HiDream.ai
有战略资本背书、也有真实产品验证的多模态 AI 平台,但公开证据还不足以证明当前独角兽以上定价在经济上站得住
HiDream.ai 有足够技术和战略含金量,值得继续深挖;但当前超过独角兽的估值,更清楚由融资动能和产品证明支撑,而不是公开经济性,证据量仍停在继续研究区间。
封面要素
公司概况
HiDream.ai,正式名称为 Beijing Zhixiang Future Technology Co., Ltd.,是一家总部在北京的多模态 AI 公司。公司于 2023 年 3 月由 Tao Mei 及一支横跨研究与商业化的运营团队创立。公开证据显示,公司产品栈覆盖基础模型、企业 API、营销工作流、影视制作工具和创作者产品;2026 年 7 月,RMB 1.5B 的 C 轮把公司推入独角兽阵营,近期融资总额超过 RMB 2.1B。投资逻辑由基准测试可信度、开源分发,以及罕见的战略性国资与传媒产业资本共同支撑。尽调难点在经济透明度:公开来源仍未披露收入质量、毛利率、烧钱速度或优先股结构,无法证明当前估值的风险已充分出清。
- 成立时间
- 2023-03-02
- 创始人
- Tao Mei, Wang Ke
- 创立地点
- Beijing, China
- 总部
- Beijing, China
- 产品
- HiDream 通过 HiHarness 提供企业 API 和多模态基础模型,通过 HiBurst 提供营销工作流,通过 Zhenzan 支持影视制作协作,并用 vivago 及相关工具服务消费级创作者。
- 客户
- 使用多模态生成和工作流工具的企业、媒体与娱乐合作方、营销团队、创作者和开发者。
- 商业模式
- B2B 与 B2C 混合变现,覆盖企业 API 服务、模型即服务部署、媒体与营销工作流,以及面向创作者的订阅或用量型产品。
- 阶段
- Series C private company
- 融资情况
- 2026 年 7 月 C 轮融资 RMB 1.5B;此前约三个月内已完成两轮融资,使近期融资总额超过 RMB 2.1B。除独角兽状态外,确切投后估值仍未披露。
执行摘要
主要优势
- 国资、银行系和影视产业资本围绕清晰的多模态 AI 平台命题组成强战略财团。
- 开源发布、基准曝光、第三方托管和可见企业 / API 分发信号,构成真实产品证明。
- API、媒体工作流和创作者工具覆盖面较广,让 HiDream 不只是单点功能 demo 故事。
主要风险
- 公开证据仍未披露经审计收入、利润率、现金消耗、留存或 cap table 条款,无法精确支撑价格。
- 当前估值只有在战略合作和活跃度指标转化为高质量经常性变现时才站得住。
- 2026 年 AI 资本市场仍开放但更挑剔;若增长证明落后于估值,不透明公司下行更大。
- 开源分发和激进生态扩张可能扩大触达,同时削弱稀缺性和定价权。
未决问题
- 准确 post-money 估值、清算优先权、反稀释保护,以及任何内部二级出售压力仍未披露。
- 尚无公开收入桥,把用户、关键客户或 API 调用量连接到经常性企业收入、留存或毛利率。
- 没有管理层财务、现金流细节和算力成本披露,DCF 级模型仍无法建立。
- 与 Shanghai Film 和 Huace 的战略渠道合作,还需要合同价值和续约证据,才能支撑溢价估值。
目录
01公司概况
1.1 身份定位、产品触点与运营范围
HiDream.ai 的官方材料把公司定义为全球多模态生成式 AI 平台,而不是单一模型实验室。About 页面称,公司成立于 2023 年 3 月,目标是打造多模态基础模型和 AI 应用;首页则强化「AI 与创意相遇」这一更宽的身份,覆盖文本、图像、视频和 3D。产品集很关键,因为它已经拆成可变现入口,而不只是研究叙事。HiHarness 定位在企业 MaaS 层,拥有 200 多个 API、100 多个重点客户和 5000 多亿次 API 调用;HiBurst 面向营销工作流;Zhenzan 面向专业影视制作;vivago 是大众创作者应用;HiDreamFans 指向线下全息营销硬件。这样的宽度支撑了一个判断:HiDream 正试图成为全栈视觉 AI 操作层。它也解释了公司为何营销「原生全模态世界模型」:现有产品集已经把模型开发接到营销、娱乐和创作者场景,而这些场景中,时间和空间一致性直接影响商业价值。[CO001, CO003, CO004, CO005, CO006, CO007]
| 指标 | 数值 / 状态 | 日期 | 置信度 | 缺口 |
|---|---|---|---|---|
| 成立时间 | March 2023 | 2023-03-02 | 高 | |
| 注册总部 | 北京海淀区 | 2026-08-11 | 高 | |
| 最新披露轮次 | RMB 1.5B Series C 轮 | 2026-07-27 | 高 | |
| 近期融资总额 | >RMB 2.1B,三轮合计 | 2026-07-31 | 高 | 适用于此前三个月窗口,不是累计融资。 |
| 估值状态 | 独角兽 / >$1B | 2026-07-24 | 高 | 未公开披露精确投后估值。 |
| 专业用户口径 | 50M+ | 2026-07-27 | 中 | 公司声称的使用量数字,不是经审计的 MAU 定义。 |
| 企业客户口径 | 40K+ | 2026-07-27 | 中 | 公司声称的覆盖面,未披露付费账户。 |
| 重点客户口径 | 100+ | 2026-08-11 | 中 | 官网使用的是重点客户,不是企业客户总数。 |
| API 调用口径 | 500B+ | 2026-08-11 | 中 | 官网没有披露计量窗口。 |
| 公开员工数 | 无法可靠核验 | 2026-08-11 | 低 | 登记快照相互冲突,不能可靠代理员工数。 |
各行混合了登记事实、融资披露和公司声称的经营指标;缺口列标出缺乏支撑的私营公司披露。
[CO001, CO002, CO009, CO010, CO019, CO023]当前 HiDream.ai 叙事里,身份、产品、客户、资本和监管负担如何相互连接。
[CO004, CO007, CO019, CO020, CO025, CO031]截至运行日期,公开资料能支撑的成熟度和牵引力指标。
用户和 API 调用指标是公司披露的运营指标,不是经审计披露。
[CO001, CO019, CO023, CO026, CO032, CO035]1.2 创始人锚点、法律实体结构与治理不透明
公开层面最清晰的领导锚点是 Tao Mei。官网、Baidu 资料和 IEEE 作者页都将其列为创始人兼 CEO,独立传记来源也把他与 JD.com 和 Microsoft Research 的过往高管经历联系起来。这一背景有助于解释公司为何同时强调大模型研究和应用商业化。不过,公开治理细节比创始人履历薄得多。官网显示 Wang Ke 为联合创始人兼 COO,工商记录则显示 Wang Ke 是北京运营实体的经理和法定代表人,Tao Mei 列为董事,并似乎是实际控制人或受益所有人。Qichacha 还显示,北京公司由一家合肥控股实体全资持有,这与地方城市资本支持的叙事方向一致,但仍不能替代清晰的股权结构表或董事会名单。公开证据没有披露当前董事会名单、委员会结构或决策权框架,因此投资判断仍高度依赖以创始人为中心来理解组织。[CO002, CO011, CO012, CO013, CO014, CO015]
| 人物 | 职务 | 背景 | 创始人-市场匹配或职能覆盖 | 关键人依赖 |
|---|---|---|---|---|
| Tao Mei | 创始人兼 CEO | 曾任 JD.com 副总裁、Microsoft Research 负责人;兼具学术和产业 AI 背景。 | 为多模态 AI 平台串起研究可信度、产品策略和投资人叙事。 | 高 |
| Wang Ke | 联合创始人 / COO 信号;北京实体法定代表人兼经理 | 官网称其为联合创始人兼 COO;登记信息显示其担任经营实体管理职务。 | 可能负责日常运营落地和外部合作,但公开资料未完整披露职责范围。 | 中 |
| Chen Ruizhe | 监事(登记) | Qichacha 将其列为北京实体监事。 | 在实体层面提供最低限度的法定治理层,但公开董事会可见度仍薄。 | 低 |
| Jiao Yang | 财务负责人(登记) | Qichacha 将其列为财务负责人。 | 显示公司配有财务职能,但不能证明公开财务披露质量。 | 低 |
本表只纳入已审阅来源中可见的公开领导层和登记职务;HiDream.ai 未公布完整董事会或管理层名单。
[CO012, CO013, CO014, CO015, CO017]1.3 资本基础、投资者组合与规模信号
公司 2026 年 7 月的融资节奏,是公开记录中最强的后期信号。多个来源一致显示,HiDream.ai 在 7 月下旬完成 RMB 1.5B 的 C 轮融资,并在约三个月内完成三轮融资,累计超过 RMB 2.1B。投资者组合格外有信息量。本轮集合了国家和省级国资基金、银行系资本,以及 Shanghai Film New Vision Fund、Huace Film & TV 等传媒产业投资者。这样的组合意味着,投资人判断 HiDream 时,更像是在投具有直接产业应用路径的国产 AI 基础设施,而不是普通应用初创公司。公开规模信号存在,但不均衡。融资新闻稿称公司拥有 5000 多万专业用户、4 万多家企业客户,覆盖 100 多个国家;首页则突出 100 多个重点客户和 5000 多亿次 API 调用。这些都是商业活动的有用指标,但仍是公司口径数字,没有审计财务、队列细节,也没有可靠员工数披露。[CO019, CO020, CO021, CO022, CO023, CO024]
| 利益相关方 | 角色 | 控制权或经济重要性 | 尽调问题 |
|---|---|---|---|
| National Social Security Fund Sichuan Revitalization Sci-Tech Innovation Fund(国家社保基金四川振兴科创基金) | Series C 轮共同领投方 | 说明国家级战略资本在支持基础模型基础设施。 | 厘清持股比例和任何治理权。 |
| ICBC Capital | Series C 轮共同领投方 | 增加大型银行系机构的可信度和资本纵深。 | 确认支持是否纯财务,还是包括融资设施。 |
| Hongyi Asset Management / Dunhong Capital 资本方 | Series C 轮共同领投方 | 与国资基金一起锚定本轮融资,并拓宽投资人财团深度。 | 索取准确出资规模和后续跟投预期。 |
| Hefei Industrial Investment | 继续跟投投资人 | 多次跟投支撑合肥产业政策协同叙事。 | 厘清合肥持股结构反映的是地方政策条件,还是经营实质。 |
| Shanghai Film New Vision Fund 电影产业基金 | 新战略投资人 | 把资本和电影制作落地场景直接连起来。 | 确认数据权利、排他性和项目管线承诺。 |
| Huace Film & TV | 新战略投资人 | 增加内容和视听产业分发杠杆。 | 审查任何与投资绑定的商业最低承诺或优先访问权。 |
| Zhixiang Future (Hefei) Technology Co., Ltd. 合肥实体 | 实体层面股东 | Qichacha 将合肥实体列为北京运营公司的唯一股东。 | 获取北京实体以上层级的股权结构表和最终受益所有权瀑布。 |
各行汇总公开备案和融资报道中可见的投资人或控制权实体;不能替代完整股权结构表。
[CO016, CO020, CO021, CO022, CO023, CO029]1.4 里程碑、合作关系与早期监管压力
对一家 2023 年才成立的公司来说,HiDream.ai 的公开里程碑记录相当密集。官网突出了一系列快速产品和合作进展:HiDream-O1-Image 的基准测试胜出、O1 开源发布周期、与 Shanghai Film 和 Huace 的战略合作、BlueFocus 与 Tencent Cloud 合作,以及 vivago 2.0 和 TikTok 相关营销认可。独立报道补充了有用细节。NetEase 和 Jiemian 都描述了与大型影视集团的战略投资关系,China Biz Insider 则描述了通过 HiBurst、vivago 和 Zhenzan 的具体商业化。结果是,公司已经在尝试把 AI 能力嵌入影视、营销和跨境电商工作流。主要反向权重来自监管,而不是诉讼。中国的生成式 AI 和合成内容标识规则并不专门针对 HiDream,但任何大规模分发多模态合成媒体的公司都会面对更高合规门槛。公司未来路线图明确指向世界模型和更广的娱乐部署,因此即便公开层面还没有公司特定执法,这一政策负担也很重要。[CO029, CO030, CO031, CO032, CO033, CO034]
| 日期 | 事件 | 类型 | 金额 / 估值 / 状态 | 参与方 | 含义 |
|---|---|---|---|---|---|
| 2023-03-02 | 北京运营实体成立 | 成立 | 实体设立 | Beijing Zhixiang Future Technology | 为 HiDream.ai 运营建立法律基础。 |
| 2024-06-01 | 与 Ciwen Media、Shanghai Film 的战略合作进入公开记录 | 合作 | 已宣布 | HiDream.ai、Ciwen Media、Shanghai Film 等合作方 | 显示早期电影产业整合。 |
| 2025-05-01 | 较早期融资历史的大致情况进入公共数据库 | 融资 | InforCapital 档案显示累计 $73M | Oriental Fortune 及其他早期投资人 | 说明公司在 2026 年巨额轮次之前已经开始资本化。 |
| 2026-05-08 | HiDream-O1-Image 开源发布 | 产品 | 8B 开源模型发布 | HiDream.ai GitHub 仓库 | 让开发者和研究人员看见模型架构。 |
| 2026-05-14 | HiDream-O1-Image-Dev-2604 开源 | 产品 | Dev-2604 发布 | HiDream.ai GitHub 仓库 | 延伸社区牵引力和基准存在感。 |
| 2026-06-11 | Science and Technology Daily 报道全球图像基准 #2 结果 | 规模 | 6 月快照中 Elo 为 1265 | HiDream.ai / Artificial Analysis 榜单 | 强化外部对图像模型质量的验证。 |
| 2026-06-14 | Shanghai Film 战略投资合作宣布 | 合作 | 战略投资合作 | Shanghai Film Co., Ltd. | 将资本与电影产业落地连接起来。 |
| 2026-06-07 | Huace 战略投资合作宣布 | 合作 | 战略投资合作 | Huace Film & TV | 扩大视听产业渠道和内容访问。 |
| 2026-07-24 | 公开报道称 HiDream.ai 跨过独角兽门槛 | 规模 | >$1B 估值状态 | 多家融资报道媒体 | 标志公司进入后期私营公司状态。 |
| 2026-07-27 | Series C 轮以 RMB 1.5B 完成 | 融资 | RMB 1.5B;近期合计 >RMB 2.1B | 国资基金、银行、媒体投资人 | 为全模态世界模型推进提供资本。 |
该时间线仅限公开可核验里程碑;公司事件的准确日期不清楚时,使用公开发布日期。
[CO002, CO019, CO023, CO024, CO029, CO030]按日期列出成立、开源发布、融资、战略合作,以及邻近监管的里程碑。
时间线使用公开发布日期;内部决策日期未披露时不作推断。
[CO002, CO019, CO023, CO024, CO029, CO032]1.5 展项
02市场分析
2.1 界定真实市场边界
不能不加边界逻辑地用整个生成式 AI 头部市场来承销 HiDream。公开证据显示,它的商业覆盖更具体:通过 HiHarness 做企业 API,通过 HiBurst 做营销生产软件,通过 Zhenzan 做娱乐工作流,通过 vivago 做创作者订阅。这些业务入口都处在多模态视觉生成市场内,但并不覆盖大多数纯文本 copilot、搜索智能体,或那些抬高通用 TAM 图表的基础设施硬件收入。实际含义是,HiDream 同时参与至少三层相邻市场:基础模型能力、开发者 / API 赋能,以及终端用户应用收入。这样的分层位置有吸引力,因为它扩大了变现选项,但也让测算更复杂。因此,正确的市场框架是多模态内容生成工作流,尤其是图像、视频、音频和创意编排软件,而不是「全部 AI」或狭窄的图像模型基准测试利基。[CM001, CM002, CM003, CM004, CM005, CM010]
| 细分 / 类别 | 纳入支出 | 排除支出 | 买方 / 付款方 | 匹配度 |
|---|---|---|---|---|
| 基础模型授权与 API 推理 | 按量计费 API 收入、模型授权、通过 HiHarness 销售的工作流推理 | GPU 销售、云基础设施转售、纯文本 copilots | 企业产品 / IT 团队 | 核心项,因为 HiHarness 将模型访问商业化。 |
| 营销与电商内容软件 | 创意生成、活动素材、商品图、通过 HiBurst 提供的广告视频工作流 | 不含生成层的通用营销技术套件 | 品牌、电商和代理商预算 | 高度匹配,因为 HiBurst 直接切入这些支出池。 |
| 电影 / 媒体制作工具 | 通过 Zhenzan 做的预演、概念美术、视频生成、动画或制作加速 | 全球票房总额、流媒体订阅或摄影机硬件 | 制片厂、制片人、媒体集团 | 上海电影和 Huace 的连接让该项高度匹配。 |
| 创作者订阅软件 | vivago 订阅、模板包、创作者工具增购 | 一般社交媒体广告支出或游戏支出 | 个人创作者 / 小团队 | 可作为分发和数据表面,但 ARPU 较低。 |
| 世界模型研发期权价值 | 如果商业化,未来仿真和更丰富的多模态控制 | 把投机性的世界模型价值当作当前收入 | 战略投资人 / 未来企业买方 | 战略重要,但还不是清晰的当前市场桶。 |
边界有意排除大多数纯文本 AI、硬件和通用云支出,同时保留 HiDream 模型加应用的混合定位。
[CM001, CM002, CM003, CM004, CM005, CM015]从广义 GenAI 品类到 HiDream 近期可服务工作流的分层市场视图。
底层故意保持定性,因为公开来源没有单独切出干净的 HiDream SAM。
[CM006, CM007, CM009, CM010, CM036, CM038]2.2 市场测算口径有用,但不能互换
现有市场研究确认了一个庞大且快速增长的背景,但彼此差异很大,因为它们测量的东西不同。Grand View 对 2026 年的估计是 USD 29.6 billion,Fortune 是 USD 161 billion,ResearchAndMarkets 则只给 2024 年基础模型收入切出 USD 4.1 billion,同时把更多价值放在平台和硬件上。对 HiDream 来说,这种分散不是噪音;它提示市场必须先拆解,才能估值。一家同时卖模型和应用的公司,不应只对标狭窄的基础模型池,但也不应继承硬件收入或所有文本生成负载。可用结论是,需求显然在扩张,软件 / 应用捕获也真实存在;但精确的 SAM/SOM 仍需要按工作流、地理和买方预算做自下而上的工作。在那之前,区间式测算比一个英雄式 TAM 数字更诚实。[CM006, CM007, CM008, CM009, CM010, CM017]
| 发布方 | 年份 | 地区 | 数值 | CAGR | 方法 | 置信度 | 局限 |
|---|---|---|---|---|---|---|---|
| Grand View Research | 2026 | 全球 | USD 29.6B 生成式 AI 市场 | 40.8% (2026-2033) | 广义生成式 AI 应用与栈估算 | 中 | HiDream 专属 SAM 用不上,范围太宽。 |
| Fortune Business Insights | 2026 | 全球 | USD 161B 生成式 AI 市场 | 29.3% (2026-2034) | 范围更宽的广义生成式 AI 市场估算 | 中 | 与 Grand View 不直接可比。 |
| ResearchAndMarkets | 2024 | 全球 | USD 4.1B 基础模型市场 | n/a | 仅模型收入,不含终端用户应用 | 中 | HiDream 还看应用,单看模型太窄。 |
| ResearchAndMarkets | 2024 | 全球 | USD 17B 开发平台市场 | n/a | 平台和工具层 | 中 | 仍比 HiHarness 本身更宽。 |
| ResearchAndMarkets | 2024 | 全球 | USD 132.3B GenAI GPU 硬件收入 | n/a | GenAI 工作负载的基础设施层 | 高 | HiDream 作为软件收入无法服务该市场。 |
各行是视角,不是可相加的 TAM 组件。保留彼此矛盾的范围,是为了避免虚假精确。
[CM006, CM007, CM008, CM009, CM010, CM036]一组相关市场量级,说明为什么单一数字无法代表 HiDream 的可服务市场。
各行比较的是不同范围,不是在同一数量上表达概率分布。
[CM006, CM007, CM008, CM009, CM016, CM037]2.3 买方、用户和付款方分层
HiDream 的产品栈横跨多种买方原型。企业 API 部署里,用户可能是应用团队,付款方则是产品、IT 或平台预算负责人。营销工作流里,创作者和运营人员使用工具,但品牌、电商或代理商负责人通常掌握预算。娱乐领域,制片人和片厂负责人为工作流加速付费,艺术家、剪辑师或动画团队才是日常用户。vivago 增加了自助订阅层,用户和付款方可能合并为一个创作者账号。分层很重要,因为采用触发点不同:企业重视集成、可靠性和治理;营销方重视素材速度和 ROI;片厂重视可控性和保真度;消费者重视广度、模板和价格。HiDream 的优势是同一个模型底座可以喂给四个入口,但一个细分市场的成功并不保证在其他市场占优。[CM020, CM021, CM022, CM023, CM024, CM025]
| 细分市场 | 买方 | 用户 | 付款方 | 工作流 | 预算负责人 | 采用触发点 |
|---|---|---|---|---|---|---|
| 企业 API / MaaS | 产品或平台团队 | 开发者 / 应用团队 | IT、产品或平台预算 | 把图像 / 视频 / 音频生成嵌入产品 | CTO / 产品 / 平台负责人 | 需要快速上线多模态功能。 |
| 营销 / 电商创意 | 品牌运营方 / 代理商 | 设计师、营销人员、卖家 | 营销或电商预算 | 更快生成素材、商品视觉和广告 | CMO / 增长 / 电商负责人 | 内容速度和 ROI 压力。 |
| 电影 / 电视 / 娱乐 | 制片厂 / 制片人 | 艺术家、剪辑师、制作团队 | 制作预算 | 预演、创意迭代、合成场景 | 制片人 / 制片厂负责人 | 需要成本更低或速度更快的内容工作流。 |
| 创作者 / 专业消费者订阅 | 个人创作者 | 同买方 | 月订阅付款方 | 短视频、梗图、模板驱动创作 | 创作者预算 | 模型和模板足够丰富,价格足够低。 |
| 战略合作 | 媒体集团 / 云合作伙伴 | 联合解决方案团队 | 企业战略 / BD 预算 | 打包分发或共同开发 | CEO / BD / 战略负责人 | 获得分发或品牌化工作负载。 |
自助产品里买方、用户和付款方会合一;企业和制片厂工作流里,三者会明显分开。
[CM020, CM021, CM022, CM023, CM024, CM025]矩阵展示 HiDream 目标细分市场中,买方、付款方、采用触发器和转换成本如何不同。
[CM021, CM022, CM023, CM024, CM031, CM032]2.4 增长驱动与约束
需求侧很强。Stanford 记录了个人和组织采用的快速上升,多个市场研究也指向软件和营销用例的持续双位数或更高增长。竞争同样强。Runway、Veo 等全球领导者拼质量和创意控制,中国视频实验室则在价格、速度和平台分发上激烈竞争。因此,HiDream 受益于快速增长的品类,但也身处残酷争夺之中。两个约束尤其突出。第一,中国监管和合成媒体标识规则会拖慢面向公众的视频与音频产品分发,或推高合规成本。第二,模型经济性仍重度暴露在算力和平台层,而这些层的规模远大于应用收入。乐观情景是,HiDream 能靠开源、工作流包装和战略投资者快速进入商业预算;怀疑情景是,公司仍需在真实可服务市场内证明细分领导力、合规执行和持久单位经济性。对一家公开披露仍更强调里程碑和合作、而不是经常性细分经济性、留存队列或毛利率历史的中国私营公司来说,这一点尤其重要。[CM011, CM012, CM013, CM027, CM028, CM029]
| 驱动 / 约束 | 方向 | 时点 | 影响 | 尽调问题 |
|---|---|---|---|---|
| GenAI 用户快速普及 | 正向 | 当前 | 降低创作者产品和试点项目的教育成本 | vivago 用户有多少转为付费? |
| 企业侧 GenAI 试验采用率达 88% | 正向 | 当前 | API 和工作流试用成功概率更高 | 40K 家企业客户中,有多少正在付费且活跃? |
| 媒体 / 娱乐需求 | 正向 | 当前至中期 | 支撑 Zhenzan 和影视合作 | 哪些生产工作流已上线,哪些还在试点? |
| 中国视频 AI 的价格优势 | 正向 / 负向 | 当前 | 能拉动需求,也会压缩利润 | 按生成分钟或图片计算,贡献利润是多少? |
| 竞品平台纵向整合 | 负向 | 当前 | 有分发的对手可能比独立实验室更快放大规模 | HiDream 不拥有大型内容应用,如何获得分发? |
| 合成内容标识与 GenAI 规则 | 负向 | 当前 | 抬高合规、审核和溯源成本 | 水印、审核、政策工具已部署到什么程度? |
| 算力依赖 | 负向 | 持续 | 硬件 / 平台成本结构会限制毛利率 | 当前推理成本和供应商依赖是什么? |
| 开源 / 社区牵引 | 正向 | 当前 | 可能扩大开发者采用和品牌认知 | 多少管线或收入来自开源用户? |
同一股力量既能推需求,也能伤经济性;定价就是最清楚的例子。
[CM011, CM012, CM027, CM028, CM033, CM035]从认知到生产部署的采用路径,覆盖 HiDream 主要细分市场。
自助创作者可能压缩同一路径,受监管企业部署则可能拉长。
[CM014, CM020, CM021, CM033, CM034, CM038]2.5 展项
03竞争格局
3.1 HiDream 实际在哪里竞争
HiDream 并不是在一条清晰赛道里竞争。公开证据把它同时放在图像模型基准测试、开源开发者心智、创作者平台软件,以及新兴影视 / 营销工作流工具中。这样的混合位置很重要,因为参照系会随买方变化。图像侧,HiDream 在质量上与闭源和开源模型领导者竞争。创作者侧,vivago 与简单消费工具以及能路由到多种最佳模型的聚合器竞争。企业和影视侧,公司对手包括工作流平台、内部自建方案,以及合规或分发更强的供应商。结果是,公司有很多获得相关性的路径,但公开来源尚未显示单一受保护滩头阵地。它最尖锐的公开优势仍是视觉模型质量加开源可信度,而不是掌握主导需求渠道。这也意味着,把 HiDream 只当作又一个图像模型或又一个视频应用来比较,会错过管理层如何试图把用户从基准测试认知,导向软件使用,再导向行业特定工作流。[CP001, CP002, CP003, CP004, CP005, CP008]
| 竞争对手 | 类别 | 规模 / 融资 | 目标客群 | 差异化 | 局限 |
|---|---|---|---|---|---|
