初创公司尽调
尽调报告 Generative AI / Video Generation Series B 2026-08-21

Shengshu Technology

可信的中国 AI 视频挑战者,势头强劲,但独角兽定价不透明

Shengshu 是更可信的中国 AI 视频创业公司之一,有强产品动能和战略背书,但 ARR、毛利率和股权结构披露缺失,在不透明独角兽定价下仍应观察而非买入。

封面要素

最近一轮融资 01
RMB 2B Series B [CO016]
公开估值区间 02
$1.0B-$2.5B [CO019, CO020]
已披露 2026 融资额 03
RMB 2.6B+ [CI026, CV003]
开发者 + 企业客户 05
10,000+ [CO023, CU004]

公司概况

Shengshu Technology 是一家在 Beijing 创立的中国 AI 视频初创公司,成立于 2023-03-06,创始团队带有清华背景。公开领导层叙事主要围绕 Tang Jiayi 和 Zhu Jun,Luo Yihang 与 Bao Fan 后来出现在运营角色中。旗舰平台 Vidu 覆盖文生视频、图生视频、参考图生视频、开发者 API,以及 Vidu Agent 和实时 Vidu S1 数字人体验等新工作流产品。到 2026 年,公司已披露的 2026 年融资超过 RMB 2.6 billion,并被 Dealroom 公开列为独角兽;但 CNBC 报道称,April 2026 Alibaba 领投轮的准确投后估值没有披露。

官网
www.shengshu-ai.com
成立时间
2023-03-06
创始人
Tang Jiayi, Zhu Jun
创立地点
Beijing, China
总部
Beijing, China
产品
Vidu 是 Shengshu 的旗舰 AI 视频平台,覆盖文生视频、图生视频、参考图生视频、开发者 API、创作者工具,以及 Vidu Agent 和实时 Vidu S1 体验等工作流产品。
客户
目标客户包括创作者、开发者、企业营销团队、媒体与娱乐用户,以及寻求 AI 视频生成和自动化的电商内容运营团队。
商业模式
创作者与企业混合模式:一边靠点数或订阅变现,一边提供按量计费的 API 访问,并销售价值更高的工作流或广告导向商业套餐。
阶段
Series B
融资情况
Series B 轮于 April 2026 完成,规模约 RMB 2 billion;此前 February 2026 完成 >RMB 600 million 的 Series A+。公开资料称 Shengshu 为独角兽,但准确投后估值仍未披露。
[CO001, CO003, CO004, CO005, CO006, CO011, CO013, CO015]

执行摘要

主要优势

  • Vidu 在独立 AI 视频基准和持续发布节奏上都有可信产品可见度。
  • Shengshu 近期完成多轮大额融资,并吸引 Alibaba Cloud 等战略投资人。
  • 公司公开说法指向可观商业使用、创作者触达和开发者 / 企业采用。
  • 商业模式覆盖创作者工具、API 访问,以及 Vidu Agent 等工作流产品。
  • 相比 Pika 等规模更小的 AI 视频同行,公司资本实力更强,商业野心也更大。

主要风险

  • 2026 年 4 月确切轮后估值、股权结构表和清算优先权结构仍未披露。
  • 未找到公开 ARR、毛利率、烧钱速度、客户集中度或净留存披露。
  • 先进计算出口管制和中国 AI 监管会直接影响成本与风险画像。
  • 公开客户证明有意义,但仍太依赖合作伙伴平台,难以证明收入质量可持续。
  • Runway、Kling、MiniMax、ByteDance 及其他资金充足的 AI 视频玩家会压缩定价和用户注意力。

未决问题

  • 最后一轮确切轮后估值和股权结构表 / 优先权条款尚无公开信息。
  • 分产品线 ARR、毛利率、烧钱速度和算力成本桥仍未披露。
  • 留存、流失、企业 ACV 和头部客户集中度公开渠道看不到。
  • 算力来源韧性和监管合规落地需要管理层级尽调,不能只靠公开信息推断。
  • 做出实时投资决定前,仍需重新拉齐直接可比公司的收入和估值数据。

目录

Chapter 01

01公司概况

1.1 身份、总部与产品范围

Shengshu Technology 是一家位于 Beijing 的生成式 AI 公司,成立于 March 6, 2023,注册地址位于 Haidian District Zhongguancun East Road 的 Dongsheng Building。多方资料把公司与清华大学研究根基,以及早于 Vidu 商业化的 U-ViT 架构工作联系起来。公司的核心产品是 Vidu,一个多模态视频生成平台,覆盖文生视频、图生视频和参考图生视频工作流,后来又延伸到 API、智能体和实时数字人产品。公司公开材料将 Shengshu 定位为同时销售 MaaS 和 SaaS,第三方画像则显示收入来自订阅和企业使用的组合。身份问题很关键,因为后续章节需要一个清晰基线:这不只是一个消费者玩具应用,而是一家研究密集型中国视频模型初创公司,有商业化平台、Beijing 法人主体,并且产品家族已经推向全球分发。[CO001, CO002, CO003, CO006, CO007, CO008]

快照 KPI 表
指标数值 / 状态日期 / 期间置信度缺口 / 备注
成立日期2023-03-062023-03-06由 Baiduwiki 公司简介和后续公司自我描述支持
总部 / 注册地址北京市海淀区当前法律实体直接披露了注册地址;北京以外的运营足迹不太确定
现任 CEOLuo Yihang2026公开公司描述与 Dealroom 描述一致,但未审阅任命日期的企业登记摘录
前任 CEO / 现任总裁Jiayu Tang20262024 年末 CNBC 仍将 Tang 列为 CEO;后续资料称其转任总裁
最新披露轮次Alibaba Cloud 领投的 B 轮2026-04CNBC 披露 RMB 2B;同篇文章未披露估值
公开估值信号$1-2.5B 区间;具体投后估值未披露2026Dealroom 区间确认独角兽状态,但不是精确点估计
公开融资下限>$380M 等值2026由已披露的 RMB 600M A+ 轮、RMB 2B B 轮,以及规模仅部分披露的更早轮次推导
采用触达200+ 个国家和地区2025-2026公司通过新闻稿发布的主张;未审阅第三方审计
创作者 / 开发者规模40M 创作者;10K+ 开发者和企业客户2026-01仅来自 Global Creativity Week 发布的公司主张
员工数70+ 名员工,~90% 研发2024-03仅审阅到较早公开披露;当前员工数仍待确认

混合了硬性法律 / 实体事实、第三方数据库区间和公司声称的运营指标。估值、累计融资、当前员工数、创作者 / 开发者规模应在管理层尽调中重新核验。

[CO001, CO002, CO005, CO016, CO019, CO020]
FO002: 公司快照逻辑

Shengshu 的研究根基如何连接 Vidu 产品、商业入口以及更长期的世界模型愿景。

[CO003, CO006, CO011, CO013, CO030, CO031]

1.2 创始人、领导层与组织设置

领导层披露有意义,但并不完全干净。2024 年底的公开英文报道将 Jiayu Tang 描述为 Shengshu 联合创始人兼 CEO;2026 年 CNBC 报道引用 Zhu Jun,称其为创始人,公司声明则称其为创始人兼首席科学家。Dealroom 的梳理最能对齐各方说法:Tang 是联合创始人兼前 CEO,Bao Fan 是 CTO,Zhu Jun 是与清华深度绑定的首席科学家,Luo Yihang 则是从 ByteDance 旗下 Volcano Engine 引入的高管,担任 CEO,负责研究、产品和商业化,Tang 转任总裁。领导层交接说明,Shengshu 在 2025-2026 期间从创始人主导的技术孵化,转向更规模化的运营结构。缺口仍在于:公开资料没有董事会构成、正式治理结构或控制权披露,关键人物集中度和治理尽调仍然悬而未决。[CO003, CO004, CO005, CO027, CO036, CO041]

领导层与创始人表
人物职务背景创始人-市场匹配 / 覆盖关键人依赖
Zhu Jun创始人兼首席科学家清华教授、高级 AI 研究者,与 U-ViT 和多模态模型研究相关支撑 Shengshu 的研究可信度和高校人才管线高——核心科学权威,也是世界模型叙事负责人
Jiayu Tang联合创始人;前 CEO;总裁清华计算机科学毕业生;在 2024 年 CNBC 报道中主导产品商业化叙事连接研究、商业化和投资人叙事高——即使头衔变化,公开创始人身份仍与他绑定
Luo YihangCEO前 ByteDance Volcano Engine 高管、清华校友,被引入负责 R&D、产品和商业化补上规模化互联网运营经验和企业级执行肌肉中高——对市场进入和规模化重要,但不是原始研究锚点
Bao FanCTO / 法定代表人出现在 arXiv 论文和公司记录中;专长是扩散 / 视频模型执行负责技术落地,并把架构转译成可交付产品高——核心技术执行看起来较集中

列举范围仅限公开可见的顶层运营团队。已审阅来源中没有找到公开董事会名单、独立董事或更广泛高管梯队。

[CO003, CO004, CO005, CO036, CO041]

1.3 融资、资本结构与利益相关方

Shengshu 自 2023 年以来频繁融资,但公开披露质量参差不齐。最硬的披露是 February 2026 超过 RMB 600 million 的 Series A+,以及 April 2026 由 Alibaba Cloud 领投的 RMB 2 billion Series B;更早的融资历史则要靠 Baiduwiki 和 Dealroom 重建。这些资料指向 June 2023 近 RMB 100 million 的天使轮、August 2023 的天使+轮、March 和 June 2024 的数亿元人民币轮次,以及 September 2025 的数亿元人民币 Series A。Dealroom 将 Shengshu 标为独角兽,并给出 $1-2.5 billion 的宽估值区间;但 CNBC 明确指出,公司拒绝披露 April 2026 轮的估值。实际尽调结论是,Shengshu 明显已经达到风险投资规模,也足够具有战略重要性,能吸引 Alibaba Cloud、Qiming、Baidu 系投资方、北京 AI 基金和多家产业伙伴;但投资人面对的准确累计募资、稀释、清算优先权、当前所有权控制仍然相当不透明。[CO015, CO016, CO017, CO018, CO019, CO020]

利益相关方与投资者图谱
利益相关方角色控制 / 经济重要性证据尽调问题
Alibaba CloudB 轮领投方(2026)战略云、算力和分发相关性;可能对下一阶段有重大影响CNBC 2026 年 4 月融资报道获取持股比例、董事会权利和商业绑定
Qiming Venture Partners多轮财务投资方显示早期多轮获得机构 VC 支持Qiming 投资组合 + Baiduwiki / Dealroom 融资历史确认确切进入轮次、后续出资策略和当前持股
Baidu Ventures / Baidu 相关资本种子轮及后续投资方组合重要战略 AI 生态赞助方,也是中国 AI 信号CNBC 2024、Baiduwiki、Dealroom厘清是否存在战略权利或商业整合
Ant Group最早战略支持方帮助孵化早期公司成立和种子融资故事Baiduwiki 和 CNBC 2024核实 Ant 是否仍活跃或已被稀释
北京 AI 产业投资基金 / 中关村科学城国资相关资本支持政策对齐和北京本地生态支持Baiduwiki 和 A+ 轮新闻稿了解是否有政策义务或报告条件
LINK-X / Xinglian CapitalA+ 轮联合领投方Alibaba 领投规模化融资前,2026 年过桥轮的财务赞助方A+ 轮新闻稿和 Baiduwiki确认确切出资金额和治理权利
Wondershare / Visual China / TORS战略产业投资者潜在下游软件、媒体和版权工作流相关性A+ 轮新闻稿评估商业试点是否变成经常性收入
Huawei Hubble2024 年工商变更披露的股东显示中国硬件 / AI 栈中更广泛的战略兴趣Baiduwiki 公司简介确认当前持股以及是否有算力或生态合作

这是公开信息的部分图谱,不是完整股权结构表。它只总结已审阅来源明确点名的投资者和战略利益相关方。

[CO015, CO016, CO017, CO018, CO020]
FO003: 快照 KPI

截至 2026 年 8 月研究日,Shengshu 的关键公开运营和融资指标显示:产品扩张很快,但披露缺口仍大。

[CO019, CO021, CO023, CO029, CO032]

1.4 里程碑、商业化与认可

Shengshu 最突出的强项是产品节奏。公司 April 2024 发布 Vidu,July 2024 全球上线,随后在 November 2024 推出 Vidu 1.5、February 2025 推出企业 API、January 2025 推出 Vidu 2.0,2025-2026 持续升级偏重参考一致性的 Q 系列,December 2025 推出 TurboDiffusion 和一键式 Vidu Agent,July 2026 推出实时 Vidu S1 模型。公司声明把这条路线图与商业化提速相连:据称 Vidu 覆盖超过 200 个国家和地区,mid-2026 已服务超过 40 million 创作者、超过 10,000 名开发者或企业客户,具名商业用户包括 ByteDance、Samsung、TAL、Alipay、JD.com、Amazon、L'Oréal、Tencent Animation、iQIYI 和 Mango TV。外部验证存在,但并不单向。Shengshu 入选 World Economic Forum Technology Pioneer,并多次被基准测试引用,说明品类地位成立;但当前 Artificial Analysis 快照把 Vidu Q3 Pro 放在上层,却还不是全球顶端。因此,商业化叙事可信,但仍然高度依赖公司自称,而不是经审计的使用量或收入披露。[CO008, CO009, CO010, CO011, CO012, CO013]

里程碑表
日期事件类型金额 / 估值 / 状态参与方含义
2022-09提出 U-ViT 架构产品研究里程碑Tsinghua / Shengshu 研究人员后续 Vidu 主张的商业化前技术基础
2023-03Beijing Shengshu Technology 注册成立并开源 UniDiffuser创立公司成立创始团队 / Tsinghua 相关研究人员公司正式起步,并留下开放研究足迹
2023-06天使轮融资近 RMB 100MAnt Group、Baidu Ventures、Zhuoyuan Capital 等投资方验证早期投资人对多模态视频团队的兴趣
2023-08天使+轮融资数千万 RMBJinqiu Fund在 Vidu 发布前延长种子期现金跑道
2024-032024 年初融资轮融资数亿 RMBQiming 及跟投方产品发布前扩大 R&D
2024-04-27Vidu 与 Tsinghua 共同正式亮相产品1080p / 最高 16s 定位Shengshu + Tsinghua University确立中国最知名早期 Sora 对手地位
2024-06Pre-A 融资和算法备案披露监管数亿 RMB;备案获批 / 通过Beijing AI fund、Baidu 等增加政策合法性和更多资本
2024-07-30Vidu 全球发布产品公开全球可用Shengshu / Vidu开始英语和跨境分发
2024-11-13Vidu 1.5 发布产品多实体一致性Shengshu / Vidu提升可控性和商业叙事
2025-01-15Vidu 2.0 发布产品10 秒内生成,成本更低Shengshu / Vidu强化速度和可负担性定位
2025-02-13Vidu API 发布产品开发者 $10 起购Shengshu / 开发者打开直接 B2B 和平台渠道
2025-06-24宣布入选 WEF Technology Pioneer规模外部认可World Economic Forum / Shengshu借非中国机构增强全球信号
2025-12发布 TurboDiffusion 和 Vidu Agent产品100-200x 加速;一键生成 15-30s 视频Shengshu + Tsinghua从模型质量推进到工作流和成本效率
2026-02-05A+ 轮融资融资>RMB 600MZhongguancun Science City、LINK-X、战略投资者较大 Alibaba 领投融资前的过桥轮
2026-04-10Alibaba Cloud 领投 B 轮融资RMB 2B;估值未披露Alibaba Cloud、TAL、Baidu Ventures 等战略投资方确认国家级战略支持
2026-07-03Vidu S1 发布产品实时 540P / 25 FPS 互动视频Shengshu / Vidu将路线图从短片生成延伸到实时虚拟人

这是已审阅公开来源的记录时间线。2024-2025 年较早轮次通常只披露“数亿 RMB”,因此精确累计融资仍是估计值,而非经审计事实。

[CO001, CO007, CO008, CO009, CO011, CO013]
FO001: Shengshu / Vidu 里程碑时间线

这条时间线梳理 Shengshu 如何从 Tsinghua 相关研究团队,演进为 Alibaba Cloud 支持、面向全球分发的 AI 视频平台。

[CO001, CO007, CO008, CO011, CO014, CO015]

1.5 负面信号与未解决披露缺口

公司概况层面的主要风险不是生死级红旗,而是披露和执行上的保留项。Notebookcheck 在 September 2025 对 Vidu 的上手测试认为,产品能生成醒目的视觉效果,但故障、瑕疵和不一致仍太多,难以支撑可靠的专业工作,凸显基准测试营销与生产可靠性之间长期存在的缺口。公开基准排名也对时间很敏感:公司 PR 在 early 2026 将 Vidu Q3 包装为中国 No.1、全球 No.2,但 August 2026 的 Artificial Analysis 快照显示,文生视频和图生视频排名都更低。治理披露更薄:已审阅的公开资料没有披露董事会、独立监督、所有权权利或经审计财务报表;连当前员工数、准确累计融资额这样的基础指标也只能部分观察。招聘门户确实显示公司在 Beijing、Shanghai、Shenzhen 和 San Francisco 招人,但不能把这误认为除 Beijing 之外的运营总部已有完整文件支持。投资人应把 Shengshu 视为一家高速推进、但仍不透明的私营公司;它最强的事实是技术交付速度,而不是机构透明度。[CO020, CO029, CO033, CO034, CO035, CO036]

1.6 图表

Chapter 02

02市场分析

2.1 市场边界与范围

分析 Shengshu 时,应把它放在狭义 AI 视频生成器市场,而不是所有视频软件市场。The Business Research Company 和 Research and Markets 定义的狭义品类,是用 AI 根据提示词、图像、幻灯片或配套结构化输入生成或转换视频的软件和服务。这个品类已经覆盖文生视频、图生视频、剪辑自动化,以及从自助订阅到 API 和企业服务的交付模式。Research and Markets 相邻的「生成式 AI 视频创作」视角一方面略窄,一方面又略宽:它强调创作工作流和云端、本地部署等部署类型,但仍覆盖从企业到个人创作者和媒体公司的终端用户。官方市场界面证实,商业品类已经走过单一演示模型阶段。Runway 销售创意 SaaS、开发者工具和机器人 / 仿真界面;Kling 提供 API、4K 生成和移动端分发;Pika 强调智能体和工作流自动化;PixVerse 主推 CLI、智能体、营销中心和 API 工作流;Jimeng 则优化中文提示词、社区二创和帧控制功能。因此,Shengshu 的实际纳入规则是:计算 AI 原生视频生成、活动创作、创作者工具和开发者 / 企业视频 API 的支出;剔除传统 NLE 套件、监控分析、CDN / 流媒体基础设施,以及从未触达视频模型的普通社交媒体广告支出。[CM001, CM002, CM003, CM004, CM020, CM021]

市场定义表
细分 / 类别纳入支出排除支出买方 / 付款方Shengshu 相关性
AI 视频生成器软件文本到视频、图像到视频、幻灯片 / 文档到视频、生成服务、编辑自动化、API 访问流媒体 / CDN 基础设施、监控分析、通用广告购买营销团队、创作者、开发者、企业Vidu 和 Vidu API 的核心市场
视频创作中的生成式 AI云端或本地创作工具、合成媒体工作流、协同制作不触及 AI 工具的非生成式编辑套件和传统制作人力大型企业、中小企业、创作者、媒体团队创作工作流的有用相邻视角
创作者生产力工具移动优先特效、社区混剪、趋势模板、快速短视频生成专业后期制作套件、代理商留用费、社交平台分发费个人创作者和进阶消费者拉新和品牌重要,但货币化深度较低
营销 / 商业视频工作流产品演示、本地化广告、宣传短视频、活动迭代、目录视频与生成工具无关的全漏斗媒体支出和代理商服务CMO、增长团队、电商运营方近期最可能变现的用例
开发者 / API 视频基础设施API 点数、编排、CLI 批处理、工作流集成、应用嵌入与视频平台无关的通用云计算支出开发者、产品团队、平台建设者支撑企业和平台分发
现状替代方案自有工作室、自由职业者、素材视频流程、人工剪辑、PowerPoint 和旁白不属于软件 TAM,但会替代支出同一批终端客户,不同预算科目战略替代池,而非直接市场收入

最清晰的边界是以生成为核心的视频软件和服务。更宽泛的 AI 视频或创意软件 TAM 只能作背景,不应不经调整就当作 Shengshu 的直接 SAM。

[CM001, CM002, CM003, CM004, CM021, CM028]

2.2 市场规模与地理形态

只有分析师采用相近边界时,公开市场估算才会聚在相对窄的区间。The Business Research Company 估算 AI 视频生成器市场在 2025 为 $0.85 billion、2026 为 $1.04 billion、2030 为 $2.07 billion,意味着 2026 到 2030 的 CAGR 为 18.9%。它相邻的生成式 AI 视频创作视角更小:2025 为 $0.39 billion、2026 为 $0.47 billion、2030 为 $0.98 billion,CAGR 为 20.4%。Fortune Business Insights 在当前基年落在两种视角之间:2025 为 $716.8 million、2026 为 $847 million,并预测到 2034 达到 $3.35 billion,CAGR 为 18.8%。分歧不小,但可以解释:有些出版方只计算纯生成器软件,有些纳入更多创作服务;部分区域拆分也会随分析、企业工作流服务或消费者应用是否纳入边界而大幅移动。不过,分部数据仍有方向性价值。Fortune 报告称,文生视频占 2026 市场的 46.25%,营销和广告是最大应用,占 33.88%;社交媒体是增长最快的应用,CAGR 为 23.5%;大型企业是最大客户类别,占 50.86%,SME 增速最快,CAGR 为 21.1%。地理口径更不一致:TBRC 称 Asia-Pacific 是 2025 AI 视频生成器工具最大区域;Fortune 则给 North America 2025 年 41.0% 份额、Asia-Pacific 20.9%,并称 China 2026 为 $49 million。承销 Shengshu 时,这种不一致是在提醒:自上而下的 China TAM 只能当方向性判断,不能当精确数字。[CM005, CM006, CM007, CM008, CM009, CM010]

TAM / SAM / SOM 规模测算视角表
发布方 / 视角基准年份地域市场类别数值复合年增长率(CAGR)方法 / 换算置信度局限
The Business Research Company2025全球AI 视频生成器市场$0.85B18.9% (2026-2030)发布方对狭义生成器市场的估计包含解决方案和服务;定义与其他发布方不同
The Business Research Company2026全球AI 视频生成器市场$1.04B18.9% (2026-2030)发布方对当年市场规模的估计生成器类别内仍偏宽;没有中国单独拆分
The Business Research Company2025全球生成式 AI 视频创作市场$0.39B20.4% (2026-2030)发布方对相邻、以创作为核心视角的估计小于生成器市场,口径也不同
Fortune Business Insights2025全球AI 视频生成器市场$716.8M18.8% (2026-2034)带应用和区域拆分的发布方估计预测期更长,增加不确定性
Fortune Business Insights2026全球AI 视频生成器市场$847M18.8% (2026-2034)用作当年 TAM 锚点的发布方估计单一发布方;非中国专属
由 Fortune Business Insights 推导2026全球文本生成视频子细分~$392Mn/a46.25% 文本生成视频占比 × $847M 总市场假设细分占比在所有地域和厂商间都一致
由 Fortune Business Insights 推导2026全球营销与广告子细分~$287Mn/a33.88% 应用占比 × $847M 总市场应用组合可能因区域和产品类别而异
由 Fortune Business Insights 推导2026归属亚太文本生成视频部分~$82Mn/a20.9% 亚太占比 × ~$392M 文本生成视频部分区域占比和文本生成视频占比来自同一报告的不同切面
Fortune Business Insights2026中国AI 视频生成器市场$49Mn/a发布方的区域国家估计单一来源的中国数据点;片段中未完全披露国家口径方法

当前公开规模数据足以框定市场区间,但不足以为 Shengshu 证明精确的中国 SAM。推导行只是对 Fortune Business Insights 已发布百分比做算术换算,应作为工作估计,而不是经审计的细分收入。

