Investment Rating - The report maintains a strong buy rating for the media industry, indicating a positive outlook for investment opportunities in the AI video sector [2]. Core Insights - The AI video generation technology is evolving rapidly, transitioning from GAN to DiT architectures, which are crucial for advancing towards AGI. This evolution is expected to significantly enhance the capabilities of AIGC (AI-Generated Content) [3][9]. - The global AI video generation market is projected to reach $296 million by 2026, with a year-on-year growth of 35.16%. The industry is exploring both consumer (C-end) and business (B-end) revenue models, with significant advancements in commercial applications expected in the near future [3][4]. Summary by Sections 1. Video Generation Evolution - Video generation integrates multiple modalities, including text, images, and audio, which enhances its complexity and expressiveness, representing the upper limit of AIGC capabilities [7]. - The technology has progressed from early GAN models to the current DiT architecture, marking a significant turning point in the industry with the introduction of models like OpenAI's Sora [9][25]. 2. Technical Progress - Current AI video generation models can produce short segments that approach professional production quality, with resolutions supporting 1080p and frame rates reaching 30fps. However, challenges remain in generating longer videos and maintaining physical realism [34][36]. - The emergence of world models is anticipated to address existing limitations in video generation, potentially leading to a new phase of technological advancement [33]. 3. Commercialization Progress - The AI video generation market is expanding rapidly, with both consumer and business segments progressing simultaneously. The C-end focuses on subscription models, while the B-end primarily utilizes APIs for applications in advertising and e-commerce [3][4]. - The industry is witnessing a shift towards integrating AI capabilities into film production, with significant projects already generating substantial revenue, such as Utopai's projects totaling approximately $110 million [3][4]. 4. Core Beneficiaries - Key companies benefiting from this trend include technology firms with proprietary algorithms, content providers with extensive asset libraries, and platforms actively integrating AI into marketing strategies [4].

Core Viewpoint - The software ETF (515230) has seen a significant increase of over 2% on January 23, with a net inflow of over 2.8 billion yuan in the past 10 days, indicating strong investor interest in the sector. The multi-modal technology is expected to be a key factor in AI applications by 2026, benefiting primarily AI video and robotics/autonomous driving sectors [1]. Group 1: Multi-modal Technology - Multi-modal technology is anticipated to be a decisive factor in AI applications by 2026, with direct beneficiaries being AI video and robotics/autonomous driving [1]. - In the AI video sector, advancements such as the resolution of physical consistency issues with Sora2 and Veo3 are expected to lead to a generative environment by Q4 2025, with further acceleration anticipated as domestic multi-modal models catch up in Q1 2026 [1]. - The robotics/autonomous driving field is expected to see practical applications in experimental environments by 2026, driven by advancements in world models like Google's Genie and Tesla's iterations [1]. Group 2: Domestic and International Developments - Internationally, multi-modal technology is projected to evolve further in 2026, moving towards a unified tokenized world model [1]. - Domestic models such as Byte's Seed and Minimax's Hai Luo are expected to catch up quickly, with related products likely to be released in the first half of 2026 [1]. - The demand for computing power and storage is expected to benefit from the implementation of multi-modal and long-memory technologies [1]. Group 3: Software ETF Overview - The software ETF (515230) tracks the software index (H30202), which reflects the market performance of the software industry, covering companies involved in application software, system software development, and related services [1]. - The index focuses on technology innovation and high-growth companies, with a concentration in the information technology sector, leaning towards a growth-oriented style [1].

