialization strategy [2][3] - The company has integrated Vidu into major platforms such as 360 Search, Baidu Search, and Amazon Cloud, expanding its global market reach [2] - In the film and animation sector, Shengshu Technology has secured rights to adapt popular web novels and collaborated on AI-generated promotional content for major films [2] Group 2: Performance Metrics - Vidu has achieved top rankings in global video generation assessments, with scores of 87.41% and 60.98% in the VBench Leaderboard, surpassing competitors like Runway and LumaAI [4] - The model also leads in specialized categories for animated and realistic styles, showcasing its strong and stable capabilities in video generation [4] Group 3: Solution Delivery and Client Engagement - The company emphasizes a complete and systematic delivery capability, which is crucial for its commercial maturity [6] - Shengshu Technology provides tailored technical teams for various industries, offering over 500 industry-specific templates and end-to-end solutions for advertising and animation [7] - The ability to embed AI capabilities into enterprise workflows and provide comprehensive lifecycle services is a key factor for clients choosing Shengshu Technology's models [7][8] Group 4: Industry Positioning - Compared to larger companies with complex business lines, Shengshu Technology's agility allows for quick responses to client needs, enhancing its competitive edge in the B2B market [8] - The focus on solution value over mere technological scarcity is becoming a significant pricing anchor in the AI industry [8] Group 5: Future Outlook - The industry is closely watching whether the AI video generation sector can carve out a unique path for commercialization [9]

作者 | 刘宝丹 编辑 | 周智宇 近日，快手正式发布可灵2.0视频生成模型及可图2.0图像生成模型，将视频及图像创作的精准度带上一 个新高度。同期，字节Seed团队正式发布Seedream 3.0 技术报告，据第三方榜单Artificial Analysis， Seedream 3.0综合性能已追平文生图SOTA模型GPT-4o，进入全球第一梯队。 作为短视频平台，字节和快手被认为是AI多模态领域的有力竞争者。经过一年多的技术追赶，双方在 AI视频生成领域都取得了不错的进展。 根据AI产品榜3月数据，在全球AI产品增速榜（仅APP）上，即梦AI 以173.57%的月活增速位居第5，是 增速最快的AI视频应用，其月活规模约2037万，而可灵AI的增速仅为36.44%，排名第14。根据快手公 布的数据，截至目前，可灵AI全球用户规模突破2200万。 AI竞赛焦点已经开始转向多模态，字节和快手在AI视频赛道的竞争也日趋激烈。 不过，当前AI视频生成领域尚未涌现类似DeepSeek在大型语言模型（LLM）领域的标杆性产品，根据 Gartner 2024年新兴技术成熟度曲线显示，该技术仍处于创新触发期，这也意味着，字 ...

Core Viewpoint - The article discusses the current state of AI video generation, highlighting the limitations of existing tools and the breakthrough achieved by Kunlun Wanwei's Skyreels-V2, which redefines video generation capabilities and offers a comprehensive filmmaking solution [1][3]. Group 1: Current State of AI Video Generation - AI video generation tools are currently limited to short clips of around 10 seconds, struggling with coherent storytelling and quality [1]. - Existing models often produce unsatisfactory visual effects and lack emotional depth in character portrayal [1][3]. - The industry is facing a technical bottleneck, with many tools unable to produce longer, cohesive narratives [1][5]. Group 2: Breakthrough of Skyreels-V2 - Skyreels-V2 is the first open-source film-grade generation model that supports unlimited video length, breaking the existing constraints of AI video generation [1][3]. - It introduces a "dual-engine" architecture that enhances three core metrics: duration extensibility, visual quality, and director control [1][3]. - The model allows for continuous storytelling, enabling the creation of long-form content that rivals traditional filmmaking [6][10]. Group 3: Technical Innovations - Skyreels-V2 employs a diffusion forced framework, integrating multi-modal large language models and reinforcement learning to overcome existing technical challenges [10][12]. - The model has a vast dataset of over 100 million samples, including 280,000 films and series, which enhances its training and output quality [14]. - It achieves high visual fidelity, supporting outputs of 720p and above, and maintains realistic motion dynamics [8][12]. Group 4: Practical Applications - Skyreels-V2 serves as a creative platform for various users, from novelists to marketers, enabling them to generate high-quality video content with minimal technical knowledge [20][22]. - It allows creators to experiment with different narrative styles and visual languages, enhancing the creative process [24][25]. - The model simplifies the filmmaking process, making it accessible to a broader audience by transforming ideas into visual narratives without the need for extensive technical skills [25].

Investment Rating - The report maintains a "Buy" rating for the company, with a target price of HKD 77.61 per share based on a 16x PE valuation for 2025 [2][6]. Core Views - The report highlights the successful launch of the Kuaishou Keling 2.0 model, which is expected to empower various industries such as advertising, UGC, and film creativity [1][4]. - The company has shown significant growth in its AI capabilities, with a user base exceeding 22 million and a monthly active user (MAU) growth of over 25 times since the launch of Keling AI [4]. - The Keling 2.0 model has improved semantic response capabilities, dynamic quality, and visual aesthetics, positioning the company as a leader in the global video generation market [4]. Financial Forecast and Investment Recommendations - The adjusted net profit forecast for the company is projected to be CNY 177 billion, CNY 194 billion, and CNY 229 billion for the years 2024, 2025, and 2026 respectively [2][5]. - The company's revenue is expected to grow from CNY 113.47 billion in 2023 to CNY 153.30 billion in 2026, reflecting a compound annual growth rate (CAGR) of approximately 11.1% [5][8]. - The report anticipates a gross margin improvement from 50.6% in 2023 to 56.2% in 2026, indicating enhanced operational efficiency [5][8]. Valuation Metrics - The report uses a PE valuation method, maintaining a 16x PE for 2025, leading to a reasonable valuation of CNY 3,107 billion or HKD 3,345 billion [6][7]. - The company's earnings per share (EPS) is projected to increase from CNY 1.48 in 2023 to CNY 4.96 in 2026, demonstrating strong earnings growth potential [5][8].

