AI 科技大本营（ID：rgznai100） 想象一下，你面前摆着一杯咖啡，你伸手去拿，在你的手真正触碰到杯子之前，你的大脑已经在"脑补"了整个过程：手臂将如何移动、杯子会是什么触 感、抬起后桌面的样子……这种对未来场景的想象和预测能力，正是人类操控世界的核心认知基石。 那么，能否赋予机器人同样的"预演能力"，先在"脑海"中模拟动作后果，再付诸执行？这就是具身世界模型要做的事情：让机器人在行动前，就能"看 见"未来。近年来，借助大规模视频生成模型（如Sora、Wan等）强大的视觉先验，这一方向取得了令人瞩目的进展。 然而，一个尴尬的问题始终悬而未决： 视频生成模型的世界由像素编织而成，而机器人的语言却是关节角度与位姿坐标，它们使用完全不同的"表征语 言"描述同一个物理世界。 为了解决上述问题，中科第五纪联合中科院自动化所团队推出 BridgeV2W，它通过一个极为优雅的设计，具身掩码（Embodiment Mask），一种由 机器人动作渲染出的"动作剪影"，将坐标空间的动作无缝映射到像素空间，从而真正打通预训练视频生成模型与世界模型之间的桥梁，让机器人学会可 靠地"预演未来"。 「以棱镜之思，折射 AI 研究 ...

编辑：何隆星 审定：黄青 核发：邹建宾 点亮 近期，字节跳动最新视频生成模型Seedance 2.0在即梦、豆包、小云雀等产品开启内测，引发国内国外广泛关注。值得注意的是，由于该模型在真实感上 的突出表现，引发不少网友对可能混淆现实的担忧。对于用户反馈，字节跳动很快做出反应。2月9日，即梦创作者社群中，平台运营人员发布消息 称："Seedance 2.0 在内测期间收获了远超预期的关注，感谢大家的使用反馈。为了保障创作环境的健康可持续，我们正在针对反馈进行紧急优化，目前暂 不支持输入真人图片或视频作为主体参考"，并表示平台深知创意的边界是尊重，产品调整后会以更完善的面貌与大家正式见面。 来源：新京报 新闻荐读 ...

近期，字节跳动最新视频生成模型Seedance 2.0在 即梦 、豆包、小云雀等产品开启内测，引发国内国外广泛关注。值得注意的是，由于 该模型在真实感上的突出表现，引发不少网友对可能混淆现实的担忧。对于用户反馈，字节跳动很快做出反应。2月9日，即梦创作者社群 中，平台运营人员发布消息称："Seedance 2.0 在内测期间收获了远超预期的关注，感谢大家的使用反馈。为了保障创作环境的健康可持 续，我们正在针对反馈进行紧急优化，目前暂不支持输入真人图片或视频作为主体参考"，并表示平台深知创意的边界是尊重，产品调整后 会以更完善的面貌与大家正式见面。 多家银行，上调存款利率 雷军：春节期间小米汽车因自身故障抛锚，可报销1500元高铁/机票费和 500元住宿费 来源：新京报 关注" 新华网财经 "视频号 更多财经资讯等你来看 往期推荐 ...

"我很庆幸，至少今天的Seedance 2.0，来自中国。"他表示。 游戏科学CEO冯骥今日发文谈及刚上线的即梦Seedance 2.0：AI理解多模态信息并整合的能力完成了一 次飞跃，令人惊叹。AIGC的童年时代，结束了。 以下为微博全文： 昨天试了试刚上线即梦的Seedance 2.0。 4、产能爆炸。一般性视频的制作成本将无法再沿用影视行业的传统逻辑，开始逐渐趋近算力的边际成 本。内容领域必将迎来史无前例的通货膨胀，传统组织结构与制作流程会被彻底重构。相信只要用过的 朋友，都能很快理解这个预测绝非杞人忧天。 5、视频全民化。基于第4点，一切之前需要考虑制作成本的展示方式都会轻易视频化，典型如电商广告 与预拍摄。虽然目前游戏研发的核心环节尚未被AI直接侵蚀，但视频本身会逐渐向定制化、实时化、 游戏化演进，也许会成为未来某种全新的娱乐方式。 它在使用手册标题的最后，写着Kill the game！（杀死比赛）。这个评价，相当客观。 AIGC的童年时代，结束了。 写几条初步用过后的感受以及预测，仅供参考： 1、领先。当前地表最强的视频生成模型，没有之一。 2、全能。AI理解多模态信息（文、画、影、音）并整合的 ...