| HiDream.ai | 多模态实验室 + 创作者 / 工作流栈 | 刚完成 Series C 轮;开源图像模型有牵引;媒体投资人加持 | 创作者、营销人员、影视 / 媒体、API 团队 | 开源图像质量、多模态品牌、vivago 聚合、影视合作 | 没有明显自有流量平台;公开企业级证据仍偏薄。 |
| Runway | 工作流平台 | 60M+ 创意用户使用 | 创作者、工作室、企业团队 | 视频 / 图像 / 音频一体化工作区,打包方式更适合企业 | 本资料集中,规模证据强于定价细节。 |
| Google Veo | 模型 + 平台型在位者 | Google 支撑的平台分发 | 电影人、叙事创作者、企业创作者 | 原生音频、高控制度、1080p / 4K、水印和安全栈 | 低端自助价格是否够低,这里证据较少。 |
| Kling | 中国视频竞品 | 第三方资料显示,属 Kuaishou 关联竞品 | 创作者、短视频、专业内容 | API 覆盖、强运动 / 电影感结果、价格有竞争力 | 自有平台优势看起来强于 HiDream;价格可能压缩利润。 |
| Hailuo | 中国创作者 / 视频竞品 | 第三方资料显示,属 MiniMax 关联竞品 | 创作者、短视频 | 逼真视觉和面向创作者的产品界面 | 第三方对比显示,同步音频弱于头部同业。 |
| 现状 / 内部自建 | 替代方案 | 既有团队和软件预算 | 代理商、设计师、产品团队 | 无供应商依赖;工具和采购路径熟悉 | 质量够用时,AI 迭代速度更慢,自动化也更少。 |
这些行聚焦 HiDream 可见产品栈最相关的购买替代,而非穷尽列出所有多模态实验室。
[CP001, CP005, CP012, CP013, CP014, CP015]X 轴为分发 / 渠道力量,Y 轴为模型加工作流广度。分数是有证据支撑的序数估计,不是市场份额统计。
各点综合了官方产品广度、可见创作者触点、基准状态,以及检索来源集里的第三方分发评论。
[CP012, CP013, CP014, CP015, CP020, CP021]3.2 同业类别与能力差异
竞争者至少分成三类。第一类是 Runway、Google Veo 等高端全球工作流或模型玩家,它们拼控制力、可靠性和企业可接受的信任姿态。第二类是 Kling、Hailuo 等中国视频原生对手,它们更直接地竞争短视频创作、API 可用性和价格敏感的国内创作者需求。第三类是混合或路由层,包括 vivago 本身,这些产品越来越多地把几个上游模型包装成一个体验。HiDream 的图像模型故事在公开基准测试中表现不错,尤其是在开源领域;但它更广的多模态位置,仍没有品牌暗示得那么稳固。它还没有展示出 Veo 那样的公开企业控制栈,也没有 ByteDance 或 Kuaishou 相关对手那样明显的大众分发循环。因此,功能广度是真实的,但竞争适配高度取决于细分市场。[CP006, CP007, CP012, CP013, CP014, CP015]
| 购买标准 | HiDream.ai | Runway | Veo | Kling | Hailuo |
|---|---|---|---|---|---|
| 开源图像可信度 | 强 | 检索资料未证实 | 检索资料未证实 | 检索资料未证实 | 检索资料未证实 |
| 创作者应用 / 自助界面 | vivago 支撑,强 | 强 | 检索资料未证实 | 可能较强,但官方来源未证实 | 可能较强,但官方来源未证实 |
| 视频原生音频 | 检索资料未证实 | 检索资料部分证实 | 强 | 检索资料未证实 | 第三方对比中较弱 |
| 企业信任 / 溯源姿态 | 起步中 | 部分证实 / 强 | 强 | 检索资料未证实 | 检索资料未证实 |
| 自有分发优势 | 公开证据有限 | 中等创作者平台 | Google 生态支撑,强 | Kuaishou 关联生态支撑,强 | 创作者界面中等,但这里自有流量证据较弱 |
| 媒体 / 影视合作证据 | 早期公开证据强 | 检索资料未证实 | 有部分电影人案例 | 检索资料未证实 | 检索资料未证实 |
「检索资料未证实」指本次检索资料没有证明,不代表竞争对手没有该能力。
[CP005, CP008, CP012, CP013, CP014, CP015]围绕本章主要购买标准,对能力和大致锁定程度做横向比较。「未知」表示抓取到的来源集中尚未证明。
[CP005, CP008, CP012, CP013, CP014, CP015]3.3 定价、包装与多栖使用
定价和包装支持更细的判断。HiDream 可以指向 vivago 可触达的自助定价和宽创作者功能集,但产品自身差异化部分来自对第三方模型的包装。获客上这很有吸引力,因为用户想要选择和快速结果;但它也提高了买方多栖使用、而不是锁进 HiDream 栈的概率。第三方比较显示,中国视频对手在单条短片价格上竞争激烈,高端西方栈则常在更高端质量或控制力上竞争。同样重要的是,公开来源没有给主要对手提供清晰、苹果对苹果的企业标价比较,这限制了对可持续定价权的硬结论。现实中,当前市场看起来仍很流动:用户按任务、价格和可靠性路由,这种行为让切换成本低于成熟软件品类。这样的动态通常奖励最能简化编排、支持和重复工作流价值的公司,而不是只发布最新模型分数的公司。[CP009, CP010, CP011, CP017, CP018, CP023]
| 公司 | 价格 / 单位 / 合同模式 | 包含能力 | 折扣或未知项 | 影响 |
|---|---|---|---|---|
| vivago / HiDream | 自助月付从 $7.9-$9.99 起;应用商店月付档位最高 $99.99 | 模板、图像 / 视频生成、聚合模型、创作者工作流 | 企业 / API 定价未公开检索到 | 创作者获客摩擦低,但定价权不清楚。 |
| Kling | 第三方对比显示,五秒 720P 视频约 RMB 4 / US$0.56 | 运动效果强、专业视频定位;对比称支持 1080P | 非官方标价;可能随套餐变化 | 激进单价给视频竞品施压。 |
| Seedance | 第三方对比显示,五秒 720P 视频约 RMB 2.3 / US$0.32 | 对比称叙事 / 视频生成能力强 | 与本章直接点名的 HiDream 竞品相邻,而非同列 | 在中国视频市场立下低价参照。 |
| Google Veo | 第三方对比显示,五秒视频约 US$5 | 高端视频,加原生音频和控制能力 | 这里未检索到官方可比标价 | 高端市场更拼质量 / 信任,而不是最低成本。 |
| OpenAI Sora 2 | 第三方对比显示,五秒视频约 US$2.5;独立产品已停用 | 历史上具备高端电影感品牌 | 独立产品下线改变其相关性 | 说明视频模型经济性重置可以有多快。 |
| Runway | 公开可见企业 / 创作者打包;未检索到精准可比单价 | 一体化创意工作区 | 本资料集价格未知 | 本记录中,竞争重点更在工作流深度,而非公开剪辑单价。 |
本表混合官方订阅数据与第三方单条视频对比;未获支持的企业标价均明确标为未知。
[CP009, CP010, CP017, CP018, CP030, CP031]概览 HiDream 当前竞争耐久度。
这些值混合了等级判断、公开基准和应用信号事实。
[CP005, CP006, CP010, CP020, CP028, CP029]3.4 护城河持久性与反向证据
HiDream 的护城河故事可信,但仍有条件。开源图像领导力可以加强开发者声誉;媒体合作可以打开差异化用例;多模态 / 世界模型品牌有助于吸引资本和注意力。反向侧同样重要。开源会加速模仿,路由层依赖上游模型访问,公开合作并不能证明持久客户捕获。与此同时,相对于流入视频 AI 的实验和资本,品类 ARR 看起来仍不高,说明市场可能在标准尚未确定前就变得拥挤。除非某个玩家控制分发、深度嵌入工作流,或产出明显更优结果,这种组合通常会压缩供应商权力。HiDream 以后可能达到这一状态,但今天的公开证据支持的是一个有前景、在扩张的竞争位置,而不是已经扎稳且可防守的位置。要得出更强结论,需要看到真实客户选择、扩张行为,以及影视或营销合作方无法轻易用类似工作流外壳路由到其他模型来替换 HiDream 的证据。[CP016, CP020, CP026, CP027, CP028, CP029]
| 护城河主张 | 威胁 | 严重性 | 缓释 / 尽调问题 |
|---|---|---|---|
| 开源图像领先 | 竞品快速复制功能或追平质量 | 中高 | 衡量开发者管线,以及社区用户转向付费产品的转化。 |
| vivago 聚合广度 | 上游模型访问条件变化,或用户变得模型无感 | 高 | 厘清依赖条款,以及由第三方模型驱动的使用占比。 |
| 媒体合作 | 试点合作未能放大为经常性预算 | 中高 | 索取生产部署、ACV 和续约证据。 |
| 多模态 / 世界模型品牌 | 叙事跑在已交付客户价值前面 | 中高 | 索取可归因于世界模型能力的具体工作流改善。 |
| 中国市场定位 | 平台自有竞品分发能力压过专业实验室 | 高 | 评估分发合作、渠道经济性和自有受众建设。 |
| 低门槛自助定价 | 锁定形成前,价格竞争先压缩利润 | 高 | 按付费档位审查单位经济性与留存。 |
严重性是分析师判断,依据是每项威胁拖慢采用或压缩议价能力的直接程度。
[CP020, CP027, CP028, CP029, CP032, CP035]3.5 展项
04财务情况
4.1 收入入口与公开价格轨道
HiDream 的公开记录让变现入口可见,但还看不到已经实现的经济性。最清晰的可观察轨道是 vivago,应用商店和官方产品页展示了月度订阅层级和点数购买。官方公司材料也让 HiHarness 的 API 业务具备可信度,产品新闻则显示 HiBurst 和影视 / 媒体工具可以嵌入付费商业工作流。这令人鼓舞,因为公司不像一个等待变现的纯研究实验室。但可见度并不对称。公开创作者定价很具体,企业 / API 真实成交价、自定义合同条款和支持义务仍不透明。因此,第一个财务结论很简单:HiDream 有几条可信收费路径,但公开记录展示价格表和活动声明,仍比展示净收入质量更清楚。[CI001, CI002, CI003, CI004, CI005, CI009]
| 收入流 | 机制 | 单位 | 当前值 / 状态 | 质量 | 尽调问题 |
|---|---|---|---|---|---|
| 企业 API / HiHarness | 基于用量的 API / 模型访问,以及潜在定制协议 | API 调用 / 合同 | 200+ 个 API 和 500B+ 次调用显示平台可变现,但未披露实际定价 | 存在证据中等,经济性证据低 | 提供企业合同、最低承诺额,以及按 API 产品线拆分的收入。 |
| 创作者订阅 | 通过 vivago 销售月度套餐 | 订阅用户 / 月 | 公开 iOS 档位为 $9.99、$29.99 和 $99.99 | 标价证据高,净收入证据低 | 提供付费用户、流失、应用商店费用和地域组合。 |
| 积分 / 应用内消费 | vivago 内一次性购买积分 | 积分 / 订单 | iOS 可见 1000 积分售价 $12.99 | 存在证据高,贡献利润证据低 | 提供积分购买频率、沉淀余额和促销使用情况。 |
| 营销工作流产品 | 借助 HiBurst 生成活动素材和电商视频 | 项目 / 账户 / 用量 | HiBurst 公开活跃度高,但未披露直接定价或抽成率 | 工作流活跃度证据中等,收入转化证据低 | 提供 HiBurst 定价模式、客户数量和收入确认政策。 |
| 影视 / 媒体工作流工具 | 通过 Zhenzan 和合作伙伴部署,可能获得项目或合同收入 | 定制合同 | 能看出商业牵引,但公开变现条款未知 | 经济性证据低 | 提供媒体合同结构、支持负担和续约证据。 |
各行把可观察收费界面与未验证经济性拆开。公司概览章节覆盖历史融资时间线;本表聚焦潜在经常性收入流。
[CI001, CI002, CI003, CI004, CI005, CI007]| 价格 / 单位 / 合同 | 标价与实际定价 | 折扣 / 未知项 | 来源 |
|---|---|---|---|
| vivago basic / plus / pro 月付:$9.99 / $29.99 / $99.99 | 公开标价 | 扣除应用商店费用、促销积分和流失后的实际净额未知 | App Store |
| vivago basic 年付:$79.99 | 公开标价 | 续费率和付费转化未知 | App Store |
| 1000 积分:$12.99 | 公开一次性价格 | 使用模式和沉淀余额未知 | App Store |
| vivago 官方起价 $7.9/月(按年付折算的营销口径) | 官方营销价位 | 适用于年付;具体套餐组合未知 | vivago.ai |
| HiHarness / 企业 API | 公开资料未知 | 这里未见公开费率表或合同最低额 | 官方首页 / 融资新闻稿 |
| HiBurst / Zhenzan | 公开资料未知 | 这里未见公开抽成率、席位或用量价格 | China Biz Insider / 公司新闻稿 |
本表有意把可见创作者标价与明确的企业未知项放在一起,避免伪精确。
[CI003, CI004, CI005, CI009, CI010]HiDream 公开产品界面上的可见用户和客户活动,可能如何转成收入。
该图为定性图,因为公开来源能看到可收费界面,但看不到管理层级的收入桥接。
[CI001, CI003, CI006, CI009, CI015, CI030]4.2 GTM、牵引力与指标质量
GTM 图景最好按混合模式来读。vivago 支持产品驱动的创作者获客,而官网、融资报道和商业用例新闻暗示 API、营销工作流和媒体部署走销售驱动或合作驱动。公开牵引信号不小:公司相关报道提到 50M+ 专业用户和 40K+ 企业客户,China Biz Insider 描述 HiBurst 生成超过 100 万条视频,并支撑超过 RMB 100 million 的 GMV。这些数字说明产品栈周围有真实经济活动。它们仍未达到财务尽调标准。客户数不是收入,GMV 不是公司收入,活动量也不揭示 CAC、回本周期、留存或合同耐久性。因此,投资判断取决于把运营产出接到确认收入,公开来源目前还没有给出这座桥。[CI006, CI007, CI008, CI011, CI012, CI030]
4.3 成本结构与单位经济性
HiDream 的成本基础更像 AI 软件和推理公司,而不是传统重资产制造商,但这不代表它便宜。算力强度、模型刷新节奏、审核或政策控制、创作者支持运营和渠道费用,可能都是经常性成本中心。市场层面证据也说明这为何重要:生成式 AI 周边基础设施层远大于基础模型收入层,第三方评论认为,如果下游变现不强,独立视频生成可能变得不经济。如果传闻中的 UiT 效率提升成立,HiDream 可能有真实架构优势;但公开证据仍太薄,无法把这一叙事转化为毛利率信心。已审阅来源没有提供按图像、视频分钟、企业客户或订阅队列划分的贡献利润率。因此,财务问题不是 HiDream 能不能收费,而是它能否以有吸引力且持久的单位经济性收费。[CI013, CI014, CI015, CI016, CI017, CI018]
| 指标 | 值 / null | 置信度 | 重要性 | 尽调问题 |
|---|---|---|---|---|
| 付费创作者 ARPU | 未公开 | 低 | 需要用它把可见标价换算成真实消费者变现 | 提供 ARPU、折扣和按套餐拆分的地域组合。 |
| 企业实际 ASP | 未公开 | 低 | 判断 API 和工作流收入能否支撑重算力交付,关键在这里 | 提供平均合同价值、最低承诺额和定价瀑布。 |
| 毛利率 | 未公开 | 低 | 推理密集型业务里,决定软件质量的核心指标 | 按业务分部和算力伙伴提供毛利率。 |
| 单个生成资产贡献利润 | 未公开 | 低 | 决定规模增长是在创造价值还是摧毁价值 | 提供按图像 / 视频拆分的贡献分析,包括审核和退款率。 |
| HiBurst 经济性代理指标 | 每年 >1M 条视频;GMV > RMB100M | 中 | 显示工作流活跃,但不能误认为 HiDream 收入 | 说明 GMV 如何转化为 HiDream 抽佣率或软件收入。 |
| 架构成本优势 | 据称训练成本为行业平均的 10%-20% | 中 | 如果能在生产环境复现,模型经济性可能明显改善 | 提供经审计或管理口径的算力成本对比,以及对服务成本的影响。 |
空值不是遗漏;它们标出阻碍承销的具体缺失财务数据。
[CI007, CI008, CI012, CI017, CI018, CI029]从内容生成活动到单位经济的定性桥接。
没有公开来源给出数字化利润率桥接,因此这些节点列出尽调必须测试的成本和质量驱动因素。
[CI007, CI008, CI017, CI018, CI027, CI029]基于公开产品和市场证据,定位 HiDream 可能的成本强度。
矩阵使用方向性证据,而非公司披露的会计数据。
[CI015, CI016, CI034, CI035, CI036]4.4 资本充足性与私募市场不透明
单看资本获取,HiDream 很强。多篇同期报道一致指向 RMB 1.5B 的 C 轮融资,以及短期窗口内超过 RMB 2.1B 的融资总额,资金用途指向全模态或世界模型开发。这正是一家仍在大举投入模型能力和商业化的公司应有的融资模式。更难的问题是充足性,而不是获取能力。公开来源没有披露账面现金、月度烧钱速度、现金跑道月数、债务义务或未来融资需求的确切时间。更糟的是,私募市场数据库并不一致,或过于稀疏:InforCapital 显示的融资历史小得多,而 PitchBook、Tracxn 和 Caplight 在本次检索集合中几乎没有可用的当前投资判断细节。结果是,资本故事在新闻标题上看起来很强,但仍缺少干净前瞻现金模型所需的透明度。更广的 2026 年 AI 独角兽环境似乎也有利,可能改善了融资接受度,但宏观风险投资胃口不能替代公司层面披露。[CI019, CI020, CI021, CI022, CI023, CI024]
| 手头现金 | 月度烧钱 | 可支撑月数 | 计划资金用途 | 下一轮融资触发点 | 债务 / 项目融资义务 |
|---|---|---|---|---|---|
| 未公开 | 未公开 | 未公开 | Series C 轮及近期资金指定投向全模态 / 世界模型与商业化产品开发 | 可能绑定模型路线图和商业化里程碑,但未披露 | 检索材料未公开披露 |
资本充足性目前只能推断:融资新闻可见,但现金流披露不可见。
[CI019, CI020, CI024, CI025, CI026, CI041]关乎早期创作者经济性的可见自助变现档位,全部以每单或每月美元表示。
这是变现输入区间,不是收入估计。企业价格实际成交仍不可得。
[CI003, CI004, CI007, CI008, CI030, CI041]4.5 财务结论与尽调阻断点
正向逻辑真实存在:HiDream 有多条变现入口、可见产品活动,以及一家年轻私营 AI 公司罕见的融资势头。负向逻辑同样真实:公开记录仍无法验证收入规模、收入结构、毛利率、留存、现金跑道,或增长在多大程度上依赖持续外部资本。这意味着公司可能商业前景不错,但财务文档仍不足。机构投资者的实际含义不是否定 HiDream,而是要求管理层把漏斗顶部活动接到收入确认和现金生成。管理层提供这座桥之前,任何精确的收入质量或毛利率扩张模型都仍属推测。[CI031, CI032, CI033, CI034, CI035, CI036]
| 缺失的私营公司指标 | 影响 | 精确尽调路径 |
|---|---|---|
| 按细分板块确认的收入和 ARR | 没有这些,变现质量仍停留在个案叙述 | 要求提供经审计收入桥接,按 API、营销产品、影视 / 媒体和创作者订阅拆分。 |
| 按产品拆分的毛利率和算力成本 | 无法判断软件质量或规模经济 | 要求提供 COGS 拆分、云合作伙伴承诺,以及按资产类型计算的贡献利润率。 |
| 现金余额、月度消耗和可支撑周期 | 无法评估融资依赖或下一轮融资时点 | 要求提供董事会现金看板、月度烧钱桥接和 12-18 个月预测。 |
| 合同质量和留存 | 无法判断收入耐久性或先落地再扩张的动力 | 要求提供续约、流失、NRR/GRR 以及按细分板块拆分的队列行为。 |
| 企业定价兑现度 | 公开标价主要只覆盖消费端 | 审查 MSA 样本、订单表和折扣表。 |
这些关键卡点,决定商业化故事能否从可信叙事升级为财务级投资案。
[CI023, CI024, CI031, CI033, CI037, CI039]4.6 展项
05产品与技术
5.1 产品套件与用户任务
HiDream 的公开产品故事宽于单个基准测试获胜模型。官方材料列出五条产品线——HiHarness、HiBurst、Zhenzan、vivago 和 HiDreamFans——覆盖企业 API、营销工作流、影视或专业内容工作流、创作者工具,甚至线下零售硬件。这样的宽度很重要,因为它揭示了管理层希望模型栈如何被消费。HiHarness 是最清晰的企业入口,围绕数百个 API、重点客户采用和大调用量来表述。vivago 是最清晰的自助创作者入口,包装多模型访问、模板和智能体编辑。HiDreamFans 把逻辑延伸到线下获客工具,HiBurst 和 Zhenzan 则锚定战略投资者似乎在意的营销和媒体工作流。统一的待完成任务不只是图像生成,而是为创作者、营销人员和媒体运营方加速多模态内容生产,并长期把同一个核心模型资产变成更广的全模态工作流平台。[CE001, CE002, CE003, CE004, CE005, CE006]
| 模块 / 资产 / 产品线 | 用户 | 状态 / 成熟度 | 差异化 | 尽调缺口 |
|---|---|---|---|---|
| HiHarness | 企业开发者 / 产品团队 | 已有商业露出 | 200+ 个 API,多模态集成的企业定位清晰 | 需要合同条款、延迟和支持指标。 |
| HiBurst | 营销人员 / 电商运营方 | 已有商业露出但披露不足 | 把核心模型接到营销内容产能和电商工作流 | 需要定价模型和具名客户结果。 |
| Zhenzan | 影视 / 专业媒体用户 | 已有商业露出但披露不足 | 面向更高端制作流程,并与媒体投资方匹配 | 需要生产部署证据和具体工作流。 |
| vivago | 创作者 / 专业消费者 / SMB | 已有商业露出并公开定价 | 多模型聚合、模板和 AI 智能体降低创作摩擦 | 需要留存、审核和渠道经济性细节。 |
| HiDreamFans | 线下零售商 / 门店运营方 | 概念上有差异化,但证据偏轻 | 把 AI 栈延伸到实体门店获客硬件 | 需要落地采用证据、硬件经济性和维护负担。 |
公开产品组合很宽,但各条线披露深度差异很大。vivago 和开源图像栈的证据远强于 HiBurst、Zhenzan 或 HiDreamFans。
[CE002, CE003, CE004, CE007, CE008, CE040]5.2 核心技术架构
最强的公开技术披露仍以图像为中心,但对一家私营 AI 公司来说异常丰富。文档将 HiDream-I1 描述为一个 17B 参数开源图像基础模型,基于稀疏 Diffusion Transformer,带动态 MoE 组件,并在统一交互前分别进行图像 / 文本编码。HiDream-O1-Image 采取了不同架构步骤,使用像素级 Unified Transformer,不依赖外部 VAE 或割裂的文本编码器。这个选择很重要,因为管理层和外部报道都明确把 UiT 与更高效率,以及未来图像 - 视频协同联系起来。代码库也让公司的产品哲学可读:一个架构应覆盖生成、编辑、个性化、长文本渲染和类似分镜的任务,同时由提示词智能体处理围绕布局和物理逻辑的潜在推理。公开证据不能证明这已经在生产中产出真正世界模型,但确实显示公司严肃尝试减少模态和任务之间的模块化交接。[CE017, CE018, CE019, CE020, CE021, CE025]
| 层 / 流程 / 组件 | 角色 | 依赖 | 风险 |
|---|---|---|---|
| HiDream-I1 稀疏 DiT + MoE 栈 | 高质量开源图像生成 | 模型训练、开源维护、部署工具 | 公开论文没有披露生产服务经济性。 |
| HiDream-E1 图像编辑扩展 | 基于 I1 系列的指令式编辑 | 共享图像栈和编辑数据 | 本报告未看到现实编辑可靠性的独立基准。 |
| HiDream-A1 图像智能体概念 | 交互式创作与精修 | 围绕图像栈的智能体编排 | 当前可能更接近路线图,而非已广泛商业化。 |
| HiDream-O1-Image UiT 核心 | 统一像素 / 文本 / 任务架构 | 高质量训练数据和优化实现 | 图像侧证明未必能完整迁移到视频或音频。 |
| Reasoning-Driven Prompt Agent | 围绕布局、物理逻辑和文字渲染改写提示词 | 外部或本地 LLM 后端选择 | 给生成工作流增加依赖和复杂度。 |
| Flask Web 界面 / 演示 / Spaces | 上手评估和引导 | 托管演示、代码仓库、Hugging Face 分发 | 演示可用不等于企业级可靠。 |
图像栈的技术故事异常可检查,但服务成本、安全过滤和生产运营仍披露不足。
[CE017, CE018, CE020, CE025, CE026, CE027]在以图像为先、但具备多模态能力的模型底座之上,HiDream 叠加创作者产品、企业 API、媒体工作流和线下硬件。
[CE003, CE008, CE009, CE017, CE020, CE024]5.3 部署与工作流就绪度
HiDream 的公开部署就绪度强于纯实验室姿态。公司在 GitHub 分发模型权重和代码,公开 Hugging Face 模型卡和 Spaces,为 I1 系列支持 Diffusers,并为 O1 发布 Flask web UI 与 prompt-agent 工具。这意味着开发者可以通过代码、演示或面向创作者的应用接触这套栈,而不是只通过一个封闭端点。创作者侧,vivago 把模型访问包装进订阅、点数、模板、图生视频功能和 AI 辅助编辑 / 聊天工作流。Business Wire 关于 vivago 2.0 的发布尤其重要,因为它把底层模型工作接到 AI 播客、社交模板和零售促销视频生成等具体用户流。缺口在于,公开来源对使用入口的描述强于对服务质量的描述。关于正常运行时间、延迟承诺、失败任务处理、支持 SLA 或企业上手承诺的信息很少。因此,产品栈看起来可部署且宽,但可靠性和支持成熟度的证据不如功能迭代速度充分。公开代码路径和演示端点也让直接测试上手摩擦更容易,即便它们不能回答企业支持问题。[CE011, CE012, CE014, CE015, CE016, CE022]
| 用户任务 | 当前工作流 | 公司方案 | 可衡量收益 | 局限 |
|---|---|---|---|---|
| 将多模态生成嵌入应用 | 自定义集成工作或外部模型 API | HiHarness API 套件 | API 广度可见,并有企业场景包装 | 未公开真实企业 SLA 或价格细节。 |
| 快速制作产品或营销活动视觉 | 手工设计、代理商工作或多工具拼接 | HiBurst + 核心模型栈 | 可能加快素材生产和迭代 | 具名客户 ROI 未公开披露。 |
| 生成媒体概念或影视周边素材 | 传统概念美术 / VFX 前期准备 | Zhenzan + 媒体工作流定位 | 契合投资方关联的影视用例 | 具体制作流程步骤仍模糊。 |
| 将提示词和照片转成可分享视觉内容 | 分散使用图像、视频和编辑应用 | vivago | 模板、图生视频、聚合头部模型、AI 编辑 / 聊天 | 创作者留存和满意度数据稀少。 |
| 为线下门店吸引客流 | 传统标牌或促销屏 | HiDreamFans | 3D 全息营销角度区别于纯软件工具 | 硬件部署经济性未公开。 |
本表把产品名转成工作流任务,避免把模型命名过度解读为商业化证明。
[CE004, CE008, CE011, CE013, CE014, CE016]| 日期 / 阶段 | 功能 / 里程碑 | 状态 | 含义 | 来源 |
