机构：中信建投证券 研究员：于伯韬 核心观点 2 月4 日，可灵更新3.0 版本，视频生成能力近翻倍，核心升级包括15s 超长生成、分镜能力、语言能 力、升级版的视频3.0 Omni，在生成时长、多模态输入、细节控制等核心指标上全面突破。 目前看快手AI 战略已进入正轨，并开始显现"飞轮效应"。可灵模型&产品力居全球第一梯队，商业化快 速起量，两年时间内已完成从"技术验证"到"商业模式跑通"的关键跨越，作为第二增长曲线弹性较大。 除可灵贡献新增长曲线外，AI 全链路赋能快手原有业务形态，提升效率。快手构建了一个以 OneRec （智能推荐）、OneSearch（生成式搜索）和 G4RL（生成式强化学习出价）为代表的AI 应用模型层， 可灵商业化快速起量，两年时间内已完成从"技术验证"到"商业模式跑通"的关键跨越。2025 年全年收 入预计约为1.4 亿美元，2025 年12 月单月收入超过2000 万美元（对应ARR 达2.4 亿美元）。运营侧， 截止25 年底，可灵AI 已在全球拥有6000 万创作者，累计生成超6 亿个视频，累计合作超3 万家企业用 户，2026 年1 月MAU 已超过1200 万。收入结 ...

Core Insights - The emergence of Seedance 2.0 marks a significant transformation in AI video generation, enabling creators to produce high-quality content with minimal technical expertise [3][10][12] - The technology has sparked a surge in the stock market for media and film companies, with notable stocks like Jiecheng Co. and Light Media hitting their daily limits [3][10] - Concerns regarding the implications of realistic AI-generated videos on intellectual property and trust in media have been raised by industry experts [3][13] Group 1: Technology and Performance - Seedance 2.0 and Kuaishou Keling 3.0 have evolved into advanced video generation models capable of understanding cinematic techniques, such as camera movement and editing [4][9] - In practical tests, Seedance 2.0 demonstrated superior performance in generating videos with a cinematic feel, outperforming Kuaishou Keling 3.0 in aspects like fluidity and detail [5][6] - Both models have shown improvements in stability and visual quality compared to their previous versions, but differences remain in semantic understanding and creative imagination [4][5] Group 2: Market Impact and Industry Response - The release of Seedance 2.0 has led to a notable increase in stock prices within the media sector, indicating strong market confidence in AI-generated content [10][12] - Industry professionals have begun to recognize the potential of Seedance 2.0 for commercial applications, particularly in short content and advertising, although concerns about its impact on traditional filmmaking persist [10][12][14] - The technology is seen as a means to lower the barriers to entry for video creation, potentially reshaping the creative landscape by diminishing the reliance on individual expertise [12][14] Group 3: Ethical and Regulatory Concerns - The rapid adoption of AI video generation technology has raised alarms about the proliferation of deepfakes and the associated trust crisis in media [13][18] - Industry insiders emphasize the need for clear regulations and standards to address issues related to content authenticity and intellectual property rights [13][18] - The potential for misuse of AI-generated content has prompted companies to implement safety measures, such as requiring user consent for using personal images in video generation [13][18]

当一项技术让见惯大场面的创作者连呼"恐怖"，当"人人可做AI导演"从畅想照进现实，视频生成的变革浪潮已呼啸而至。 字节跳动旗下视频生成模型Seedance 2.0自2月7日低调亮相，旋即在海内外创作圈与资本市场掀起双重海啸。2月10日，A股影视板块应声沸腾：其中，捷成 股份20CM涨停，光线传媒（核心股）20CM涨停，文投控股、欢瑞世纪、上海电影、横店影视、博纳影业10CM涨停。 行业顶端的创作者们感受最为直接与震撼。游戏科学创始人、《黑神话：悟空》制作人冯骥深度体验Seedance 2.0后，在微博上毫不吝啬地给出"地表最强， 没有之一"的评价，并断言"AIGC的童年时代，结束了"。 然而，在这场技术狂欢中，冯骥也发出了冷静的预警："逼真的假视频将变得毫无门槛，而现有的知识产权与审查体系会面临空前冲击。"这种担忧并非孤 例，知名科技博主"影视飓风"创始人Tim（潘天鸿）在测评视频中，先后六次用"恐怖"来形容Seedance 2.0的能力与影响。 为亲身验证这股技术浪潮的边界与冲击，2月10日，《每日经济新闻》记者分别拿到了Seedance 2.0和同样于近期（2月4日）上线的快手可灵AI 3.0内测名 额， ...

版面之外，才是真相。 本文来自微信公众号： 版面之外 ，作者：画画，头图来自：视觉中国 不知道你的朋友圈是否和我一样，被一些 AI 生成的视频刷屏了。 以下文章来源于版面之外 ，作者画画 版面之外 . 画面里，镜头自然地从全景推向特写，光影流转，动作顺滑得不像话。最神的是，场景竟然能根据旁 白的节奏精准切换。这不是好莱坞大片的花絮，而是字节跳动新发布的视频模型 Seedance 2.0。 说实话，过去我们让AI 画个艺术图、搞个几秒钟的动作视频，早就见怪不怪了。但这次，连平时从 不关注科技的朋友都在转发。那种惊奇、兴奋，甚至带点陌生感的情绪，混合在一起。 仔细想想，这种情绪很关键。以前我们看到OpenAI的Sora、谷歌的Gemini出彩时，感觉更多是仰 望，那是别人家孩子又考了满分。但这次，感觉变了，有一种自己孩子出息了的感觉，让人忍不住想 比较一下：这个好像不比那些眼球顶级的差啊？ 这个瞬间让人恍惚。还记得 2024 年初 Sora 横空出世吗？东京街头那条漫步视频，逼真得让人惊到 了下巴。当时国内科技圈讨论的全是差距，是又要落后一个时代的焦虑。 到了2026年初，Seedance 2.0的刷屏，把那种沉 ...