[CM005, CM006, CM007, CM008, CM010, CM011]
FM001: Shengshu 市场规模视角:TAM / SAM / SOM(2026 工作视图)

基于 Fortune Business Insights 数据搭建三层工作视角,隔离出与 Shengshu 当前品类最相关的文本到视频市场和亚太归因切片。

这是一个受约束的工作视角,不是 Shengshu 实际已入账机会。底层是区域代理,并非已证明可获取份额。中国特定的企业和 API 变现没有公开披露。

[CM010, CM011, CM036, CM037]
FM002: 相邻 AI 视频口径下的当前收入视角

已发布和推导出的当前市场估计显示,评估 Shengshu 应采用区间,而不是单一头条 TAM。

中点使用公开值或直接算术转换。低 / 高值是在这些估计周围设置的括号区间,用来呈现口径不确定性,而不是单独经审计的发布方数字。

[CM005, CM006, CM007, CM008, CM011, CM012]

2.3 买方、用户与付款方架构

市场不是一个单一创作者池。它至少拆成六个商业上不同的分部,各自有不同预算负责人和采用触发点。第一,营销团队和代理公司购买速度、迭代量和活动创意的成本压缩;分析师数据表明营销和广告已经是最大应用,因此这是近期最重要的分部。第二,零售和电商运营方需要动态产品展示、本地化短视频和目录级素材生产。第三,媒体、娱乐和工作室用户看重更高端的可控性、一致性和生产工作流。第四,个人创作者和准专业消费者以免费增值或点数层进入,对 UX、趋势工具和移动端分发高度敏感。第五,开发者和工作流构建者买的是 API、CLI 工具和编排原语,而不是面向消费者的订阅。第六,企业创新、教育和内部内容团队用 AI 视频做培训、产品说明和内部沟通。官方竞品界面让这种分层不需要猜测就能看见:Runway 区分创意、开发和机器人产品线;Kling 把消费者创作与 API 和移动端打包;Pika 主推智能体和应用;PixVerse 暴露 CLI、智能体、营销中心和 API;Jimeng 优化中文提示和社区二创循环。因此,Vidu 同时在 B2C 和 B2B 走廊竞争,但经济上重要的买方大概率是企业营销人员、开发者和媒体团队,而不是随手试用的免费创作者。[CM015, CM016, CM017, CM018, CM019, CM020]

细分市场 / 买方图谱
细分市场买方用户付费方 / 预算负责人工作流采用触发点Shengshu 相关性
营销团队和代理商CMO、创意总监、增长负责人设计师、活动经理、效果营销团队营销预算广告快速迭代、本地化、A/B 创意生成生产成本更低、周转更快近期价值最高的 B2B 通道
电商卖家和品牌GM、电商运营负责人、平台团队商品运营和内容团队电商 / 增长预算产品演示、商品目录短视频、规模化促销视频需要大批量产品叙事与 Vidu 智能体 / API 工作流高度契合
媒体、工作室和娱乐工作室负责人、制片人、创新负责人剪辑师、艺术家、后期团队制作 / 内容预算预演、场景生成、特效、品牌叙事控制力和一致性提升战略重要,但采购更慢
个人创作者 / 专业消费者创作者本人创作者本人个人订阅或点数钱包短视频内容、追热点、实验低摩擦 UX、特效、移动端访问漏斗顶部大,但变现深度较低
开发者和平台CTO、产品负责人、开发者工程师和自动化构建者产品 / 基础设施预算将生成嵌入应用、流水线或内部工具API 可用性和可靠性对持久 B2B 收入很重要
企业培训 / 内部内容团队L&D 负责人、产品营销负责人、运营培训师、赋能员工、内部沟通团队HR、赋能或运营预算培训、入职、讲解视频、内部活动不靠工作室也能规模化做多媒体次级但可信的扩展通道

买方类型来自公开产品包装和已发布用例细分推断。确切 ACV 和采购权限门槛仍未公开。

[CM015, CM016, CM017, CM018, CM019, CM020]
FM003: 买方细分重点与打包方式图

按预算负责人、核心价值驱动、产品入口、变现模式和当前细分信号,映射主要客户原型。

[CM012, CM013, CM014, CM020, CM021, CM027]
FM004: AI 视频采用路径:从试用到合规部署

这个品类越来越靠把用户从新奇生成拉进集成化、合规的工作流来变现。

[CM004, CM020, CM021, CM027, CM029, CM032]

2.4 需求驱动因素与工作流迁移

需求同时被媒体消费量和工作流成熟度拉动。Pew 发现,美国成年人中 83% 使用 YouTube、68% 使用 Facebook、47% 使用 Instagram、33% 使用 TikTok;YouTube 自己的 2024 U.S. impact report 称,平台创作者生态为 U.S. GDP 贡献 $55 billion,并且 YouTube 在 2021 到 2023 之间向创作者、艺术家和媒体公司支付超过 $70 billion。这个宏观背景解释了为什么 AI 视频工具不需要先替代电影;它们只要降低持续短视频、活动、教育和产品视频生产的成本即可。第二个需求驱动是能力成熟。a16z 在 March 2025 写道,过去六个月视频质量和可控性大幅进步,中国模型 Hailuo 和 Kling 到 January 2025 已在月度网站访问量上超过 Sora,供应商差异化正在围绕提示词遵循、唇形同步和镜头控制出现。官方产品页也强化了这一点:厂商正在交付参考控制、智能体剪辑、语音 / 唇形同步、现成广告模板、CLI 批量执行和 API 优先编排等工作流功能。这个演进把品类从新奇玩具推向工作流软件。约束仍然重要。OpenAI 关闭 Sora 消费者产品并在 2026 计划停用 API,说明平台可用性和商业化路径会快速变化;即便是前沿实验室,市场领导地位也不是永久的。[CM023, CM024, CM025, CM026, CM027, CM028]

增长驱动因素与约束表
驱动因素 / 约束方向时间影响尽调问题
社交 / 视频消费仍然庞大正向当前支撑低成本视频制作工具的结构性需求衡量 Shengshu 使用中,复购型创作者或活动产出与实验性使用各占多少
创作者经济分成池继续扩大正向当前下游能变现时,更多创作者和小企业有理由为工具付费要求按创作者细分查看队列转化和留存
质量和控制力在 2024 年末 / 2025 年初快速提升正向近期企业试点更可信,也降低新奇感折价按用例对比 Vidu 相对 Kling、Runway 和 Hailuo 的胜率
工作流打包(API、智能体、CLI、模板)越来越多正向当前平台可从新奇生成转向粘性工作流软件量化 API 或团队套餐相对单用户套餐的收入占比
中国生成式 AI 合规义务负向当前抬高公开部署的审核、备案、日志和标识成本核验 Shengshu 备案、安全流程和标识落地
中国 2025 年 AI 标识规则负向当前公开分发合成媒体需要可见和技术披露控制审计水印、元数据留存和客户合规工具
美国先进计算出口管制负向当前可能限制中国厂商获得领先芯片,并增加算力成本波动审查 GPU 来源、云依赖和应急规划
市场数据不透明、区域估计相互冲突负向当前自上而下的 TAM 可能夸大确定性并错估市场份额在依赖 TAM 数学前,要求自下而上的收入和客户队列证据

核心市场约束不是纯学术问题,而是运营问题。合规、算力来源和变现深度都会影响技术强的模型能否变成持久生意。

[CM023, CM024, CM025, CM026, CM029, CM030]

2.5 监管、算力与跨境访问

监管对 Shengshu 是一阶市场变量,因为它的本土市场在中国,产品输出又是面向公众的合成媒体。中国《生成式人工智能服务管理暂行办法》于 August 15, 2023 生效,并明确适用于生成文本、图像、音频、视频和配套内容的公共服务。已审阅监管资料对运营义务讲得异常清楚:提供者必须处理合法训练数据和 IP 来源、个人信息保护、内容安全、投诉处理、透明度,以及必要时的备案或安全评估要求。China Law Translate 对该办法的译文在 Article 12 里把视频特定义务说得很直白:提供者应按深度合成规则标注生成图像和视频;Article 14 进一步要求移除违法内容并处置滥用用户。March 2025 中国发布标识规则后,这些义务进一步变硬;规则将于 September 1, 2025 生效,要求可见标识加技术标识,并禁止删除或隐匿这些标识。跨境供给又加了一层约束。BIS 在 May 2026 发布的指引重申,面向中国或 D:5 总部实体的受管先进计算物项仍需 U.S. 出口许可,即便这些实体在中国境外接收物项也是如此。一家中国视频模型公司面对的是直接输入成本和硬件访问风险,不是抽象的地缘政治脚注。[CM029, CM030, CM031, CM032, CM033, CM034]

2.6 Shengshu 相关 SAM 与未解决规模缺口

最站得住脚的 Shengshu 规模测算视角,不是一个无差别的数十亿美元全球 AI 视频 TAM。更干净的近期视角,应从 Fortune 的 $847 million 2026 AI 视频生成器市场出发,再剥离出 46.25% 的文生视频切片,得到估计 $392 million 的全球文生视频收入池。套用 Fortune 的 20.9% Asia-Pacific 份额,意味着约 $82 million 的 APAC 归因文生视频切片;同一报告又单独提到 China 在 2026 为 $49 million。另一个不同但互补的视角,是剥离 33.88% 的营销和广告切片,得到约 $287 million 的 2026 年活动生产工作流支出。这些计算不能得出 Shengshu 的实际 SOM;它们只是说明,公司最容易立刻变现的竞技场,全球大概率只有几亿美元,在可观察 China / APAC 切片内还要小得多。如果市场份额、企业 ACV 和全球扩张足够强,这个规模仍可能支撑一个独角兽;但公开数据还不足以揭示中国企业特定支出、免费转付费、API 收入结构,或非中国买方采用中国视频模型的意愿。投资人应把这里的 TAM 工作当作三角校验,而不是客户和收入尽调的替代品。[CM036, CM037, CM038, CM039, CM040]

2.7 图表

Chapter 03

03竞争格局

3.1 格局与竞争类别

Shengshu 所处的 AI 视频格局拥挤但仍不成熟,直接、相邻和替代型竞争者彼此重叠。直接同业包括已经交付文生视频或图生视频、创作者界面,并至少具备 API、企业或生产工作流功能之一的平台。按这个标准,最接近的对手是 Runway、Kling、Hailuo / MiniMax、Pika、Luma、Jimeng、PixVerse 和 Wan。第三方证据支持这个分组。Artificial Analysis 把 Hailuo、Kling、Sora、Vidu 和 Wan 放在同一个基准家族中比较;a16z 的 March 2025 消费者排名把 Hailuo、Kling 和 Sora 点名为新近重要的网页产品,同时把 Runway 放在 Brink List。市场里也有重要替代品,而不只是直接同业:Stable Video Diffusion 和 Wan 等开放模型或研究替代方案,大型 AI 实验室或云厂商的内部自建,以及依靠剪辑师、代理公司或内部创意团队手工制作视频的现状。Shengshu 的战略问题因此不是「还有谁能生成一个片段?」,而是「哪个对手掌握了买方最想要的质量、工作流深度、价格清晰度、本地化和分发组合?」[CP001, CP002, CP003, CP004, CP022, CP023]

FP001: 竞争定位图(工作流深度 vs 分发杠杆)

基于证据支撑的两个维度,对主要竞争对手做序数定位:x 轴是工作流深度,y 轴是分发杠杆。该图是综合判断,不是基准测试图。

工作流深度分数为分析师给出的序数,依据包括有公开记录的 API、协作、控制功能、制作导出和工作流模块。分发杠杆分数反映公开的用户 / 分发信号,例如 App 导向、生态邻近性、Adobe 入口位置,或据报道的创作者覆盖。

[CP003, CP005, CP008, CP010, CP014, CP016]

3.2 直接同业画像与弹药级别

最重要的直接同业并不站在同一起跑线上。Runway 是本组审阅对象中资本最充足的专业玩家:据报道它在 February 2026 以 $5.3 billion 估值完成 $315 million Series E,声称拥有 60 million-plus 创意用户,并把世界模型叙事从媒体延伸到机器人。Hailuo 背后的 MiniMax 看起来甚至更宽,是一个多模态平台:Sacra 将其描述为估值 $4 billion、累计融资约 $1.15 billion 的公司,并拥有很强的企业 API 牵引,Hailuo 只是其中一个界面。Luma 也资本充足且越来越偏企业,Owler 报道其累计融资 $1.1 billion,并在 November 2025 完成 $900 million 融资;官方页面则把 Luma 呈现为专业创意智能体工作流,而不是新奇玩具。Pika 财务规模小得多,Sacra 报道其融资 $135 million、估值 $470 million,但它用易用包装和 Adobe Firefly 分发来补偿。Kling、Jimeng 和 Wan 在已审阅资料中披露的投资人细节较少,但产品姿态显示它们与中国生态有很强相邻性。因此,Shengshu 同时面对三类资本层级:数十亿美元级专业玩家、宽口径中国多模态平台,以及更轻的消费者 / 创作者挑战者。[CP005, CP006, CP007, CP008, CP009, CP010]

竞争对手画像表
竞争对手类别规模 / 融资目标细分差异化局限
Shengshu / Vidu直接同业私营;有独角兽级融资和估值信号,但公开财务细节仍不完整创作者、广告主、开发者、企业中国原生视频质量,API 加创作者界面,产品迭代快治理、价格兑现和全球商业深度比功能营销更不透明
Runway直接同业 / 全球专业厂商2026 年 2 月 $315M Series E 轮,估值 $5.3B;声称 60M+ 创意用户专业创作者、工作室、开发者、企业团队、机器人 / 游戏相邻场景覆盖创意、开发和机器人界面,工作流广度深更昂贵的专业定位;面临激烈的前沿模型竞争
Hailuo / MiniMax直接同业 / 中国原生多模态平台Sacra 引用 MiniMax 约 $4B 估值和约 $1.15B 累计融资消费级创作者及企业 API 用户强消费端牵引信号、多模态平台广度、1080p / 镜头控制能力声明Hailuo 专属定价和企业客户细节在已审查公开来源中仍不够透明
Kling直接同业 / 中国原生平台在位者Kuaishou 支持的生态玩家;本报告未使用新的独立融资披露消费者、创作者、API 用户、移动端用户原生 4K、音频-图像-视频栈、移动端分发英文公开定价和企业披露较薄
Pika直接同业 / 消费创作者挑战者Sacra 引用约 $135M 融资和约 $470M 估值,并有更高投机上行休闲创作者、社交创作者、专业消费者、Adobe 相邻用户易上手 UX、病毒式特效、智能体内容创作、Adobe Firefly 分发资本体量小于 Runway / Luma,专业工作流深度也更弱
Luma直接同业 / 专业工作流挑战者Owler 引用约 $1.1B 累计融资和 2025 年 11 月 $900M 融资创意专业人士、团队、API 构建者、企业创意智能体工作流、关键帧控制、HDR/EXR 导出、团队工作区消费者品牌心智似乎低于更有病毒传播力的创作者应用
Jimeng直接同业 / 中国本地化创作者平台ByteDance 生态相邻;本报告未使用独立融资来源中文创作者和社区用户中文提示词流畅、首末帧控制、社区二创公开企业 / API 细节比 Runway、Luma 或 MiniMax 更薄
PixVerse直接同业 / 工作流和 API 平台已审查来源未披露私营规模营销人员、开发者、创作者、企业API、CLI、智能体、画布、营销中心工作流广度公开披露的实际定价和规模指标仍有限
Wan / 开源式替代方案替代 / 相邻模型层已审查来源显示更像社区或平台支持模型,而非明确披露 SaaS 规模开发者、研究人员、技术用户托管 SaaS 产品之外,免费或低摩擦实验托管工作流支持较弱,企业 / 合规包装也更弱

竞争行混合了直接同业和一个替代层,因为买方可以用托管 SaaS、广义多模态平台或开源模型替代方案完成同一任务。Runway、Pika、MiniMax 和 Luma 的规模 / 融资细节强于 Kling、Jimeng、Wan 或 PixVerse。

[CP001, CP004, CP005, CP007, CP009, CP014]
FP003: 关键对手的护城河 / 准备度 KPI

用紧凑的序数视角看主要同行当前最强的地方。

这些项目概括相对强项,不是经审计的评分;应用作尽调提示,而不是最终排名。

[CP007, CP010, CP015, CP018, CP020, CP021]

3.3 能力与包装对比

能力宽度正在趋同,但包装方式仍然分化。Runway 营销创意、开发者和机器人界面;Luma 销售智能体、团队工作区和 API;PixVerse 暴露 API、CLI、画布和营销中心;Pika 强调易用特效、智能体内容创作和移动社交二创;Kling 组合视频、图像、声音、特效和原生 4K;Jimeng 强调中文提示词流畅度和首尾帧控制;Hailuo 所属的 MiniMax 技术栈则把消费者应用和企业 API 结合起来。实际看,买方选择的是产品哲学,而不只是模型输出。专业团队可能偏好 Runway 或 Luma,因为工作流连续性、导出格式、共享点数或 API 控制,比新奇感更重要。消费者创作者可能偏好 Pika、Jimeng 或 Hailuo,因为界面和病毒循环更容易。API 优先构建者可能偏好 Vidu、PixVerse、MiniMax 或 Luma。Artificial Analysis 和官方文档也暗示,前沿竞争现在同时横跨质量、速度和价格,这会压缩任何只靠声称「我们有视频 AI」建立的护城河。Shengshu 的挑战在于,多数核心功能已经在别处存在;机会在于靠中国原生产品适配和高质量输出,赢下具体用户群。[CP011, CP012, CP013, CP014, CP016, CP017]

功能 / 能力矩阵
购买标准ViduRunwayKlingHailuoLumaPikaJimengPixVerse
文本生成视频
图像生成视频 / 参考工作流
公开 API 或开发者入口有限 / 官网非主入口已审查来源未知
移动应用重点有限 / 已审查来源显示非主入口有消费端应用已审查来源未知创作者应用 / 社区重点已审查来源未知
音频 / 唇形同步 / 声音功能是(Vidu Q3 原生音频)官网未知;更广的多模态栈在其他地方支持音频部分 / 偏工作流已审查来源未知
专业工作流深度中低中低中高
中国本地化提示词与合规适配
团队 / 企业级套餐通过 MiniMax 平台实现部分已审阅来源中未知

「未知」和「部分」单元格反映的是来源缺口,不是确认缺失。该矩阵依据官方产品页、平台文档和关联基准测试系列搭建, 而不是厂商自写的对比博客。

[CP011, CP012, CP013, CP014, CP016, CP017]
定价 / 套餐对比
平台入门档中阶 / Pro 档企业级 / API 信号定价模式启示
Runway免费;Standard 为 $12/monthPro 为 $28/month;Max 为 $76/month企业销售和开发者入口积分制订阅,叠加高用量付费档自助阶梯清晰,叠加高端 Pro 定位
Pika免费计划每月 80 积分据 Sacra,付费档为 $8、$28 和 $76/monthAdobe 分发;API / 主页入口已存在免费增值积分模式触达门槛低,增购设计激进
Luma$30/month 计划含 10,000 积分$90 和 $300 计划分别含 40,000 和 150,000 积分团队、企业级和 API 计划积分制计划,叠加按视频计价的 APIPro / 工作流打包强,用量经济账透明
PixVerse已审阅来源中,公开文档披露积分消耗,不给简单的消费者标价用量随模型和操作变化API 平台和文档公开可用API 和平台按积分计价开发者吸引力强;消费者价格透明度较低
Kling定价页可访问,但已审阅的抓取结果中公开文本细节很少已审阅公开文本中未知首页呈现 API 和企业级信号可能按积分或用量计价,但引用来源未支撑公开定价不透明,采购尽调负担更重
Hailuo / MiniMax更广的 MiniMax 产品栈中有消费者档位Sacra 在更广平台口径下提到 MiniMax Plus / Max / UltraSacra 提到按模态计价的企业 API消费者订阅与按用量计费的企业定价并存更宽的多模态产品栈可能支持视频以外的交叉销售
Jimeng / Wan有公开消费者访问入口信号,但已审阅来源未确认详细定价Unknown企业级 / API 细节有限或不清楚未知 / 未支撑这些中国本土入口的价格比较仍不完整

各平台积分体系不统一,名义套餐价格不能很好代表输出经济性。企业折扣、第三方模型打包和用量上限, 都可能显著改变客户实际成本。

[CP006, CP010, CP017, CP018, CP021, CP024]
FP002: 按商业入口划分的竞品能力广度

聚合观察哪些竞争对手在创作者端、开发者端、企业工作流和中国本地化上最强。

[CP011, CP012, CP014, CP016, CP017, CP020]

3.4 分发、切换成本与多平台使用

分发能力可能和模型质量一样重要。Pika 进入 Adobe Firefly,让它不必从零搭建传统企业销售,就能接触大型专业创意生态。Runway 既有自己的大型创作者基础,也有外部企业合作;TechCrunch 还报道它在把算力容量和用例扩展到游戏与机器人。Luma 的团队工作区、共享上下文、API 和企业计划,把它推向工作流系统位置,而不是一次性片段生成。PixVerse 也类似地打包画布、CLI、营销中心和 API 界面。相比之下,许多面向创作者的产品仍然很容易多平台使用,因为套餐按点数计费、用户界面以提示词为中心,内容也可以用不高的切换成本在别处重新生成。当平台掌握团队协作、管线集成、品牌资产、审核控制或企业承诺时,锁定效应才更强。因此,Shengshu 的防御性大概率不会来自随意型创作者订阅,而会来自 API、企业部署,以及它能持续守住的任何本地化合规或分发优势所形成的更深工作流绑定。[CP020, CP021, CP024, CP025, CP028, CP029]

3.5 护城河耐久性与反向竞争证据

反向证据很直接:Shengshu 没有干净的功能垄断。Runway 和 Luma 正在向专业和企业工作流更深处推进;Kling、Hailuo 和 Jimeng 具备文化本地产品适配和强消费者势能;PixVerse 在推进工作流编排和 API 界面;Stable Video Diffusion 和 Wan 等开放模型替代方案,降低了在任何托管 SaaS 平台之外试验的成本。即便 OpenAI 的 Sora 不再是活跃消费者产品,它也帮助重置了买方对前沿视频模型能力的预期。公开披露同样不均衡。官方网站通常最擅长功能营销,最弱的是已实现企业定价、净留存或经审计使用量,这意味着任何严肃比较里都会有部分行停留在未知或二手来源。竞争结论因此是混合的:Shengshu 有合理机会在中国原生视频质量和商业速度上获胜;但除非它能把模型质量与持久分发、企业工作流粘性和合规执行结合起来,并且做得比生态更大或资本更深的对手更好,否则护城河并不稳。[CP022, CP023, CP027, CP031, CP032, CP033]

护城河耐久性 / 竞争风险表
护城河主张威胁严重性缓释措施 / 尽调问题
原始视频质量更好Vidu、Kling、Hailuo、Runway 等前沿质量正在收敛按用例验证胜率,不看泛泛的基准测试话术
创作者牵引会形成持久护城河积分制多平台并用让创作者切换成本保持低位按队列衡量留存,并看用户是否迁移到付费团队 / API 计划
中国本地化 UX 足以打国际市场全球信任、合规和品牌门槛可能压住中国以外采用率中高审查国际客户构成和本地化合规工具
API 访问保证企业粘性许多同行现在也开放 API、文档或工作流集成入口测试集成深度、支持 SLA 和模型连续性承诺
只靠低价赢市场积分不标准;资本更厚的对手还能继续压价跨厂商标准化每可用秒或每次营销活动成果的成本
功能广度形成护城河功能重叠正在扩大;分发和工作流场景可能更关键识别哪些功能真正改变采购决策或留存
Sora 仍是主导威胁面向消费者的 Sora 在 2026 年停止服务,减轻了直接商业压力,但没有降低基准预期跟踪 OpenAI 是否通过新入口或合作伙伴渠道重新进入
开放模型与托管 SaaS 无关Stable Video Diffusion 和 Wan 类替代品会压低价格预期,并增加内部自建选项中高量化哪些买方细分真正需要托管 SLA / 合规,而不是开放式实验