|---|---|---|---|---|
| 2025-04-07 | HiDream-I1 开源发布 | 已发布 | 让基础图像栈可被公开检查 | GitHub README |
| 2025-04-28 | HiDream-E1-Full 编辑模型开源 | 已发布 | 从生成扩展到指令式编辑 | GitHub README / HF 模型卡 |
| 2025-07-16 | HiDream-E1.1 更新 | 已发布 | 显示编辑能力继续迭代 | GitHub README / HF 模型卡 |
| 2026-05-08 | HiDream-O1-Image 开源 | 已发布 | 公开引入基于 UiT 的 8B 架构 | GitHub O1 仓库 |
| 2026-05-14 | HiDream-O1-Image-Dev-2604,带提示词优化器 | 已发布 | 给文生图工作流加入推理 / 提示词优化 | GitHub O1 仓库 |
| 2025-06-20 | vivago 2.0 在 Web 和移动端全球发布 | 已发布 | 把模型能力推入更宽的创作者产品 | Business Wire |
日期反映公开发布标记,不是内部研发开始日期。这种节奏支撑快速推进的路线图,但对运营稳定性说明较少。
[CE011, CE017, CE020, CE024, CE027, CE040]创作者、营销人员和开发者从不同界面进入,但最终汇聚到模型执行、迭代和付费工作流使用。
[CE004, CE009, CE011, CE012, CE016, CE028]公开成熟度在图像生成和创作者工具上最强,企业级保障和广泛全模态证明在来源集中仍不充分。
[CE017, CE021, CE023, CE024, CE031, CE040]5.4 信任、合规与路线图平衡
信任和合规是 HiDream 产品兴奋感的主要反向权重。中国生成式 AI 规则和合成媒体标识框架显然适用于面向公众的图像、视频和音频服务,因此 vivago 和 HiHarness 这类产品需要合法数据、用户协议、隐私、投诉处理和标识能力。HiDream 至少为 vivago 发布了全球隐私页面,但已审阅来源集合没有显示成熟的公开信任中心、第三方认证、详细审核指标或企业保证材料。路线图图景也不均衡。公开发布显示迭代很快——I1、E1、O1、O1-Dev、提示词智能体和 vivago 2.0——但最清晰的证明仍围绕图像生成、编辑和创作者工具。关于更广全模态世界模型、视频协同或未来无限长度视频的主张,在战略上很有意思,但作为客户可用产品现实,证据还不够。这让 HiDream 拥有有吸引力的产品逻辑,同时留下治理和可靠性尽调负担。[CE030, CE034, CE035, CE036, CE037, CE038]
| 控制 / 认证 / 质量指标 | 状态 | 范围 | 缺口 |
|---|---|---|---|
| 生成式 AI 服务合规义务 | 中国监管规则中明确 | 面向公众的文本 / 图像 / 音频 / 视频服务 | 需要落地证据,而不只是适用性判断。 |
| 合法数据与 IP 要求 | 中国监管规则中明确 | 训练和服务提供 | 检索材料未见公开的不侵权 / 可运营性资料包。 |
| 生成媒体标识义务 | 合成内容规则中明确 | 图像 / 视频 / 音频输出 | 未见 HiDream 对自身水印或元数据栈的公开说明。 |
| 用户协议与投诉处理义务 | 中国监管规则中明确 | 所有公开生成式 AI 用户 | 未公开投诉指标或升级处理 SLA。 |
| 隐私政策可见面 | 已面向 vivago 全球用户发布 | 面向消费者产品治理 | 仅能看到政策,力度弱于认证或审计证据。 |
| 基准质量信号 | 公开排行榜与基准声明 | 图像模型质量与开源比较 | 基准领先不能证明正常运行时间、抗滥用能力或企业就绪度。 |
HiDream 有可见的质量和政策信号,但公开证据更能证明模型性能,较少证明信任保障已经落地。
[CE021, CE031, CE036, CE037, CE038, CE039]HiDream 的产品栈既依赖代码分发、演示平台、应用商店和监管合规,也依赖模型质量。
[CE024, CE033, CE036, CE037, CE038, CE039]5.5 展项
06客户情况
6.1 分层与进入入口
HiDream 的客户基础应按接入入口分层,而不是按一个总量标题来读。公开证据指向至少五类不同客户:使用 HiHarness 的企业 API 团队,使用 HiBurst 式工作流的营销和电商运营方,通过战略合作进入的影视或媒体机构,经由 vivago 进入的创作者或专业消费者用户,以及可能被 HiDreamFans 瞄准的线下零售运营方。每个细分里的买方、用户和付款方都不同。创作者往往自付,或通过免费点数试用。企业使用可能从技术用户开始,预算在产品或 IT 负责人手里。媒体合作可能混合资本、共同开发和未来工作流部署。这很重要,因为「40,000 家企业客户」和「50 million 专业用户」等头部数字几乎肯定把行为、价格点和合同质量都很不同的群体打包在一起。最干净的公开获客渠道是 vivago,因为它在网页和移动端都可见,展示定价层级,并能把免费试验转化为付费套餐或点数。[CU001, CU002, CU003, CU004, CU006, CU008]
| 细分板块 | 买方 / 用户 / 付款方 | 用例 | 规模 | 收入 / 战略价值 | 缺口 |
|---|---|---|---|---|---|
| 企业 API / HiHarness | 产品或 IT 负责人 / 开发者 / 公司预算 | 将多模态生成集成进产品和工作流 | 公开声称 100+ 个重点客户 | 可能是价值最高的经常性 B2B 层 | 未披露付费账户数或合同规模。 |
| 营销 / 电商运营方 | 品牌或代理商负责人 / 运营方 / 营销预算 | 批量内容、营销活动素材、短视频促销 | HiBurst 活动加 BlueFocus 合作 | 将模型接入可变现营销工作流 | 具名付费客户名单未公开。 |
| 影视 / 媒体机构 | 制片厂或媒体集团 / 制作团队 / 公司预算 | IP 开发、AI 内容生产、营销、标准、实验室 | 上海电影与 Huace 的具名战略证明 | 对高端工作流和行业可信度有战略价值 | 商业规模和经常性支出未披露。 |
| 创作者 / 专业消费者 | 个人用户 / 本人 / 自费或免费计划 | 文生视频、图生视频、模板、AI 辅助创作 | Web、iOS 和 Android 端均有入口并公开定价 | 最大的可见获客漏斗,也是最清晰的公开变现路径 | 缺少留存和付费转化数据。 |
| 商业 SMB 用户 | 店主或营销人员 / 创作者型用户 / SMB 预算 | 促销视频、产品视觉、头像内容 | Business Wire 披露的零售商和品牌用例 | 可能从消费产品桥接到更高价值商业用途 | 没有具名 SMB 案例研究或支出数据。 |
本细分将战略证明、自助式采用和企业 API 使用拆开,因为三者对应的客户质量和毛利画像完全不同。
[CU001, CU002, CU003, CU012, CU013, CU017]HiDream 通过创作者、伙伴和企业入口吸引客户;要实现持久变现,最终仍需证明工作流价值。
[CU006, CU023, CU024, CU031, CU032, CU039]6.2 采用轨迹与具名证明
公司的公开采用叙事有足够内容,值得认真对待,但细节还不足以支撑干净投资判断。融资报道反复声称拥有 5000 多万专业用户、4 万多家企业客户,并覆盖 100 多个国家和地区。这些数字提示规模,而不是商业化前实验;可见创作者入口、应用商店定价和 HiBurst 活动声明也强化了这一点。具名证明在合作关系中最强。BlueFocus 是最清晰的营销行业证明。Shanghai Film 和 Huace 显示,HiDream 已在战略层面打入影视机构。这些很有意义,因为它们说明大型行业参与者对实际工作流有兴趣。同时,这并不等同于带合同金额、席位扩张或经常性使用的公开采购案例。因此,公开记录支持真实采用和严肃战略交易对手,但仍更偏生态证明,而不是经典企业客户披露。[CU004, CU005, CU012, CU013, CU014, CU016]
| 指标 | 数值 | 日期 | 来源 | 置信度 | 含义 | 缺失分母 |
|---|---|---|---|---|---|---|
| 重点客户 | 100+ | 2026-08-11 访问 | HiDream 主页 | 中 | 企业渗透的方向性证据 | 重点客户定义及付费状态未知 |
| 专业用户 | 50M+ | 2026-07-27 | 融资报道 | 中 | 显示触达范围和使用规模较广 | 不是付费、活跃或留存用户指标 |
| 企业客户 | 40K+ | 2026-07-27 | 融资报道 | 中 | 表明企业端获客漏斗可观 | 缺少活跃、付费或 ACV 分母 |
| 地域覆盖 | 100+ 个国家 / 地区 | 2026-07-27 | 融资报道 | 中 | 显示国际足迹 | 没有地区收入或客户拆分 |
| vivago iOS 评分 | 59 个评分给出 4.6 | 2026-08-11 访问 | App Store | 中 | 正面但样本小的公开 iOS 满意度信号 | 评分不等于付费留存 |
| HiBurst 活动代理指标 | 每年 1M+ 条视频;支撑 RMB100M+ GMV | 2026-06-18 | China Biz Insider | 中 | 显示围绕客户内容生产的营销工作流活跃度 | GMV 不是 HiDream 收入或续约 |
轨迹表有意拆分企业 / 客户数量、创作者信号和工作流产出代理指标,避免把不可比指标当作可互换。
[CU003, CU004, CU009, CU014, CU015, CU021]| 客户 | 细分板块 | 部署 / 用例 | 生产 vs 试点 | 结果 | 局限 |
|---|---|---|---|---|---|
| BlueFocus | 营销技术 / 代理商网络 | AIGC 商业内容生产、模型共建、多模态营销,以及影视 / 短视频应用 | 战略合作 / 部署意向 | 说明 HiDream 能吸引中国头部营销合作方 | 未公开合同金额、部署数量或续约数据 |
| Shanghai Film Co., Ltd. | 影视与影院生态 | IP 共同开发、影院场景营销、沉浸式体验、AIGC 实验室、AI 制作工作流探索 | 战略投资合作 / 从试点到生产的意向 | 影视和院线的具名公开机构证明 | 未披露经常性支出或生产吞吐 |
| Huace Film & TV | 影视制作 | 资本 + 技术 + 产业合作,支撑 AI 制作生态 | 战略合作 / 部署意向 | 影视剧内容和 IP 工作流有具名公开佐证 | 商业深度和持续使用未量化 |
具名佐证真实且仍有效,但更多是合作伙伴级证明,不是带有 ROI 和留存细节的典型企业客户背书材料。
[CU016, CU017, CU018, CU019, CU020]公开需求从可见产品或合作界面起步;只有信任和结果站得住,之后才会变成经常性支出。
[CU006, CU008, CU012, CU023, CU024, CU025]公开披露对初始采用信号很强,对长期续约证据很弱,因此客户故事仍偏前端。
这些不是留存百分比。100 表示保留下来的公开来源集至少包含一个针对该时间跨度的可见披露信号;0 表示未找到保留下来的公开披露。
[CU023, CU025, CU026, CU027, CU039]6.3 留存、耐久性与证据缺口
客户章节最大的缺口是耐久性。公开来源显示 HiDream 能吸引用户和合作伙伴,但没有显示这些用户是否会留下、扩张,并贡献高质量收入。App Store 评分方向上积极,vivago 详细产品页描述了免费套餐、多个层级、点数和智能体增购路径,暗示公司有意设计产品驱动漏斗。但这些信号仍不揭示付费转化、流失、队列行为、净留存率(NRR)或合同期限,也不揭示所谓企业客户基础中有多少是活跃或战略上重要的。即便最强的可见具名证明——BlueFocus、Shanghai Film 和 Huace——也没有披露收入贡献或续约节奏。结果是,客户故事有广度和势头,但关于重复经济性的公开证据有限。尽调上,下一组问题应聚焦队列质量,而不只是客户数量大小。[CU025, CU026, CU027, CU029, CU031, CU033]
| 指标 | 数值 / 空值 | 客群 | 置信度 | 尽调要求 |
|---|---|---|---|---|
| iOS 评分信号 | 59 个评分,4.6 分 | 创作者 / 准专业用户 | 中 | 按套餐层级拆出订阅人数、活跃用户和续费 |
| Android 规模主张 | 产品页称 Android 触达 1000 万+ | 创作者 / 准专业用户 | 中 | 提供活跃 MAU、下载到付费转化率和退款率 |
| 社区规模主张 | vivago 社区有数百万用户 | 创作者 / 准专业用户 | 中 | 提供 DAU/MAU、社区留存和高频用户活跃度 |
| 企业 NRR | null | 企业 API / 工作流客户 | 高 | 按开发者自助、SMB 和较大企业账户拆出 NRR |
| 企业 GRR / 流失 | null | 企业 API / 工作流客户 | 高 | 披露客户数流失、总留存和重新激活情况 |
| 平均合同期限 | null | 企业 / 媒体客户 | 高 | 提供合同期限、试点与生产部署拆分、续约节奏 |
| 合作伙伴来源收入集中度 | null | 媒体 / 营销合作 | 高 | 展示 BlueFocus、Shanghai Film、Huace 及同类合作伙伴的收入贡献 |
公开资料更能显示获客和可见活动,较难证明客户耐久度。空值表示缺少披露,不是缺少分析。
[CU009, CU011, CU012, CU025, CU026, CU027]证据对已具名战略合作方和可见创作者获客通道最强,对留存或合同质量的可见度最弱。
[CU009, CU014, CU016, CU017, CU018, CU019]6.4 扩张与集中风险
HiDream 有几条合理的扩张循环,但每条都带着集中风险。免费或低成本创作者漏斗可以扩张到订阅、点数和更重的智能体使用。媒体和营销合作在工作流规模化上线后,可以扩张为更高价值部署。企业 API 采用在信任和支持跨过采购门槛后可以增长。反向侧是,这些循环在年轻品类里都很脆弱。创作者需求可能快速流失,合作驱动收入可能波动很大,如果保证材料仍薄,企业转化也可能停滞。更广市场证据强化了谨慎:AI 视频行业变现仍早,中国赢家也常靠分发和价格,而不是长期验证过的留存。这并不否定 HiDream 的采用进展;只是说明当前公开证据更支持客户势头,而不是客户耐久性。投资判断的任务是把广度和粘性分开。[CU028, CU030, CU032, CU033, CU034, CU035]
| 扩张驱动因素 | 集中风险 | 影响 | 尽调路径 |
|---|---|---|---|
| 免费计划和每日积分 | 庞大的创作者漏斗顶部可能转化不佳 | 抬高使用量,却不能证明收入可持续 | 要求提供从免费用户到订阅用户再到重度用户的漏斗指标 |
| 分层订阅和积分 | 消费者支出波动大,对退款敏感 | 推高流失和客服风险 | 核查退款、拒付以及扣除渠道费后的净收入 |
| vivago 内 AI 智能体加购 | 高频用户功能可能拉高 ARPU,但覆盖人群更窄 | 加购可能真实存在,但会更集中 | 要求提供 Chat Agent 和 HiDreamClaw 功能附加率 |
| 媒体合作 | 收入可能集中在少数战略对手方 | 部署节奏不均,会扭曲客户质量判断 | 要求提供合作伙伴来源收入、部署数量和续约条款 |
| 企业 API 和工作流采用 | 信任和采购摩擦可能拖慢大合同 | 企业客户组合可能低于叙事预期 | 核查安全材料包、SLA 姿态和管线阶段转化 |
| 跨境营销工作流 | 平台 / 生态依赖会放大集中度和政策风险 | 渠道变化可能迅速扰动需求 | 按地区和平台梳理渠道依赖 |
HiDream 有多条可见扩张回路,但现有资料显示,大多数仍需要证明黏性和经济质量。
[CU023, CU024, CU028, CU030, CU031, CU032]6.5 展项
07风险
7.1 按严重程度排序的风险栈
HiDream 的风险画像不是由一个生死级红旗主导,而是由几项相关的中高暴露主导;它们叠加时可能变成关键风险。最重要的是面向公众的多模态生成监管合规、企业信任披露不足、竞争性价格压力,以及战略交易对手相对于持久经常性收入的集中风险。这些比简单需求风险更重要,因为公开证据已经支持采用广度、基准测试可信度和融资能力。问题在于,这些优势都不会自动证明增长安全、有粘性或能高效变现。即使模型声誉很强,如果合规成本上升、买方因信任包装犹豫,或创作者和合作伙伴渠道变现低于预期,公司仍可能损失价值。因此,看 HiDream 风险的正确镜头,是可见缓释后的剩余严重程度,而不是原始理论风险。风险之间的相关性是核心投资判断担忧。今天还没有单一缓释因素能独自清掉整组风险。[CR001, CR021, CR024, CR025, CR039, CR040]
| 规则 / 许可 / 案件 | 辖区 | 状态 | 可能性 | 严重性 | 缓释措施 | 剩余暴露 | 尽调路径 |
|---|---|---|---|---|---|---|---|
| 《生成式 AI 暂行办法》 | 中国 | 明确适用于面向公众的多模态生成 | 高 | 高 | 法律触点可见,内部政策工作也可能已经展开 | 高 | 要求提供算法备案 / 评估状态、内容审核工作流和治理负责人 |
| 《数据安全法》 | 中国 | 基础法律适用于数据处理;重要数据义务可能构成重大负担 | 中高 | 高 | 隐私页面和可能存在的内部控制 | 高 | 要求提供数据分级政策、重要数据认定和风险评估节奏 |
| 《网络安全法》 | 中国 | 基础网络运营者义务明确适用 | 中高 | 高 | 公开产品页显示已有运营控制,但看不出成熟度 | 高 | 要求提供安全计划、日志、漏洞管理和事件报告流程 |
| AI 内容标识规则 | 中国 | 商业合成媒体大概率受影响 | 中 | 中高 | 可能的产品层标识或元数据控制 | 中高 | 核查输出标识落地和审计流程 |
| 媒体和营销工作流中的 IP / 版权暴露 | 中国及跨境商业使用 | 未发现公开案件,但使用场景强度真实存在 | 中 | 高 | 开源许可清晰度和客户工作流边界划定 | 中高 | 要求提供来源政策、赔偿姿态、权利过滤机制和投诉日志 |
行顺序按剩余投资风险的重要性排列,不按抽象法律位阶。核心不确定性在执行成熟度,不在法律是否存在。
[CR002, CR003, CR004, CR005, CR006, CR007]剩余风险并不集中在简单的产品需求,而是卡在合规执行、信任披露不足、竞争带来的利润率压力和合作伙伴集中度上。
[CR001, CR007, CR013, CR021, CR026, CR039]7.2 监管、法律与信任暴露
最重的有据义务来自中国 AI、网络安全和数据法律。《暂行办法》明确覆盖中国境内面向公众的文本、图像、音频和视频生成。《数据安全法》和《网络安全法》又增加了安全运行、日志、事件响应和风险评估等基线职责。对 HiDream 来说,多模态很重要,因为每增加一种输出类型,滥用和合规暴露面都会变宽。开源许可和商业友好信息会加速采用,但也提高了外界对来源、权利清洁度和下游使用的期待。公开记录确实显示了隐私和条款页面,以及 GitHub 和 Hugging Face 上可见的许可透明度。它没有显示的是大型客户通常想要的企业信任包:公开认证、DPA 流程、SLA 姿态或事件透明度。因此,监管和法律栈可读,但缓释成熟度只部分可读。这种错配会形成可见尽调折价。[CR002, CR003, CR004, CR005, CR006, CR007]
| 失效模式 | 可能性 | 严重性 | 缓释成熟度 | 剩余暴露 | 未解决缺口 |
|---|---|---|---|---|---|
| 企业信任材料包薄弱,拖慢或挡住大合同 | 中高 | 高 | 中低 | 高 | 看不到信任中心、DPA 材料包、认证或 SLA 摘要 |
| 大型 API 或应用界面发生安全或滥用控制事件 | 中 | 高 | Unknown | 高 | 未发现公开事件或复盘记录 |
| 营销或媒体工作流出现输出完整性 / 品牌安全失效 | 中 | 高 | Unknown | 中高 | 没有公开证据显示政策 QA 指标或客户护栏 |
| 提示词智能体或工作流自动化失效,扩散低质量输出 | 中 | 中 | 中 | 中 | 需要审查提示词层和模型层控制架构 |
| 商业可靠性没有跟上基准测试声誉 | 中 | 中高 | 中 | 中高 | 需要可用性、延迟和支持质量指标 |
本表关注实际运营失效:即便模型测试表现好,这些问题也会侵蚀转化和留存。
[CR010, CR011, CR012, CR017, CR018, CR030]HiDream 的几类风险彼此勾连:政策、信任、伙伴和竞争问题,会一起传导到转化质量、成本结构和估值容忍度。
[CR012, CR021, CR026, CR028, CR031, CR039]7.3 运营、依赖与执行风险
从运营上看,HiDream 对硬件供应剧本的暴露,似乎小于对可靠性、算力经济性和商业化质量的暴露。公司首页声称有 5000 多亿次 API 调用和 100 多个重点客户,这意味着宕机、滥用控制失败或薄弱支持工具都会产生实质下游影响。中国算力环境在改善,但这并不能消除成本或可用性风险;它更多只是改变压力出现的位置。中国视频 AI 价格竞争仍很激烈,公开品类证据仍指向行业层面 ARR 不成熟。BlueFocus、Shanghai Film 和 Huace 等战略合作创造了有价值证明,但也凸显出,公司需要把少数重要关系转成可重复商业项目。人的瓶颈因此在执行:安全、合规、客户成功、解决方案架构和法务协同,需要与模型和产品野心同步成熟。规模会迅速放大小流程弱点。[CR017, CR018, CR019, CR020, CR022, CR023]
| 依赖 | 对手方 | 角色 | 集中度 | 失效情景 | 严重性 | 缓释措施 | 剩余暴露 |
|---|---|---|---|---|---|---|---|
| App 和 Web 分发 | 应用商店和 Web 基础设施 | 创作者获客和变现通道 | 中 | 政策变化、审核升级或分发摩擦冲击增长 | 中高 | 维持多端分发和直接 Web 使用 | 中 |
| 战略媒体和营销合作伙伴 | BlueFocus、Shanghai Film、Huace 及同类标杆客户 | 验证、行业可信度和潜在部署扩张 | 中高 | 少数标杆客户关系未能转化为可复制收入 | 高 | 扩大具名客户基础,披露生产环境参考案例 | 高 |
| 算力和网络环境 | 国内算力提供商和基础设施生态 | 推理可用性和成本基础 | 中高 | 容量或成本飙升伤及服务水平和利润率 | 高 | 利用国家算力建设和效率提升 | 中高 |
| 监管机构和政策解释者 | CAC 及其他机构 | 制定运营、备案、标识和事件规则 | 高 | 规则收紧速度快于产品控制成熟 | 高 | 搭建持续合规职能和产品化控制 | 高 |
| 开源生态用户 | 开发者、下游部署者、fork | 加速采用,但削弱控制 | 中 | fork 被误用或低质量衍生部署会反噬声誉 | 中高 | 明确许可、文档,并与下游误用切割 | 中高 |
这里的依赖包括监管机构、开源用户等非传统卡点,因为它们的行为会实质影响变现和信任。
[CR013, CR019, CR020, CR026, CR027, CR034]| 角色 / 职能 | 依赖或缺口 | 可能性 | 严重性 | 缓释措施 | 尽调路径 |
|---|---|---|---|---|---|
| 合规 / 政策负责人 | 需要持续把 AI、网络安全、标识和数据规则转成产品运营动作 | 中高 | 高 | 专门合规负责人,并接入产品流程 | 询问谁负责监管路线图、备案和审计准备 |
| 安全 / 隐私负责人 | API 和应用界面需要事件响应、日志、访问控制和客户保障 | 中 | 高 | 集中安全职能,汇报线清晰 | 要求提供组织架构图、认证路线图和事件演练 |
| IP / 法务运营 | 媒体和商业使用需要来源溯源、投诉处理和合同姿态 | 中 | 高 | 专业律师和权利治理工作流 | 要求提供权利审查流程和权利主张 / 升级日志 |
| 客户成功 / 解决方案架构 | 需要把战略标杆客户转成生产工作流采用 | 中高 | 中高 | 专门的部署和续约负责人 | 核查试点到生产转化指标和人员配比 |
| 财务 / 分析 | 需要看清价格实现、单位经济和合规成本叠加 | 中 | 中高 | 利润率分析和情景规划 | 要求提供产品级贡献利润率和支持成本曲线 |
这些是根据公开义务和披露缺口推断出的角色依赖型执行风险,不是证明 HiDream 缺少这些职能。
[CR021, CR024, CR025, CR035, CR036, CR038]HiDream 依赖监管机构、算力生态、应用触点、战略客户和开源用户;这些外部方一变,商业结果也会很快改变。
[CR019, CR020, CR026, CR027, CR034, CR035]7.4 缓释、监控与终止标准
HiDream 确实有缓释因素。作为一家年轻模型公司,它在产品范围、开源许可和部分技术栈上异常透明。中国算力建设和公司的强资本获取能力,也降低了近期生存风险。不过,决定性缺失证据仍是运营证明:信任材料、达到留存级别的客户质量、集中度披露,以及能顶住合规和价格压力的单位经济性。因此,投资者应监控一组简短的公开和私下触发点。如果公司能展示企业信任材料、战略试点稳定转成生产,并且没有毛利率坍塌证据,几项当前折价会收窄。相反,如果监管要求变硬、合作伙伴集中度上升,或价格压力迫使不经济增长,就应快速削减投资逻辑。这是一组只有主动监控、并用证据而不是乐观来清除时才可管理的风险。随着公司从验证走向规模,每一轮融资后,信心门槛都应提高。[CR030, CR031, CR033, CR034, CR037, CR038]
| 风险 | 可监测触发因素 | 阈值 / 事件 | 行动含义 |
|---|---|---|---|
| 监管合规缺口 | 出现执法、下架、内容移除证据,或备案准备度缺失 | 出现任何重大执法行动,或无法展示运营控制 | 暂停投资测算,直到合规准备度得到证明 |
| 企业信任缺口 | 企业市场推进继续,但信任材料仍缺位 | 大客户没有 DPA、认证路线图、安全材料包或 SLA 姿态 | 假设 B2B 转化更慢、收入组合质量更低 |
| 价格压缩 / 利润率压力 | 商业增长伴随更重折扣或上升的支持负担 | 反复降价、创作者占比过高,或有量增却无利润率证据 | 下调收入质量假设,并施加更高风险折扣 |
| 战略对手方集中 | 收入或部署价值集中在少数标杆客户合作 | 头部合作伙伴集中度仍高,具名客户没有扩宽 | 把客户故事视为集中且脆弱 |
| 跨境和数据治理缺口 | 全球使用扩张,但本地化数据或政策控制没有跟上 | 管理层无法梳理数据流、区域控制或事件义务 | 提高合规折扣,限制投资测算信心 |
否决标准设计成可监测、与决策相关,而不只是制造警报。
[CR028, CR029, CR031, CR037, CR038, CR039]7.5 展项
08估值
8.1 投资建议与价格纪律
HiDream.ai 通过了严肃后期 AI 公司第一道测试:业务真实到足以让国资基金、银行系资本、影视产业战略方、开源开发者和第三方基础设施伙伴都在公开记录中出现。问题在第二道测试,而这道测试对估值更关键。公开证据没有披露经审计收入、ARR、毛利率、烧钱速度、留存或清算条款,因此市场几乎要求投资人主要靠融资势能、产品基准和活跃度代理指标来承销一个独角兽以上的价格。融资势能、产品基准和活跃度代理指标有用,但还不足以支持买入。因此建议是继续研究。如果尽调拿到高质量经常性企业收入和毛利证明,评级可以很快上调;如果拿不到,当前估值最好先看作一个仍需经济性验证的故事。换句话说,只有证据变强之后,这个机会才可投;不能因为市场热度高就投。[CV001, CV002, CV003, CV004, CV006, CV007]