Core Insights - Kuaishou's AI product, Keling AI, has rapidly increased its monthly revenue from 150 million yuan to 140 million yuan in less than a year, with an annual recurring revenue (ARR) projected to reach 240 million USD (approximately 1.68 billion yuan) by December 2025 [1][4] - The launch of the "Motion Control" feature during the New Year holiday significantly boosted Keling AI's downloads, leading to a 28.15% increase in Kuaishou's stock price by January 14 [1][4] Revenue Growth - Keling AI's revenue surpassed 140 million yuan in December 2025, showing a noticeable acceleration in commercialization compared to 150 million yuan in Q1 2025 [4] - The growth in revenue is attributed to increased demand for video generation models and Kuaishou's ongoing investment in computational power [4][5] Competitive Landscape - The competition in the domestic video generation sector is intensifying, with Alibaba Cloud's Wanxiang 2.6 and Volcano Engine's Seedance 1.5 Pro targeting the same market as Keling AI [3][6] - Keling AI's revenue structure indicates that nearly 70% comes from paid subscriptions by professional users, such as content creators and marketers [6] Future Directions - Kuaishou plans to focus its capital expenditures on computational upgrades and technological advancements over the next 1-2 years [5] - To expand its user base, Keling AI aims to enhance its product offerings for ordinary users and integrate social interaction features [9][10] Product Innovation - Keling AI has launched several innovative models, including the "Keling O1" and "Keling 2.6," which support advanced features like audio-visual synchronization and multi-angle generation [6][7] - The recent success of the "Motion Control" feature, which allows users to create dynamic videos from static images, exemplifies Keling AI's strategy to simplify user experience and broaden its appeal [10]

Core Insights - The article discusses the emergence of AI video generation technology, highlighting the journey of Wang Changhu and his company, Aishi Technology, which aims to democratize video creation for everyone [1][9][32] Group 1: Industry Overview - The AI industry is experiencing a significant transformation, comparable to the impact of steam engines, electricity, and computers, with AI models becoming a universal technology [1][4] - 2022 marked the beginning of the AIGC era, with tools like Midjourney and ChatGPT showcasing the potential of AI, leading to a new wave of interest and investment in AI technologies [4][5] - The AI video generation sector was initially viewed skeptically by experts, who doubted its viability within five years, yet Aishi Technology has successfully launched a leading AI video model [6][10] Group 2: Company Strategy - Aishi Technology's strategy focuses on video generation, a less popular but potentially more impactful area compared to text and image models, allowing the company to carve out a unique niche [10][11] - The company aims to make video creation accessible to everyone, transforming ordinary users into directors of their own content, thus lowering the barriers to entry in video production [13][14] - The product "拍我AI" (Shoot Me AI) exemplifies this philosophy, enabling users to create videos with simple prompts, significantly enhancing user engagement and creativity [15][16] Group 3: Growth Mechanism - Aishi Technology has developed a self-reinforcing growth cycle, where high-quality models attract users, who in turn provide valuable data that enhances the models and products [20][22] - The company’s approach combines advanced technology with user-friendly design, resulting in a product that appeals to a broad audience and encourages content creation [21][22] Group 4: Organizational Efficiency - Aishi Technology emphasizes extreme organizational efficiency and precise technical judgment to compete with larger firms, aiming for ten times the efficiency of its competitors [24][25] - The company adopts a flat organizational structure to facilitate rapid decision-making and resource allocation, enhancing overall productivity [26][27] - Continuous innovation and attracting top talent are crucial for maintaining a competitive edge in the rapidly evolving AI landscape [28]