本品类最耐久的护城河似乎来自分发、工作流集成和企业信任,而不是孤立的功能清单。

[CP023, CP024, CP025, CP027, CP032, CP033]

3.6 图表

Chapter 04

04财务情况

4.1 收入界面与变现通道

公开证据支持的是多通道变现模型,而不是单一订阅产品。Shengshu 自有材料称公司同时运营 MaaS 和 SaaS;Series A+ 新闻稿提到的产品生态覆盖 Vidu MaaS、Vidu SaaS、Vidu App 和 Vidu Agent。API 发布稿给出了最具体的定价证据:Vidu API 开放无需申请,入门访问从 $10 起,基础定价为每点数 $0.05;一个 4 秒视频根据功能和画幅消耗 4 到 40 点数。这是典型的按量计费开发者漏斗。在此之上,Agent 发布把视频生成重新包装成面向广告、TVC、电商和短视频媒体的工作流产品,指向比纯提示词娱乐更高价值的商业预算。Dealroom 和公司材料合在一起暗示,创作者订阅和点数并不是全部;企业客户和平台级服务似乎也很重要。收入模型的核心争论因此不是 Shengshu 是否变现,而是收入中有多少来自低 ARPU 自助创作者支出,又有多少来自质量更高的 API 和企业收入。[CI001, CI002, CI003, CI004, CI005, CI006]

收入流表
收入流机制单位当前值 / 状态质量尽调问题
创作者订阅 / 积分自助访问 Vidu 工具和高级功能套餐费加积分通过创作者计划和定价入口公开存在;实际组合未披露按计划档位拆分付费创作者、ARPU 和流失
API 用量通过 Vidu API 平台向开发者按用量开放每次生成对应美元和积分公开 API 起步门槛为 $10,且每积分 $0.05中高提供月度 API GMV、活跃账户和头部客户集中度
企业级 / B2B 服务面向企业的专属支持和集成合同或按用量计费已披露 B2B 服务团队;实际合同金额未披露分享 ACV 区间、部署周期和续约率
工作流产品(Vidu Agent)面向品牌和营销人员的一键广告及营销活动视频生产用量、订阅或企业套餐产品已发布,明确瞄准商业场景量化 Agent 是作为高级档变现,还是扩大企业支出
合作伙伴 / 平台级服务集成进合作伙伴应用和开发者工具收入分成或用量有具名合作伙伴案例,但经济条款未披露中低提供合作伙伴收入分成条款和分渠道集中度
App / 移动消费者入口桌面工作流之外的消费者或创作者分发订阅或积分产品生态中有公开提及,但确切贡献未知拆分移动端 MAU、付费率和应用商店渠道费用

Shengshu 似乎同时沿 SaaS 和 MaaS 两条线变现。待回答的问题是结构:创作者量可能很大, 但投资判断取决于有多少收入来自更耐久的 API 和企业用量。

[CI001, CI002, CI003, CI005, CI006, CI007]
FI001: 收入模式桥

Shengshu 看起来如何把使用量转换成多条收入流。

[CI001, CI002, CI003, CI005, CI011, CI012]

4.2 定价架构与 GTM 代理指标

Vidu 的公开定价架构混合了消费者和开发者动作。英文定价页确认了创作者套餐界面和大量高级工具,但可访问页面没有用可读文本给出清晰价格阶梯;相比之下,API 发布稿直接给出了每点数价格和最低支出细节。这种不对称本身就是有用信号:标价足够公开,可以支撑自助漏斗入口;但实际经济性大概率取决于功能组合、企业安排和页面外条款。API 发布稿还称 Shengshu 配备专门 B2B 服务团队,说明它有明确企业动作,而不是纯 PLG 创作者业务。Series A+ 发布稿中的客户例子——Pollo AI、PhotoGrid、OpenArt、Hubx、Fal.ai、Eachlabs、Freepik 和 GensPark——说明 GTM 可以借开发者平台和伙伴应用延伸,而不只是直接终端客户销售。财务含义是,Shengshu 的分发杠杆可能强于纯独立应用,但每个严肃账户的支持和集成成本也可能更高。即时 API 访问拓宽了漏斗顶部;它能否带来高效回本,取决于转化、使用深度和支持负担,这些都没有公开披露。[CI003, CI004, CI012, CI013, CI014, CI015]

定价 / 变现表
产品 / 渠道价格 / 单位 / 合同标价与实际价格折扣 / 未知项来源
Vidu API最低入门 $10;每积分 $0.05;4 秒视频消耗 4 至 40 积分公开标价式入门定价批量折扣、企业条款和有效净价未披露PR Newswire API 发布稿
Vidu 创作者定价公开定价页存在,并有创作者计划入口标价页面可见;已审阅的抓取结果未能完整读取可访问的详细价格梯度英文页面动态内容稀疏;实际 ARPU 未知Vidu 定价页面
Vidu 中国定价中国定价页存在公开入口已确认已审阅的抓取结果能提取的可读详细计划有限Vidu.cn 定价页面
Vidu Agent已审阅来源未找到独立公开费用不清楚是打包、按量计费,还是作为增购高级项定价不透明很关键,因为 Agent 瞄准更高价值商业用例PR Newswire Vidu Agent 发布稿
合作伙伴 / 平台用量经济条款未公开披露Unknown收入分成条款和最低承诺未公开Series A+ 发布稿和合作伙伴案例
竞争参考:Runway / Luma / Pika / MiniMax / Kling已审阅的大多数同行采用积分制或按用量计费多家同行有公开标价积分与输出质量不同,不能直接比较同行官方定价页和第三方跟踪源

Shengshu 的公开定价足以证明有变现,但不足以推断实际利润率或合同价值。竞品页面显示,本品类普遍如此: 价格会公开,但真正的单位经济性藏在用量、分辨率和支持复杂度后面。

[CI003, CI004, CI016, CI019, CI020, CI021]

4.3 牵引信号与收入质量

Shengshu 披露的牵引令人鼓舞,但高度依赖自报。最强商业叙事来自 Series A+ 和 Global Creativity Week 发布稿:公司称 2025 年用户和收入都增长超过 10x,Vidu 已覆盖超过 200 个国家和地区,服务超过 40 million 创作者和超过 10,000 名开发者及企业客户,生成视频超过 500 million,且商业项目占产出的超过 70%。如果准确,商业项目占比尤其有意义,因为它暗示 Vidu 用于可变现工作,而不只是实验。问题不是这些说法不可信,而是它们未经审计。公司仍没有披露 ARR、MRR、按产品划分的收入结构、留存、企业集中度或队列转化。Notebookcheck 的上手评测在这里是有用的反向校验:如果生产可靠性仍落后于基准营销,部分漏斗顶部创作者活动未必能干净转成持久付费使用。因此,收入质量应被视为有希望但未证实——如果伙伴和企业主张成立,它强于纯消费者玩具;但距离尽调级证据仍有差距。[CI008, CI009, CI010, CI028, CI029, CI030]

FI002: 单位经济性桥

决定 Shengshu 可能贡献利润率的公开可见因素。

每个节点都来自公开产品和品类证据,但 Shengshu 尚未披露量化这座桥所需的数值。

[CI017, CI021, CI022, CI028, CI037]

4.4 成本结构、单位经济与利润率压力

围绕 Shengshu 的品类经济性很清楚,即便 Shengshu 自己的毛利率没有披露。前沿 AI 视频高度消耗算力,公有云和同业定价证据都说明原因。Google 官方智能体定价体现了更宽泛的事实:多模态模型使用按量计费,并靠基于支出的承诺、缓存 token 和差异化模型费率来优化,而不是固定成本式基础设施。Luma、Runway、Pika、MiniMax 和 Vidu 都使用点数、按量收费或两者结合——因为更长、更高分辨率、更可控的输出,交付成本显著更高。MiniMax 文档和 Sacra 画像进一步显示,严肃多模态厂商要承担大量模型训练和推理开销;Pika 的 Sacra 画像还明确警告,用户需求上行后,会出现「不可持续的算力经济性」。Shengshu 不能只靠收入增长。关键财务问题是,企业收入结构、工作流自动化和伙伴渠道,能否把单位算力实现收入提升得比定价压力和模型质量军备竞赛挤压得更快。没有毛利率、重试率、GPU 来源或每生成秒成本数据,单位经济仍是一个结构化未知数。[CI017, CI018, CI019, CI020, CI021, CI023]

单位经济性表
指标数值 / 状态置信度重要性尽调问题
毛利率决定视频增长会转化为软件式经济性,还是被算力吃掉按创作者、API 和企业细分提供毛利率
收入增长2025 年 >10x(公司声称)方向性信号强,但需要经审计的分母和绝对基数提供 2024-2026 年月度收入桥
创作者转付费转化漏斗顶端规模只有转化为经常性支出才有价值按地区和计划档位提供付费率
企业 ACV收入质量高度取决于平均合同规模和集中度提供 ACV 分布和前 20 大账户
API ARPU / 支出深度用量型业务账户数可能好看,但变现深度偏弱按账户提供月度支出分桶
商业项目占比输出中 >70%(公司声称)如果商业用量比兴趣型产出更能变现,就是正面信号定义哪些算商业,并展示与收入的相关性
单个企业部署支持成本专属 B2B 服务团队可以提高赢单率,但会压缩贡献利润率提供实施成本和持续支持负担
每生成秒 / 片段的算力成本毛利率、价格底线和定价权的核心驱动按模型和分辨率档位提供混合推理成本
重试 / 失败率可靠性差会毁掉可用产出的经济性和客户 ROI提供任务失败率、退款率和审核拒绝率

这张表刻意保留大量 null,因为公开披露没有揭开真实单位经济性引擎。上述每个缺失字段都会直接影响 Shengshu 是否是可持续的软件业务。

[CI008, CI009, CI017, CI021, CI028, CI029]

4.5 资本充足性与融资依赖

资本充足性是 Shengshu 公开财务故事中最强的一块。官方和一线报道显示,February 2026 完成 >RMB 600 million Series A+,随后 April 2026 完成 RMB 2 billion Series B,这个节奏说明公司在短窗口内融到了一大笔新资金。CNBC 还指出,Series B 资金拟用于支持「通用世界模型」,意味着融资中相当一部分会投向前沿模型 R&D 和算力,而不只是商业化扩张。这一点很重要,因为世界模型雄心可能远早于收入消耗资本。正面解读是,Shengshu 短期内看起来不太可能有融资压力。负面解读是,近期融资规模可能掩盖同样沉重的成本野心,尤其是在公司试图同时竞争模型质量、实时系统、国际扩张和企业部署时。公开资料没有披露在手现金、月度烧钱、债务或现金跑道。所以,虽然 Shengshu 看起来资金充足,但还不能称为资本效率高。更合适的判断是「有缓冲,但不透明」。[CI025, CI026, CI027, CI035]

资本充足性表
指标数值 / 状态置信度重要性尽调问题
近期大额融资2026 年 2 月 Series A+ >RMB 600M;2026 年 4 月 Series B RMB 2B证明公司新近获得可用于算力、R&D 和 GTM 的资本核对募集总额与净到账金额,以及交割现金到账日期
账面现金判断现金跑道需要它,不能只看融资头条提供最新一轮融资以来的月末现金余额
月度烧钱判断现金跑道和融资依赖的核心输入按 R&D、算力、销售和 G&A 拆分现金消耗
现金跑道(月)没有烧钱速度,无法判断资本充足性提供基准和压力情景现金跑道模型
计划资金用途世界模型开发、模型扩展、商业化、更广平台建设决定本轮资金买来的是效率,还是只是更大的野心提供董事会批准的资金用途计划
债务 / 项目融资义务未公开披露债务可能显著改变风险和现金灵活性确认银行授信、云承诺和表外义务

近期融资看似充足,但缺少现金和烧钱披露,无法真正判断资本充足性。融资规模是缓冲垫,不是效率证明。

[CI025, CI026, CI027, CI035]
公开财务缺口表
缺失的私有指标影响具体尽调路径
经审计收入和 ARR阻碍对规模和估值效率的投资判断要求按产品提供经审计或董事会口径的月度收入桥
按产品线拆分毛利率阻碍评估算力经济性和定价权要求按创作者、API、企业和合作伙伴渠道拆分毛利率
现金余额和烧钱阻碍分析现金跑道和下一轮融资时点要求提供 Series B 交割以来的月度资金仪表盘
企业 ACV 和集中度阻碍评估收入质量和流失风险审查合同清单、ACV 分布和头部客户占比
按计划拆分付费转化和流失阻碍 PLG 效率分析按月份和地区审查队列仪表盘
云 / GPU 承诺阻碍分析固定成本和下行情景风险审查供应商合同、预留容量和预付款义务
合作伙伴收入分成条款阻碍评估渠道质量审查关键具名合作伙伴和平台客户的经济条款

这些不是锦上添花的细节;它们是把一个有前景的增长故事转成可支撑融资的投资判断所需的最低信息集。

[CI027, CI028, CI035, CI036]
FI003: 近期披露资本和牵引区间

公开可观察的财务和商业量级;由于收入本身未披露,这里混合融资和活动信号。

融资区间按 low=A+、mid=两个已披露 2026 年轮次的粗略均值、high=2026 年 A+ 和 B 轮已披露资本合计展示。商业占比和客户数项目是公司口径,未经审计。

[CI003, CI009, CI026, CI029, CI030]
FI004: 资本强度 / 现金流图

大额融资轮并不自动意味着自由现金充裕。

[CI023, CI024, CI025, CI026, CI027]

4.6 财务结论与尽调阻塞点

Shengshu 的财务姿态有足够吸引力,值得更深入尽调,但信息不完整,尚不足以形成确信。公开看,这家公司像是一家快速扩张的 AI 视频公司,具备真实变现界面、强近期融资、自称全球覆盖,并且可能向质量更高的商业使用迁移。这些是生成式媒体赛道风险投资赢家需要的正确要素。但缺失变量恰好决定它究竟是可持续软件业务,还是一门利润率薄、算力昂贵的故事:按产品线毛利率、创作者转付费、企业 ACV、伙伴收入分成、客户集中度、支持成本、现金消耗和下一轮融资触发条件。与公开公司备案体系中可见的报告纪律相比,Shengshu 披露仍然稀疏。工作结论是:收入质量有可信度,资本强度肯定高,资本充足性当前较强;但在管理层提供真实财务模型和队列证据之前,承销应保持有条件。[CI027, CI028, CI029, CI035, CI036, CI037]

4.7 图表

Chapter 05

05产品与技术

5.1 产品定义与模块地图

从客户工作流看,Vidu 不只是一个 AI 模型,而是一套创意操作栈。公开界面现在覆盖经典提示词生成模式,比如文生视频和图生视频;也覆盖用于角色或物体连续性的重参考生成、原生音频叙事、面向广告和商业的工作流自动化、给开发者的 API 访问,以及基于 Vidu S1 的实时互动数字人生成。Q3 页面尤其有帮助,因为它围绕最终用途而不是实验室指标来描述产品:可同时生成对白、音效和音乐的成片;单次输出更长的 16 秒片段;镜头语言控制;多语言输出;以及适配漫画剧、短剧和叙事广告。S1 / stream 界面又扩展了这个定义,把 Vidu 定位成语音控制数字角色的实时互动层,而不只是批量片段渲染。这个宽度很重要,因为 Shengshu 似乎在解决几个相邻任务:创意构思、生产提速、本地化、广告素材生成、互动数字人和开发者嵌入。因此,模块地图更像平台路线图,而不是单模型功能清单。[CE001, CE002, CE003, CE004, CE005, CE017]

产品模块 / 资产矩阵
模块 / 资产主要用户状态 / 成熟度差异化尽调缺口
Vidu 核心生成应用创作者、营销人员、编辑已上线 / 公开入口成熟文本、图像和参考驱动生成都收进同一产品家族需要按模式拆分的生产使用量,以及按工作流拆分的留存
Vidu Q3叙事创作者、广告团队、短剧制作方已上线 / 高阶公开产品原生音视频、16 秒单次生成、镜头控制、多语言输出需要实测成功率和音画同步错误指标
Vidu S1 / 流式交互交互式头像构建者、创作者、数字人用户新推出 / 公开产品仍早期540p、25 FPS 的实时语音驱动交互,提供语音选项和 API 路径需要真实负载下的延迟、并发和正常运行时间数据
Vidu API / MaaS开发者、企业产品团队已上线 / 商业化活跃开箱即用的视频生成,提供快速接入、模板、对口型和 MCP 支持需要配额、速率限制和企业部署控制文档
Vidu Agent品牌方、广告主、电商团队测试版 / 发布初期的工作流层脚本、镜头规划和自动组装,产出 15-30 秒平台可用视频需要发布口径之外的采用证据
模板 / 工作流工具中小企业、创作者、合作伙伴平台已上线 / 扩展中降低提示词复杂度,并把可复用场景打包需要模板使用结构,以及不同模板家族的质量差异
vidu-cli 与智能体技能开发者、AI 智能体操作者已上线 / 面向实操用户通过 CLI、npm、cargo 和智能体技能封装做程序化集成需要活跃开发者和支持负担的遥测数据

Shengshu 现在更像一个由多个相邻 SKU 拼成的平台,而不是一次单一模型发布。批量生成和 API 工具链看起来最成熟,实时数字人的公开推出仍更早期。

[CE001, CE002, CE003, CE004, CE005, CE013]
FE002: 客户工作流 / 运营流程

Vidu 公开可见的典型工作流:从输入到输出。

[CE017, CE018, CE024, CE027, CE038]

5.2 核心模型架构与工程栈

Shengshu 的技术栈有异常清晰的技术根基。最初的 Vidu 论文把模型描述为以 U-ViT 为主干的扩散系统,并明确声称可生成 1080p、单次最长 16 秒,还展示了强一致性和可控视频实验,如 canny-to-video、视频预测和主体驱动生成。后续公司发布显示,商业化层不是替换核心架构,而是在其上叠加可控性、一致性和推理加速。Vidu 1.5 增加多实体一致性、多角度一致性和高级镜头控制。Vidu 2.0 随后强调全栈推理加速器、更低延迟、更低成本和模板抽象。到 late 2025,TurboDiffusion 和 SageAttention 又把工程叙事推进到推理基础设施,开源材料详细说明稀疏注意力、量化、采样步蒸馏和示例部署模式。关键结论是,Vidu 的差异化不是一个算法技巧,而是多模态视频生成、参考一致性、控制、音频和系统工程的复利。[CE006, CE007, CE008, CE009, CE010, CE011]

技术 / 运营架构表
层 / 组件角色依赖风险
U-ViT 扩散骨干核心视频生成架构Shengshu 研究和模型训练管线前沿视频质量可能容易被同行快速追上
一致性和控制层多实体、多角度、镜头和参考控制模型调优、语义理解、参考处理营销口径可能跑在生产可靠性前面
原生音频生成同步生成声音和视频,产出完成态片段多模态同步和音频工具链音画同步失败可能伤害专业用例
推理加速栈为可部署视频生成降低延迟和成本TurboDiffusion、SageAttention、SLA、rCM、GPU 内核速度主张可能依赖特定硬件和提示词条件
API / MaaS 服务层向开发者和企业开放生成能力平台基础设施、认证、速率控制、计费、支持企业级可观测性和安全姿态未知
工作流自动化层模板、Agent、对口型、TTS、MCP 路由产品编排和场景打包模式越多,复杂度可能推高 QA 负担
实时流层语音驱动的数字人交互低延迟渲染、声音克隆、会话状态实时并发和滥用控制挑战未公开

Shengshu 的运营架构看起来是分层的,而不是单体式。它最强的技术叙事,是把核心多模态生成与加速、包装层组合起来,让系统更可用。

[CE006, CE007, CE008, CE009, CE010, CE011]
FE001: 产品架构图

Shengshu 的产品栈看起来分层清晰:上层是面向用户的工作流,下层是模型和加速基础设施。

[CE001, CE005, CE006, CE011, CE013, CE021]
FE003: 关键依赖图

技术栈既依赖自研建模,也依赖外部基础设施或开发者生态。

[CE010, CE011, CE029, CE030, CE031]

5.3 部署、集成与运营工作流

部署证据显示,Shengshu 试图像生产平台一样运作,而不只是研究实验室。API 平台、CLI 工具、GitHub 仓库和帮助中心内容展示了多个运营层:开发者可以访问 API 界面、自动化上传和任务、选择模型版本和时长、管理唇形同步与文本转语音任务,并以编程方式取回输出。MaaS API 更新又增加了一个重要集成信号:Model Context Protocol 支持让 Claude 和 Cursor 等工具可以借对话式工作流选择视频生成模式,而不是手工编排 API。战略上,这很重要,因为它让 Vidu 更容易嵌入智能体和工作流工具,而不只是独立网页 UI。支持和入门引导基础设施仍比大型企业软件规范更轻,但 CLI、技能、文档和客户支持界面的存在,说明 Shengshu 已经围绕核心模型交付了一个生态。技术栈的这一部分今天已有商业用途,尽管看起来仍更偏初创公司原生,而不是企业级加固。[CE005, CE012, CE013, CE014, CE015, CE024]

工作流 / 用例表
用户任务当前工作流公司方案可衡量收益限制
把提示词变成短视频片段输入提示词、等待、下载、在外部编辑Vidu 的文生视频和图生视频一个平台支持多种创作模式提示词精确服从和物理一致性仍不确定
让角色 / 产品在多镜头中保持一致手动画分镜或反复重生成参考转视频和多实体一致性工具广告活动和短叙事的连续性更强公开评测显示,实操中一致性仍可能破裂
快速制作可投放的社交或电商视频人工策划、镜头清单、剪辑、配音、导出Vidu Agent,加模板和对口型 API缩短生产周期,减少手工组装采用情况、QA 负担和企业审批流程未披露
把视频本地化到多种语言分别完成配音、字幕和剪辑Q3 原生音频和 MaaS 60+ 语言对口型后期交接更少对口型准确率或翻译质量没有公开错误率数据
把视频生成嵌入应用或工作流工具围绕模型端点自建编排API 平台、MCP 支持、CLI 和智能体技能开发者接入和自动化更快企业控制、SLA 和监控深度未知
运行实时交互式数字人头像、语音和渲染栈彼此分离Vidu S1 实时语音驱动交互从离线生成走向同步体验公开推出不久,运营韧性尚未验证

工作流契合度最强的是快速创意迭代、场景模板和嵌入式开发者使用。实时和企业关键用例仍有执行风险,因为可靠性和控制指标没有公开。

[CE003, CE004, CE005, CE008, CE015, CE018]

5.4 差异化、基准测试与成熟度轨迹

Vidu 的产品差异化,最强处在一致性、速度和工作流封装的交叉点。公开发布反复强调主体或实体一致性、镜头控制、长时单次生成、原生音频,以及基于参考素材的生产,而不是泛泛的文本生成视频新鲜感。Artificial Analysis 排行榜提供了第三方证据:Vidu 仍是全球第一梯队竞争者,即便还不能明确说是全球最好。围绕 TurboDiffusion 的 GitHub 和新闻稿材料也加强了一个判断:Shengshu 正在系统层面攻 AI 视频类别的核心瓶颈之一——延迟。与此同时,路线图从 2024 年的主体一致性,到 2025 年初的 Vidu 2.0 速度和模板,再到 Q3 的音频能力、MaaS 口型同步 / MCP,以及 2026 年实时 S1,显示公司正从离线短片生成,连贯地走向更丰富的生产和交互平台。成熟度的限制在于,最亮眼的主张仍高度依赖公司自控材料;基准位置和公开上手评测说明性能领先真实存在,但结论尚未落定。[CE008, CE009, CE010, CE015, CE019, CE025]