| 维度 | 评估 | 证据基础 | 决策含义 |
|---|---|---|---|
| 建议 | 继续研究 | 公司真实存在,但价格支撑不完整,也没有公开材料打通收入质量证据链。 | 只有管理层愿意开放财务和股权结构材料包,才推进尽调。 |
| 置信度 | 中 | 融资、产品和基准证据可见,但估值论证仍依赖间接证明。 | 披露质量改善前,任何投资测算都应视为暂定。 |
| 风险评级 | 高 | 2026 年 AI 融资市场更挑剔,估值却依赖未披露经济性。 | 要求强下行保护,否则推迟进入。 |
| 估值立场 | 偏高 | 独角兽以上定价可见,但公开材料对内在价值的支撑,弱于对市场热度的支撑。 | 没有变现质量证据,不要支付稀缺性溢价。 |
| 决策含义 | 对价格敏感的尽调案例 | 如果确切投后估值仅略高于 $1B,且收入质量很强,这个估值区间仍可辩护;若条款明显更贵,当前证据太薄。 | 用尽调结果决定继续跟踪、谈判,还是退出。 |
本表有意围绕价格敏感性,而不只看公司质量。核心问题是,可见的独角兽叙事是否有足够公开经济性支撑,足以证明入场合理。
[CV001, CV002, CV003, CV014, CV016, CV030]| 论点 | 支持证据 | 什么会改变判断 |
|---|---|---|
| HiDream 不只是演示实验室 | 公司刚完成 RMB 1.5B 融资,已有 200+ 个 API、100+ 个重点客户、开源发布,以及电影 / 媒体战略投资方。 | 经验证的经常性收入和留存,会把这个判断从合理推断提升为可写进投资模型的结论。 |
| 战略资本可能带来真实渠道价值 | Shanghai Film、Huace 和国资背景投资方能打开产业用例和数据合作,普通实验室很难拿到。 | 需要收入归因、部署案例,以及这些关系能产出长期合同的证据。 |
| 开源和基准证明带来期权价值 | HiDream-O1-Image 已进入开源前列,第三方分发入口也已经存在。 | 还需要证明开源采用能转化为付费企业客户或 API 变现,而不只是声量。 |
| 当前估值可能仍跑在基本面前面 | 公司没有公开收入、ARR、利润率或现金消耗,而上市可比公司的交易倍数远低于私募 AI 融资倍数。 | 完整财务包可以解释溢价;拿不到之前,折价应当保留。 |
| 融资动能不等于内在价值 | 2026 年行业证据显示,赢家仍能拿到资本,但资金更挑剔;过去的生成式 AI 领军者也可能停滞或被重组融资。 | 更干净的资本结构和可见的单位经济模型,会降低对融资观感的依赖。 |
反论点不是 HiDream 缺乏产品质量,而是估值支撑仍过度依赖推断出来的经济性。
[CV004, CV006, CV009, CV010, CV011, CV012]从融资势头和产品证明,推导到受经济性缺口约束的「继续研究」建议。
仅为概念综合。该流程解释决策逻辑,不预测因果估值模型。
[CV001, CV003, CV006, CV009, CV014, CV030]8.2 融资背景与估值历史
最新一轮融资无论放在本土还是全球 AI 语境里都很大。HiDream 在 2026 年 7 月下旬完成 RMB 1.5 billion Series C 轮,多个来源称公司约三个月内三轮合计融资超过 RMB 2.1 billion,并在此过程中跻身独角兽。财团的重要性几乎不亚于金额。国家级、省级国资基金、银行资本以及影视战略方同时出现,说明投资人既把 HiDream 看作战略基础设施,也看作应用型媒体科技平台。资本组合能支撑其相对普通创作者应用的估值溢价。但估值历史仍然稀疏。公开记录没有显示早期轮次的准确投前或投后估值、累计融资额,也没有披露最新一轮之下的优先权结构。因此,该估值只能证明市场愿意出清,不等于内在价值。用于承销时,估值历史有参考价值,但不完整。[CV001, CV002, CV003, CV004, CV005, CV025]
| 日期 / 阶段 | 公开估值信号 | 实际披露内容 | 解读 |
|---|---|---|---|
| 2023 成立期 | 未披露公开估值 | 登记记录显示,北京运营主体于 2023-03-02 成立,但公开资料里看不到早期融资估值。 | 可作为运营公司锚点,不能当作估值证据。 |
| Series C 前约三个月的密集融资 | 近期合计超过 RMB 2.1B,但逐轮估值细节未公开 | Seedtable、The SaaS News、Media OutReach 和 TMTPost 均指向约三个月内完成三轮融资。 | 动能很清楚;早期估值和稀释不清楚。 |
| 2026-07-27 Series C | 独角兽 / >$1B | 融资 RMB 1.5B,披露战略财团,未披露确切投后估值。 | 确认当前市场胃口和独角兽下限,但不是精确公允价值。 |
| 2026-08 公开承做判断 | 没有比独角兽以上更干净的标记 | 公开记录仍缺少确切估值、历史累计融资额,以及该轮后的优先权条款。 | 任何投资人都必须在尽调中补齐缺失经济性,才能把该估值当作可投标记。 |
这里故意保持稀疏,因为公开记录本身稀疏。重点不是精确估值时间线,而是融资能见度跑在估值透明度前面。
[CV001, CV002, CV003, CV008, CV039]| 可比对象 | 日期 / 事件 | 披露金额 / 估值 | 相关性 | 局限 |
|---|---|---|---|---|
| HiDream.ai Series C 轮 | 2026-07-27 后期轮 | 融资 RMB 1.5B;独角兽 / >$1B 估值状态 | 该资产当前最直接的成交价格。 | 确切投后估值和稀释条款仍未披露。 |
| Runway Series C 延长轮 | 2023-06 融资延长 | 融资 $141M,估值 $1.5B | 可作为生成式视频同业参考:创作者 / 企业媒体工具在强 AI 动能下可能达到的成交位置。 | 交易较早,资本市场环境也不同。 |
| Synthesia Series E | 2026-01 融资轮 | 融资 $200M,估值 $4B | 说明品类领先的企业 AI 视频平台,在企业定位更清晰时可以拿到高溢价。 | Synthesia 的企业披露强得多,业务组合也不同。 |
| Stability AI 2024-25 融资回顾 | 2024 年融资加 2025 年少数股权交易 | 总融资约 $225M-$231M,法律和现金消耗压力仍在 | 一个警示性的开源媒体 AI 类比:产品相关性消除不了融资或法律风险。 | 不是一笔干净的 2026 年单轮定价交易,且依赖市场数据拼接。 |
这些交易按战略相关性选择,不是为了拼出干净估值阶梯。样本混合了一个当前直接标记,以及相邻私有生成式媒体融资。
[CV001, CV003, CV025, CV026, CV027]8.3 可比公司、情景区间与 DCF 缺口
估值三角测算原则上简单,落地却很别扭。当前市场里,公开市场创意和软件可比公司的收入倍数大约在 0.2x 到 6.2x;AI 融资数据集仍显示,私募轮次的收入倍数中位数高得多,超过 24x。HiDream 夹在两者之间:技术和战略信号强于下行中的图库平台,但公开财务证明远少于能够支撑公开市场溢价倍数的企业软件公司。HiDream 没有公开披露收入或 ARR,经典 DCF 就站不住脚。唯一负责任的方法是情景分析:问清什么证据能支撑独角兽底价,什么证据能支撑更高区间,什么失败会压垮溢价。在 $1 billion 底价下,如果收入为 $50 million,需要的收入倍数约 20x;如果收入为 $150 million,则约 6.7x。估值对一个缺失数字极其敏感。[CV014, CV015, CV016, CV017, CV018, CV019]
| 可比对象 | 指标 | 倍数 / 估值 / 状态 | 相关性 | 局限 |
|---|---|---|---|---|
| Adobe | 2026 市值 / 2025 收入 | $108.5B / $23.76B = ~4.6x 收入 | 大型创意软件龙头,具备生成式 AI 敞口和已变现的创意工作流。 | 规模、业务多元化和盈利能力都远高于 HiDream。 |
| C3 AI | 2026 市值 / 2025 收入 | $1.61B / $0.30B = ~5.4x 收入 | 上市纯 AI 软件标的,可用来观察软件式 AI 情绪。 | 不是媒体生成公司,增长也已经放缓。 |
| Duolingo | 2026 市值 / 2025 收入 | $6.41B / $1.03B = ~6.2x 收入 | 高增长订阅平台,显示在变现清晰时,上市软件倍数的上沿。 | 语言学习比多模态媒体 AI 更可读,也更经常性。 |
| Kuaishou | 2026 市值 / 2023 收入 | $23.53B / $16.01B = ~1.5x 收入 | 中国消费内容平台,具备视频经济性和分发规模。 | 广告和平台业务宽得多;不是基础模型公司。 |
| Shutterstock | 2026 市值 / 2025 收入 | $0.21B / $0.98B = ~0.2x 收入 | 与创意资产变现和授权敏感性直接相关。 | 公开市场定价反映的是结构性压力和弱增长,不是前沿 AI 上行空间。 |
| Getty Images | 2026 市值 / 2026 TTM 收入 | $0.18B / $0.98B = ~0.18x 收入 | 图像业务暴露在 IP 问题下,Getty 是重要的权利与授权类比。 | Getty 是成熟资产授权平台,不是高增长模型开发者。 |
公开创意媒体倍数远低于私募 AI 融资中位数。这个缺口正是 HiDream 未披露收入基数如此关键的核心原因。
[CV016, CV017, CV018, CV019, CV020, CV021]| 情景 | 核心假设 | 估值 / 回报逻辑 | 概率信号 |
|---|---|---|---|
| 熊 | HiDream 的收入质量仍不透明,价格压力加剧,市场开始按更接近上市创意公司或低层级 AI 可比公司的方式给生成式媒体业务定价。 | ~$0.6B-$0.9B。若公司无法证明软件式经济性,或更大规模战略退出路径出现前融资环境收紧,独角兽下限会被打穿。 | 没有财务披露、企业转化证据弱,或出现降价轮 / 结构性稀释融资迹象。 |
| 基准 | HiDream 把基准和战略动能转化为足够商业证据,守住独角兽下限,但仍拿不出稀缺性溢价所需的干净经济性。 | ~$0.9B-$1.3B。该区间认为当前估值只在低端大致可守,且前提是后续尽调补齐收入质量缺口。 | 战略采用持续推进,且没有明显负面意外;但仍没有经审计收入桥。 |
| 牛 | HiDream 披露强劲经常性企业收入、清晰利润率改善,以及开源和电影 / 媒体工作流向长期合同的可复制转化。 | ~$1.5B-$2.4B。要高于当前下限拿溢价,增长质量必须开始更像品类龙头,而不是有前景但不透明的前沿媒体实验室。 | 经审计财务、留存证明、股权结构清晰,以及世界模型或电影工作流实现规模化变现的证据。 |
这里故意不把 DCF 作为主要方法。公开记录仍缺少足够扎实的收入、利润率、burn、capex 和营运资本输入,撑不起财务级折现现金流模型。
[CV030, CV031, CV032, CV033, CV034, CV035]在不同假设收入水平下,要支撑独角兽估值底线所需的收入倍数。
本图并不声称 HiDream 达到其中任何收入水平。它说明独角兽估值底线对未披露收入数字有多敏感。
[CV003, CV024, CV030, CV032]相对于公开可见独角兽估值底线的悲观、基准和乐观估值带。
区间是判断带,不是模型定价结果。它们用于框定尽调纪律,而不是替代数据室模型。
[CV003, CV035, CV036, CV037, CV038, CV043]按 IC 视角给 HiDream 打分,覆盖产品证明、披露质量、估值支撑和风险。
分数是判断性综合,不是算法。低分反映估值支撑缺失,而不是模型能力弱。
[CV004, CV006, CV009, CV014, CV029, CV030]8.4 尽调清单、投资逻辑失效点与退出准备度
剩下的工作是实操问题,不是哲学问题。投资人需要准确的投后估值、优先股堆叠、按分部拆开的收入桥、留存数据、毛利率证据,以及足够的现金流细节,才能判断这到底是类软件业务,还是一个套着软件外壳、极耗资本的世界模型实验室。战略上行空间已经可见:与 Shanghai Film 和 Huace 的合作能帮助 HiDream 把基准可信度转成专有内容工作流;开源分发也能扩大开发者触达。风险在于,战略信号可能比变现更容易营销。没有监管文件级别的披露,未来任何交叉轮融资或 IPO 式流程仍会面临问题:稀释、盈利质量,以及用户或 API 活动能否转成耐久的企业现金流。因此结论仍是继续研究,而不是回避:HiDream 有值得承销的部分,但下一步应是尽调材料包,不是信仰式跳跃。价格仍需要证据。[CV025, CV026, CV027, CV029, CV030, CV031]
| 触发器 | 阈值 | 如何传导到论点 | 行动含义 |
|---|---|---|---|
| 确切估值附带重优先权 | 尽调发现多倍清算优先权、惩罚性反稀释,或大量内部人二级出售悬顶。 | 表面可接受的名义估值,可能对新普通股或后期优先股投资人变得没有吸引力。 | 继续推进前,扣除悬顶后重新切分上行空间。 |
| 收入质量弱于活动指标暗示 | 管理层无法把重点客户、API 调用或用户数桥接到经常性付费收入和留存。 | 会打破战略和产品证明足以支撑软件式溢价的论点。 | 转向熊情景或暂停。 |
| 价格压力跑赢模型优势 | HiDream 必须持续打折才能赢得用量,而利润率仍未被证明。 | 把基准胜利变成低质量增长,而不是长期现金流。 | 没有抵消性利润率证据,就不要支付独角兽以上倍数。 |
| 战略合作无法转化为变现 | 电影 / 媒体标志仍只是叙事资产,而不是合同或客群证明。 | 拿掉了相对普通开源实验室给溢价的关键理由。 | 下调估值区间,推迟进入。 |
| 披露追上前就启动退出流程 | 在没有申报级财务和股权结构细节时,就启动 crossover 或类 IPO 流程。 | 公开或准公开市场投资人会打披露折扣,或直接退出。 | 在材料改善前维持继续研究。 |
每个触发器都可在尽调或后续披露中核查,而且会直接改变估值支撑,而不只是泛泛影响公司质量。
[CV029, CV030, CV039, CV041, CV042, CV043]| 主题 | 缺失证据 | 重要性 | 负责人或尽调路径 |
|---|---|---|---|
| 确切投后估值和股权结构 | 精确估值、股份类别、清算优先权、反稀释机制,以及任何内部人二级交易条款 | 决定该轮按名义价格对新投资人是否真的有吸引力。 | 数据室里的财务 + 法律尽调。 |
| 收入质量桥 | 分部收入、客户集中度、留存、续约、流失,以及 API 或用户活动与付费用量之间的联系 | 没有这条桥,估值靠的是活动代理指标,而不是变现质量。 | 管理层问答加财务工作流。 |
| 利润率和算力经济性 | 分部毛利率、GPU 承诺、托管成本,以及模型效率提升能经受真实部署的证据 | 检验 HiDream 是在变得软件化,还是仍然吃资本。 | 技术尽调加 CFO 模型复核。 |
| 合作伙伴变现 | Shanghai Film、Huace 及相关战略渠道的合同金额、部署范围和续约证据 | 说明战略投资人是真实收入渠道,还是主要撑叙事。 | BD / 产品尽调,配合合同审查。 |
| IPO / crossover 准备度 | 申报级披露、治理包、董事会结构和控制体系状态 | 未来更高估值或退出,取决于投资人是否相信业务经得起公开市场审视。 | 法律、审计和治理尽调。 |
这些问题是从动人叙事走向可承销资产的最短路径。多数问题答清前,当前建议不应改变。
[CV030, CV031, CV039, CV040, CV043, CV044]8.5 展示材料
免责声明
本报告是基于公开证据的尽调快照,不构成投资建议。重要的财务、法律、技术和合同事实仍未公开;做出任何投资决定前,应直接向管理层和一手文件核验。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | HiDream.ai says it was founded in March 2023. | 高 | SO001, SO002 |
| CO002 | Qichacha records Beijing Zhixiang Future Technology Co., Ltd. as established on 2023-03-02 in Haidian District, Beijing. | 高 | SO003, SO004 |
| CO003 | HiDream.ai positions itself as a global leader in multimodal generative artificial intelligence. | 中 | SO001 |
| CO004 | The company says it develops multimodal foundation models and AI applications for marketing, film and television production, and social-media content creation. | 中 | SO001 |
| CO005 | The official homepage says HiDream.ai has a base model that simultaneously supports text, image, video, and 3D modalities. | 中 | SO002 |
| CO006 | The official homepage says HiDream.ai’s proprietary data and model parameters together exceed 200 billion. | 中 | SO002 |
| CO007 | HiDream.ai publicly markets five product lines: HiHarness, HiBurst, Zhenzan, vivago, and HiDreamFans. | 中 | SO002 |
| CO008 | The homepage describes HiHarness as an enterprise-grade multimodal AI platform with more than 200 APIs. | 中 | SO002 |
| CO009 | The homepage says HiHarness is trusted by more than 100 key accounts. | 中 | SO002 |
| CO010 | The homepage says HiHarness has processed more than 500 billion API calls. | 中 | SO002 |
| CO011 | HiDream.ai says its core team brings together research and management talent from top universities and companies including Microsoft, Tencent, and ByteDance. | 中 | SO001 |
| CO012 | Multiple public sources identify Tao Mei as founder and chief executive of HiDream.ai. | 高 | SO001, SO006, SO007 |
| CO013 | Baidu Baike and IEEE-profile sources describe Tao Mei as a former JD.com vice president and former Microsoft Research leader. | 高 | SO006, SO007 |
| CO014 | The official site’s news feed refers to Wang Ke as HiDream.ai co-founder and COO. | 中 | SO001 |
| CO015 | Qichacha lists Wang Ke as manager and legal representative, Tao Mei as director, Chen Ruizhe as supervisor, and Jiao Yang as finance lead for the Beijing operating entity. | 中 | SO003 |
| CO016 | Qichacha shows the Beijing entity is wholly owned by Zhixiang Future (Hefei) Technology Co., Ltd. with registered capital of CNY10 million. | 中 | SO003 |
| CO017 | Aiqicha describes the company as a 2024 technology-SME and a 2025 national high-tech enterprise. | 中 | SO004 |
| CO018 | Aiqicha says the company has 23 patent records, four software copyrights, and one brand project in the public record. | 中 | SO004 |
| CO019 | HiDream.ai completed a RMB 1.5 billion Series C round in late July 2026. | 高 | SO008, SO009, SO011 |
| CO020 | The Series C was co-led by the National Social Security Fund Sichuan Revitalization Sci-Tech Innovation Fund, ICBC Capital, Hongyi Asset Management, and Dunhong Capital. | 高 | SO008, SO009, SO011 |
| CO021 | The Series C also included investors such as Xiamen ITG Capital, Shanghai Film New Vision Fund, Hubei Yangtze River Industry Investment Group, Huace Film & TV, Hangyuan Capital, Chuangyunhai Capital, Huafu Investment, Yuhang Financial Holding, Bank of Communications Capital, and Wakamatsu Fund. | 中 | SO008, SO009 |
| CO022 | Returning investors in the Series C included Hefei Industrial Investment, Fortune Capital, Kingpo or GPC Capital, Jinhua Capital, Zhongzhe Capital, and Caixin Capital. | 中 | SO008, SO009, SO011 |
| CO023 | Public coverage says HiDream.ai raised more than RMB 2.1 billion across three rounds completed within roughly three months. | 高 | SO008, SO009, SO010, SO011 |
| CO024 | The latest financing pushed HiDream.ai above a $1 billion valuation and into unicorn status according to multiple reports. | 高 | SO008, SO009, SO010 |
| CO025 | Media OutReach says the new capital will fund native omni-modal world models plus product and commercial ecosystem expansion. | 中 | SO009 |
| CO026 | Media OutReach says HiDream.ai serves over 50 million professional users and more than 40,000 enterprise customers across more than 100 countries and regions. | 中 | SO009 |
| CO027 | The official homepage publishes a more conservative commercial proof set of 100-plus key accounts and 500-plus billion API calls rather than the broader user totals in the financing press release. | 中 | SO002, SO009 |
| CO028 | Qichacha and Aiqicha do not provide a clean public headcount, with Qichacha showing zero insured staff in the 2025 annual report and Aiqicha summarizing 32 insured people in one snapshot. | 中 | SO003, SO004 |
| CO029 | Shanghai Film announced a strategic investment cooperation with HiDream.ai focused on IP development, cinema-scene marketing, joint AIGC labs, and talent development. | 中 | SO014 |
| CO030 | Jiemian reported that Huace Film planned a strategic investment and cooperation with HiDream.ai around AI audiovisual production, content assets, and talent training. | 中 | SO015 |
| CO031 | The official About page highlights Shanghai Film, Huace, BlueFocus, Tencent Cloud, Hubei Yangtze River Film Group, and TikTok Shop award milestones as part of the company’s current partner narrative. | 中 | SO001 |
| CO032 | Science and Technology Daily reported that HiDream-O1-Image-1.5 ranked second globally on Artificial Analysis on 2026-06-11 with a 1265 Elo score. | 中 | SO016 |