将一头已灭绝6500万年的远古沧龙做成菜需要几步？ 来自上海的餐厅主理人辛西娅展示了她的步骤：颊肉经历过黄油与香草的低温慢煮，继而在烈火宽油中迎来川菜的试炼；接着，野生鸡油菌和黑松露由黄 油煎制，为这道菜增加了丰富的层次；金黄的脆米粉则作为铺底，颊肉被放置其上，一道"龙吟之心"便被摆在了头戴恐龙头套的"地狱厨师"面前。 这一片段源于B站UP主@黄浦江三文鱼 模仿《地狱厨师》制作的系列视频"把远古沧龙做成六道菜"，每期视频时长都在6分钟以上，且从头到尾都由AI生 成。 这类视频悉数由AI生成，并在信息流中抢占着用户注意。在抖音，"AI美食"话题收获了超2亿次播放；小红书上，AI吃播则与ASMR结合，自成一类内容 模式，动辄收获上万次点赞，且涌现出不少专注于此内容品类的账号。 一面是AI抢入美食赛道，另一面，AI创作的边界问题也在显现。这一切对创作者而言是机遇还是挑战？ AI入侵美食赛道 早前，DS初步爆火互联网时，美食赛道上诞生的一种玩法实际上是"用AI创造料理"，即由AI自由发挥创造菜谱。今年2月，UP主@洛杉矶嬴政W 突发奇 想，要求AI为自己制作一道人类从未见过的料理。在AI的指导下，UP主兢兢业业一步步 ...

Core Viewpoint - The article discusses the importance of understanding the underlying principles of world models, emphasizing that relying solely on data-driven approaches ("炼丹") is insufficient for creating effective AI systems. It advocates for the integration of human-understandable structures and logic into AI models to enhance their interpretability and reliability [2][63]. Group 1: Development of 3D Generation - The evolution of 3D generation has transitioned from early attempts at creating 3D models from single images to the current era of large models capable of generating high-quality 3D content from textual descriptions [7][16]. - The emergence of "open world" 3D generation began around 2023 with the Dreamfusion project, which allowed for the generation of 3D models without category restrictions, marking a significant shift in the field [11][12]. - Current trends in 3D generation focus on achieving finer details, structured outputs for easier editing, and better alignment between generated models and input images [19][20]. Group 2: Challenges and Opportunities in 3D Generation - The article highlights a dilemma faced by the 3D generation field, particularly in light of advancements in video generation technologies that can produce content without the complex 3D modeling processes [24][28]. - Despite the rise of video generation, 3D content creation retains its value due to its ability to provide physical realism, spatial consistency, and detailed control over content [29][34]. - The potential crisis for 3D generation lies in the increasing capabilities of video generation models, which are beginning to exhibit controllable features, raising questions about the necessity of 3D in future content creation [34][38]. Group 3: The Role of 3D in World Models - The article categorizes world models into three types: macro models for societal understanding, personal experience models for exploration, and embodied models for machine intelligence, with 3D being essential for interactive virtual environments [43][44][45]. - For embodied intelligence, understanding human interaction with the physical world necessitates 3D modeling to accurately capture and simulate these interactions [48][50]. - The transition from digital to physical manufacturing processes, such as 3D printing, underscores the foundational role of 3D data in creating tangible products [52]. Group 4: Technical Approaches in AI - The article contrasts explicit and implicit approaches in AI development, with explicit methods relying on clear geometric and physical modeling, while implicit methods depend on data-driven neural networks [56][57]. - The need for explainability in AI systems is emphasized, suggesting that a balance between performance and interpretability is crucial for user trust and safety [58][63]. - The discussion concludes that 3D and 4D modeling are vital for providing a comprehensible framework for understanding complex AI systems, thereby enhancing user confidence [59][63].

欧盟方面表示，监管机构担心谷歌可能通过对出版商和内容创作者施加不公平条款，或为自身提供对相 关内容的特权访问，从而在训练大型模型时获取竞争者难以复制的数据优势。 外界认为，欧盟正试图在全球科技竞争中巩固对平台行为的规则引导权。 欧盟委员会认为，谷歌可能在创作者无法真正选择的情况下，使用上传至YouTube的视频训练自家的 Gemini与Veo3模型，而创作者在上传内容时被要求授予谷歌广泛的数据使用许可，使得"同意"带有默认 性质，缺乏现实的选择空间。 同时，谷歌禁止第三方公司使用YouTube视频训练模型，除非版权持有人明确授权，这使谷歌可能在训 练数据层面形成天然壁垒，进一步激化外界对其市场支配力的担忧。 对此，谷歌回应称，相关投诉可能抑制本已竞争激烈的市场创新，并强调其已与新闻和创意产业保持合 作，帮助他们适应AI带来的行业变化。 尽管谷歌公司否认有任何滥用市场地位的行为，但欧盟此次行动仍被视为欧洲近年来针对美国科技企业 监管升级的又一次体现。 欧盟委员会近日宣布将对谷歌展开正式调查，重点评估其在训练Gemini等人工智能（AI）模型时，使 用在线出版商内容以及YouTube创作者视频的方式是否违反了欧洲 ...