路线图 / 发布 / 研发阶段表
日期 / 阶段功能 / 里程碑状态含义来源
July 2024 发布窗口Vidu 公开发布已发布成立后商业化入口很快上线CNBC 发布报道 / 公司材料
September 2024非人类形态主体一致性已发布为影视制作和品牌内容带来早期连续性差异化PR Newswire 主体一致性发布稿
November 2024Vidu 1.5,支持多实体一致性和高级控制已发布可控性更强,并定位 1080pPR Newswire Vidu 1.5 发布稿
January 2025Vidu 2.0,生成更快、更便宜,并加入模板已发布速度和易用性成为明确产品优先级PR Newswire Vidu 2.0 发布稿
February 2025API 发布已发布平台向开发者和企业集成开放PR Newswire API 发布稿
December 2025TurboDiffusion 开源已发布加速能力成为公开工程资产和开发者信号PR Newswire 与 GitHub TurboDiffusion
December 2025Vidu Agent已发布 / 带测试版属性的工作流入口从原始生成走向结构化生产工作流PR Newswire Agent 发布稿
2026 Q3 阶段原生音视频和更长叙事生成已发布更适配广告和故事工作流Vidu Q3 页面
July-August 2026MaaS 对口型、模板、MCP 和 Vidu S1 实时交互已发布 / 早期平台扩展到智能体集成和实时头像PR Newswire MaaS 更新和 S1 发布稿

发布顺序是连贯的:先做连续性,再做速度和价格,然后开放开发者接入,接着补上音频和实时交互。对于一家商业视频基础模型公司,这是一条合理的产品成熟路径。

[CE008, CE009, CE010, CE015, CE025, CE028]
FE004: 产品成熟度 / 能力图

能力成熟度在批量生成上最强,在工作流和实时层推进最快。

[CE005, CE015, CE023, CE025, CE028, CE029]

5.5 信任、安全、隐私与质量控制

Shengshu 已经露出一些信任与控制界面,但还称不上能单独降低企业采用摩擦的公开保障包。帮助中心明确包含内容审核、积分、订阅和封号恢复,说明审核、反欺诈和账号处置工作流确实存在。Vidu S1 页面另行提示该功能涉及个人信息处理;这很关键,因为产品支持图片上传、音色选择,以及面向数字人交互的声音克隆。不过,已审阅公开材料没有显示公开状态页、公开事故历史、SOC 2 或 ISO 认证、详细模型卡、红队报告或量化安全性能指标。因此,Notebookcheck 的测试不只是产品评测,也是质量控制警讯:如果提示词遵循、物理一致性和参考还原在真实使用中仍明显失灵,客户信任和重复采用可能落后于基准营销。信任叙事方向上存在,但实质仍不完整。[CE020, CE021, CE022, CE023, CE033, CE034]

信任 / 质量 / 合规表
控制 / 指标状态范围缺口
内容审核帮助入口已有用户帮助和面向政策的运营公开证据只能证明议题存在,不能证明审核质量或政策深度
封禁账户和可疑活动恢复已有账户安全和执法工作流未发现公开欺诈或滥用率报告
S1 个人信息处理披露已有实时头像、图片上传和语音相关交互所审来源未发现详细公开隐私架构或留存披露
支持中心 / 帮助台已有上手、计费、点数、订阅、集成未发现公开 SLA 或事故历史披露
公开基准可见度已有模型质量和价格的第三方比较背景基准排名不等于企业级可靠性或安全性
公开模型卡 / 红队报告未发现安全和评估透明度所审公开来源未见
公开认证或信任中心证据未发现安全 / 合规保障所审来源未见 SOC 2、ISO 27001 或同等证据
公开状态页 / 正常运行时间透明度未发现运营韧性所审公开来源未见

Shengshu 有可见的运营控制,但相比大型企业买家通常对基础设施或工作流供应商的期待,它的公开信任材料偏薄。

[CE020, CE021, CE022, CE023, CE033, CE034]

5.6 产品技术结论与阻碍

Shengshu 过了尽调最重要的产品技术门槛:这里有一个真实、快速迭代的产品平台,不只是前沿模型融资故事。证据支持一个技术上严肃的栈:可见研究内核、连续产品发布、面向生产的 API 和工具,以及对延迟、一致性、工作流封装的具体攻关。这比只靠演示网站讲故事的创业公司强得多。阻碍在更上一层。买方和投资人仍无法充分确认部署可靠性、安全治理、企业级控制,以及技术优势到底有多少可持续,还是会被资本更厚的同行快速复制。因此,产品信心应在广度和节奏上给高,在可防守技术领先上给中,在信任成熟度上仅给中低,直到 Shengshu 公开更多生产级控制证据。[CE019, CE020, CE023, CE025, CE032, CE035]

5.7 图表

Chapter 06

06客户情况

6.1 按买方、用户和付款方划分的客户分层图

Shengshu 的客户基础更像分层结构,而不是单一客群。流量端是创作者和自助用户,用 Vidu 做社交内容、短视频和实验。高价值端是开发者、企业客户、营销人员和合作平台,把 Vidu 嵌进自己的工作流或最终产品。公司反复面向广告、电影、动画、电商、移动广告、文旅和教育营销,说明 Vidu 为不同付款方解决不同任务:创作者购买访问权和积分,开发者购买 API 用量,企业或商业团队购买活动产能或工作流加速。这一区分很重要,因为用户数不等于收入质量。公开来源暗示按使用量看最大群体是创作者,而战略上最重要的群体是能以更高客单价产生经常性用量的平台合作伙伴和 B2B 客户。因此,客户分层很宽、全球化,也具商业吸引力,但披露还不够细,无法看出业务真正锚在哪里。[CU001, CU002, CU003, CU018, CU024, CU027]

客户分群表
分群买方 / 用户 / 付费方用例规模收入 / 战略价值缺口
自助式创作者用户,通常也是付费方社交短片、创意实验、短叙事、模板按公司说法规模很大拉动认知和点数需求,但 ARPU 可能较低需要付费转化、流失和使用频率数据
开发者 / API 用户买方和用户把 Vidu 嵌入应用、工作流或创作工具按公司说法,开发者和企业客户合计 10,000+使用量可程序化扩张,战略价值更高需要活跃 API 账户和支出分层
企业营销和电商团队买方和付费方;终端用户是团队或代理商产品广告、本地化活动、虚拟试穿、叙事广告制作规模可观但未披露如果可复用,变现潜力优于消费者使用需要 ACV、续约和部署深度证据
合作伙伴平台 / 市场买方或渠道合作伙伴;终端用户是其客户在更宽的创作者或开发者产品中提供 Vidu公开证据显示可信且在增长高效分发和间接获客需要渠道分成经济性和依赖数据
硬件 / 生态渠道合作伙伴战略合作伙伴PC 工作流捆绑和设备级创作赋能公开证据有限分发杠杆和品牌背书需要出货搭载率和实际使用数据
制作团队 / 工作室 / 叙事创作者用户;可能直接购买,也可能通过平台购买电影、动画、短剧、长篇叙事工作流公开提及,但量化很薄如果部署属实,可成为有价值的标杆需要带结果的实名案例

分群结构像一只杠铃:一端是创作者规模,另一端是更高价值的 B2B 或伙伴渠道。核心尽调任务,是判断用户多快能从杠铃左端走到右端。

[CU001, CU002, CU018, CU019, CU024, CU027]
FU001: 客户旅程图

从发现到重复使用和扩张的公开可见路径。

[CU001, CU002, CU019, CU024, CU036]

6.2 采用轨迹与公开使用规模

无论按哪种公开标准看,采用曲线都很快,虽然仍来自公司自控口径。Shengshu 称 Vidu 首月用户达到 100 万,三个月内突破 1000 万,第四个月生成超过 1 亿条视频,第八个月参考生视频超过 1 亿次,后来总生成视频超过 5 亿条,服务超过 4000 万创作者,以及超过 1 万名开发者和企业客户。这些数字很大,说明 Vidu 不是小众原型。更重要的是,公司称目前超过 70% 的输出来自商业项目;如果准确,说明使用场景已明显超出业余实验。但公开记录缺少判断客户质量的分母:这些用户中付费占比、账户活跃频率、活跃客户定义、企业部署深度,以及合作平台与 Vidu 自有阵地分别贡献多少使用量。因此,采用故事最好被视为规模亮眼但透明度不完整。[CU003, CU004, CU005, CU006, CU016, CU026]

客户增长 / 采用轨迹表
指标日期来源置信度含义缺失分母
用户里程碑1 million 用户发布后第一个月内MaaS 更新新闻稿初期采用很快活跃用户占比,而不只是注册用户
用户里程碑10 million 用户前三个月内Vidu 2.0 发布稿 / MaaS 更新新闻稿证明有明显破圈付费占比和地区结构
使用里程碑100 million 条视频第 4 个月前MaaS 更新新闻稿漏斗顶端互动很重该阶段每活跃用户视频数和商业占比
功能采用里程碑100 million 次参考转视频生成第 8 个月前MaaS 更新新闻稿说明连续性工作流有共鸣使用该功能的不同付费用户数
创作者规模40 million+ 创作者截至 January 2026Global Creativity Week 新闻稿创作者触达很大月活占比和付费方转化
B2B 规模10,000+ 开发者和企业客户截至 January 2026Global Creativity Week 新闻稿具备真实 B2B 和平台存在感企业、开发者和伙伴之间的拆分
商业项目结构70%+ 产出来自商业项目截至 January 2026Global Creativity Week 新闻稿如果定义一致,是正向质量信号商业项目的精确定义和收入相关性
地理覆盖200+ 个国家和地区公司当前定位API 发布稿 / 公司材料表明地域集中度较低按地域拆分的收入结构

这些数字说明采用速度很快,但仍只是使用规模指标。它们没有直接回答 Shengshu 的核心用户是否留存、扩张,或付费水平是否足以支撑隐含估值。

[CU003, CU004, CU005, CU006, CU016, CU026]
FU002: 采用 / 部署漏斗

公开采用路径在漏斗顶部证据充分,伙伴部署也越来越可见,但续约阶段披露很弱。

[CU003, CU004, CU005, CU016, CU032]

6.3 具名客户证明与背书质量

具名客户证明存在,但质量差异很大。最强证据不是 Fortune 500 标志墙,而是合作伙伴或平台页面公开把 Vidu 打包成可用模型或嵌入能力。OpenArt 有专门的 Vidu 视频生成器页面。each::labs 公开提供 Vidu 模型家族和 API 访问,把 Vidu 放进统一开发者平台。Shengshu 自己的发布称 PhotoGrid 已把 Vidu 能力嵌入其产品;直接检查 PhotoGrid 的 AI 视频页面 HTML 也看到了 Vidu Q3 卡片,尽管可读性提取没有完整保留。相比之下,Pollo AI、Odin 和长片制作团队等具名案例在战略上有意思,但从已审阅公开来源看,独立可验证性弱得多。Lenovo 是另一个有用类别:它更像渠道或生态证明,而不是客户证明,说明 Shengshu 能借硬件和 PC 工作流获得分发。总体看,客户证明真实存在,但创作者或开发者平台上的证据最强,完整企业案例研究较弱。[CU007, CU008, CU009, CU010, CU011, CU012]

具名客户验证表
客户 / 合作伙伴客群部署 / 使用场景生产环境 / 试点结果局限
OpenArt创作者平台 / 交易市场面向 OpenArt 用户开放的 Vidu 视频生成器专页大概率已在生产环境上架第三方将 Vidu 公开封装成产品,把触达范围扩到 Vidu 自有产品之外未披露使用量、留存或商业条款
each::labs开发者平台each::labs 目录中通过统一 API 接入 Vidu 模型家族大概率已在生产环境上架证实 Vidu 已通过多模型开发者平台分发未披露客户规模或支出贡献
PhotoGrid大型创作者平台Shengshu 称 PhotoGrid 已嵌入 Vidu 能力;PhotoGrid 的 AI 视频界面和原始 HTML 检查显示 Vidu Q3 可用可能已进入生产环境,但未充分量化表明 Vidu 可借大众化编辑平台触达大量创作者Readability 提取结果不完美,公开成效指标缺失
Lenovo硬件 / 生态渠道将 Vidu 生成式视频能力与 Lenovo PC 和智能硬件生态做战略捆绑合作阶段已确认证明除纯软件入口外,还存在渠道分发路径不能证明终端客户重复使用或附加率
Pollo AI / Odin / 叙事制作团队公司声称的合作伙伴和终端市场验证据公司公告,涵盖图像-音频-视频工作流、虚拟试穿和长篇叙事项目混合 / 不清楚显示跨垂直场景采用潜力已审阅来源中的独立公开验证仍然薄弱

具名客户验证最强的场景,是第三方平台把 Vidu 以产品形态公开呈现;如果只是 Shengshu 引用终端使用场景,却没有客户侧文档或量化结果,证据就弱得多。

[CU007, CU008, CU009, CU010, CU011, CU012]
FU003: 客户证明矩阵

客户证明类型之间的证据质量差异很大,独立性和验证深度尤其不同。

[CU014, CU020, CU026, CU030, CU031]

6.4 留存、耐久性与重复使用缺口

耐久性是最大的未解客户问题。公开来源提供了大量规模信号和一些具名合作伙伴证据,但几乎没有直接留存证据。公司没有披露 NRR、GRR、logo 流失、合同期限、续约率、队列曲线或满意度指标。最好的公开重复使用代理指标是商业项目占比:据称超过 70% 的生成输出来自商业项目;如果口径稳定,这会暗示产品具有重复可用性。支持和计费界面也说明公司有服务重复客户的运营脚手架,而不只是一次性新鲜流量。但 Notebookcheck 上手评测中的可靠性担忧指向相反方向:如果参考还原、提示词遵循和场景一致性仍频繁失灵,客户使用可能很广但很浅。正确的尽调姿态,是把留存视为重大开放问题,目前只有粗略估计代理指标,而不是已经成立的正面结论。[CU015, CU016, CU017, CU021, CU025, CU032]

留存 / 重复使用 / 满意度表
指标数值 / 空缺客群置信度尽调要求
净收入留存率企业 / API要求提供按企业、合作伙伴平台和 API 队列拆分的 NRR
总收入留存率企业 / API要求提供 GRR 和客户留存瀑布图
创作者流失自助式创作者要求提供月活到付费的留存,以及按套餐拆分的流失
API 重复支出开发者 / 合作伙伴平台要求提供按账号年龄和月使用量分桶拆分的支出队列
合同期限 / 续约企业营销与电商团队提供合同期限中位数、试点转生产率和续约时点
商业项目占比生成输出的 70%+(公司声称)混合精确定义商业输出,并把它同收入或重复支出挂钩
满意度 / 投诉只有混合信号混合低-中汇总 NPS、CSAT、退款率和支持工单严重程度趋势
支持基础设施是否存在帮助、计费、点数和审核入口已经存在混合展示支持质量是否与续约和扩张相关

本表故意保持稀疏,因为公开记录缺少留存级披露。70% 商业占比是有用的重复使用代理指标,但不能替代队列数据。

[CU015, CU016, CU017, CU025, CU032, CU034]
FU004: 估计留存 / 重复队列:客户细分代理指标

按细分估计的重复使用代理指标;纳入该指标是因为 Shengshu 不公布实际留存队列。

下方百分比不是公司披露的留存率。它们是保守代理估计,依据包括客户类型、自助式与嵌入式工作流依赖的差异、>70% 商业项目口径,以及缺少公开流失或续约数据。

[CU015, CU016, CU017, CU021, CU032]

6.5 扩张循环与集中度风险

Shengshu 看起来有几条先落地再扩张的路径,但披露不足以判断经济性。创作者可能先在自助 Vidu 界面上手,再升级到付费积分或高级功能,之后进入更结构化的工作流。开发者或小平台可以从 API 起步,借 Vidu 更宽的模型家族加深使用,最终在市场或应用中规模化部署。品牌和商业团队可以被 Vidu Agent、口型同步或现成产品视频工作流拉进来。这些循环在战略上有吸引力,因为公司可以从消费者或创作者使用,扩张到更高价值的 B2B 场景。主要反作用力是集中度不透明。公开来源没有披露头部客户占比、企业 ACV、合作伙伴收入集中度,也没有说明关键生态渠道在不同视频模型间切换的摩擦有多低。业务的地理集中度可能低于许多创业公司,但对合作伙伴平台的依赖仍可能变成有意义的商业风险。[CU018, CU019, CU020, CU021, CU022, CU023]

扩张与集中度风险表
扩张驱动因素集中度风险影响尽调路径
创作者转付费套餐创作者流失率或付费率未知大漏斗可能变现良好,也可能只是浅层流量要求提供创作者队列看板和套餐转化漏斗
API 向合作伙伴平台放量渠道合作伙伴可能同时接入多个视频模型能拉动使用量快速增长,也可能带来快速切换审阅合作伙伴合同、排他条款和头部合作伙伴占比
企业营销工作流采购和信任要求可能拖慢扩张限制产品从创作者新奇体验进入更大预算审阅销售周期、安全审查摩擦和赢单 / 输单数据
通过 Agent / 对口型实现商业和广告自动化广告活动使用可能是阶段性的,不像订阅收入可能脉冲式且有季节性要求提供广告活动复现和复购数据
硬件 / 设备渠道扩张附加率和终端用户激活可能偏弱合作新闻可能夸大实际使用量要求提供 Lenovo 渠道的激活、留存和收入分成
地域广度国家集中度较低,但可能带来本地化成本广泛覆盖可降低地域风险,但会增加支持复杂度提供按地域拆分的收入和按区域拆分的支持成本
头部客户集中度公开未披露可能实质影响收入耐久性要求提供前 10 大客户和合作伙伴在收入与使用量中的占比

最有吸引力的扩张循环来自合作伙伴平台和结构化商业工作流。最大未知数是,这些循环会转化成可持续支出,还是很容易被用户切换掉。

[CU018, CU019, CU020, CU021, CU022, CU023]

6.6 客户结论与尽调重点

客户结论在覆盖广度和渠道创造力上偏正面,但在耐久性和支出质量上仍不完整。Shengshu 很可能拥有真正庞大的创作者漏斗顶部,以及越来越可信的合作伙伴平台分发;这是公开证据中最重要的两个正面点。真实采用的最好证明,是 OpenArt、each::labs 等外部平台公开把 Vidu 暴露给自己的用户,同时 Shengshu 自己的发布指向创意平台和商业用例的更广泛采用。最弱的部分是经典尽调变量:付费转化、续约、合同深度、ACV、客户集中度和独立 ROI 证据。因此,投资人可以对客户触达和生态相关性给予中等信心,但应把收入耐久性和背书质量继续作为开放尽调关口。[CU020, CU021, CU026, CU032, CU035]

6.7 图表

Chapter 07

07风险

7.1 核心风险图谱及其重要性

Shengshu 的风险栈高度耦合。监管事件可能变成客户问题;算力短缺可能变成产品质量和利润率问题;基准下滑可能变成融资问题。因此,最重要的洞见不是某个孤立风险,而是风险之间的传导路径。中国特定 AI 规则让合规成为产品要求,而不只是法律后置项。美国出口管制让基础设施获取变成战略变量,而不只是采购细节。竞争激烈意味着公司不能为了更安全就简单放慢速度,因为模型和工作流领先本身就是商业论点的一部分。这制造了典型前沿 AI 张力:Shengshu 必须快到足以保持相关性,同时建立足够成熟的审核、标识、隐私、支持和企业控制,避免事故或执法。对投资人而言,实际结论是下行情景彼此相关;一旦发生,很可能同时打到多个经营维度。[CR001, CR003, CR006, CR014, CR020, CR031]

FR001: 风险热力图

按发生可能性、影响、缓释成熟度和剩余敞口,对 Shengshu 最突出的风险做相对排序。

[CR006, CR012, CR014, CR015, CR018, CR021]

7.2 监管与法律风险

Shengshu 面临的中国监管负担很重,而且是持续的,不是假设风险。2023 年《生成式人工智能服务管理暂行办法》直接适用于面向公众的视频生成服务,要求数据来源合法、禁止内容可控、隐私受保护、投诉可处理、服务安全稳定。2025 年标识措施更进一步,要求生成视频带显式标识、元数据带隐式标识;当应用提供生成式 AI 服务时,互联网应用分发平台还要做检查。这些标识要求建立在更早的深度合成和算法推荐规则之上,也会影响备案和安全评估预期。实务上,Shengshu 有三类相关法律暴露:第一,内容合规和审核失败;第二,如果训练或推理输入处理不当,可能引发数据来源、隐私或 IP 挑战;第三,算法备案或标识文件不完整带来的流程风险。监管环境并不禁止增长,中国规则也明确谈到鼓励创新,但合规会成为永久运营成本;如果公司做错,还会有服务暂停风险。[CR001, CR002, CR003, CR004, CR005, CR023]

监管 / 法律风险台账
规则 / 案件法域状态可能性严重性缓释措施剩余暴露尽调路径
美国对涉华实体的先进计算出口许可美国 / 跨境已生效,并在 2026 年 5 月指引中重申严重资本缓冲、加速研究和谨慎供应商筛选高,因为算力获取仍可能收紧或变得更贵审阅 GPU 来源、云合同和替代产能计划
生成式 AI 暂行办法中国2023 年以来生效内容控制、隐私流程、投诉处理和内部合规运营高,因为违规可能导致整改令或服务暂停审阅数据来源、审核 SOP、用户协议和备案状态
AI 生成内容标识办法中国2025 年 9 月生效在生成视频和导出流程中内置显式与隐式标识高,因为视频产品和应用分发渠道直接受影响验证产品、元数据和合作伙伴平台导出中的标识落地
深度合成与算法推荐合规中国底层规则体系已生效中-高维护算法备案、安全评估准备度和文档中-高,因为义务可能随产品范围扩大审阅备案历史、监管沟通和评估触发条件
数据来源、隐私与知识产权暴露中国与跨境持续性结构风险中-高同意处理、合法数据采集、用户协议和内部审计轨迹高,因为公开训练数据披露仍然薄弱审阅训练数据治理、下架历史和隐私控制
标识要求带来的应用商店或分发审核摩擦中国平台生态自 2025 年规则体系起生效中-高预先打包标识材料和合作伙伴分发文档中,因为它可能拖慢上线或合作伙伴审批用当前产品版本测试应用审核和合作伙伴审核流程

法律负担不是单点问题,而是累积叠加:出口管制、内容规则、标识、深度合成规则和数据来源暴露会相互作用。只要 Shengshu 合规落地慢,其中任何一项都可能卡住增长。

[CR001, CR002, CR003, CR004, CR005, CR006]

7.3 运营、质量、隐私与安全风险

Shengshu 的产品本身就创造了有意义的运营和信任风险。AI 视频生成特别容易暴露质量问题,因为用户立刻能看到提示词漂移、物理错误、连续性断裂或口型同步弱。Notebookcheck 的上手评测显示,这不只是理论问题。Vidu 自身产品方向也在扩大风险面:原生音频、口型同步、声音克隆、实时虚拟人互动和基于模板的自动化,都带来新的审核、隐私、滥用和声誉负担。公开来源显示 Vidu 有内容审核界面和账号处置流程,这比完全没有可见控制要好;但它们没有显示大型买方常期待的模型评估深度、企业安全保证或事故透明度。结果是,每个新功能都可能提升商业价值,同时抬高安全部署的成本和难度。[CR009, CR010, CR011, CR012, CR013, CR026]

运营 / 质量 / 安全风险台账
失效模式可能性严重性缓释成熟度剩余暴露未解决缺口
视频生成可靠性达不到提示词、物理规律或连续性预期低-中未公开成功率、退款率或 QA 指标
内容审核失败,或对有害输出漏判中-高公开政策入口存在,但有效性未披露
过度拦截或审核摩擦伤害创作者和合作伙伴平台中-高中-高未公开申诉、误判或处理时长指标
语音克隆、头像和对口型功能带来冒充或隐私滥用风险中-高低-中未发现公开滥用率或红队文档
服务不稳定、排队延迟或并发限制损害工作流信任低-中未披露公开正常运行时间、延迟和 SLA 指标
企业安全或信任文档薄弱,拖慢采购中-高中-高中-高未发现公开的信任中心式保障材料