| CO033 | The 36Kr benchmark feature said HiDream-O1-Image-1.5 outranked Google Nano Banana 2, NVIDIA Cosmos3-Super-Text2Image, and ByteDance Seedream 4.0 in its June 2026 snapshot. | 中 | SO017 |
| CO034 | GeekPark reported that the open-source HiDream-O1-Image reached 1187 Elo over more than 3000 comparison pairs and topped the open-source image leaderboard. | 中 | SO018 |
| CO035 | The current Artificial Analysis leaderboard snapshot fetched on 2026-08-11 shows HiDream-O1-Image-1.5 at number 11 globally with 1227 Elo and a listed price of $80 per 1000 images. | 中 | SO019 |
| CO036 | China Biz Insider says HiBurst became one of TikTok’s top five official service providers, producing more than one million e-commerce marketing videos annually and supporting more than RMB100 million of GMV. | 中 | SO020 |
| CO037 | China Biz Insider says vivago has covered more than 100 countries and more than 40 million users, while Zhenzan has generated over 5000 minutes of short comic drama content. | 中 | SO020 |
| CO038 | The official HiDream-O1-Image GitHub repository says the open-source 8B model and prompt agent were released on 2026-05-08 and the Dev-2604 variant on 2026-05-14. | 中 | SO025 |
| CO039 | China’s generative AI interim measures impose truthfulness, legality, and public-order duties on providers of generative AI services. | 中 | SO021, SO023 |
| CO040 | The 2025 AI-generated-content identification rules increase labeling and traceability requirements for synthetic media distributed in China. | 中 | SO022, SO024 |
| CM001 | HiDream addresses a narrower commercial market than “all generative AI”: it sells or enables multimodal content-generation workflows across APIs, marketing tools, film-production tooling, and consumer creator software. | 高 | SM001, SM002, SM003 |
| CM002 | HiHarness positions HiDream in model-as-a-service and enterprise workflow infrastructure rather than only consumer creation. | 中 | SM001, SM003 |
| CM003 | HiBurst positions HiDream against e-commerce and marketing-production budgets, not just R&D budgets. | 中 | SM003, SM021 |
| CM004 | Zhenzan and the media-investor mix point to film, episodic video, and entertainment production as explicit target segments. | 中 | SM003, SM022 |
| CM005 | vivago extends HiDream into a lower-ARPU but broader creator-consumer layer through subscriptions, templates, and model routing. | 中 | SM016, SM003 |
| CM006 | Grand View’s broad generative-AI market lens values the category at USD 22.2B in 2025 and USD 29.6B in 2026 with 40.8% CAGR to 2033. | 中 | SM004 |
| CM007 | Fortune Business Insights publishes a much larger broad-market lens: USD 103.58B in 2025 and USD 161B in 2026 with 29.3% CAGR to 2034. | 中 | SM005 |
| CM008 | The spread between Grand View and Fortune shows that generic “GenAI TAM” numbers are definition-sensitive and not directly usable as HiDream’s serviceable market. | 中 | SM004, SM005 |
| CM009 | ResearchAndMarkets separates the stack and estimates 2024 foundation-model revenues at only USD 4.1B, versus USD 17B for development platforms and USD 132.3B for GPU hardware. | 中 | SM006 |
| CM010 | For HiDream, the most relevant sizing lens sits between the broad application market and the narrower foundation-model revenue pool because the company participates in both models and applications. | 中 | SM001, SM003, SM006 |
| CM011 | Stanford AI Index says generative AI reached roughly 53% population-level adoption within three years, indicating unusually fast user education for products like creator tools. | 中 | SM007 |
| CM012 | The same Stanford report says organizational AI adoption reached 88%, reinforcing that enterprise buyers are actively experimenting with deployment. | 中 | SM007 |
| CM013 | Stanford reports industry produced over 90% of notable AI models in 2025, which favors venture-backed labs that can ship quickly and commercialize directly. | 中 | SM007 |
| CM014 | HiDream’s open-source releases and API posture suggest a developer-distribution strategy alongside direct products. | 中 | SM001, SM020 |
| CM015 | HiDream’s current market boundary excludes text-only productivity AI and most horizontal enterprise copilots; its evidence base is overwhelmingly visual and multimodal. | 高 | SM001, SM002, SM003 |
| CM016 | The company also should not be sized against pure GPU or infrastructure spend even though infrastructure economics shape margins and entry barriers. | 中 | SM006, SM025 |
| CM017 | Grand View identifies media and entertainment as a dominant generative-AI end market, matching HiDream’s film and creator positioning. | 中 | SM004, SM003 |
| CM018 | Grand View also says software captured 64.1% of 2025 generative-AI revenue, which is directionally favorable for software-led companies like HiDream. | 中 | SM004, SM001 |
| CM019 | Fortune highlights marketing and advertising as a fastest-growing vertical, which aligns with HiBurst and social-content use cases. | 中 | SM005, SM003 |
| CM020 | Media OutReach frames HiDream as already serving 50M+ professional users and 40K+ enterprise customers, implying cross-segment reach from self-serve creators to enterprises. | 中 | SM003, SM021 |
| CM021 | The buyer, user, and payer often separate in HiDream’s target workflows: creators use the tools, marketing or studio heads own budgets, and IT/product teams may own API integration. | 中 | SM001, SM003, SM016 |
| CM022 | Status-quo substitutes include human creative agencies, stock media, editing suites, and incumbent ad-production workflows, not only rival model vendors. | 中 | SM004, SM008 |
| CM023 | Direct AI substitutes span high-end enterprise tools (Runway, Veo), mass-market Chinese video models (Kling, Hailuo), and aggregators or routing layers. | 中 | SM011, SM012, SM013, SM014, SM016 |
| CM024 | Runway’s official positioning around video, images, audio, enterprise, and 60M+ creatives shows the level of creator-platform maturity HiDream is chasing in parts of the market. | 中 | SM012 |
| CM025 | Google Veo’s native audio, filmmaker positioning, and 1080p/4K outputs show that global leaders are competing on integrated multimodal control, not just clip generation. | 中 | SM011 |
| CM026 | Kling and Hailuo confirm that Chinese competitors are active across API and creator surfaces, increasing pressure on HiDream in its domestic and regional markets. | 中 | SM013, SM014, SM009 |
| CM027 | China Biz Insider reports a domestic cost advantage for Chinese video models versus Google and Sora, which can accelerate adoption in cost-sensitive segments such as ads and short video. | 中 | SM009 |
| CM028 | Forbes argues Chinese video-AI labs benefit from vertical integration, aggressive pricing, and local distribution ecosystems, creating a structurally different market than Western subscription-only creator tools. | 中 | SM008, SM009 |
| CM029 | Forbes further argues Google and Runway are better positioned in agency, enterprise, and Hollywood segments, implying HiDream must choose where to meet Western-grade control and compliance requirements. | 中 | SM008, SM011, SM012 |
| CM030 | OpenAI’s discontinuation of Sora as a standalone web/app product weakens one direct creator competitor but also highlights the difficulty of monetizing pure standalone video generation. | 中 | SM015, SM008 |
| CM031 | vivago’s model-aggregation strategy suggests user demand is shifting from allegiance to one model toward routing across models for quality, speed, or price. | 中 | SM016, SM010 |
| CM032 | Atlas Cloud’s comparison explicitly recommends multi-model selection by task, which supports the view that workflow orchestration can matter as much as base-model leadership. | 中 | SM010, SM016 |
| CM033 | Chinese generative-AI service rules and synthetic-content labeling rules add friction to distribution, especially for video and audio products that generate public-facing media. | 高 | SM023, SM024, SM025 |
| CM034 | These rules matter more to HiDream than to a text-only AI app because its products create image, video, and audio outputs that are more likely to require provenance, moderation, or watermarking controls. | 中 | SM001, SM023, SM024 |
| CM035 | Infrastructure intensity remains a market constraint: model companies still depend on compute economics set by the much larger GPU and platform layers. | 中 | SM006, SM025 |
| CM036 | HiDream benefits from strong secular adoption and funding signals, but public evidence still does not isolate a clean SAM or SOM for its own addressable segment. | 中 | SM004, SM005, SM006 |
| CM037 | Contradictory market estimates should be preserved rather than averaged because they describe different scopes, geographies, and stack layers. | 中 | SM004, SM005, SM006 |
| CM038 | HiDream’s serviceable near-term market is best underwritten as visual and multimodal creation budgets in enterprise marketing, commerce content, film/TV production, and creator subscriptions. | 高 | SM001, SM003, SM016 |
| CM039 | The company’s market relevance to “world models” is strategically important but still largely pre-revenue or pre-disclosure compared with its current visual-generation products. | 中 | SM002, SM022 |
| CM040 | Because China and U.S. model performance gaps have narrowed to low single digits on some benchmarks per Stanford, HiDream competes in a market where distribution and productization may matter almost as much as raw model quality. | 中 | SM007, SM008 |
| CP001 | HiDream competes across two fronts at once: base-model capability and creator-facing workflow software. | 高 | SP001, SP009, SP023 |
| CP002 | HiDream-I1 is a 17B-parameter open-source image foundation model with public code and weights. | 高 | SP002, SP005 |
| CP003 | HiDream-O1-Image extends the open-source strategy with an 8B UiT-based image model and MIT-licensed codebase. | 高 | SP003, SP006 |
| CP004 | Hugging Face evidence of 100 Spaces using HiDream-I1-Full indicates real developer experimentation beyond a press-release narrative. | 中 | SP004, SP026 |
| CP005 | HiDream’s strongest public competitive proof today is image-model quality and open-source adoption rather than closed video-platform dominance. | 高 | SP002, SP003, SP004, SP007, SP008 |
| CP006 | Science and Technology Daily reported HiDream-O1-Image-1.5 at #2 globally in a June 2026 snapshot, while later leaderboard snapshots place it lower, showing that competitive rank is real but volatile. | 高 | SP007, SP008 |
| CP007 | GeekPark separately described HiDream-O1-Image as the top open-source model on Artificial Analysis at 1187 Elo, reinforcing that its image moat is strongest in the open-source lane. | 中 | SP006, SP007 |
| CP008 | vivago pushes HiDream into creator-platform competition by offering templates, subscriptions, and easy multimodal creation rather than only raw APIs. | 中 | SP009, SP010, SP011 |
| CP009 | vivago’s official positioning explicitly depends on aggregating outside models such as Sora, Veo, and Kling alongside HiDream’s own model. | 中 | SP009 |
| CP010 | That aggregation broadens feature breadth for users but weakens exclusivity because the product partly rides on rivals’ model quality and availability. | 中 | SP009, SP020 |
| CP011 | App-store evidence shows vivago has real consumer traction signals, but the public proof is still early rather than category-dominant. | 中 | SP010, SP011 |
| CP012 | Runway competes as a mature creator and enterprise platform with video, image, and audio tooling used by 60M+ creatives. | 中 | SP013 |
| CP013 | Veo competes on premium multimodal control with native audio, filmmaker positioning, 1080p/4K outputs, and watermarking via SynthID. | 中 | SP014 |
| CP014 | Kling is a direct Chinese video competitor with API distribution and stronger professional-video positioning in third-party comparisons. | 中 | SP015, SP019 |
| CP015 | Hailuo is another direct Chinese creator/video competitor, though third-party comparison sources suggest weaker synchronized-audio capability than leading peers. | 中 | SP016, SP019 |
| CP016 | Sora’s discontinuation as a standalone product removed one direct consumer threat but also highlighted weak standalone video-generation economics. | 中 | SP017, SP018 |
| CP017 | Chinese video-model rivals currently compete on lower per-clip pricing than Google or Sora according to third-party comparisons. | 中 | SP019 |