Core Viewpoint - Alibaba has launched the new Tongyi Wansxiang 2.6 model, which is the most comprehensive video generation model globally, covering various capabilities such as text-to-video, image generation, and audio-driven video creation [1]. Group 1: Video Generation Capabilities - The Wansxiang 2.6 model introduces multi-audio driven video capabilities, along with features like audio-visual synchronization and multi-shot storytelling, which were not available in Sora 2 [2]. - The model demonstrates significant improvements in artistic style control, realistic portrait generation, and understanding of historical and cultural semantics in image generation [3][8]. - The model's video generation capabilities include video reference generation, maintaining subject consistency, and natural audio-visual synchronization, which enhances the overall user experience [11][12]. Group 2: Performance Testing - Initial tests show that Wansxiang 2.6 performs well in video subject consistency and prompt understanding, achieving a near 1:1 replication of the subject's appearance and matching lip movements accurately [11]. - The model's ability to generate multi-shot narratives is effective, with smooth transitions and coherent storytelling across different shots, although some abstract actions may still pose challenges [17][18]. - The model's aesthetic quality in video generation has improved, showcasing a cinematic feel and strong visual appeal, particularly in complex scenes like cyberpunk cityscapes [14][24]. Group 3: Image Generation Enhancements - Wansxiang 2.6 has made advancements in image generation, particularly in style transfer, portrait generation, and bilingual text handling, demonstrating a better grasp of new aesthetic styles [19][22]. - The model successfully generated a food promotional poster with clear bilingual text and an appealing layout, indicating its reliability in aesthetic judgment [25][27]. - Overall, the model's performance is commendable, with minor flaws in multi-character dialogue and complex action understanding, but it is deemed usable for daily short video creation and secondary creation tasks [28][29].

Core Viewpoint - Alibaba has launched the upgraded Wanxiang 2.6 model, which is the first video model in China to support character role-playing, enhancing video creation efficiency and quality [1]. Group 1: Product Features - The Wanxiang 2.6 model includes features such as audio-visual synchronization, multi-camera generation, and sound-driven capabilities, making it the most comprehensive video generation model globally [1]. - The model allows for a maximum video length of 15 seconds, which is the highest in China, and introduces new functionalities like role-playing and storyboard control [1]. - Users can create professional-level videos by simply uploading personal videos and inputting prompts, enabling them to act as movie protagonists in high-quality cinematic scenes [1]. Group 2: Market Position - The previous version, Wanxiang 2.5, was the first audio-visual synchronization video generation model released in China and ranked first in the LMArena evaluation [1]. - The enhancements in Wanxiang 2.6 further improve image quality, sound effects, and adherence to instructions, catering to professional film-level requirements [1].

Core Insights - Alibaba has launched the next-generation Wanxiang 2.6 model, which is the first video model in China to support character role-playing functionality, enhancing capabilities for professional film production and image creation [1][2] Group 1: Product Features - The Wanxiang 2.6 model includes features such as audio-visual synchronization, multi-camera generation, and sound-driven capabilities, making it the most comprehensive video generation model globally [1] - The model has improved video quality, sound effects, and instruction adherence, achieving a maximum video length of 15 seconds, which is the highest in China [1] - New functionalities include character role-playing and storyboard control, allowing for one-click completion of videos with single or multiple characters, as well as automatic multi-camera switching for professional tasks [1][2] Group 2: Technological Innovations - The model integrates multiple innovative technologies for multi-modal joint modeling and learning, capturing emotional, postural, and visual features from reference videos, along with acoustic characteristics like tone and speech rate [1] - The high-level semantic understanding in storyboard control allows the creation of professional-grade multi-camera segments with a cohesive narrative and atmosphere [2] Group 3: User Accessibility - Users can experience Wanxiang 2.6 directly on the official website, and enterprise users can access the model API through Alibaba Cloud [2] - The model family supports over ten visual creation capabilities, including text-to-image, image editing, text-to-video, and video editing, widely applied in AI comics, advertising design, and short video creation [2]

Core Viewpoint - Alibaba has launched the next-generation Wanxiang 2.6 series model, which is a comprehensive upgrade aimed at professional film production and image creation scenarios. This model is the first in China to support character role-playing functionality and includes features such as audio-visual synchronization, multi-camera generation, and sound-driven capabilities, making it the most feature-rich video generation model globally [1]. Group 1 - The Wanxiang 2.6 model is designed for professional film production and image creation [1]. - It is the first video model in China to support character role-playing functionality [1]. - The model includes advanced features such as audio-visual synchronization, multi-camera generation, and sound-driven capabilities [1]. Group 2 - The Wanxiang 2.6 model has been made available on Alibaba Cloud's Bailian and the Wanxiang official website [1].