视频本身会放大运营风险:质量问题很显眼;产品扩到音频、头像和商业场景后,每一次错误的成本都会抬高。

[CR009, CR010, CR011, CR012, CR013, CR026]

7.4 合作伙伴、依赖、客户与财务风险

依赖层是地缘政治、财务和客户风险汇合处。Shengshu 需要先进算力、前沿工程人才和持续产品加速,才能与 Runway、Kling、Hailuo 等资金充足的对手竞争。美国出口管制执行让 GPU 获取和相关基础设施更脆弱,而 Shengshu 的世界模型雄心可能让算力需求持续高企。与此同时,公开可见客户证明在 OpenArt、each::labs 等合作伙伴平台上最强,商业上有用,但也引入平台依赖和潜在切换风险。财务不透明进一步放大问题:公司已完成大额融资,但公开来源没有披露烧钱速度、毛利率或客户集中度。也就是说,投资人能看到可能的压力点——算力供给、合作伙伴平台集中、定价压缩和利润率稀释——却无法精确量化。这是典型情形:资本缓冲降低近期生存风险,但不能消除模型风险或执行风险强度。[CR006, CR007, CR008, CR014, CR015, CR016]

合作伙伴 / 依赖风险台账
依赖项对手方角色集中度失效场景严重性缓释措施剩余暴露
先进算力和受限芯片GPU 供应商 / 云服务商 / 出口商训练和推理容量出口摩擦或更严格许可拖慢模型迭代并推高成本严重加速研究、供应商多元化和资本缓冲
平台分发合作伙伴OpenArt、each::labs、PhotoGrid 等类似渠道下游获客和工作流嵌入中-高多平台并用或合作伙伴调整优先级,会削弱分发和定价权强化自有产品价值并分散渠道中-高
中国监管体系CAC、应用平台、相关主管部门规则解释、备案、标识、投诉处理新规或更严格执法拖慢上线或触发整改合规运营和文档准备
研究与加速生态清华相关加速工作、内部系统团队效率提升和延迟控制人才流失或加速停滞会恶化成本结构中-高继续在系统加速方向发表成果并招聘中-高
资本提供方和融资市场现有投资者和下一轮市场支撑前沿规模研发的资金增长仍高,但在下一次融资需求出现前,变现或基准测试地位走弱利用当前缓冲证明利润率、客户和控制能力中-高
客户 / 合作伙伴集中度可见性未披露的头部账户和渠道收入耐久性unknown隐性集中度会造成突发收入或流失冲击内部报告、多元化和合同保护

最严重的依赖风险是算力。最容易被低估的依赖风险,是合作伙伴平台带来的渠道集中度——这些平台的用户未必对 Vidu 本身有忠诚度。

[CR006, CR007, CR008, CR015, CR016, CR017]
FR003: 依赖关系图

Shengshu 最关键的外部依赖横跨监管机构、基础设施、渠道和人才生态。

[CR006, CR008, CR015, CR024, CR039]

7.5 人员、执行与否决标准

最后一层风险是组织执行。Shengshu 的产品节奏显示工程速度很强,但也抬高了管理系统、合规运营、企业支持和人才留存的要求。公司似乎已经从 Jiayu Tang 主导的创始人产品动能,进入 Luo Yihang 作为公开 CEO 声音的更广泛运营阶段;这并不自动是负面,但确实需要理解治理连续性和决策权。公司还需要继续吸引模型研究、系统加速、安全或审核运营、企业交付等专业人才,而资金池很深的对手也在争抢同一批人。这些风险可以监控,因此应直接进入否决标准,而不是停留在模糊担忧。本章的核心破题事件包括出口管制冲击、监管执法、基准或产品质量下滑、客户耐久性失败,以及有证据显示前沿雄心跑得快过运营纪律。[CR021, CR022, CR030, CR041, CR042]

人才 / 执行风险台账
角色 / 职能依赖或缺口可能性严重性缓释措施尽调路径
前沿模型研究负责人面对更大对手,需要守住质量和路线图推进速度中-高近期发布节奏和清华相关工作说明团队有能力,但留人仍是关键审阅组织架构、流失率和薪酬竞争力
系统 / 加速工程需要对冲算力稀缺和成本压力中-高TurboDiffusion 及相关工作是可见缓释因素审阅路线图归属、团队厚度和对关键个人的依赖
合规与审核运营必须跟上中国标识、内容和隐私规则已有帮助中心和审核入口,但深度不明核查人员配置、SOP、审计轨迹和事件演练
企业销售和客户成功能力这是把产品牵引力转成稳定合同的关键中高产品在向商业工作流扩展,但支持深度未披露核查销售周期、安全审查通过率和 CSM 覆盖
领导层连续性 / 治理公开 CEO 发声主体已逐步从 Tang Jiayu 转向 Yihang Luo中高交接可能健康,但治理清晰度很关键核查董事会结构、授权边界和创始人角色连续性

Shengshu 的人才风险不是泛泛的招聘难,而是集中在稀缺的前沿 AI 和合规能力上;这些能力直接决定成本、产品质量和监管准备度。

[CR021, CR022, CR030, CR039, CR042]
缓释措施与终止标准表
风险可监测触发项阈值 / 事件行动含义
出口管制冲击GPU 或云供给受限关键算力流失、许可被拒或成本大幅跳升暂停投资论证,直到替代算力和预算得到证明
监管执法CAC 或相关执法行动警告、强制整改、算法备案失败或服务暂停信号提高尽调强度,并重置上线或增长假设
产品质量下滑基准测试和实地表现恶化排名明显下滑、用户投诉增加,或退款 / 重新生成负担上升下调增长假设,并重新检验留存逻辑
客户持久性失效续约乏力或合作伙伴集中度冲击发现显著流失、合作伙伴下架,或收入暴露集中重估收入质量和估值支撑
资本强度超支烧钱速度或毛利率不达预期在客户经济性跑稳前就出现融资需求除非新的战略资本条款异常有利,否则视为投资逻辑破裂
治理 / 执行失效领导层或控制权失序关键人员离职、事件久拖未决,或合规积压显性化继续推进前要求治理整改

这些终止标准刻意设成事件驱动。Shengshu 的主要风险只有从背景不确定性转成可观察的运营或监管失灵时,才会成为投资问题。

[CR020, CR031, CR041, CR042]
FR002: 风险传导图

核心风险事件如何层层传导到商业和融资结果。

[CR014, CR020, CR031, CR041]

7.6 图表

Chapter 08

08估值

8.1 建议与价格纪律

核心估值问题不是 Shengshu 是否有意思,而是当前公开证据是否支持支付一个精确的后期价格。在这个更窄的问题上,答案仍是否定的。CNBC 证实 Shengshu 2026 年 4 月由 Alibaba 领投的融资,但也称公司拒绝披露估值;Dealroom 只给出大约 $1 billion 到 $2.5 billion 的宽泛公开独角兽区间。与此同时,Shengshu 有足够多正面因素,不能简单放弃:公司披露 2026 年 2 月完成超过 RMB 600 million 的 A+ 轮融资,2026 年 4 月完成约 RMB 2 billion 的 B 轮融资,2025 年用户和收入增长 10x,产品面已经覆盖创作者工具、Vidu API 和面向广告的 Vidu Agent。 这组因素让 Shengshu 值得放进观察名单,但仍然对价格敏感。如果投资人按 $1.5 billion 到 $1.7 billion 的内部标记来承销,等于隐含假设公司已经把产品质量和商业使用转化为可观收入基础。没有经审计收入、毛利率、股权结构表或优先权披露,公开证据无法支撑这一假设。因此建议是观察而不是买入:保持覆盖活跃,但除非管理层打开收入桥、毛利率画像、客户集中度和稀释堆叠,否则不要在不透明标记下投入资本。实操规则很简单:价格越好、财务包越干净,Shengshu 越能从吸引人的故事转向可承销资产。[CV001, CV002, CV003, CV004, CV005, CV006]

投资建议摘要表
维度评估证据依据决策含义
投资建议观察公司有潜力,但确切估值和核心经济性仍未验证。保持主动跟踪;不要在不透明的独角兽定价上出手。
置信度独立报道足以框出估值区间,但不足以支持精确估值。只有在披露经审计或达到申报级别的财务资料后,才上调判断。
风险评级算力、监管、竞争和财务不透明可能同时压缩估值。要求估值折扣,并设置更硬的尽调门槛。
估值立场偏高私人成长期溢价有合理性,但公开证据尚未证明支撑独角兽区间上沿所需的收入。将 ~$1.5B+ 视为价格敏感区间,而不是显然便宜。
当下回报轮廓只有较低入场价才有非对称性如果 Shengshu 证明企业级经济性,仍有可观上行;但按传闻后期估值看,现有证据留下的安全边际很薄。投资前更偏好更低价格,或明显更充分的披露。

该表关注价格敏感性,而不只看公司质量。公司可以很强,但当前公开证据仍不足以支撑精确的后期估值。

[CV001, CV002, CV003, CV017, CV021, CV022]
投资逻辑 / 反向逻辑表
立场论点证据依据什么会改变判断
投资逻辑产品能力和基准测试可信度是真实支撑。Vidu 出现在外部 AI 视频榜单上,并持续发布重大版本。如果基准测试持续下滑,或用户质量问题显现,判断会被削弱。
投资逻辑商业使用可能强于业余创作者应用。公司称商业项目占比 >70%,开发者和企业客户超过 10,000 个。续约、ACV 和留存披露会显著增强该论点。
投资逻辑资本实力降低短期生存风险。2026 年披露融资超过 RMB 2.6B,加上 Alibaba 支持,为算力和 GTM 留出缓冲。如果出现重度烧钱或苛刻优先权证据,缓冲价值会下降。
反向逻辑当前确切估值没有公开验证。CNBC 称估值未披露;Dealroom 只给出宽泛独角兽区间。董事会批准的融资备忘录或经审计股权结构包可以解决该问题。
反向逻辑收入质量仍是最关键缺口。未发现公开 ARR、毛利率、烧钱速度、集中度或净留存披露。带有队列和毛利数据的收入桥接表会显著提高置信度。
反向逻辑监管和算力是结构性折价,不是背景噪音。中国 AI 规则叠加美国现行先进计算管制,可能放慢增长并推高成本。有文档支撑的合规体系和有韧性的算力采购计划,会收窄折价。

投资逻辑的吸引力来自产品和资本;反向逻辑集中在估值支撑和证据质量。

[CV001, CV002, CV003, CV005, CV008, CV010]
FV001: 建议逻辑

从产品和资本实力到观察建议的决策链;经济性数据缺失和价格不透明构成约束。

该图是概念性的 IC 风格综合,不是数学模型。节点标签把关键投资逻辑压缩成可跟踪因素。

[CV003, CV004, CV005, CV010, CV011, CV021]

8.2 可比区间与可支撑范围

避免伪精确的最干净方法,是用三组参照来三角定位 Shengshu:直接 AI 视频可比公司、更广义 AI / 软件公开市场倍数,以及 Shengshu 自己披露的商业信号。直接私营可比公司中,Runway 处于高端,2026 年 2 月估值 $5.3 billion、累计融资约 $860 million;Luma 完成 $900 million C 轮后估值达到 $4 billion;MiniMax 后来达到 $4 billion 增长期标记,累计融资约 $1.1 billion;Pika 则小得多,估值约 $470 million,2024 年收入 $7.6 million。从可见规模和战略资本看,Shengshu 应高于 Pika;但即便与围绕 Runway 或 MiniMax 的二手报道相比,它仍缺少已披露收入深度和投资级财务透明度。 公开市场锚点是有用的现实校验。Multiples.vc 2026 年 8 月软件数据显示,公开 AI 与设计工程软件约按未来十二个月收入 4.0x 到 4.2x 交易;媒体与娱乐软件概览也提醒投资人,高毛利创意订阅与低得多毛利的云渲染或基础设施业务,收入质量差异巨大。Shengshu 不是公开 SaaS 公司,如果商业使用转化顺利,应获得私营增长溢价。但这一溢价不能无限。$1.5 billion 股权价值意味着在 10x 下需要约 $150 million 可持续收入,在 12x 下需要约 $125 million,在 14x 下需要约 $107 million。$2.0 billion 标记要求更高。这些门槛对快速扩张的 AI 视频领导者并非不可能,但公开证据没有确认 Shengshu 已经达到。正确结论不是公司一定高估,而是当前估值论证建立在未经验证的收入假设上。[CV008, CV009, CV013, CV014, CV015, CV016]

可比估值表
可比对象指标估值 / 状态参考意义局限
Shengshu(公开区间)Dealroom 公开档案公开估值区间 ~$1.0B-$2.5B;2026 年 4 月精确投后估值未披露目标公司的直接当前区间区间很宽,且不是已确认融资估值
Runway最新私募融资2026 年 2 月以 $5.3B 估值融资 $315M;累计融资 ~$860M资金最充足的美国直接 AI 视频可比公司,也是溢价参考比 Shengshu 披露了更多规模、投资方阵容和全球企业验证
Luma AI最新私募融资2025 年 11 月以 $4.0B 估值融资 $900M直接的高端 AI 视频可比公司,且释放大型战略资本信号收入仍未公开披露
MiniMax后期私募 / 成长期估值在早前 $2.5B 轮次后,后期成长期估值为 ~$4.0B;累计融资 ~$1.1B中国多模态可比公司,资本基础大,且已有公开市场路径产品范围比 Shengshu 更宽,公开阶段也更靠后
Pika私募融资 + 收入快照估值 ~$470M,融资 ~$135M,2024 年收入 $7.6M有用的底部可比,展示较小 AI 视频玩家的形态规模早很多,收入质量也可能不同
公共市场 AI / 设计软件组合NTM 收入倍数2026 年 8 月公开可比公司为 ~4.0x-4.2x NTM 收入校准已披露公共软件公司的交易区间私有 AI 视频龙头可以享有溢价,但不能没有证据

这些可比对象用于框定 Shengshu,而不是假装公司完全相同。关键结论是相对位置:Shengshu 应该比 Pika 享有溢价,但公开记录尚不足以给予它与 Runway、Luma 或 MiniMax 同等的置信度。

[CV002, CV013, CV014, CV015, CV016, CV017]
FV002: 估值敏感性

基于当前公开记录,展示不同可持续收入与倍数组合下隐含股权价值的敏感性。

该图只是框架工具,不是管理层披露的预测。条形标签显示,要支撑越来越激进的私募估值,需要多少可持续收入。

[CV017, CV020, CV022, CV023]

8.3 情景区间与下行传导

情景分析很重要,因为 Shengshu 最好的公开指标是活动信号,而不是经审计收入指标。乐观情景下,Shengshu 所称 >70% 商业项目占比被证明是真实的收入质量信号,开发者和企业客户基础转化为耐久 API 与 B2B 合同,产品领先保持在 AI 视频基准前列,管理层也能证明毛利率在算力强度很高的情况下向软件化水平靠拢。在这条路径下,大约 $1.8 billion 到 $3.0 billion 的区间有可辩护性;如果从低端进入,也能支撑有吸引力的上行空间。 基准情景更保守。它假设 Shengshu 真实存在且在增长,但收入质量在自助创作者、平台合作伙伴和企业账户之间参差不齐;算力和合规成本仍然很重;私营市场给公开 AI / 软件倍数一定溢价,但不会把 Shengshu 当作披露充分的稀缺资产。在这条路径上,大约 $0.9 billion 到 $1.6 billion 更可支撑。悲观情景不是崩盘故事,而是倍数压缩和不透明故事。如果商业使用无法转化为耐久 ARR,如果出口管制收紧,如果监管摩擦拖慢发布,或如果竞争定价压薄利润率,大约 $0.5 billion 到 $0.9 billion 的区间就有可能成立。这就是为什么下行主要通过经济性缺失传导,而不只是产品无关。[CV005, CV008, CV009, CV010, CV017, CV018]

乐观 / 基准 / 悲观情景表
情景假设估值 / 回报逻辑概率信号关键风险
乐观商业项目占比被证明真实,收入运行率达到约 $140M-$180M+,企业 / API 组合增强,产品质量保持在品类前列。按高于公共 AI / 软件可比公司的私人成长期溢价倍数,对应 ~$1.8B-$3.0B;只有从低端价格入场才有吸引力上行。如果公司当前声称的使用指标能顺利转成稳定收入,该情景就可能成立。需要比当前公开证据更干净的毛利、留存和合规证明。
基准收入运行率落在约 $80M-$120M,增长仍强,但创作者、合作伙伴和企业渠道表现分化,市场给予中等私有溢价。~$0.9B-$1.6B;仅凭公开证据,这是当前最能支撑的区间。最符合“产品信号强、经济性不完整”的组合。毛利受压、伙伴依赖和流失披露不足,仍可能把结果推低。
悲观收入质量不达预期,企业验证仍薄,算力或监管摩擦上升,倍数向较低软件水平压缩。~$0.5B-$0.9B,意味着高价后期入场有实质下行。如果不透明延续到下一轮融资窗口,该情景可信。产品仍有相关性,但经济性始终无法证明,估值因此重置。

这些区间服务于入场纪律,而不是追求模型精确估值。它们把缺失的收入证据转成具体的估值后果。

[CV005, CV009, CV017, CV024, CV025, CV027]
投资逻辑破裂与终止触发项表
触发项阈值对投资逻辑的传导行动含义
企业 / API 证明停滞下一轮融资前仍没有有说服力的 ARR 桥接、留存数据或 ACV 证据打破“商业使用质量高,而非只是广而浅”的论点不要按溢价倍数承销。
模型质量下滑相比 Runway、Kling、MiniMax 或其他头部同行,基准测试或评价持续恶化削弱支撑后期定价的产品驱动溢价将估值区间下修至基准 / 悲观情景。
出口管制冲击或算力挤压实际可获得先进算力更受限,或推理经济性明显恶化同时把产品势能转化为毛利和迭代风险扩大折价,并要求证明算力韧性。
监管事件出现与中国 AI 规则相关的内容、标识、数据来源或合规失误将政策风险转化为客户信任和分发摩擦暂停投资论证,直到整改得到验证。
披露仍然稀薄尽管持续融资,仍没有经审计财务包、股权结构表或优先权视图即便公司本身表现良好,也无法做可靠回报测算维持观察 / 继续研究,而不是买入。

这些触发项可监测,而非抽象风险。每一项都会直接改变估值案例背后的收入、倍数或稀释假设。

[CV008, CV010, CV011, CV013, CV027, CV035]
FV003: 估值 / 回报区间

Shengshu 在不同收入质量和倍数假设下的悲观、基准、乐观估值区间。

区间用于把公开证据质量转化为估值纪律。最后一项不是已确认融资估值,而是呈现 Dealroom 当前可见的公开独角兽区间。

[CV002, CV030, CV031, CV032, CV033]

8.4 退出准备度、尽调问题与最终观点

从尽调角度看,Shengshu 尚未准备好退出;虽然从战略兴趣角度,它可能具备退出可能。公司显然会发产品、吸引资本,也会营销商业用例。缺的是投资委员会承销具体价格所需的一整套材料:按产品线拆分的 ARR、净留存、企业集中度、毛利率、算力承诺、现金消耗、董事会结构,以及 2025-2026 年快速融资形成的优先股堆叠。SEC EDGAR、HKEXnews 等监管申报基础设施展示了公开投资人最终要求的标准;Shengshu 离这一级别披露还很远。 最终观点很直接。Shengshu 是更可信的中国 AI 视频公司之一,值得继续覆盖,因为产品、资本和商业化故事都真实到足以重要。但今天的公开证据支持兴趣,而不是确信。只有在管理层提供经审计或达到监管申报级别的经济性,或进入价格低到足以在不透明下创造安全边际时,才应上调。此前,观察是纪律性的答案:公司很强,估值证明不完整。[CV011, CV012, CV026, CV034, CV035, CV036]

最终尽调问题清单
主题缺失证据为何重要负责人或尽调路径
收入桥接ARR、确认收入、产品线组合,以及创作者 / API / 企业贡献没有这些,估值只能依赖活动代理指标,而不是财务事实财务尽调和管理层资料包。
收入质量净留存、流失、ACV 区间、合作伙伴集中度和前十大客户占比决定 Shengshu 应享有软件类倍数,还是接受更低的平台 / 消费混合折价商业尽调加队列复盘。
毛利与算力分渠道毛利率、GPU 承诺、重试率和每可用秒成本AI 视频估值很大程度取决于规模是改善经济性,还是只增加算力消耗技术 + 财务工作流。
股权结构和优先权清算优先权、反稀释条款、董事会权利,以及 2025-2026 轮后持股如果优先权悬顶吃掉上行,头条估值仍可能没有吸引力法律顾问和数据室审查。
合规准备度算法备案、标识落地、数据来源控制和事件历史监管风险是估值折价的一部分,尤其是与中国相关的 AI 视频政策顾问和产品信任尽调。
退出准备度适合跨阶段投资者、战略方或 IPO 承销的经审计财务资料或申报级资料包这是从有趣故事变成可投资产的最短路径任何条款清单工作前,先发起董事会层面的尽调请求。

这些问题是最低数据集,决定 Shengshu 能否从高关注标的转成可承销的后期投资。

[CV011, CV012, CV034, CV036, CV041, CV042]
FV004: 投资 KPI

IC 风格打分,衡量 Shengshu 在市场位置、证据质量、经济性可见度和估值支撑上的表现。

分数带有主观判断,且为相对分,采用 1-10 分,分数越高越好。分数只总结公开证据。

[CV008, CV009, CV010, CV011, CV024, CV027]

8.5 图表

附录 A: 融资和估值背景

公开记录里,Shengshu 融资披露最有力的证据集中在 2026 年:2026 年 2 月宣布 >RMB 600 million Series A+,2026 年 4 月完成由 Alibaba Cloud 领投的约 RMB 2 billion Series B。Dealroom 公开将 Shengshu 归入 $1 billion-$2.5 billion 独角兽区间,CNBC 则明确称 2026 年 4 月那轮融资未披露估值。两条证据合在一起,支撑强战略资本叙事;但没有私人尽调材料,仍不足以做精确的后期轮投资核算。[CO015, CO016, CO019, CO020, CV003, CV021]

免责声明

本报告由 AI 研究智能体生成,仅供信息参考,不构成财务建议。估值、客户和商业化指标基于 runDate 当日可得公开来源,其中许多来自公司自述或二手资料。任何投资决策都应依赖管理层直接披露和独立尽调。