Core Insights - Runway AI has launched a new AI model aimed at creating videos with consistent characters, objects, and backgrounds, marking significant progress in the competition for faster and lower-cost film production [1][2] - The new model, Gen-4, will be released to paid users and includes features for generating more cohesive video scenes, allowing users to create 720p resolution clips of five and ten seconds [1][2] Group 1: Product Development - Runway's new AI model challenges OpenAI's Sora by providing users with more coherent video outputs, amidst increasing competition from tech companies [2] - The CEO of Runway, Cris Valenzuela, emphasized the goal of meeting Hollywood standards and the quality expected by professional filmmakers [3] - The software has been used in various projects, including scenes for Amazon's "David's House," visual effects for Madonna's concert tour, and advertisements for Puma [6] Group 2: Technical Improvements - The latest AI model improves output by maintaining a series of details such as positioning, character consistency, and overall video aesthetics [5] - Valenzuela noted that the company is focusing on industry-specific terminology during model training to make the prompt-writing process more intuitive for filmmakers [7] Group 3: Future Goals - The first goal of the AI model is to render videos, while the second phase aims to create engaging stories that resonate with viewers [8]

AI视频生成赛道已到了技术成熟和商业化落地的关键时刻。 本文为IPO早知道原创 作者｜Stone Jin 微信公众号｜ipozaozhidao 据IPO早知道消息，原字节跳动AI大将、火山引擎高管骆怡航于近日加入生数科技，担任CEO一职， 全面负责公司研发、产品、商业化及团队管理工作。 图 生数 骆怡航博士毕业于清华大学自动化系，深耕云计算及AI领域十余年，拥有深厚的技术背景、产业生态 理解和成熟的商业化经验，还具有丰富的海外拓展经历。 在加入生数科技之前，他担任字节跳动火 山引擎AI应用产品线一号位，汇报火山引擎总裁，全权负责产品线的战略、产品和商业化。 据悉， 该条产品线由骆怡航博士从 0组建，涵盖多个传统AI、大模型及大模型应用产品，管理规模数百人， 服务全球多个行业及国家的近万家客户，该产品线当前是火山引擎的重点产品线之一，也是大模型业 务的主力产品线。更早期，他在字节跳动负责AI解决方案与商业合作，曾参与到字节跳动早期的AI中 台规划与建设，见证并推动了字节跳动在AI领域的发展。 事实上，从整个行业发展来说，骆怡航选择加入生数科技，某种程度上也意味着AI视频生成赛道已到 了技术成熟和商业化落地的关键 ...

发布日期：2025年3月11日 本报告由中信建投证券股份有限公司在中华人民共和国（仅为本报告目的，不包括香港、澳门、台湾）提供。在遵守适用的法律法规情况下 ，本报告亦可能由中信建投（国际）证券有限公司在香港提供。同时请务必阅读正文之后的免责条款和声明。 前言：可灵，不可忽视的国产AI新力量 1 证券研究报告行业动态报告 可灵：头部AI视频产品 ——AI产品深度拆解（系列1） 分析师：杨艾莉 yangaili@csc.com.cn 010-85156448 SAC 编号：S1440519060002 SFC 编号：BQI330 分析师：杨晓玮 yangxiaowei@csc.com.cn SAC 编号：S1440523110001 近期，在国新办举行的新闻发布会上，政府工作报告起草组成员、国务院研究室副主任陈 昌盛提到，中国AI产业跑出加速度，除了大家熟知的Deepseek、宇树科技之外，还有可灵 视频，在国际上的评价已经超过Sora，可灵的价值正在被逐步认知。 可灵是由快手在24年初推出的AI视频生成模型，推出后在国内和海外都引起好的反响，尤 其是网页端80%以上的访问量都来自海外，呈现"墙内开花墙外香"的趋 ...

Core Viewpoint - OpenAI plans to integrate its AI video generation tool Sora into ChatGPT, aiming to expand the tool's accessibility and functionality while maintaining the simplicity of ChatGPT [2][3][4]. Group 1: Sora Integration and Expansion - OpenAI intends to make Sora accessible within ChatGPT, although the version may not offer the same level of control as the standalone web application [3]. - The integration of Sora into ChatGPT could drive user engagement and potentially encourage upgrades to premium subscriptions for more frequent video generation [3][4]. - OpenAI is actively seeking mobile engineers to develop a standalone Sora mobile application, enhancing user experience and accessibility [4]. Group 2: Future Developments - OpenAI is working on expanding Sora's capabilities to include image generation, potentially allowing users to create more realistic photos [5]. - The company is also developing a new version called Sora Turbo, which powers the current Sora web application [6].