| CP018 | Atlas Cloud’s comparison argues teams increasingly multi-home across Chinese video models by task, which reduces lock-in for any single model vendor. | 中 | SP020 |
| CP019 | HiDream’s Shanghai Film and Huace relationships provide more strategic distribution relevance than a pure tooling startup would have. | 中 | SP021, SP022 |
| CP020 | Those media partnerships are still narrower than the owned traffic and built-in distribution advantages that ByteDance and Kuaishou enjoy. | 中 | SP018, SP019, SP021, SP022 |
| CP021 | Forbes frames Chinese video AI competition as structurally advantaged by vertical integration, platform data, and state-backed patience capital. | 中 | SP018 |
| CP022 | Compared with that structure, HiDream looks more like a specialized multimodal lab plus workflow product company than a giant consumer distribution platform. | 中 | SP001, SP018, SP023 |
| CP023 | Internal build remains a real substitute on the enterprise API layer because buyers can integrate several model vendors rather than commit to one stack. | 中 | SP001, SP020 |
| CP024 | Status-quo substitutes still matter: agencies, designers, editors, and traditional production software remain competitors where AI quality or trust is insufficient. | 中 | SP012, SP018 |
| CP025 | Competitive pressure is bifurcated: Google and Runway are strongest on enterprise trust and control, while Chinese rivals are strongest on price and creator-distribution economics. | 高 | SP013, SP014, SP018, SP019 |
| CP026 | HiDream’s public differentiation stack currently combines open-source image leadership, multimodal/world-model branding, creator aggregation, and media partnerships. | 中 | SP001, SP003, SP009, SP019, SP021, SP022, SP023 |
| CP027 | Each of those moat pillars has an offsetting risk: open source can commoditize, aggregation depends on rivals, branding is pre-deployment, and partnerships may not equal repeat demand. | 中 | SP003, SP009, SP018, SP021, SP022 |
| CP028 | HiDream’s open-source posture likely improves developer goodwill and technical reputation relative to closed-only peers. | 高 | SP002, SP003, SP004, SP005, SP026 |
| CP029 | However, open source alone is unlikely to create durable pricing power in creator or enterprise markets without stronger distribution, customer outcomes, or proprietary data loops. | 中 | SP004, SP013, SP018 |
| CP030 | vivago pricing from app-store evidence shows HiDream can compete on accessible subscription entry points even if public enterprise pricing remains undisclosed. | 中 | SP009, SP010 |
| CP031 | Publicly retrieved sources do not support a clean, apples-to-apples list-price comparison for Runway, Veo, and HiDream enterprise products, which limits hard win/loss underwriting. | 中 | SP013, SP014, SP023 |
| CP032 | The absence of robust public customer win/loss case studies means HiDream’s competitive conversion rate versus peers is still unproven. | 中 | SP019, SP021, SP022, SP023 |
| CP033 | China Biz Insider says industry ARR across major video-model companies remained below USD 1B as of January 2026, implying rivalry is intensifying before the category is fully monetized. | 中 | SP019 |
| CP034 | That early monetization stage raises displacement risk because buyers can experiment broadly before long-term vendor standards are set. | 中 | SP019, SP020 |
| CP035 | HiDream’s competitive durability is therefore moderate rather than entrenched: technically promising, commercially broadened, but still exposed to platform-rich and lower-cost rivals. | 中 | SP005, SP009, SP018, SP019, SP021 |
| CP036 | The company’s world-model narrative may become a moat only if it yields materially better physics, duration, or workflow outcomes than current peers. | 中 | SP023, SP018, SP019 |
| CP037 | Until then, HiDream’s clearest battlefront is winning creator, marketing, and film workflows with better packaging and partnerships around strong visual models. | 中 | SP001, SP009, SP012, SP021, SP022 |
| CP038 | The competitive map is not winner-take-all yet because model users increasingly route across vendors by task, price, and control needs. | 中 | SP009, SP018, SP020 |
| CP039 | HiDream is better positioned against image-generation and creator-tool peers than against giant platform incumbents with owned demand channels. | 中 | SP006, SP008, SP018, SP019 |
| CP040 | The main adverse evidence is that many of HiDream’s visible advantages are easy for rivals to copy or counter unless proprietary distribution or customer embedment becomes clearer. | 中 | SP018, SP019, SP020, SP023 |
| CI001 | HiDream has at least four visible monetization surfaces in public sources: enterprise API usage, marketing workflow products, film/media tooling, and consumer creator subscriptions or credits. | 高 | SI001, SI004, SI008, SI009 |
| CI002 | HiHarness makes API and model-as-a-service revenue plausible because the homepage describes 200+ APIs, 100+ key accounts, and 500B+ API calls. | 中 | SI001 |
| CI003 | vivago creates a clear self-serve monetization rail through monthly subscriptions and credit purchases visible on the App Store. | 高 | SI006, SI009 |
| CI004 | App-store pricing shows a public creator ladder from roughly USD 9.99 monthly to USD 99.99 monthly, plus credits upsell. | 中 | SI006 |
| CI005 | Business Wire and the official vivago page indicate the creator product is also marketed to businesses, implying monetization can straddle B2C and SME use cases. | 中 | SI008, SI009 |
| CI006 | Media OutReach and theSaaSnews both cite 50M+ professional users and 40K+ enterprise customers, which are traction signals but not audited revenue disclosures. | 中 | SI004, SI005 |
| CI007 | China Biz Insider says HiBurst has generated over one million videos annually and supported GMV above RMB 100 million, suggesting meaningful commercial workflow activity if accurate. | 中 | SI015 |
| CI008 | HiBurst-linked GMV is a customer-commerce output metric, not HiDream revenue, so it should not be treated as top-line without a take-rate bridge. | 中 | SI015 |
| CI009 | Public sources do not disclose realized enterprise/API pricing, contract minimums, or discounting for HiHarness or Zhenzan. | 中 | SI001, SI004, SI023 |
| CI010 | That means the visible price record is strongest for creator subscriptions and weakest for higher-value enterprise contracts. | 中 | SI001, SI006, SI009 |
| CI011 | HiDream’s GTM appears hybrid: self-serve creator acquisition through vivago plus sales-led or partnership-led enterprise and media deployments. | 中 | SI004, SI008, SI009, SI018 |
| CI012 | Public customer metrics imply commercial breadth, but there is no disclosed CAC, payback, paid-conversion rate, churn, or cohort retention in the retrieved source set. | 中 | SI004, SI005, SI006 |
| CI013 | Registry and filing sources support the company’s legal existence and operating base, but they do not expose audited income statements or a clean employee-cost view. | 高 | SI002, SI003 |
| CI014 | Aiqicha’s insured-person and qualification data can hint at operational substance, yet the figures are not a reliable substitute for headcount, payroll, or burn. | 中 | SI003 |
| CI015 | Cost structure is likely dominated by compute, model refresh, moderation/compliance, and app-store or channel fees rather than physical capex or inventory. | 中 | SI017, SI022, SI023 |
| CI016 | ResearchAndMarkets’ large gap between foundation-model revenue and GPU hardware spending underscores why multimodal AI software can face margin pressure even as demand grows. | 中 | SI022, SI023 |
| CI017 | China Biz Insider’s claim that UiT compressed training costs to 10%-20% of industry average supports a potential architectural cost advantage, but the statement is still media-reported rather than audited. | 中 | SI015 |
| CI018 | Even if training is more efficient, serving cost and gross margin remain unverified because public sources do not disclose inference cost per image, per video minute, or per enterprise account. | 中 | SI015, SI022 |
| CI019 | Recent capital access is strong: multiple sources report a RMB 1.5B Series C and more than RMB 2.1B raised across roughly three months. | 中 | SI004, SI005, SI018, SI019 |
| CI020 | The stated use of funds is to advance native omni-modal or world-model development and commercial products, implying a continued high-investment roadmap rather than near-term cash harvesting. | 中 | SI004, SI018 |
| CI021 | InforCapital’s snapshot of only $73M raised across three rounds conflicts with the larger 2026 press reports, indicating some private-market databases are stale or incomplete for current underwriting. | 中 | SI010, SI018 |
| CI022 | PitchBook, Tracxn, and Caplight are too sparse in this retrieved set to resolve valuation, secondary pricing, or financing history cleanly. | 中 | SI011, SI012, SI013 |
| CI023 | That data-quality problem is itself financially important because it leaves investors dependent on company-framed or media-reported numbers for core private-company economics. | 中 | SI011, SI012, SI013, SI019 |
| CI024 | No public source in this set provides a verified cash balance, monthly burn, or runway for HiDream. | 高 | SI002, SI003, SI004, SI018 |
| CI025 | Accordingly, capital adequacy can only be inferred from fundraising success and product breadth, not from explicit cash-flow disclosure. | 中 | SI019, SI020, SI024 |
| CI026 | The absence of disclosed debt, credit lines, or project-finance obligations should not be read as proof that none exist; it only shows they are not visible in the reviewed public record. | 中 | SI002, SI003, SI012 |
| CI027 | Forbes and OpenAI’s Sora discontinuation together support the broader adverse point that standalone video-generation economics can be poor without stronger platform monetization or enterprise attach. | 高 | SI023, SI024 |
| CI028 | China Biz Insider’s estimate that combined ARR across major video-model companies was below USD 1B in January 2026 reinforces that category monetization may lag technical excitement. | 中 | SI025 |
| CI029 | HiDream’s consumer list pricing therefore proves monetization intent, but not that the company has solved retention or gross-margin quality in creator video. | 中 | SI006, SI023, SI025 |
| CI030 | The most defensible public traction indicators remain activity metrics, customer counts, and visible price cards rather than verified revenue or ARR. | 中 | SI001, SI004, SI006 |
| CI031 | Baidu Baike and product press collectively show a broad product matrix, which increases monetization optionality but also complicates revenue-mix analysis. | 中 | SI014, SI004, SI008 |
| CI032 | Because HiDream spans creators, marketers, film/media teams, and API buyers, revenue quality could vary materially by segment and contract model. | 中 | SI001, SI008, SI015 |
| CI033 | Public sources are not sufficient to judge whether revenue is usage-based, subscription-based, project-based, or licensing-heavy by percentage mix. | 中 | SI004, SI009, SI014 |
| CI034 | The privacy-policy and app-based product surface imply ongoing compliance, moderation, and support obligations that likely rise with creator scale. | 中 | SI016, SI017, SI006 |
| CI035 | Film and enterprise deployments may require bespoke support or solutioning, which would make service delivery less software-pure than a simple API story suggests. | 中 | SI008, SI018, SI019 |
| CI036 | The positive financial case is that HiDream already has multiple chargeable surfaces, meaningful activity proxies, and unusually strong late-stage funding support. | 高 | SI001, SI004, SI006, SI019 |
| CI037 | The skeptical case is that public evidence still cannot verify revenue scale, gross margin, customer retention, or runway, so underwriting remains capital-story heavy. | 中 | SI011, SI012, SI013, SI024 |
| CI038 | On current public evidence, HiDream looks commercially credible but financially opaque in the dimensions that matter most for institutional diligence. | 中 | SI019, SI023, SI024 |
| CI039 | The next diligence step must be a management-grade bridge from users, API calls, and generated assets to recognized revenue, gross profit, and cash burn. | 中 | SI001, SI004, SI015 |
| CI040 | Until that bridge exists, any precise revenue or margin model for HiDream would be more speculative than evidenced. | 中 | SI011, SI012, SI013, SI024 |