证据索引

结论
编号陈述可信度来源
CO001 Beijing Shengshu Technology Co., Ltd. was established on 2023-03-06 in Beijing. SO006, SO016
CO002 The company registered address is Unit 801A, 8th Floor, Building AB / Dongsheng Building, 8 Zhongguancun East Road, Haidian District, Beijing. SO006
CO003 Public sources consistently tie Shengshu core team and technical origin to Tsinghua University and the U-ViT / multimodal research lineage. SO005, SO006, SO009
CO004 CNBC identified Jiayu Tang as Shengshu co-founder and CEO in November 2024. SO002
CO005 By 2026, Dealroom and company materials describe Luo Yihang as CEO, Tang as president, Zhu Jun as founder/chief scientist, and Bao Fan as CTO / legal representative. SO005, SO006, SO004
CO006 Vidu is Shengshu flagship multimodal video platform spanning text-to-video, image-to-video, and reference-to-video creation. SO001, SO012
CO007 Shengshu and Tsinghua formally unveiled Vidu on 2024-04-27 as a long-duration, high-consistency video model positioned around up to 1080p generation and multi-shot coherence. SO006, SO007, SO009
CO008 Vidu launched globally on 2024-07-30 with Chinese and English prompt support. SO008, SO007
CO009 Vidu 1.5 launched in November 2024 and introduced multiple-entity consistency plus stronger camera and control features. SO018
CO010 Vidu 2.0 launched in January 2025 and was marketed as generating clips in under 10 seconds at materially lower cost than earlier versions. SO017, SO007
CO011 Shengshu launched the Vidu API in February 2025, offering text-to-video, image-to-video, and reference-to-video access for developers and enterprises with instant purchase starting at $10. SO012
CO012 Vidu Q3 added native audio-video generation, up to 16-second single-pass clips, multilingual output, and native 1080p narrative features by 2026. SO013, SO025
CO013 Vidu S1, announced in July 2026, shifted Shengshu beyond offline clip generation into real-time voice-driven avatar interaction at 540p and 25 FPS, with API availability for developers. SO011
CO014 TurboDiffusion, co-released with Tsinghua in December 2025, was presented as delivering 100-200x faster video generation and cutting an example 1080p 8-second render from about 900 seconds to about 8 seconds. SO014
CO015 Shengshu announced a Series A+ round of over RMB 600 million on 2026-02-05, co-led by Zhongguancun Science City and LINK-X Capital / Xinglian Capital with multiple strategic investors joining. SO004, SO006
CO016 Alibaba Cloud led Shengshu Series B financing of approximately RMB 2 billion in April 2026, with TAL Education and Baidu Ventures also participating. SO003, SO006
CO017 Earlier public funding history includes a nearly RMB 100 million angel round in June 2023, an angel-plus round in August 2023, several-hundred-million-yuan 2024 rounds, and a several-hundred-million-yuan Series A in September 2025. SO005, SO006
CO018 Publicly named investors and strategic backers across Shengshu financing history include Ant Group, Baidu Ventures / Baidu-linked capital, Qiming Venture Partners, the Beijing AI Industry Investment Fund, Huawei Hubble, Zhongguancun Science City, and Alibaba Cloud. SO003, SO004, SO005, SO006, SO010
CO019 Dealroom labels Shengshu Technology a unicorn and provides a public valuation band of $1-2.5 billion. SO005
CO020 CNBC reported that Shengshu declined to disclose valuation alongside the April 2026 Series B round, leaving the current exact post-money figure unverified. SO003
CO021 Shengshu said both users and revenue grew more than 10x during 2025. SO004
CO022 Company materials say Vidu operates in more than 200 countries and regions. SO004, SO012, SO016
CO023 At Global Creativity Week in January 2026, Shengshu claimed Vidu had expanded to more than 40 million creators, over 10,000 developers and enterprise customers, and more than 500 million generated videos. SO013
CO024 Shengshu publicly named commercial users including ByteDance, Samsung, TAL Education, Alipay, HONOR, JD.com, Alibaba 1688, Amazon, Meituan, L’Oréal, and Anta. SO004
CO025 Shengshu also names Tencent Animation & Comics, China Literature, CCTV Animation, iQIYI, Jiangxi Film Group, and Mango TV among Vidu entertainment partners or users. SO004
CO026 Shengshu announced in June 2025 that it had been selected as a 2025 Technology Pioneer by the World Economic Forum. SO016
CO027 Baiduwiki reports that Shengshu had more than 70 employees as of March 2024 and that nearly 90% of staff were R&D personnel. SO006
CO028 Baiduwiki says the company held 65 patents and had established four wholly owned subsidiaries by 2025. SO006
CO029 Shengshu jobs portal exposes recruiting categories for Beijing, Shanghai, Shenzhen, and San Francisco, indicating hiring activity across multiple cities even though Beijing is the only clearly documented headquarters. SO024
CO030 Dealroom describes Shengshu revenue model as primarily subscriptions with a significant portion from corporate clients. SO005
CO031 Vidu public product surfaces now include a web app, API platform, enterprise entry points, creator programs, and agent-style workflow products, consistent with a hybrid SaaS-plus-platform model. SO001, SO011, SO015, SO025
CO032 CNBC reported in November 2024 that Vidu was already generating revenue from advertisers, animators, and other businesses, with monthly customer usage ranging from RMB 100,000 to RMB 1 million. SO002
CO033 As of August 2026, Artificial Analysis snapshots place Vidu Q3 Pro around rank 18 in text-to-video and around rank 19 in image-to-video, with visible per-minute pricing near $9.60. SO022, SO023
CO034 ShengShu January 2026 Global Creativity Week release claimed Vidu Q3 ranked No.1 in China and No.2 globally on Artificial Analysis at that time. SO013
CO035 Notebookcheck September 2025 hands-on review concluded that Vidu could create impressive visuals but was still too glitch-prone and inconsistent for dependable professional production. SO021
CO036 Reviewed public sources do not disclose Shengshu board composition, independent directors, or detailed governance rights. SO002, SO003, SO004, SO005, SO006
CO037 Public Chinese-language summaries indicate Shengshu completed 2024 algorithm-filing procedures for image and video generation scenarios, showing at least baseline regulatory processing in China. SO006
CO038 Shengshu increasingly frames Vidu and its adjacent research as infrastructure for broader world-model and physical-AI applications rather than only creator tools. SO003, SO011, SO014
CO039 The company 2026 messaging links real-time interactive video, AI avatars, and general world-model ambitions into a single roadmap that could extend beyond media into robotics and embodied AI. SO003, SO011
CO040 A conservative public lower-bound funding estimate exceeds $380 million equivalent once the disclosed RMB 600 million Series A+ and RMB 2 billion Series B are combined with earlier 2023-2025 rounds whose exact sizes are only partially disclosed. SO003, SO004, SO005, SO006
CO041 The founding-team narrative is directionally consistent on Tsinghua roots but not on title labels, with Zhu Jun, Jiayu Tang, and Luo Yihang each occupying different parts of the public founder-versus-operator story over time. SO002, SO003, SO005, SO006
CM001 TBRC defines the AI video generator market as software or systems that use AI to generate or enhance video from inputs such as text, images, or audio. SM004
CM002 TBRC's adjacent generative-AI-in-video-creation market covers cloud and on-premise deployment, and applications spanning marketing, education, entertainment, and social media for enterprises, SMEs, and individual creators. SM003
CM003 Research and Markets outlines TAM and segmentation frameworks for both AI video generator and generative AI in video creation categories, reinforcing that publisher market boundaries include end-user and workflow definitions rather than one single universal scope. SM001, SM002
CM004 Official competitor sites show the commercial category now bundles self-serve creation, APIs, agents, CLI workflows, enterprise packaging, and in some cases mobile distribution rather than only a single prompt box. SM008, SM011, SM012, SM022, SM023
CM005 TBRC sizes the AI video generator market at $0.85 billion in 2025, $1.04 billion in 2026, and $2.07 billion in 2030, implying 18.9% CAGR from 2026 to 2030. SM004
CM006 TBRC sizes the generative-AI-in-video-creation market at $0.39 billion in 2025, $0.47 billion in 2026, and $0.98 billion in 2030, implying 20.4% CAGR from 2026 to 2030. SM003
CM007 Fortune Business Insights values the AI video generator market at $716.8 million in 2025, $847 million in 2026, and $3.35 billion by 2034 with 18.8% CAGR from 2026 to 2034. SM005
CM008 The public spread between roughly $0.39 billion, $0.717 billion, and $0.85 billion current market lenses is best explained by boundary differences rather than a settled single consensus TAM. SM003, SM004, SM005
CM009 TBRC describes Asia-Pacific as the largest region in 2025 for the AI video generator market, while its adjacent video-creation report names North America largest and Asia-Pacific fastest growing. SM003, SM004
CM010 Fortune Business Insights says North America held 41.0% of AI video generator revenue in 2025, Asia-Pacific held 20.9%, and China was valued at $49 million in 2026. SM005
CM011 Fortune Business Insights reports text-to-video accounted for 46.25% of the AI video generator market globally in 2026. SM005
CM012 Fortune Business Insights reports marketing and advertising as the largest application segment at 33.88% of the AI video generator market in 2026. SM005
CM013 Fortune Business Insights reports social media as the fastest-growing application segment, at 23.5% CAGR. SM005
CM014 Fortune Business Insights reports large enterprises as the largest customer class at 50.86% share in 2026, while SMEs are the fastest-growing segment at 21.1% CAGR. SM005
CM015 Runway organizes its market surface into Creative, Dev, and Robotics platforms built on the same core models and says its tools are used by 60 million plus creatives. SM008
CM016 Runway pricing progresses from free exploration to paid creator tiers and enterprise sales, demonstrating a freemium-to-team-to-enterprise commercial ladder. SM009
CM017 Pika positions itself around AI video creation, workflow automation, agents, MCP connectivity, and mobile-style effects, signaling a creator-first but workflow-aware segment. SM011
CM018 Kling exposes video, image, sound, effects, API, native 4K output, and mobile apps, indicating both consumer and enterprise/developer packaging. SM012
CM019 Jimeng emphasizes Chinese-language prompts, text or image to video generation, first-frame and last-frame control, and a community for inspiration and remixing. SM013
CM020 PixVerse combines text and image to video, agent-driven editing, marketing-hub workflows, CLI execution, lip sync, and API surfaces, showing that commercial competition is already workflow-bundled. SM021, SM022, SM023
CM021 Across official vendor pages, the market clearly monetizes through two lanes: low-friction subscription or credits for creators and higher-touch API or enterprise contracts for teams and developers. SM009, SM012, SM021, SM022
CM022 Chinese-language prompt optimization and localized product/compliance surfaces are visible differentiators in Chinese platforms such as Jimeng and Kling. SM012, SM013
CM023 Pew Research Center found 83% of U.S. adults use YouTube, 68% use Facebook, 47% use Instagram, and 33% use TikTok. SM004, SM025
CM024 YouTube's 2024 U.S. impact report says the creator ecosystem contributed $55 billion to U.S. GDP in 2024, supported 490,000 full-time-equivalent jobs, and that YouTube paid more than $70 billion to creators, artists, and media companies during 2021-2023. SM024
CM025 a16z wrote in March 2025 that the prior six months delivered major progress in AI video quality and controllability, and that Kling and Hailuo had surpassed Sora in monthly visits by January 2025. SM006
CM026 a16z described provider fragmentation around differentiated strengths such as Sora's versatility, Hailuo's prompt adherence, and Kling's camera-movement control and lip sync. SM006
CM027 Official vendor pages suggest the category is shifting from novelty generation toward integrated workflow software by adding templates, references, lip sync, agents, API orchestration, and CLI tooling. SM008, SM011, SM012, SM023
CM028 The narrow market is pulled most directly by marketing, e-commerce, media or entertainment, education, social media, and developer workflow use cases rather than by a single homogeneous creator persona. SM003, SM004, SM005, SM023
CM029 China's Interim Measures for the Administration of Generative AI Services took effect on 2023-08-15 and apply to public generative AI services that produce text, images, audio, video, and other content. SM015, SM016
CM030 The Interim Measures require providers to address lawful training data, personal-information protection, content governance, transparency, complaint handling, and where relevant filing or security-assessment obligations. SM015, SM016
CM031 China Law Translate's text of the Interim Measures says generated images and video must be labeled under deep-synthesis rules, and providers must stop generation or transmission of illegal content and act against abusive users. SM015, SM016
CM032 China's 2025 AI-labeling rules take effect on 2025-09-01, require visible labels and technical identifiers for AI-generated content, and forbid deleting, tampering with, fabricating, or concealing those labels. SM017, SM018
CM033 BIS guidance published in May 2026 reaffirmed that licenses are still required for exports of covered advanced-computing items to China or other D:5/Macau headquartered entities even if the recipient is physically outside those jurisdictions. SM019, SM020
CM034 The persistence of U.S. advanced-computing export controls creates ongoing compute-access and cost uncertainty for China-based frontier video-model vendors. SM019, SM020
CM035 OpenAI discontinued the Sora web and app experiences on 2026-04-26 and says the Sora API will be discontinued on 2026-09-24. SM010
CM036 A defensible 2026 global text-to-video SAM proxy is approximately $392 million, calculated as 46.25% of Fortune's $847 million AI video generator market estimate. SM005
CM037 Applying Fortune's 20.9% Asia-Pacific share to that text-to-video slice implies an APAC-attributed text-to-video pool of roughly $82 million in 2026. SM005
CM038 Applying Fortune's 33.88% marketing-and-advertising share to the 2026 AI video generator market implies roughly $287 million of current spend tied to campaign-production workflows. SM005
CM039 Public market data still do not isolate China-only enterprise spend, free-to-paid conversion, API revenue mix, or customer concentration for AI video startups, so top-down TAM remains directional rather than audit-grade. SM001, SM002, SM003, SM004, SM005
CM040 Contradictory regional leadership signals and differently scoped analyst categories should be preserved as diligence caveats rather than normalized into a false-precision single number. SM003, SM004, SM005
CP001 Vidu's closest direct peer set includes Runway, Kling, Hailuo or MiniMax, Pika, Luma, Jimeng, PixVerse, and Wan, alongside historical benchmark pressure from Sora. SP003, SP004, SP005, SP008, SP010, SP011, SP012, SP013, SP015, SP019, SP025
CP002 Artificial Analysis publicly compares Hailuo, Kling, Sora, Vidu Q3 Pro, and Wan within the same video benchmark family, indicating a shared practical comparison set. SP001, SP002, SP003
CP003 a16z wrote in March 2025 that Hailuo, Kling, and Sora debuted on its web rankings, Runway made the Brink List, and Hailuo and Kling had surpassed Sora in monthly visits by January 2025. SP004
CP004 The competitive field splits into professional workflow suites, creator-social apps, China-native mass platforms, API or workflow platforms, and substitutes such as open models or manual production. SP004, SP019, SP024
CP005 Runway markets Creative, Dev, and Robotics platforms and says it is used by more than 60 million creatives. SP005, SP006
CP006 Runway's self-serve pricing ladder currently spans free, $12 Standard, $28 Pro, and $76 Max plans with credits and bundled model access. SP005, SP006
CP007 TechCrunch reported in February 2026 that Runway raised a $315 million Series E at a $5.3 billion valuation. SP007
CP008 Runway's public positioning extends beyond creative generation into developer and robotics surfaces, and TechCrunch says it is expanding from media and advertising into gaming and robotics. SP005, SP007
CP009 Pika positions itself around AI video creation, workflow automation, agents, MCP connectivity, and mobile or viral effects. SP008
CP010 Sacra reports Pika at about $470 million valuation with $135 million raised, a freemium model with paid tiers at $8, $28, and $76 per month, and Adobe Firefly distribution. SP009
CP011 Kling's official surface spans video generation, image generation, sound generation, effects, API, native 4K claims, and iOS and Android distribution. SP010
CP012 Jimeng emphasizes Chinese-language prompt understanding, text or image to video generation, first-frame and last-frame control, and a community remix loop. SP011
CP013 Wan's public site confirms it as an AI video generation model, and Artificial Analysis lists Wan 2.2 and Wan 2.5 alongside Vidu, Hailuo, Kling, and Sora in the same comparison family. SP003, SP012
CP014 Hailuo's official site markets top-tier quality and versatile references, while Sacra says the broader MiniMax stack offers Hailuo-02 1080p video with physics consistency and camera controls plus API access. SP013, SP014
CP015 Sacra reports MiniMax at roughly $4 billion valuation and about $1.15 billion funding, with Hailuo positioned as one consumer and enterprise-video surface inside a broader multimodal company. SP014
CP016 Luma markets itself as a creative-agent platform for professional teams, emphasizing shared context, team workspaces, collaboration, and production-ready workflows. SP015, SP016
CP017 Luma's API exposes Ray3.2 video generation, up to 16 keyframes in one clip, 1080p output, video-to-video up to 20 seconds, and HDR/EXR exports for professional pipelines. SP016, SP017
CP018 Luma's official plans include $30, $90, and $300 monthly tiers with 10,000, 40,000, and 150,000 credits, plus per-video or per-second credit schedules on the pricing page. SP015, SP017
CP019 Owler reports Luma AI has raised about $1.1 billion in total funding and that its latest round was $900 million in November 2025. SP018
CP020 PixVerse's public surfaces include marketing hub, CLI, agent, canvas, lip sync, and enterprise-ready API workflows. SP019, SP021
CP021 PixVerse's docs expose API platform and credit-based pricing or usage schedules, confirming a developer-forward commercialization approach. SP020, SP021
CP022 OpenAI says the Sora web and app experiences were discontinued on April 26, 2026 and the Sora API will be discontinued on September 24, 2026. SP022
CP023 Stability AI's Stable Video Diffusion research confirms that open text-to-video and image-to-video model paths exist outside managed SaaS platforms. SP024
CP024 Runway, Pika, Luma, and PixVerse all offer public self-serve plans or credits, which makes casual experimentation and cross-testing relatively easy. SP006, SP009, SP017, SP020
CP025 Lock-in is likely higher when a platform owns APIs, team workspaces, enterprise commitments, shared credits, or production pipeline context rather than isolated clip generation. SP015, SP016, SP021
CP026 Runway and Luma appear strongest on professional workflow depth, Pika on accessibility or social remix, Kling and Hailuo and Jimeng on China-native distribution, and PixVerse on API plus workflow modularity. SP004, SP005, SP008, SP010, SP011, SP013, SP015, SP019
CP027 Competitive advantage in AI video now depends on workflow context, pricing clarity, distribution, and production tooling in addition to raw generation quality. SP003, SP015, SP019
CP028 Pika's Adobe Firefly distribution gives it an enterprise-adjacent channel even though its brand tone is consumer and creator friendly. SP009
CP029 TechCrunch says Runway has both a recent Adobe partnership and compute expansion via CoreWeave, signaling stronger ecosystem and infrastructure depth than smaller startups. SP007
CP030 Luma's team, business, enterprise, and API surfaces indicate a push toward professional accounts and persistent workflow adoption rather than pure consumer virality. SP015, SP016, SP017
CP031 a16z's ranking suggests China-native video products such as Hailuo and Kling have already become globally visible consumer destinations, reducing the assumption that Western brands dominate top-of-funnel attention. SP004
CP032 Sora's discontinuation weakens OpenAI as a directly monetizing video-platform rival at the run date even though it remains an important quality benchmark. SP004, SP022
CP033 Open or lower-friction alternatives such as Stable Video Diffusion and Wan increase substitute pressure and make consumer-tier pricing less defensible for closed platforms. SP012, SP024
CP034 Some leading competitors are aggregating broader workflow surfaces rather than selling a single model endpoint, which compresses moat claims based only on model access. SP005, SP015, SP019
CP035 Shengshu's plausible right to win is strongest where China-native quality, localization, creator surfaces, and API delivery matter simultaneously, but it faces equally local rivals and better-known global workflow brands. SP010, SP011, SP013, SP015, SP025
CP036 Public competitor disclosures remain uneven because official sites richly describe features but often omit realized enterprise pricing, audited usage, enterprise customer counts, or retention data. SP005, SP010, SP011, SP013, SP015, SP019
CP037 The peer set spans materially different war-chest tiers, with Runway and MiniMax in multibillion-dollar territory, Luma heavily funded, and Pika much smaller by public funding and valuation signals. SP007, SP009, SP014, SP018
CP038 A serious competitor map for Shengshu must include direct peers, open-model substitutes, and the status quo of manual production rather than only startup-to-startup feature comparisons. SP004, SP023, SP024
CI001 Shengshu publicly presents Vidu as both a SaaS and MaaS business with revenue surfaces spanning creator plans, API access, and enterprise-oriented workflow products. SI001, SI004, SI006
CI002 Shengshu's product ecosystem in public company materials includes Vidu MaaS, Vidu SaaS, Vidu App, and Vidu Agent. SI006
CI003 The Vidu API launch disclosed immediate access starting at $10 and base pricing of $0.05 per credit, with a four-second video costing 4 to 40 credits depending on feature type and aspect ratio. SI004, SI025
CI004 Vidu maintains public creator-plan pricing surfaces in both English and Chinese, but the accessible reviewed pages do not expose a fully readable list-price ladder in text. SI002, SI012
CI005 Vidu Agent is explicitly positioned for advertisements, TVCs, music videos, short-form content, and e-commerce product videos, pointing at higher-value commercial budgets. SI005
CI006 Dealroom describes Shengshu's revenue model as primarily subscription based, with a significant portion coming from corporate clients. SI009
CI007 Shengshu's Series A+ announcement says the product ecosystem serves content creators and industry clients globally through MaaS, SaaS, App, and Agent surfaces. SI006
CI008 Shengshu said in its Series A+ release that users and revenue both grew more than 10x in 2025. SI006
CI009 Shengshu said at Global Creativity Week that Vidu serves more than 40 million creators and more than 10,000 developers and enterprise customers, with more than 500 million videos generated. SI010
CI010 Company materials and CNBC reporting support a global reach claim of more than 200 countries and regions and meaningful enterprise-oriented use cases across creative and commercial sectors. SI004, SI005, SI008, SI010
CI011 Shengshu's likely revenue mix includes lower-ARPU creator self-serve spend plus higher-quality API and enterprise revenue. SI004, SI005, SI009
CI012 The API launch shows Shengshu is pursuing both self-serve PLG adoption and a staffed B2B service motion for enterprise integrations. SI004
CI013 Named clients and partners in the Series A+ release include Pollo AI, PhotoGrid, OpenArt, Hubx, Fal.ai, Eachlabs, Freepik, and GensPark, indicating distribution through other software platforms as well as direct end users. SI006
CI014 Company materials position Vidu usage across interactive entertainment, advertising, film, animation, cultural tourism, retail, education, e-commerce, and mobile ads. SI004, SI006
CI015 The API launch explicitly states that Shengshu maintains a dedicated B2B service team to support businesses integrating Vidu into workflows. SI004
CI016 Public Vidu pricing surfaces prove that pricing exists, but the reviewed pages do not provide enough detail to infer realized net pricing, volume discounts, or enterprise contract terms. SI002, SI012, SI023
CI017 AI video businesses like Shengshu almost certainly face substantial compute-related gross-margin pressure because generation is usage metered and resource intensive across public cloud and competitor pricing models. SI014, SI018, SI019, SI020, SI021
CI018 Google's official agent pricing shows multimodal AI usage is charged in metered units with discounts and caching considerations, illustrating how frontier AI services optimize inference economics rather than selling flat-cost delivery. SI014
CI019 MiniMax and Pika both show credit or usage-based pricing logic because heavier or higher-quality AI media generation consumes materially more compute resources. SI018, SI019
CI020 Luma and Runway also use credit-based or tiered usage pricing, reinforcing that the category's monetization model is generally linked to output intensity rather than flat unlimited access. SI020, SI021