以下文章来源于锌刻度 ，作者黎炫岐 锌刻度 . 专注科技互联网原创报道 重新定义"吃"的边界。 文 ｜ 黎炫岐 编辑 ｜ 陈邓新 来源｜ 锌刻度（ID： znkedu ） 封面来源 ｜ 小红书 由Veo生成 被咬开时发出清脆声响的玻璃水果、镶嵌着宝石的首饰盒、播放着音乐的水晶球，甚至还有毛绒玩具labubu和金条……各种你能想到或者想不到的，都正成 为AI吃播的"食材"，被AI主播们塞入嘴里，轻松咀嚼。 这是一场风靡国内国外的热潮。在国外，Tiktok上一位叫leilanikovac的博主发了一条AI吃熔浆的视频，点赞数突破81.7万，另一位博主在三天内发了11条切 水果的视频后，粉丝数突破8万；而在国内，各大短视频平台和社交平台上，已有不少相关账号出现，点赞量破万的也不在少数。 当真人吃播面临种种道德和法律困境，猎奇食物逐渐从吃播的饭桌前消失，AI吃播却脑洞大开，主打一个万物皆能吃。 锌刻度了解到，目前大部分AI吃播视频都由Veo3生成。这是今年5月底，Google DeepMind发布的一款视频生成模型。这款模型的最大亮点是AI原生可以一 键直接生成与画面相匹配的声音。而这正是吃播的关键。 AI吃播的流量 ...

Core Insights - The first frame in video generation models serves as a "conceptual memory buffer" rather than just a starting point, storing visual entities for subsequent frames [3][9][48] - The research highlights that video generation models can automatically remember characters, objects, textures, and layouts from the first frame and reuse them in later frames [9][10] Research Background - The study originates from a collaborative effort by research teams from UMD, USC, and MIT, focusing on a phenomenon in video generation models that had not been systematically studied [5][8] Methodology and Findings - The proposed method, FFGo, allows for video content customization without modifying model structures or requiring millions of training samples, needing only 20-50 carefully curated examples [18][21] - FFGo can achieve state-of-the-art (SOTA) video content customization with minimal data and training time, demonstrating significant advantages over existing methods like VACE and SkyReels-A2 [21][29] Technical Highlights - FFGo enables the generation of videos with multiple objects while maintaining identity consistency and action coherence, outperforming previous models that were limited to fewer objects [22][31] - The method utilizes Few-shot LoRA to activate the model's memory mechanism, allowing it to leverage existing capabilities that were previously unstable and difficult to trigger [30][44] Implications and Future Directions - The research suggests that video models inherently possess the ability to fuse multiple reference objects, but this potential was not effectively utilized until now [39][48] - FFGo represents a paradigm shift in how video generation models can be used, emphasizing smarter usage over brute-force training [52]