| CI041 | The broader 2026 venture environment remained receptive to AI unicorn formation, which likely helped frontier-AI companies such as HiDream raise capital quickly. | 中 | SI026, SI027, SI028 |
| CI042 | That favorable funding backdrop explains financing receptivity but does not substitute for company-level revenue, margin, or runway disclosure. | 中 | SI026, SI027, SI024 |
| CE001 | The official homepage says HiDream.ai has a base model that simultaneously supports text, image, video, and 3D, with proprietary data and model parameters exceeding 200 billion. | 中 | SE001 |
| CE002 | The About page frames HiDream as a company building multimodal foundation models and AI applications for marketing, film and television, and social-media creation workflows. | 中 | SE002 |
| CE003 | Official materials publicly list five product lines: HiHarness, HiBurst, Zhenzan, vivago, and HiDreamFans. | 高 | SE001, SE002 |
| CE004 | HiHarness is positioned as an enterprise integration layer with 200+ APIs. | 中 | SE001 |
| CE005 | The homepage says HiHarness serves 100+ key accounts. | 中 | SE001 |
| CE006 | The homepage says HiHarness has processed 500+ billion API calls. | 中 | SE001 |
| CE007 | HiDreamFans is described as a holographic intelligent commercial marketing machine for offline store customer acquisition. | 中 | SE001 |
| CE008 | vivago is publicly positioned as a model-aggregation and visual-creation platform rather than a single-model front end. | 中 | SE016, SE017 |
| CE009 | vivago includes HiDreamClaw AI Agent and AI Chat Agent as higher-order orchestration layers on top of model access. | 中 | SE016 |
| CE010 | vivago advertises 300+ templates updated weekly and a free plan with daily-refresh credits. | 中 | SE016 |
| CE011 | Business Wire says vivago 2.0 launched globally on web and mobile with image-to-video and AI podcast generation features. | 中 | SE017 |
| CE012 | Business Wire says vivago 2.0 packages 300+ templates that can produce professional-looking content in roughly three seconds. | 中 | SE017 |
| CE013 | Business Wire says vivago 2.0 is marketed both to individual creators and to businesses such as retailers using promo-video and virtual-outfit workflows. | 中 | SE017 |
| CE014 | The App Store page exposes public creator pricing tiers of $9.99, $29.99, and $99.99 monthly plus a 1000-credit purchase at $12.99. | 中 | SE018 |
| CE015 | The App Store page shows vivago at 4.6 stars from 59 ratings on the fetched page. | 中 | SE018 |
| CE016 | The Google Play page confirms vivago is positioned around AI image and video creation powered by HiDream-O1-1.5. | 中 | SE019 |
| CE017 | HiDream-I1 is a 17B-parameter open-source image generative foundation model. | 高 | SE003, SE005 |
| CE018 | The HiDream-I1 paper describes a sparse Diffusion Transformer using dual-stream encoding and dynamic Mixture-of-Experts before single-stream multimodal interaction. | 中 | SE005 |
| CE019 | HiDream-I1 is distributed in Full, Dev, and Fast variants. | 高 | SE005, SE006 |
| CE020 | The HiDream-I1 research stack expands beyond text-to-image into HiDream-E1 image editing and HiDream-A1 interactive image creation and refinement. | 中 | SE005 |
| CE021 | The Hugging Face model card says HiDream-I1 achieves strong prompt-following scores on GenEval and DPG benchmarks and is released under an MIT license. | 中 | SE006 |
| CE022 | The model card includes Diffusers usage instructions, indicating active attention to mainstream deployment tooling. | 高 | SE006, SE003 |
| CE023 | The fetched Hugging Face model card showed 100 Spaces using HiDream-I1-Full. | 中 | SE006 |
| CE024 | HiDream maintains public Hugging Face demo surfaces for HiDream-I1-Dev, HiDream-E1-Full, HiDream-O1-Image, and HiDream-O1-Image-Dev. | 高 | SE007, SE008, SE009, SE010 |
| CE025 | HiDream-O1-Image is documented as a pixel-level Unified Transformer that avoids external VAEs and disjoint text encoders. | 中 | SE004 |
| CE026 | HiDream-O1-Image supports text-to-image, image editing, and subject-driven personalization at up to 2048×2048 resolution. | 中 | SE004 |
| CE027 | The O1 repository says HiDream open-sourced both undistilled and distilled 8B variants on May 8, 2026, plus later Dev releases with a prompt refiner. | 中 | SE004 |
| CE028 | The O1 repository includes a Reasoning-Driven Prompt Agent that rewrites user instructions after reasoning through layout, subject attributes, physical logic, and text rendering. | 中 | SE004, SE028 |
| CE029 | The O1 web UI is described as a single-file Flask application with the same integrated prompt agent. | 中 | SE004 |
| CE030 | GeekPark reports HiDream-O1-Image reached 1187 Elo and ranked first among open-source models on Artificial Analysis. | 中 | SE012, SE014 |
| CE031 | Science and Technology Daily reported HiDream-O1-Image-1.5 reached #2 globally at 1265 Elo in a June 2026 Artificial Analysis snapshot. | 高 | SE013, SE014 |
| CE032 | GeekPark quotes HiDream leadership saying UiT enables stronger training synergy between image and video, linking today's image stack to a broader omni-modal roadmap. | 中 | SE012 |
| CE033 | DeepWiki adds a lightweight developer-oriented documentation layer around the HiDream-I1 repository, which can reduce onboarding friction for external users. | 中 | SE011, SE003 |
| CE034 | Qichacha and Aiqicha show the product stack sits inside a real operating entity with disclosed ownership and IP records, but not within a public-company style technical assurance regime. | 中 | SE022, SE023 |
| CE035 | Aiqicha lists the company as a high-tech enterprise with patent and software-copyright records, which supports an IP-building narrative but not necessarily freedom-to-operate. | 中 | SE023 |
| CE036 | China's generative-AI service rules explicitly apply to public services generating text, images, audio, and video, making HiDream's creator and API products directly subject to compliance duties. | 中 | SE024 |
| CE037 | The same legal text requires lawful data sourcing, IP respect, user agreements, safe and stable service, complaint handling, and protection of user input and usage records. | 中 | SE024 |
| CE038 | China's synthetic-content labeling rules require explicit and implicit labeling controls for generated media, which is especially relevant to HiDream's image and video products. | 中 | SE025 |
| CE039 | The public vivago privacy page shows HiDream has at least published a privacy-governance surface for global users, but the reviewed public record does not show certifications, SLAs, or trust-center depth. | 中 | SE026, SE024 |
| CE040 | Public product evidence is strongest for image-generation architecture, creator workflow breadth, and demo accessibility, while reliability metrics, enterprise support detail, and video/world-model production maturity remain under-disclosed. | 中 | SE004, SE006, SE016, SE017, SE024 |
| CU001 | HiDream’s visible customer base spans at least five segments: enterprise API buyers, marketing or commerce operators, film/media partners, creator/prosumer users, and offline retail use cases. | 中 | SU006, SU016, SU019 |
| CU002 | The buyer, user, and payer split differs by segment: creators often self-pay, enterprise teams split users from budget owners, and media partnerships may combine strategic and commercial objectives. | 中 | SU006, SU011, SU016 |
| CU003 | The official homepage says HiHarness serves 100+ key accounts, providing at least a directional signal of enterprise customer presence. | 中 | SU006 |
| CU004 | Multiple financing reports say HiDream products serve more than 50 million professional users and more than 40,000 enterprise customers in over 100 countries and regions. | 中 | SU007, SU008, SU009 |
| CU005 | The same sources frame those customer counts as broad commercial reach rather than as a narrow pilot footprint. | 中 | SU007, SU009, SU010 |
| CU006 | vivago gives HiDream a direct self-serve acquisition channel through web and mobile rather than relying only on enterprise sales. | 中 | SU013, SU014, SU016 |
| CU007 | The vivago product pages emphasize agent-led creation, templates, and multi-model routing, suggesting the company is optimizing for adoption simplicity rather than only raw model access. | 中 | SU013, SU014, SU017 |
| CU008 | The App Store page gives visible conversion rails through monthly plans and credits purchases. | 中 | SU011 |
| CU009 | The App Store page shows a 4.6 rating from 59 ratings on the fetched page, which is a positive but still small public iOS satisfaction signal. | 中 | SU011 |
| CU010 | The Google Play page confirms an Android distribution surface but, in the retrieved text, provides less directly readable satisfaction detail than the iOS page. | 中 | SU012 |
| CU011 | The more detailed vivago product page claims Android scale above 10 million and describes free, paid, and agent-access tiers, which implies a sizable top-of-funnel if accurate. | 中 | SU014, SU025 |
| CU012 | Business Wire says vivago 2.0 connects millions of users in a creative community and is marketed to both creators and brands. | 中 | SU016 |
| CU013 | Business Wire also gives concrete business-use examples such as fashion retailers and brick-and-mortar shops using AI visuals and promo videos. | 中 | SU016 |
| CU014 | China Biz Insider says HiBurst has generated over one million videos annually and supported GMV exceeding RMB 100 million, indicating workflow adoption if accurate. | 中 | SU019 |
| CU015 | HiBurst’s reported GMV support should be treated as a customer-commerce outcome proxy, not as direct HiDream revenue or guaranteed retention evidence. | 中 | SU019 |
| CU016 | BlueFocus is the clearest named marketing-sector proof in the retrieved source set, with a reported strategic cooperation focused on AIGC content production and multimodal marketing. | 中 | SU001 |
| CU017 | Shanghai Film provides named media-sector proof through strategic investment cooperation around IP development, cinema-scene marketing, and joint AIGC lab work. | 高 | SU002, SU004 |
| CU018 | Huace provides named film/TV proof through a strategic cooperation and planned strategic investment with HiDream. | 高 | SU003, SU005 |
| CU019 | These named proofs are strategically important, but they are closer to partnership or deployment signals than to classic disclosed paying-customer case studies. | 中 | SU001, SU002, SU003 |
| CU020 | The strongest public customer proof today is therefore ecosystem and partnership evidence rather than traditional enterprise-logo procurement evidence. | 中 | SU001, SU002, SU003, SU006 |
| CU021 | Official and financing coverage suggest a global customer footprint across more than 100 countries and regions. | 中 | SU007, SU009 |
| CU022 | That global footprint still lacks disclosed geographic revenue or paying-customer splits, so it should not be over-read as diversified recurring enterprise revenue. | 中 | SU007, SU009, SU022 |
| CU023 | The customer journey likely begins with free or low-friction creator experimentation, then expands into subscriptions, credits, or higher-value workflow usage if outputs prove useful. | 中 | SU011, SU014, SU016 |
| CU024 | For enterprise and media users, the entry path is more likely strategic partnership, workflow trial, or API evaluation than simple app-store conversion. | 中 | SU001, SU002, SU006 |
| CU025 | Public sources do not disclose NRR, GRR, logo churn, average contract length, or cohort retention for any customer segment. | 高 | SU007, SU011, SU022 |
| CU026 | App-store ratings and product-community claims provide only partial customer-quality evidence because they cannot distinguish free curiosity from retained paying use. | 中 | SU011, SU014, SU016 |
| CU027 | There is no public denominator for how many of the 40,000 enterprise customers are paying, active, or meaningful in revenue terms. | 中 | SU007, SU008, SU009 |
| CU028 | The company’s customer mix likely still depends significantly on creator and prosumer channels because those are the surfaces with the clearest public pricing and usage evidence. | 中 | SU011, SU014, SU016 |
| CU029 | Film and marketing partnerships may improve strategic value and future ACV potential, but the retrieved public record does not quantify renewal behavior, revenue contribution, or seat expansion. | 中 | SU001, SU002, SU003 |