CI021 Longer, higher-resolution, or more controlled video outputs consume more credits or cost more across peer pricing pages, implying Shengshu's usable margin is highly sensitive to feature mix and generation quality. SI003, SI018, SI019, SI020
CI022 Vidu's low-friction API entry price broadens the funnel but does not prove strong payback or retention economics because spend depth per account is undisclosed. SI004, SI025
CI023 Runway's recent CoreWeave capacity expansion and large financing round show how serious AI-video vendors need both capital and infrastructure to sustain growth. SI022
CI024 MiniMax's public docs and Sacra profile imply ongoing heavy R&D and inference burden by advertising multimodal breadth, high resolution video, and large-model capabilities. SI017, SI018
CI025 CNBC says Shengshu's April 2026 Series B funding will support development of a general world model bridging digital and physical domains. SI007
CI026 Public sources support at least RMB 2.6 billion of disclosed 2026 capital inflow via a >RMB 600 million Series A+ and a RMB 2 billion Series B. SI006, SI007
CI027 Despite those large rounds, public sources do not disclose Shengshu's cash on hand, monthly burn, runway months, or debt obligations. SI006, SI007
CI028 No audited public revenue, ARR, MRR, or gross-margin disclosure was found in reviewed sources. SI006, SI007, SI009
CI029 Shengshu's strongest traction metrics—10x revenue growth, 40 million creators, 10,000 plus developers and enterprise customers, 500 million videos, and >70% commercial projects—are company-claimed rather than audited. SI006, SI010
CI030 If the >70% commercial-project share is accurate, it is a positive signal for revenue quality because output is skewing toward monetizable use cases rather than pure experimentation. SI010
CI031 Immediate self-serve API access can increase funnel volume, but it may also lower pricing discipline and raise support burden if usage depth and account quality are weak. SI004, SI015
CI032 Vidu Agent and template-driven automation move Shengshu toward advertising, commerce, and brand-production budgets rather than only consumer experimentation. SI005, SI011
CI033 Creator-plan and app surfaces likely support B2C monetization, but public data do not reveal realized payer conversion or ARPU. SI001, SI002, SI006
CI034 Corporate clients, partner platforms, and enterprise customers imply that Shengshu may have a higher-quality revenue mix than a pure consumer video toy if those relationships convert into recurring contracts. SI006, SI009, SI010
CI035 The best current financial verdict is that Shengshu looks meaningfully funded and commercially promising, but still too opaque on revenue quality, margin path, and cash burn for hard underwriting. SI006, SI007, SI009, SI010
CI036 Public-company filing infrastructure such as SEC EDGAR and HKEXnews highlights how much more disciplined public financial disclosure is than the private disclosure currently available for Shengshu. SI015, SI016
CI037 There is no stable public gross-margin proxy for Shengshu because usable output economics depend on resolution, retries, model choice, support burden, and contract mix. SI014, SI019, SI020, SI021
CI038 Margin pressure is a real risk for Shengshu because peer analyses explicitly warn about compute-heavy economics and pricing compression across Chinese AI markets. SI018, SI019
CE001 Vidu's public product surface spans text-to-video, image-to-video, reference-led generation, templates, API access, native audio-video generation, and real-time interaction layers. SE001, SE002, SE003, SE004, SE012
CE002 Vidu Q3 is positioned for finished storytelling output with audio and video generated together, up to 16-second single-pass clips, camera-language control, multilingual output, and use cases such as narrative ads and short series. SE002
CE003 Vidu S1 / stream is positioned as a real-time, voice-driven interactive video model with 540p and 25 FPS generation, voice options, and an API path. SE003, SE013, SE018
CE004 Vidu Agent automates creative planning, shot sequencing, and assembly into complete 15-30 second videos for ads, commerce, and short-form content. SE011
CE005 Shengshu exposes a credible deployment surface through platform.vidu.com, CLI tooling, GitHub repos, and workflow-oriented MaaS updates rather than relying only on a web demo. SE004, SE012, SE018, SE021, SE026
CE006 The Vidu paper describes the model as a diffusion system with U-ViT as its backbone, capable of producing 1080p videos up to 16 seconds in a single generation. SE014, SE002
CE007 The paper reports initial controllable-video experiments including canny-to-video generation, video prediction, and subject-driven generation. SE014
CE008 Vidu 1.5 added multiple-entity consistency, multiple-angle consistency, advanced camera control, stronger semantic understanding, and 1080p output claims. SE008
CE009 Vidu 2.0 emphasized sub-10-second generation, a full-stack inference accelerator, lower claimed cost, and reusable templates. SE007
CE010 TurboDiffusion was co-released with Tsinghua and publicly framed as delivering 100-200x acceleration for video diffusion models, including a Vidu example cutting an 1080p 8-second render from roughly 900 seconds to about 8 seconds. SE010, SE019
CE011 The open-source TurboDiffusion materials identify SageAttention, Sparse-Linear Attention, and rCM as core parts of the acceleration stack. SE010, SE019, SE020
CE012 Shengshu's public open-source and tooling materials provide practical installation or usage detail rather than pure marketing language. SE018, SE019, SE020, SE021
CE013 The vidu-cli repository exposes programmatic task flows for text2video, img2video, headtailimg2video, character2video, lip-sync, and text-to-speech operations. SE021
CE014 vidu-cli documents model versions 3.0, 3.1, 3.2, and 3.2_a with durations extending as high as 16 seconds and 1080p output for supported task types, plus 2K and 4K image-generation options. SE021
CE015 The 2026 MaaS API update added lip sync in 60+ languages, 324 preset voices, 4K support for long videos, creative templates, and MCP integration with tools like Claude and Cursor. SE012
CE016 The API launch shows Shengshu packaging reference-to-video, image-to-video, and text-to-video together as a multimodal enterprise and developer platform. SE006
CE017 In workflow terms, Vidu is best understood as a creative operating stack that turns prompts, references, audio, and templates into publishable video assets or embedded API responses. SE001, SE002, SE004, SE011
CE018 Public product surfaces show Vidu serving several user groups at once, including creators, marketers, enterprise integrators, and interactive-avatar users. SE001, SE002, SE003, SE011, SE012
CE019 Third-party benchmarks place Vidu Q3 among visible global contenders in both text-to-video and image-to-video, supporting technical relevance even if category leadership is contested. SE015, SE016, SE023
CE020 Notebookcheck's independent hands-on review found Vidu visually promising but still too glitch-prone and inconsistent for dependable professional use. SE017
CE021 Vidu's public help surfaces explicitly include content moderation, credits, subscriptions, payments, and blocked-account recovery, showing that operational control layers exist. SE005, SE025
CE022 The Vidu S1 page explicitly warns that the feature involves personal information processing before use. SE003
CE023 Reviewed public sources did not surface SOC 2, ISO 27001, public model cards, red-team reports, or quantified safety-performance disclosures for Vidu. SE001, SE005, SE025
CE024 Deployment evidence spans the API platform, GitHub repos, CLI distribution, help surfaces, and agent-oriented integration points, indicating that Shengshu is building an ecosystem around the core model. SE004, SE012, SE018, SE021, SE025, SE026
CE025 The public release trajectory from subject consistency in 2024 to speed, API access, audio, MCP, and real-time S1 by 2026 forms a coherent maturity arc from experimental generation toward production workflows and live interaction. SE009, SE007, SE006, SE002, SE012, SE013
CE026 Reference-to-video, multi-entity consistency, and first-to-last-frame cinematic transitions are the clearest public examples of Vidu differentiating around continuity rather than generic one-shot prompting. SE008, SE012
CE027 Q3's native audio and narrative positioning reduce downstream stitching and make Vidu more suitable for comic drama, narrative ads, and short-series workflows. SE002
CE028 S1 shifts Vidu beyond offline clip generation toward synchronous digital-human interaction and live content experiences. SE003, SE013, SE018
CE029 Official GitHub repositories for Vidu S1 and vidu-cli, plus the open-source TurboDiffusion repo, provide the required practitioner-accessibility signal that developers can engage with Shengshu's ecosystem directly. SE018, SE019, SE021
CE030 The publicly visible stack depends on high-performance GPUs, specialized attention kernels, acceleration frameworks, and API infrastructure rather than only model weights. SE010, SE019, SE020, SE021
CE031 TurboDiffusion's repo notes that checkpoints and paper are not finalized and that prompt behavior may depend on prompt style and hardware assumptions, so some acceleration claims remain partly experimental. SE019
CE032 The strongest public open-source artifacts are around acceleration and tooling rather than the full proprietary Vidu model, implying that Shengshu is selectively open rather than broadly open-sourcing its core model stack. SE018, SE019, SE021
CE033 Public support and help surfaces exist, but reviewed materials do not reveal the kind of status, incident, or uptime transparency expected from more mature enterprise platforms. SE005, SE025
CE034 No public status page or equivalent uptime-transparency surface was found in reviewed sources. SE001, SE005, SE025
CE035 The best product-tech verdict is that Shengshu has a credible and broad platform with real engineering depth, but its public trust and production-assurance package still lags its product ambition. SE017, SE021, SE023, SE025
CE036 Claims around 1080p output and 16-second generation appear in both the Vidu paper and later public product surfaces, making them more credible than a single isolated marketing statement. SE014, SE002, SE008
CE037 Consistency and camera-control features recur across subject consistency, Vidu 1.5, and later MaaS or Q-series materials, suggesting they are a central product theme rather than a one-off feature claim. SE009, SE008, SE012
CE038 MCP integration makes Vidu easier to embed into agentic workflows because the service can route among video-generation methods without requiring manual API orchestration by the end user. SE012, SE021
CU001 Shengshu's public customer base spans creators, marketers, developers, enterprise customers, partner platforms, and interactive-avatar users rather than a single buyer type. SU001, SU004, SU005, SU011, SU023
CU002 Buyer, user, and payer roles differ across Shengshu's segments: creators often both use and pay, developers integrate the API, and enterprise or commerce teams purchase workflow throughput. SU004, SU005, SU014
CU003 Company materials position Vidu as serving users in more than 200 countries and regions. SU002, SU004
CU004 Shengshu says Vidu serves more than 40 million creators and more than 10,000 developers and enterprise customers. SU002
CU005 Shengshu says more than 500 million videos have been generated on the platform and that more than 70% of output comes from commercial projects. SU002
CU006 Public company materials describe milestone adoption of 1 million users in the first month, 10 million users in three months, more than 100 million videos by month four, and 100 million reference-to-video generations by month eight. SU018, SU019
CU007 Named customer proof is strongest where third-party platforms publicly package Vidu rather than where Shengshu only cites logos or examples. SU002, SU003, SU007, SU008, SU009
CU008 OpenArt publicly exposes a dedicated Vidu video-generator page to its own users, providing live platform-level proof that Vidu is distributed outside Shengshu's first-party surfaces. SU003, SU008
CU009 each::labs publicly offers a Vidu model family with API access in its catalog, which is strong evidence of partner-platform distribution to developers. SU003, SU009, SU026, SU027, SU028
CU010 PhotoGrid is a large creator platform, and Shengshu says PhotoGrid embedded Vidu capabilities; public PhotoGrid surfaces show a relevant AI-video workflow and raw-page inspection surfaced a Vidu Q3 tile. SU002, SU007, SU024
CU011 Pollo AI is a real creator or marketer platform named by Shengshu, but the reviewed Pollo AI public site does not explicitly confirm Vidu as an underlying model. SU002, SU010
CU012 The Lenovo partnership is better interpreted as channel or ecosystem proof than pure end-customer proof because it validates distribution potential more than repeat usage. SU006
CU013 Shengshu's claims about Odin virtual try-on, long-form narrative projects, and other named use cases are strategically interesting but lightly corroborated in reviewed public sources. SU002
CU014 Relative proof quality is highest for OpenArt and each::labs, medium for PhotoGrid, and lower for Pollo AI, Odin, or unnamed production teams. SU007, SU008, SU009, SU010, SU002
CU015 No public NRR, GRR, churn, renewal-rate, contract-length, or cohort-retention data was found in reviewed sources. SU002, SU012, SU014
CU016 The >70% commercial-project share is the strongest public repeat-usage proxy because it suggests Vidu output is tied to real work if measured consistently. SU002
CU017 Public help, billing, credits, and support surfaces imply that Shengshu has repeat-use operational infrastructure rather than only a demo experience. SU012, SU022
CU018 The largest user segment is likely creators, while API, partner-platform, and enterprise channels likely matter more for monetization quality. SU002, SU004, SU014
CU019 Partner platforms such as OpenArt and each::labs can act as land-and-expand channels by exposing Vidu to downstream creator and developer audiences. SU008, SU009, SU018
CU020 Platform-distribution proof is more robust than end-brand proof in Shengshu's current public customer record. SU007, SU008, SU009, SU006
CU021 Top-customer concentration, enterprise ACV, and partner-revenue concentration are not publicly disclosed. SU002, SU003, SU014
CU022 Partner platforms likely introduce concentration and switching risk because they can multi-home across multiple AI-video providers. SU007, SU008, SU009, SU010
CU023 Self-serve creator adoption and partner platforms probably face less procurement friction than direct large-enterprise deployments because Vidu's public trust package remains relatively light. SU008, SU009, SU012
CU024 Vidu's customer use cases span advertising, film, animation, e-commerce, mobile ads, social content, and education, supporting cross-vertical relevance. SU001, SU004, SU005, SU011
CU025 Reliability problems identified by Notebookcheck—such as prompt-following gaps, distortions, and scene inconsistency—could reduce repeat usage or brand trust. SU015
CU026 Shengshu's strongest customer scale and adoption metrics remain company-claimed rather than independently audited. SU002, SU018
CU027 Dealroom's company profile is consistent with a mixed customer base where corporate clients matter alongside broader subscription-like usage. SU014
CU028 Agent, lip-sync, native audio, and MCP integration expand Shengshu's customer value proposition toward commerce teams, creative operations, and agentic-developer workflows. SU005, SU011, SU018
CU029 The customer story is broadest across creative and marketing workflows rather than concentrated in a single vertical. SU001, SU005, SU011
CU030 Public named customer proof is still weaker than a mature enterprise case-study base because many logos are cited without customer-side ROI or deployment detail. SU002, SU003, SU006
CU031 Production-versus-pilot maturity is clear for public platform listings like OpenArt and each::labs, plausible but less explicit for PhotoGrid, and unclear for several cited brand or narrative use cases. SU007, SU008, SU009, SU002
CU032 The public customer record is strongest on top-of-funnel adoption and weakest at the retention and renewal stage. SU002, SU012, SU014, SU015
CU033 Geographic breadth likely reduces country-level concentration risk even though it does not eliminate channel or top-customer concentration risk. SU003, SU004
CU034 Enterprise deployments remain impossible to underwrite fully without contract length, renewal, implementation burden, or support-intensity disclosure. SU004, SU012, SU021
CU035 The best customer verdict is that Shengshu has broad top-of-funnel reach and credible ecosystem distribution, but still lacks public proof on paid durability and concentration quality. SU002, SU008, SU009, SU021
CU036 A plausible land-and-expand path runs from creator discovery into paid features, API usage, partner-platform distribution, and eventually enterprise campaign workflows. SU001, SU004, SU005, SU018
CU037 Procurement friction is likely lower for creators and partner platforms than for direct large enterprises because creators can self-serve while enterprises need more control evidence. SU004, SU012, SU022
CU038 Strategic partner value lies not only in direct revenue but also in distribution leverage, validation, and lower customer-acquisition cost if partners continue surfacing Vidu prominently. SU006, SU008, SU009
CR001 China's 2023 interim measures directly apply to public generative AI video services and require lawful data sources, prohibited-content controls, privacy obligations, complaint handling, and safe stable services. SR003, SR004
CR002 China's 2025 AI-labeling measures require explicit labels on AI-generated video and implicit labels in file metadata, and they also impose responsibilities on app distribution platforms and service providers. SR005, SR006
CR003 Deep-synthesis and algorithmic-recommendation rules remain part of the legal foundation governing labeling, filing, and risk control for AI-video services in China. SR007, SR008, SR009, SR006
CR004 Chinese regulatory remedies for noncompliance can include warnings, demanded corrections, handling by relevant departments, and potentially suspension of the related service. SR003, SR005
CR005 The Chinese regime is not purely prohibitive because the interim measures explicitly combine development encouragement with security obligations. SR003, SR004
CR006 BIS's May 2026 guidance confirms that a license is still required to export covered advanced-computing items to D:5- or Macau-linked entities even when the recipient is located in a third country. SR001, SR002
CR007 The BIS guidance makes ownership and headquarters diligence a live operational requirement because screening must account for the ultimate parent, not just the immediate recipient's location. SR001, SR002
CR008 For Shengshu, export-control risk is economically material because frontier AI video and world-model ambitions depend on high-end compute for both training and inference. SR001, SR016, SR028
CR009 Vidu has visible moderation and enforcement surfaces, including a content-moderation API page and help-center references to blocked accounts, suspicious activity, and terms violations. SR010, SR011
CR010 Moderation and labeling obligations create operational friction because every new content mode—video, audio, avatars, exported files, or partner distribution—needs policy and implementation coverage. SR005, SR006, SR011, SR027
CR011 Chinese rules explicitly require safe, stable, and sustained services, which turns uptime, queueing, and service continuity into regulatory as well as customer-experience obligations. SR003, SR004
CR012 Notebookcheck's hands-on review found Vidu visually impressive but too glitch-prone and inconsistent for dependable professional use, which is direct evidence of product-quality risk. SR012
CR013 Reviewed public sources do not show a mature trust-center style package with public incident history, quantified safety metrics, or detailed enterprise assurance artifacts. SR010, SR011, SR029
CR014 Compute access, compute cost, and acceleration capability are intertwined risks for Shengshu because they influence product quality, speed, gross margin, and competitiveness simultaneously. SR001, SR015, SR016, SR028
CR015 Partner platforms such as OpenArt, each::labs, and PhotoGrid create useful distribution, but also add dependency risk if those channels multi-home or deprioritize Vidu. SR021, SR022, SR023, SR024
CR016 Public sources do not disclose top-customer share, top-partner share, enterprise ACV, or concentration by channel. SR017, SR018, SR025
CR017 Shengshu's public customer proof is stronger in partner platforms than in fully documented enterprise case studies, increasing uncertainty around revenue durability and channel quality. SR021, SR022, SR023, SR018
CR018 Recent fundraising materially reduces immediate solvency pressure, but does not prove capital efficiency or remove the need to translate model lead into durable economics. SR016, SR017, SR025
CR019 Because Shengshu competes in a frontier category with rapid model iteration, competition acts as a risk amplifier on cost, customer acquisition, and benchmark pressure rather than only a market-share concern. SR019, SR020, SR016
CR020 If benchmark position or output quality visibly slips, the impact can cascade into weaker customer retention, slower fundraising, and lower valuation support. SR019, SR020, SR012
CR021 Shengshu's product cadence implies heavy dependence on scarce frontier-model, systems-acceleration, and compliance talent. SR015, SR026, SR027
CR022 The change in public CEO voice from earlier Jiayu Tang-led releases to later Yihang Luo-led releases creates at least a governance and leadership-continuity diligence question. SR013, SR018, SR025, SR030
CR023 Compliance operations must continuously track evolving Chinese requirements across generative AI, deep synthesis, algorithmic recommendation, and labeling rather than treat them as one-time setup work. SR003, SR005, SR007, SR009
CR024 Cross-border commercial expansion is complicated not only by export controls but also by the need to reconcile Chinese operating obligations with global partner and customer expectations. SR001, SR003, SR021, SR022
CR025 Publicly visible support and help surfaces are useful mitigants, but they do not resolve the deeper question of whether Vidu can reliably satisfy demanding commercial workloads. SR010, SR012
CR026 Vidu's audio, lip-sync, and real-time avatar capabilities increase misuse risk around impersonation, harmful content, or privacy-sensitive outputs. SR014, SR026, SR027
CR027 Voice cloning and avatar interactions are especially sensitive because public pages already acknowledge personal-information processing and moderation needs for these features. SR011, SR027
CR028 Quality failures such as inconsistent subjects, incorrect camera execution, or prompt noncompliance can turn directly into refunds, regeneration cost, churn, and reputational damage. SR012, SR026
CR029 Thin enterprise assurance evidence increases procurement friction because security, privacy, and uptime review standards are typically stricter in business deployments than in creator experimentation. SR010, SR011, SR021, SR022
CR030 Public financial opacity means investors still cannot cleanly model whether Shengshu's frontier ambitions are scaling ahead of or behind operating discipline. SR016, SR017, SR025
CR031 Shengshu's major risks are correlated because export controls can pressure compute, compute can pressure product quality and margins, and those can in turn pressure customers and financing. SR001, SR012, SR016, SR019
CR032 The best overall risk judgment is that Shengshu is a high-upside but high-correlation risk asset rather than a simple product-growth story. SR001, SR003, SR012, SR016
CR033 Shengshu does have visible mitigants today, including moderation surfaces, a support layer, large recent funding, and public acceleration work aimed at lowering latency and cost. SR010, SR011, SR015, SR025
CR034 Chinese AI rules explicitly seek to encourage innovation while managing security, so the main risk is execution of compliance rather than an automatic ban on growth. SR003, SR004
CR035 Data provenance and IP exposure remain significant because Chinese rules require lawful sources for training data and noninfringement, while Shengshu's public training-data disclosure remains limited. SR003, SR004
CR036 Limited public detail on training-data governance leaves uncertainty about how much legal or takedown exposure Shengshu could face if scrutiny rises. SR003, SR004, SR013
CR037 Services with strong public-opinion or social-mobilization characteristics may face additional filing or security-assessment obligations under the Chinese rule stack. SR003, SR009
CR038 The 2025 labeling measures extend risk to distribution because app platforms are expected to check whether generative-AI applications have the required labeling materials. SR005, SR006
CR039 TurboDiffusion and related acceleration work partially mitigate compute cost and latency risk, but they also underscore how central specialized infrastructure and systems research are to Shengshu's viability. SR015, SR028
CR040 Partner channels can compress pricing power because multi-model platforms can expose end users to alternative video providers with limited switching friction. SR021, SR022, SR023, SR024
CR041 Practical thesis-break triggers include export denial, regulatory enforcement, benchmark deterioration, material churn or partner delisting, and financing need before durable customer economics are proven. SR001, SR005, SR012, SR016
CR042 Mitigation maturity is uneven: regulation and moderation have visible surfaces, capital has a temporary buffer, but customer concentration, enterprise assurance, and leadership-continuity evidence remain thinner. SR010, SR011, SR016, SR017