Core Viewpoint - The competition in AI video generation is intensifying, with Runway's new model Gen-4.5 surpassing Google's Veo3 in benchmark tests, while domestic competitor Kuaishou's new model Keling O1 has also been launched, marking a significant moment in the industry [3][19]. Group 1: Model Performance - Runway's Gen-4.5 achieved a score of 1247 in the Artificial Analysis benchmark, making it the top model in text-to-video generation, followed closely by Google's Veo3 with a score of 1226 and Kuaishou's Keling 2.5 at 1225 [7][9]. - Gen-4.5 demonstrates advancements in understanding and executing complex sequential instructions, allowing users to specify detailed shot scheduling, scene composition, event timing, and subtle atmospheric changes [9][15]. Group 2: Technical Innovations - The model has made breakthroughs in pre-training data efficiency and post-training techniques, achieving unprecedented physical and visual accuracy in generated videos [9][15]. - Runway claims that objects in the generated videos move with realistic weight and dynamics, and liquid flows according to appropriate physical laws, enhancing the realism of the generated content [15][18]. Group 3: Market Position and Future Outlook - Runway, founded in 2018, has reached a valuation of $3.55 billion, with its first video model Gen-1 launched in February 2023, followed by Gen-2 in July, which integrated text-to-video and image-to-video functionalities [18]. - The competitive landscape is expected to become more challenging for Runway starting in 2024, with Google's Veo series solidifying its leading position and other competitors like Kuaishou and MiniMax gaining traction [19].

Core Insights - The competition in AI video generation has intensified with the recent launch of Runway's Gen-4.5 model, which has surpassed Google's Veo3 in benchmark tests [1][3] - Simultaneously, domestic competitor KuaLing AI announced the release of its new model, KuaLing O1, claiming to be the first unified multimodal video model [1][3] Benchmark Performance - Runway's Gen-4.5 achieved a score of 1247, ranking first in the Artificial Analysis leaderboard, followed closely by Google's Veo3 with a score of 1226 and KuaLing's model at 1225 [3][4] - The leaderboard indicates a tight competition, with only a one-point difference between Veo3 and KuaLing 2.5 [3][4] Model Features and Advancements - Gen-4.5 has made significant advancements in pre-training data efficiency and post-training techniques, excelling in understanding and executing complex sequential instructions [5][7] - The model demonstrates improved capabilities in adhering to precise prompts, realistic physical motion effects, style control, and visual consistency [5][7] Physical Realism and Limitations - Runway claims that Gen-4.5 achieves unprecedented physical and visual accuracy, with objects moving realistically and fluid dynamics rendered appropriately [7][11] - However, the model still faces challenges in causal reasoning and object permanence, with occasional discrepancies in the expected behavior of generated objects [11] Company Background and Market Position - Runway, founded in 2018, has reached a valuation of $3.55 billion as of 2023, showcasing rapid growth in the AI video generation sector [11] - The CEO of Runway highlighted the achievement of surpassing a trillion-dollar company with a team of just 100 people, emphasizing focus and hard work [11] Future Outlook - The AI video generation market is expected to become increasingly competitive, particularly with the anticipated release of Google's next-generation model, Veo4, in 2025 [12] - The sustainability of Gen-4.5's leading position is uncertain, especially with KuaLing O1 entering the market as a strong competitor [12]

Core Insights - The FFGo method revolutionizes the understanding of the first frame in video generation models, identifying it as a "conceptual memory buffer" rather than just a starting point [1][26] - This research highlights that the first frame retains visual elements for subsequent frames, enabling high-quality video customization with minimal data [1][6] Methodology - FFGo does not require structural changes to existing models and can operate effectively with only 20-50 examples, contrasting with traditional methods that need thousands of samples [6][24] - The method leverages Few-shot LoRA to activate the model's memory mechanism, allowing it to recall and integrate multiple reference objects seamlessly [16][22] Experimental Findings - Tests with various video models (Veo3, Sora2, Wan2.2) demonstrate that FFGo significantly outperforms existing methods in multi-object scenarios, maintaining object identity and scene consistency [4][17] - The research indicates that the true mixing of content begins after the fifth frame, suggesting that the first four frames can be discarded [16] Applications - FFGo has broad applications across multiple fields, including robot manipulation, driving simulation, aerial and underwater simulations, product showcases, and film production [12][24] - Users can provide a single first frame with multiple objects and a text prompt, allowing FFGo to generate coherent interactive videos with high fidelity [9][24] Conclusion - The study emphasizes that the potential of video generation models has been underutilized, and FFGo provides a framework for effectively harnessing this potential without extensive retraining [23][24] - By treating the first frame as a conceptual memory, FFGo opens new avenues for video generation, making it a significant breakthrough in the industry [24][26]