近日，字节跳动推出的Seedance 2.0视频生成模型在全球范围内引发广泛关注。这一模型采用双分支扩散变换器架构，实现了原生音视频同 步生成，用户只需编写详细提示或上传一张图片，即可在60秒内生成带有原生音频的多镜头序列视频。 与传统AI视频模型相比，Seedance 2.0在三大关键能力上实现突破：多镜头叙事一致性、精准指令可控性以及物理规律模拟能力。其独特的 多镜头叙事功能能够根据单个提示自动生成多个相互关联的场景，并保持角色、视觉风格和氛围的一致性。更令人惊叹的是，该模型展现 出一定的世界理解能力，例如仅凭建筑物正面照片即可生成其背面视角的合理画面。那么Seedance 2.0的横空出世对行业有何影响？对AI投 资又有何启示？ 01 行业影响：内容生产关系的重构 首先，最直接的是创作门槛的降低与产能提升。Seedance 2.0将视频制作的技术门槛从"专业工具操作能力"转移至"创意策划能力"。根据媒 体公开报道，传统需要5人团队3天完成的工作，现在单人30分钟即可通过文本结合参考素材生成成片，内容产能从每周3-5条飙升至每天20- 30条。这种效率提升将推动UGC（用户生成内容）向"UAC（AI辅助创 ...

在科技与资本交织的浪潮中，总有一些时刻被定义为"奇点"。 如果说去年Deepseek的横空出世标志着文本大模型的元年，那么2026年的初春，随着字节跳动旗下即 梦AI发布Seedance 2.0模型，视频生成领域正式迎来了属于它的"工业革命"。 酝酿已久的AI应用行情终于出现了。 作者 | 紫枫 编辑 | 小白 更为震撼的是模型对物理世界的深刻理解。但Seedance 2.0生成的视频中，人物的运动符合重力与惯 性，光影随着物体的移动自然流转，甚至连衣服褶皱的动态变化都逼近实拍，以往模型中常见的"AI 抽搐"不再存在。 此外，模型还展现了原生音画同步的能力，无需任何音频输入，仅凭画面内容就能生成高度匹配的音 效和人声，这种多模态的深度融合让视频的完成度瞬间提升到了交付级标准。 然而，技术的双刃剑效应也随之显现。实测显示，该模型仅凭一张正面照片，就能通过算法"脑补"出 建筑背后的真实构造；仅靠人脸照片，就能生成与其本音高度相似的声音。这种能力意味着，一旦个 人影像数据进入训练集，模型可以完整复刻一个人的形态与声音，连至亲之人都难辨真假。 在巨大的舆论压力与安全隐患下，豆包和即梦平台紧急全面封禁了真人人脸的上传功 ...

Market Performance - The Hang Seng Index closed up 0.58%, the Hang Seng Tech Index rose 0.62%, and the Hang Seng China Enterprises Index increased by 0.81% [1] - The optical communication sector led the gains, with Longi Fiber Optics rising by 8.8%, Dongfang Electric up over 7%, and Weichai Power increasing by 3.69% [1] Sector Highlights - The cultural media sector saw significant gains, with Digital Kingdom surging over 29%, Reading Group rising over 15%, and DaMai Entertainment and China Ruyi both increasing by over 5% [1] - Biopharmaceutical stocks strengthened, with WuXi AppTec rising over 4%, Rongchang Bio up 3.64%, and Kailai Ying increasing by 3.5% [2] Individual Stock Movements - Notable individual stock movements included Digital Kingdom with a rise of 29.31%, Mao Yi Chuang up 21.92%, and Wan Ka Yi Lian increasing by 20% [2] - The education sector experienced declines, with Dongfang Zhenxuan down over 9% and New Oriental-S falling over 4% [4] New Listings - Le Xin Outdoor debuted with a remarkable increase of 102.29%, closing at 24.78 HKD, and at one point, it surged over 130% during trading [4] AI and Technology Developments - The stock of Zhipu AI, referred to as the "first global large model stock," saw a peak increase of over 24%, with a two-day cumulative rise exceeding 56% [5] - The release of the AI video generation model Seedance 2.0 by ByteDance is expected to significantly reduce production costs and enhance efficiency in the AI drama and short drama sectors, potentially accelerating industry growth [7]

Core Viewpoint - ByteDance's AI video generation model Seedance 2.0 has announced a temporary halt on supporting the input of real human materials as reference content due to concerns raised by a tech blogger regarding its ability to generate videos using real voices without consent [1]. Group 1 - Seedance 2.0 received significantly more attention than expected during its internal testing phase, prompting the company to make urgent optimizations based on user feedback [1]. - The platform emphasizes the importance of respecting creative boundaries and aims to present a more refined product after adjustments [1]. - As of the latest update, ByteDance has not provided an official response to the concerns raised [2].

Group 1 - The A-share media sector experienced significant gains, with multiple stocks such as Light Media, Wanda Film, and others hitting the daily limit up [1] - The "longest Spring Festival holiday" in history has officially started its online pre-sale, attracting considerable attention [1] - The AI video generation model Seedance 2.0 has gained immense popularity across the internet [1]

第一财经记者了解到，在即梦创作者社群中，有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。截至发稿，字节官方尚未予以回应。 近日，因科技博主Tim反映字节跳动旗下AI视频生成模型Seedance2.0大量训练"影视飓风"相关视频，可 以在无提示信息的情况下，自动生成带有Tim声音的视频，Seedance2.0官方宣布暂不支持输入真人素材 作为主体参考的功能。 （文章来源：第一财经） ...