CV001 CNBC reported that Shengshu did not disclose valuation alongside the April 2026 Alibaba-led financing round. SV003
CV002 Dealroom labels Shengshu a unicorn and provides a public valuation band of roughly $1 billion to $2.5 billion. SV004
CV003 Public sources support at least RMB 2.6 billion of disclosed 2026 capital inflow via a >RMB 600 million Series A+ and a RMB 2 billion Series B. SV002, SV003
CV004 Shengshu said in its Series A+ release that users and revenue both grew more than 10x in 2025. SV002
CV005 Shengshu said at Global Creativity Week that Vidu serves more than 40 million creators and more than 10,000 developers and enterprise customers, has generated more than 500 million videos, and sees more than 70% of generated content used in commercial projects. SV006
CV006 The Vidu API launch disclosed immediate access starting at $10 and base pricing of $0.05 per credit. SV005
CV007 Vidu Agent is explicitly aimed at ad production, TVCs, short-form content, and e-commerce workflows, indicating a push toward higher-value commercial budgets. SV007
CV008 Artificial Analysis leaderboards place Vidu among actively benchmarked text-to-video and image-to-video models, supporting the view that Shengshu has real product credibility rather than only marketing presence. SV008, SV009
CV009 OpenArt, each::labs, and PhotoGrid provide visible external distribution or integration proof for Vidu. SV010, SV011, SV012
CV010 Advanced-computing controls for China-linked entities remain an active valuation risk because they can affect training and inference access or cost. SV013, SV014
CV011 No public ARR, gross margin, burn, customer concentration, board-level cap table, or liquidation-preference disclosure was found in reviewed Shengshu sources. SV003, SV004, SV028, SV029, SV030
CV012 Public filing systems such as SEC EDGAR and HKEXnews illustrate the disclosure standard that late-stage investors eventually require and that Shengshu has not yet met publicly. SV028, SV029, SV030
CV013 Runway raised $315 million at a $5.3 billion valuation in February 2026 and has raised about $860 million in total. SV016, SV017
CV014 Pika sits far below the top tier of AI-video valuations, with public sources pointing to about a $470 million valuation, about $135 million total funding, and about $7.6 million of 2024 revenue. SV019, SV020
CV015 Luma reached a $4 billion valuation in November 2025 with a $900 million financing round, placing it in the premium band for direct AI-video peers. SV022, SV023
CV016 MiniMax provides a higher-ceiling China-based comparable, with public sources pointing to an earlier $2.5 billion round, a later roughly $4 billion growth mark, and about $1.1 billion total funding. SV024, SV025
CV017 Multiples.vc shows public AI and design-engineering software trading around 4.0x to 4.2x next-twelve-month revenue in August 2026. SV026
CV018 Multiples.vc describes media-and-entertainment software economics as highly mixed, with mature creative-tool subscriptions often at 80%-90% margins and cloud-rendering models nearer 40%-60%. SV027
CV019 Relative to comparables, Shengshu appears stronger than Pika on capital and commercial ambition but less financially proven in public than Runway, Luma, or MiniMax. SV004, SV016, SV020, SV023, SV025
CV020 Direct AI-video private rounds imply a legitimate scarcity premium above public software multiples, but the size of that premium depends on disclosed revenue quality and strategic credibility. SV016, SV017, SV019, SV023, SV024, SV026
CV021 Any specific current Shengshu mark around $1.5 billion to $1.7 billion should be treated as an assumption or analyst shorthand rather than a disclosed public financing fact. SV003, SV004
CV022 A $1.5 billion equity value implies about $150 million of sustainable revenue at 10x, about $125 million at 12x, or about $107 million at 14x. SV026
CV023 A $2.0 billion equity value implies about $200 million of sustainable revenue at 10x, about $167 million at 12x, or about $143 million at 14x. SV026
CV024 If Shengshu’s claimed commercial-project share and developer-plus-enterprise count translate into durable paid usage, a low-unicorn to mid-unicorn valuation band becomes more defensible. SV005, SV006, SV010, SV011
CV025 Because public customer proof is strongest in partner platforms and company statements rather than disclosed retention metrics, revenue quality remains a structured unknown. SV004, SV006, SV009, SV010, SV011, SV012
CV026 Alibaba participation and large recent financing materially reduce near-term survival risk, but they do not answer whether Shengshu already earns software-quality returns on compute and support spend. SV002, SV003
CV027 Shengshu deserves a discount to pure high-margin software on valuation because compute intensity, regulatory obligations, and incomplete disclosure all raise the risk-adjusted cost of capital. SV013, SV014, SV026, SV027
CV028 The positive investment thesis is that Shengshu combines credible product quality, rapid shipping velocity, strategic capital, and commercial workflow relevance in a fast-growing AI-video market. SV001, SV002, SV006, SV007, SV008, SV009
CV029 The anti-thesis is that exact valuation, ARR, margin, burn, retention, and preference-overhang data remain unavailable, making the current private-price debate more narrative-led than evidence-led. SV003, SV004, SV028, SV029, SV030
CV030 A defensible bull case requires Shengshu to prove something like roughly $140 million to $180 million plus of durable commercial revenue, continued category leadership, and better margin quality than the public record currently shows. SV006, SV008, SV009, SV026, SV027
CV031 The current public record most comfortably supports a base-case valuation band around $0.9 billion to $1.6 billion, assuming strong but not yet fully proven commercial economics. SV004, SV026, SV027
CV032 A bear-case band around $0.5 billion to $0.9 billion becomes plausible if revenue quality disappoints or risk discounts rise before disclosure quality improves. SV013, SV014, SV026, SV027
CV033 A 2x outcome from an assumed ~$1.5 billion entry likely requires an exit north of $3 billion, which in turn probably requires Shengshu to close much of the disclosure and scale gap versus Runway, Luma, or MiniMax. SV016, SV023, SV025, SV026
CV034 Shengshu is not yet exit-ready for crossover-style underwriting because no public prospectus-equivalent, audited package, or cap-table disclosure is available. SV028, SV029, SV030
CV035 Reasonable thesis-break triggers are stalled enterprise proof, benchmark deterioration, tighter export controls, regulatory incidents, or continued opacity into the next financing cycle. SV008, SV009, SV013, SV014
CV036 Final diligence must center on ARR bridge, retention and concentration, gross margin, compute commitments, compliance implementation, and the cap-table/preference stack. SV003, SV013, SV028, SV029, SV030
CV037 The best current recommendation is Track rather than buy. SV003, SV004, SV026
CV038 Confidence in that recommendation is medium: the public record is strong enough to bracket a range, but not strong enough to underwrite an exact price with high conviction. SV003, SV004, SV026
CV039 Risk rating is high because product, regulatory, compute, competition, and disclosure risks can all transmit into valuation simultaneously. SV013, SV014, SV027
CV040 Valuation stance is stretched because Shengshu may well deserve a unicorn mark, but the upper end of the visible range still outruns the evidence quality available publicly. SV004, SV017, SV023, SV026
CV041 The recommendation can improve if management provides audited or filing-grade financial evidence, or if the entry price falls enough to create margin of safety despite the opacity. SV003, SV004, SV028, SV029, SV030
CV042 The final valuation verdict is that Shengshu is a compelling company to watch, but not yet a late-stage price that the public record can fully underwrite. SV003, SV004, SV026, SV027
来源
编号出版方标题引文
SO001 Vidu AI Video Generator for Text, Image & Reference Videos | Vidu AI
SO002 CNBC Chinese AI startup Shengshu launches image-to-video tool, rivaling Sora
SO003 CNBC Alibaba leads $290 million investment for building a new kind of AI model as LLM limits emerge
SO004 PR Newswire / ShengShu Technology ShengShu Technology Completes Series A+ Funding of Over RMB 600 Million
SO005 Dealroom Shengshu Technology — Unicorn company profile
SO006 Baiduwiki Beijing Shengshu Technology Co., Ltd._Baiduwiki
SO007 Baiduwiki Vidu(a video large model jointly released by Beijing Shengshu Technology Co., Ltd. and Tsinghua University)_Baiduwiki
SO008 EyeShenzhen Shengshu AI launches video tool globally
SO009 arXiv Vidu: a Highly Consistent, Dynamic and Skilled Text-to-Video Generator with Diffusion Models
SO010 Qiming Venture Partners Portfolio | Qiming Venture Partners
SO011 PR Newswire / ShengShu Technology ShengShu Technology Unveils Vidu S1, Bringing Real-Time Interactive Generation to AI Video
SO012 PR Newswire / ShengShu Technology ShengShu Technology Lays Foundation for Scalable AI Video Generation with Launch of Vidu API Offering Instant Access and Industry-leading Speed for Enterprises & Developers
SO013 PR Newswire / ShengShu Technology Vidu Showcases "China Speed" in Advancing AI Video Into Production at Global Creativity Week
SO014 PR Newswire / ShengShu Technology ShengShu Technology and Tsinghua University Unveil TurboDiffusion, Ushering in the Era of Real-Time AI Video Generation
SO015 PR Newswire / ShengShu Technology Vidu Launches One-Click AI Video Creation Agent to Redefine Ad Production
SO016 PR Newswire / ShengShu Technology ShengShu Technology Selected as 2025 Technology Pioneer by the World Economic Forum
SO017 PR Newswire / ShengShu Technology ShengShu Technology Announces Vidu 2.0, Offering the Industry Fastest Generative Video
SO018 PR Newswire / ShengShu Technology Vidu 1.5 Launch Marks New Emergence in Multimodal AI, to Introduce Groundbreaking Consistency Controls that Reshape the Future of AI Video Production
SO019 PR Newswire / ShengShu Technology Vidu Introduces Subject Consistency in AI Video Creation
SO020 PR Newswire / ShengShu Technology ShengShu Technology Partners with Lenovo to Bring Vidu Generative Video Solution to Lenovo PCs and Smart Hardware Ecosystem
SO021 Notebookcheck AI generated videos with consistent characters and scenes? Hands-on test of Vidu.com
SO022 Artificial Analysis Text to Video Leaderboard - Top AI Video Models
SO023 Artificial Analysis Image to Video Leaderboard - Top AI Video Models
SO024 ShengShu Careers 欢迎加入生数科技
SO025 Vidu Vidu Q3 AI Video Model with Native Audio | Vidu AI
SM001 Research and Markets Global Generative Artificial Intelligence (AI) in Video Creation Market Report 2026
SM002 Research and Markets Global Artificial Intelligence (AI) Video Generator Market Report 2026
SM003 The Business Research Company Global Generative Artificial Intelligence (AI) in Video Creation Market Report 2026
SM004 The Business Research Company Global Artificial Intelligence (AI) Video Generator Market Report 2026
SM005 Fortune Business Insights AI Video Generator Market Size, Share | Growth Report [2034]
SM006 Andreessen Horowitz Top 100 Gen AI Consumer Apps - 4th Edition
SM007 Artificial Analysis Video Model Comparisons
SM008 Runway Runway | Building Real-World Intelligence
SM009 Runway AI Image and Video Pricing from $12/month | Runway AI
SM010 OpenAI What to know about the Sora discontinuation
SM011 Pika Pika
SM012 Kling AI Kling AI: Next-Gen AI Video & Image Generator
SM013 Jimeng AI 即梦AI - 即刻造梦
SM014 Wan AI Wan AI: Leading AI Video Generation Model
SM015 Regulations.AI Interim Measures for the Administration of Generative AI Services - Overview
SM016 China Law Translate 生成式人工智能服务管理暂行办法 / Generative AI Interim Measures
SM017 Xinhua / State Council Information Office China requires labeling of AI-generated online content
SM018 Harris Sliwoski China's New AI Labeling Rules: What Every China Business Needs to Know
SM019 Bureau of Industry and Security Homepage | Bureau of Industry and Security
SM020 Greenberg Traurig BIS Clarifies License Requirement for Advanced Computing Items to D:5-Headquartered Entities
SM021 PixVerse Platform Docs Pricing - PixVerse Platform Docs
SM022 PixVerse Platform PixVerse Platform - One of the best AI video API provided by PixVerse
SM023 PixVerse The Generative Core Behind Digital Worlds and Experiences | PixVerse
SM024 YouTube 2024 U.S. YouTube Impact Report
SM025 Pew Research Center Americans’ Social Media Use
SP001 Artificial Analysis Text to Video Leaderboard - Top AI Video Models
SP002 Artificial Analysis Image to Video Leaderboard - Top AI Video Models
SP003 Artificial Analysis Video Model Comparisons
SP004 Andreessen Horowitz Top 100 Gen AI Consumer Apps - 4th Edition
SP005 Runway Runway | Building Real-World Intelligence
SP006 Runway AI Image and Video Pricing from $12/month | Runway AI
SP007 TechCrunch AI video startup Runway raises $315M at $5.3B valuation, eyes more capable world models
SP008 Pika Pika
SP009 Sacra Pika valuation, funding & news
SP010 Kling AI Kling AI: Next-Gen AI Video & Image Generator
SP011 Jimeng AI 即梦AI - 即刻造梦
SP012 Wan AI Wan AI: Leading AI Video Generation Model
SP013 Hailuo AI Hailuo AI: MiniMax H3 LIVE NOW. Top-Tier Quality, Versatile References
SP014 Sacra MiniMax valuation, funding & news
SP015 Luma Creative agents for creative professionals | Luma
SP016 Luma Build with Luma APIs | Luma
SP017 Luma Plans & Pricing | Luma
SP018 Owler Luma AI Funding
SP019 PixVerse The Generative Core Behind Digital Worlds and Experiences | PixVerse
SP020 PixVerse Platform Docs Pricing - PixVerse Platform Docs
SP021 PixVerse Platform PixVerse Platform - One of the best AI video API provided by PixVerse
SP022 OpenAI What to know about the Sora discontinuation
SP023 CNBC Chinese AI startup Shengshu launches image-to-video tool, rivaling Sora
SP024 Stability AI Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets
SP025 Vidu AI Video Generator for Text, Image & Reference Videos | Vidu AI
SI001 Vidu AI Video Generator for Text, Image & Reference Videos | Vidu AI
SI002 Vidu Vidu Pricing Plans for AI Video Generation | Vidu AI
SI003 Vidu Vidu API
SI004 PR Newswire / ShengShu Technology ShengShu Technology Lays Foundation for Scalable AI Video Generation with Launch of Vidu API Offering Instant Access and Industry-leading Speed for Enterprises & Developers
SI005 PR Newswire / ShengShu Technology Vidu Launches One-Click AI Video Creation Agent to Redefine Ad Production
SI006 PR Newswire / ShengShu Technology ShengShu Technology Completes Series A+ Funding of Over RMB 600 Million
SI007 CNBC Alibaba leads $290 million investment for building a new kind of AI model as LLM limits emerge
SI008 CNBC Chinese AI startup Shengshu launches image-to-video tool, rivaling Sora
SI009 Dealroom Shengshu Technology — Unicorn company profile
SI010 PR Newswire / ShengShu Technology Vidu Showcases China Speed in Advancing AI Video Into Production at Global Creativity Week
SI011 Vidu Vidu Q3 AI Video Model with Native Audio | Vidu AI
SI012 Vidu CN Vidu AI视频生成定价方案
SI013 Vidu CN Platform Vidu API 平台
SI014 Google Cloud Cost of building and deploying AI models in Agent Platform
SI015 SEC EDGAR Adobe Inc. 10-K XBRL Viewer
SI016 HKEXnews Listed Company Information Title Search
SI017 MiniMax API Docs Models - MiniMax API Docs
SI018 Sacra MiniMax valuation, funding & news
SI019 Sacra Pika valuation, funding & news
SI020 Luma Plans & Pricing | Luma
SI021 Runway AI Image and Video Pricing from $12/month | Runway AI
SI022 TechCrunch AI video startup Runway raises $315M at $5.3B valuation, eyes more capable world models
SI023 Kling AI Developer Platform Kling AI: Next-Generation AI Creative Studio
SI024 Notebookcheck AI generated videos with consistent characters and scenes? Hands-on test of Vidu.com
SI025 Vidu Vidu API
SE001 Vidu AI Video Generator for Text, Image & Reference Videos | Vidu AI
SE002 Vidu Vidu Q3 AI Video Model with Native Audio | Vidu AI
SE003 Vidu Vidu S1 AI Video Model | Vidu AI
SE004 Vidu Vidu API
SE005 Vidu Vidu Help Center: AI Video Guides, Billing & Support
SE006 PR Newswire / ShengShu Technology ShengShu Technology Lays Foundation for Scalable AI Video Generation with Launch of Vidu API Offering Instant Access and Industry-leading Speed for Enterprises & Developers
SE007 PR Newswire / ShengShu Technology ShengShu Technology Announces Vidu 2.0, Offering the Industry Fastest Generative Video
SE008 PR Newswire / ShengShu Technology Vidu 1.5 Launch Marks New Emergence in Multimodal AI, to Introduce Groundbreaking Consistency Controls that Reshape the Future of AI Video Production
SE009 PR Newswire / ShengShu Technology Vidu Introduces Subject Consistency in AI Video Creation
SE010 PR Newswire / ShengShu Technology ShengShu Technology and Tsinghua University Unveil TurboDiffusion, Ushering in the Era of Real-Time AI Video Generation
SE011 PR Newswire / ShengShu Technology Vidu Launches One-Click AI Video Creation Agent to Redefine Ad Production
SE012 PR Newswire / ShengShu Technology ShengShu Technology Announces Updates to Vidu's Model-as-a-Service (MaaS) API, with Lip Sync, Templates, and MCP
SE013 PR Newswire / ShengShu Technology ShengShu Technology Unveils Vidu S1, Bringing Real-Time Interactive Generation to AI Video
SE014 arXiv Vidu: a Highly Consistent, Dynamic and Skilled Text-to-Video Generator with Diffusion Models
SE015 Artificial Analysis Video Arena - Top AI Video Models | Artificial Analysis
SE016 Artificial Analysis Video Arena - Top AI Video Models | Artificial Analysis
SE017 Notebookcheck AI generated videos with consistent characters and scenes? Hands-on test of Vidu.com
SE018 GitHub / ShengShu GitHub - shengshu-ai/Vidu-S1: Vidu S1: A Real-Time Interactive Video Generation Model
SE019 GitHub / Tsinghua ML GitHub - thu-ml/TurboDiffusion: TurboDiffusion: 100–200× Acceleration for Video Diffusion Models
SE020 GitHub / Tsinghua ML GitHub - thu-ml/SageAttention: Quantized attention acceleration repository
SE021 GitHub / ShengShu GitHub - shengshu-ai/vidu-cli: a client for vidu
SE022 CNBC Chinese AI startup Shengshu launches image-to-video tool, rivaling Sora
SE023 PR Newswire / ShengShu Technology Vidu Showcases China Speed in Advancing AI Video Into Production at Global Creativity Week
SE024 Dealroom Shengshu Technology — Unicorn company profile
SE025 Intercom / Vidu Team Home | Vidu Help Center
SE026 Vidu CN Platform Vidu API 平台
SU001 Vidu AI Video Generator for Text, Image & Reference Videos | Vidu AI
SU002 PR Newswire / ShengShu Technology Vidu Showcases China Speed in Advancing AI Video Into Production at Global Creativity Week
SU003 PR Newswire / ShengShu Technology ShengShu Technology Completes Series A+ Funding of Over RMB 600 Million
SU004 PR Newswire / ShengShu Technology ShengShu Technology Lays Foundation for Scalable AI Video Generation with Launch of Vidu API Offering Instant Access and Industry-leading Speed for Enterprises & Developers
SU005 PR Newswire / ShengShu Technology Vidu Launches One-Click AI Video Creation Agent to Redefine Ad Production
SU006 PR Newswire / ShengShu Technology ShengShu Technology Partners with Lenovo to Bring Vidu's Generative Video Solution to Lenovo PCs and Smart Hardware Ecosystem
SU007 PhotoGrid AI Video Generator – Seedance, Veo, Kling & More | PhotoGrid
SU008 OpenArt Vidu Video Generator - Text & Image to Video in Seconds
SU009 each::labs Vidu Models & APIs | each::labs
SU010 Pollo AI Pollo AI: The Ultimate AI Creative Suite for Marketers & Creators
SU011 Vidu Vidu Q3 AI Video Model with Native Audio | Vidu AI
SU012 Vidu Vidu Help Center: AI Video Guides, Billing & Support
SU013 CNBC Chinese AI startup Shengshu launches image-to-video tool, rivaling Sora
SU014 Dealroom Shengshu Technology — Unicorn company profile
SU015 Notebookcheck AI generated videos with consistent characters and scenes? Hands-on test of Vidu.com
SU016 Artificial Analysis Video Arena - Top AI Video Models | Artificial Analysis
SU017 Artificial Analysis Video Arena - Top AI Video Models | Artificial Analysis
SU018 PR Newswire / ShengShu Technology ShengShu Technology Announces Updates to Vidu's Model-as-a-Service (MaaS) API, with Lip Sync, Templates, and MCP
SU019 PR Newswire / ShengShu Technology ShengShu Technology Announces Vidu 2.0, Offering the Industry Fastest Generative Video
SU020 PR Newswire / ShengShu Technology Vidu 1.5 Launch Marks New Emergence in Multimodal AI, to Introduce Groundbreaking Consistency Controls that Reshape the Future of AI Video Production
SU021 Vidu Vidu API
SU022 Intercom / Vidu Team Home | Vidu Help Center
SU023 Vidu Vidu S1 AI Video Model | Vidu AI
SU024 PhotoGrid Online AI Photo Editor & Free Collage Maker
SU025 Vidu CN Platform Vidu API 平台
SU026 each::labs vidu-q2 API Models | each::labs
SU027 each::labs vidu-2.0 API Models | each::labs
SU028 each::labs vidu-q1 API Models | each::labs
SR001 Bureau of Industry and Security Guidance Regarding Enforcement of License Requirements for Advanced Computing Items for Entities Headquartered in Country Group D:5 and Macau
SR002 Greenberg Traurig Enforcement Pause Has Limits: BIS Clarifies Ongoing License Requirement for Advanced Computing Items to China-Linked Entities
SR003 CAC 生成式人工智能服务管理暂行办法
SR004 China Law Translate Interim Measures for the Management of Generative Artificial Intelligence Services
SR005 CAC 关于印发《人工智能生成合成内容标识办法》的通知
SR006 China Law Translate Measures for Labeling of AI-Generated Synthetic Content
SR007 The State Council / gov.cn 互联网信息服务深度合成管理规定
SR008 China Law Translate Provisions on the Administration of Deep Synthesis Internet Information Services
SR009 China Law Translate Provisions on the Management of Algorithmic Recommendations in Internet Information Services
SR010 Vidu Vidu Help Center: AI Video Guides, Billing & Support
SR011 Vidu API Content Moderation | Vidu API
SR012 Notebookcheck AI generated videos with consistent characters and scenes? Hands-on test of Vidu.com
SR013 PR Newswire / ShengShu Technology ShengShu Technology Lays Foundation for Scalable AI Video Generation with Launch of Vidu API Offering Instant Access and Industry-leading Speed for Enterprises & Developers
SR014 PR Newswire / ShengShu Technology ShengShu Technology Announces Updates to Vidu's Model-as-a-Service (MaaS) API, with Lip Sync, Templates, and MCP
SR015 PR Newswire / ShengShu Technology ShengShu Technology and Tsinghua University Unveil TurboDiffusion, Ushering in the Era of Real-Time AI Video Generation
SR016 CNBC Alibaba leads $290 million investment for building a new kind of AI model as LLM limits emerge
SR017 Dealroom Shengshu Technology — Unicorn company profile
SR018 PR Newswire / ShengShu Technology Vidu Showcases China Speed in Advancing AI Video Into Production at Global Creativity Week
SR019 Artificial Analysis Video Arena - Top AI Video Models | Artificial Analysis
SR020 Artificial Analysis Video Arena - Top AI Video Models | Artificial Analysis
SR021 OpenArt Vidu Video Generator - Text & Image to Video in Seconds
SR022 each::labs Vidu Models & APIs | each::labs
SR023 PhotoGrid AI Video Generator – Seedance, Veo, Kling & More | PhotoGrid
SR024 Pollo AI Pollo AI: The Ultimate AI Creative Suite for Marketers & Creators
SR025 PR Newswire / ShengShu Technology ShengShu Technology Completes Series A+ Funding of Over RMB 600 Million
SR026 Vidu Vidu Q3 AI Video Model with Native Audio | Vidu AI
SR027 Vidu Vidu S1 AI Video Model | Vidu AI
SR028 GitHub / Tsinghua ML GitHub - thu-ml/TurboDiffusion: TurboDiffusion: 100–200× Acceleration for Video Diffusion Models
SR029 Vidu AI Video Generator for Text, Image & Reference Videos | Vidu AI
SR030 PR Newswire / ShengShu Technology Vidu Launches One-Click AI Video Creation Agent to Redefine Ad Production
SV001 Vidu AI Video Generator for Text, Image & Reference Videos | Vidu AI
SV002 PR Newswire / ShengShu Technology ShengShu Technology Completes Series A+ Funding of Over RMB 600 Million
SV003 CNBC Alibaba leads $290 million investment for building a new kind of AI model as LLM limits emerge
SV004 Dealroom Shengshu Technology — Unicorn company profile
SV005 PR Newswire / ShengShu Technology ShengShu Technology Lays Foundation for Scalable AI Video Generation with Launch of Vidu API Offering Instant Access and Industry-leading Speed for Enterprises & Developers
SV006 PR Newswire / ShengShu Technology Vidu Showcases China Speed in Advancing AI Video Into Production at Global Creativity Week
SV007 PR Newswire / ShengShu Technology Vidu Launches One-Click AI Video Creation Agent to Redefine Ad Production
SV008 Artificial Analysis Text to Video Leaderboard - Top AI Video Models
SV009 Artificial Analysis Image to Video Leaderboard - Top AI Video Models
SV010 OpenArt Vidu Video Generator - Text & Image to Video in Seconds
SV011 each::labs Vidu Models & APIs | each::labs
SV012 PhotoGrid AI Video Generator – Seedance, Veo, Kling & More | PhotoGrid
SV013 Bureau of Industry and Security Guidance Regarding Enforcement of License Requirements for Advanced Computing Items for Entities Headquartered in Country Group D:5 and Macau
SV014 Greenberg Traurig Enforcement Pause Has Limits: BIS Clarifies Ongoing License Requirement for Advanced Computing Items to China-Linked Entities
SV015 Runway AI Image and Video Pricing from $12/month | Runway AI
SV016 TechCrunch AI video startup Runway raises $315M at $5.3B valuation, eyes more capable world models
SV017 Tracxn Runway funding and investors
SV018 Pika Pika
SV019 Sacra Pika valuation, funding & news
SV020 GetLatka Pika Revenue 2024: $7.6M ARR, $470M Valuation
SV021 Luma Plans & Pricing | Luma
SV022 Owler Luma AI Funding
SV023 CNBC Luma AI raises $900 million in funding round led by Saudi AI firm Humain
SV024 Sacra MiniMax valuation, funding & news
SV025 InforCapital MiniMax - AI Model Development, $1.1B Raised | InforCapital
SV026 Multiples.vc Public Software Valuation Multiples — August 2026 - Multiples.vc - Public Comps and Valuation Multiples
SV027 Multiples.vc Media & Entertainment Software Sector Overview
SV028 SEC EDGAR EDGAR Search Results
SV029 SEC EDGAR EDGAR Search Results
SV030 HKEXnews Listed Company Information Title Search