| CU030 | Platform and app-store channels matter materially to customer acquisition because vivago is distributed across web, iOS, and Android. | 高 | SU011, SU012, SU013 |
| CU031 | The AI-agent features inside vivago create a plausible land-and-expand path from simple generation to deeper editing or conversational workflow usage. | 中 | SU013, SU014, SU017 |
| CU032 | The free plan, daily credits, and multiple monthly tiers indicate a product-led funnel designed to maximize sampling and then monetize heavier or more demanding users. | 中 | SU014, SU025, SU011 |
| CU033 | Procurement friction is still likely on the enterprise side because the public record shows little trust-center, security, or SLA detail tied to named customers. | 中 | SU006, SU015, SU023 |
| CU034 | Customer concentration risk is heightened by the possibility that a large share of visible demand sits in fast-moving creator or partner-driven channels rather than sticky multi-year enterprise contracts. | 中 | SU016, SU023, SU024 |
| CU035 | China Biz Insider’s description of the AI video sector as early commercial-stage is an adverse signal for customer durability across all vendors, including HiDream. | 中 | SU023 |
| CU036 | Forbes’ argument that Chinese video-AI winners rely heavily on platform distribution and creator-economy loops suggests HiDream’s customer durability may depend on channel partnerships as much as on model quality. | 中 | SU024, SU001 |
| CU037 | Qichacha and Aiqicha support the existence of a substantive operating entity, but they add little direct visibility into customer concentration or contract quality. | 中 | SU020, SU021 |
| CU038 | InforCapital’s company profile helps contextualize HiDream’s growth story but does not resolve customer retention or named-customer quality. | 低 | SU022 |
| CU039 | The most supportable customer verdict is that HiDream has real adoption surfaces and real named strategic proofs, but still lacks public evidence on retention and paying-account quality. | 中 | SU001, SU002, SU007, SU011, SU023 |
| CU040 | The next diligence step must be a customer-quality bridge from headline counts and app engagement to paid account cohorts, ACVs, renewal, and concentration. | 中 | SU007, SU011, SU022 |
| CR001 | HiDream’s highest residual risks are regulatory compliance, trust/assurance under-disclosure, pricing pressure, and partner-driven revenue quality rather than headline demand absence. | 中 | SR001, SR003, SR022, SR023, SR030 |
| CR002 | China’s Interim Measures for Generative AI Services apply to public-facing generative AI services that provide text, image, audio, or video content in China. | 高 | SR001, SR002 |
| CR003 | Because HiDream markets image, video, and audio-capable products, it sits squarely inside the scope of China’s public-facing generative-AI governance stack. | 高 | SR001, SR002, SR010 |
| CR004 | The Interim Measures explicitly require lawful data and base-model sourcing, respect for intellectual property, and protection of personal rights. | 高 | SR001, SR002 |
| CR005 | China’s Data Security Law requires data processors to establish data-security controls and, for important data, periodic risk assessments. | 高 | SR005, SR028 |
| CR006 | China’s Cybersecurity Law requires network operators to maintain secure and stable operation, retain logs, and respond to vulnerabilities or incidents. | 高 | SR004, SR028 |
| CR007 | China’s 2025 AI-content labeling regime raises compliance burden for any company commercializing large volumes of synthetic image or video output. | 中 | SR003, SR029 |
| CR008 | Serving multiple modalities increases the company’s risk surface because it must govern more content types, more misuse vectors, and more product-policy edge cases than a single-medium model vendor. | 中 | SR001, SR003, SR010 |
| CR009 | The public record shows privacy and terms pages for vivago, which is better than having no legal surface at all. | 中 | SR007, SR008, SR009 |
| CR010 | The same public record does not surface a visible enterprise trust center, DPA package, certification list, or SLA summary. | 中 | SR007, SR008, SR010 |
| CR011 | That trust-packaging gap matters more because HiHarness claims 100+ key accounts and 500+ billion API calls. | 中 | SR010, SR004 |
| CR012 | A security or privacy incident would likely transmit into enterprise conversion friction faster than into creator awareness, because enterprise buyers explicitly ask for assurance evidence. | 中 | SR004, SR005, SR010 |
| CR013 | HiDream’s open-source posture lowers developer adoption friction but reduces control over downstream forks, misuse, and derivative deployments. | 中 | SR011, SR012, SR013, SR015 |
| CR014 | The Hugging Face README explicitly frames HiDream-I1 as MIT-licensed and commercial-friendly, which heightens expectations that downstream commercial users can rely on it safely. | 高 | SR013, SR011 |
| CR015 | If training-data provenance or rights hygiene later proves weaker than users assumed, the commercial-friendly positioning would amplify reputational and customer fallout. | 中 | SR002, SR013, SR020, SR021 |
| CR016 | Film and TV partnerships raise IP sensitivity because those workflows naturally touch known characters, classic content libraries, and protected creative assets. | 中 | SR020, SR021, SR002 |
| CR017 | Marketing deployments raise brand-safety and output-integrity risk because generated assets can be published externally at scale and tied to customer campaigns. | 中 | SR019, SR003 |
| CR018 | The prompt-agent workflow exposed in the public O1 image repo suggests automation convenience, but also implies that prompt-layer failures can propagate through broader creation workflows. | 中 | SR014, SR012 |
| CR019 | HiDream’s infrastructure exposure is material because rapid multimodal generation at large scale depends on compute availability, network capacity, and power economics. | 中 | SR006, SR010, SR018 |
| CR020 | China’s push to expand national compute infrastructure helps the company’s environment, but also underscores how strategic and capacity-constrained compute remains. | 中 | SR006, SR029 |
| CR021 | The risk is not only compute scarcity but compute economics: more generation volume can still compress margins if price competition outruns efficiency gains. | 中 | SR022, SR023, SR018 |
| CR022 | China Biz Insider’s description of the video-AI sector as early commercial-stage is an adverse signal for revenue durability across all vendors, including HiDream. | 中 | SR022 |
| CR023 | Forbes’ emphasis on price and vertical integration in the Chinese video-AI race implies that growth can be purchased uneconomically if customer retention is weak. | 中 | SR023, SR022 |
| CR024 | Public financing coverage shows scale ambition and adoption breadth, but it does not disclose margins, burn, or cost-to-serve by product. | 中 | SR016, SR017, SR018 |
| CR025 | That means financial/model risk is currently more about hidden unit economics and compliance-cost stacking than about immediate capital scarcity. | 中 | SR016, SR018, SR022 |
| CR026 | The customer base appears broad, but public named proof clusters around a small number of strategic film and marketing partners, creating concentration risk if those logos over-index strategic narrative. | 中 | SR019, SR020, SR021, SR030 |
| CR027 | App-store, web, and partnership distribution make platform policy a real dependency even without a single dominant distributor. | 中 | SR007, SR008, SR009 |
| CR028 | Cross-border reach into 100+ countries and regions raises the odds that product, privacy, and content-governance obligations will diverge across geographies. | 中 | SR030, SR028, SR029 |
| CR029 | The public record does not show how HiDream segments or localizes cross-border data handling for those geographies. | 中 | SR007, SR030 |
| CR030 | No public incident log, audit summary, or certification packet was found in the retrieved source set, so assurance maturity remains under-evidenced. | 中 | SR007, SR008, SR010 |
| CR031 | Absence of visible incident disclosure is not evidence of absence of incidents; it is evidence that diligence must verify security reporting and postmortem practices directly. | 中 | SR004, SR005, SR030 |
| CR032 | Because HiHarness claims 500+ billion API calls, reliability or abuse-control failures would have a wide blast radius across developers and downstream applications. | 中 | SR010 |
| CR033 | Open-source benchmark success increases expectations for product quality and raises reputational downside if commercial outputs underperform benchmark reputation. | 中 | SR024, SR025, SR013 |
| CR034 | The same open-source success also shortens moat half-life by making HiDream’s strongest image capability easier for rivals or downstream integrators to study and copy. | 中 | SR013, SR015, SR024 |
| CR035 | People and execution risk centers on compliance, privacy, security, and enterprise-assurance leadership rather than on simple research talent scarcity alone. | 中 | SR004, SR005, SR010 |
| CR036 | Strategic deployments in media and marketing also require solution architecture and customer-success capacity if pilots are to become production workflows. | 中 | SR019, SR020, SR021 |
| CR037 | Visible mitigation evidence today includes public legal pages, open-source licensing transparency, and some official explanation of product scope and usage scale. | 中 | SR007, SR008, SR010, SR011, SR013 |
| CR038 | Visible mitigation evidence does not yet include public enterprise-grade security documentation, independent audits, or disclosed policy-governance metrics. | 中 | SR007, SR008, SR010 |
| CR039 | The cleanest thesis-break triggers are regulatory enforcement, inability to show trust artifacts, visible price compression without margin proof, and concentration in a few strategic counterparties. | 中 | SR003, SR022, SR023, SR019, SR020 |
| CR040 | The most supportable current verdict is that HiDream’s risk profile is manageable for tracking or conditional underwriting, but not yet de-risked enough to treat scale claims as enterprise-quality revenue. | 中 | SR001, SR010, SR022, SR030 |
| CR041 | Qichacha and Aiqicha confirm the operating entity but do not reduce core underwriting risks around compliance execution, customer concentration, or cost structure. | 中 | SR026, SR027 |
| CR042 | Rimon and Cambridge both describe China’s AI regime as layered and evolving, reinforcing that compliance is a continuing operating function rather than a one-time filing exercise. | 中 | SR028, SR029, SR001 |
| CV001 | HiDream.ai completed a RMB 1.5 billion Series C round in late July 2026. | 中 | SV003, SV004, SV006 |
| CV002 | Public coverage says HiDream raised more than RMB 2.1 billion across three rounds completed within roughly three months before the Series C announcement. | 中 | SV003, SV004, SV005, SV006 |
| CV003 | The latest financing pushed HiDream into unicorn status above a $1 billion valuation, but no exact public post-money valuation is disclosed in the reviewed source set. | 中 | SV003, SV004, SV005 |
| CV004 | The Series C syndicate mixed state-backed funds, bank-affiliated capital, and film or TV strategics such as Shanghai Film New Vision Fund and Huace Film & TV. | 中 | SV003, SV004, SV006 |
| CV005 | The stated use of proceeds is native omni-modal world-model development plus expansion of HiDream’s product and commercial ecosystem. | 中 | SV003, SV004, SV006 |
| CV006 | HiDream’s official site says HiHarness offers more than 200 APIs, more than 100 key accounts, and more than 500 billion API calls. | 高 | SV001, SV002 |
| CV007 | The company’s financing press chain says HiDream serves more than 50 million professional users and more than 40,000 enterprise customers across more than 100 countries and regions. | 中 | SV003, SV004 |
| CV008 | Registry sources show Beijing Zhixiang Future Technology Co., Ltd., the operating entity behind HiDream.ai, was established on 2023-03-02 in Beijing’s Haidian District. | 高 | SV007, SV008 |
| CV009 | GeekPark and Artificial Analysis show that HiDream-O1-Image reached leading open-source image-generation positions in 2026. | 中 | SV009, SV010 |
| CV010 | The official GitHub repository says HiDream-O1-Image was open-sourced on 2026-05-08 with an 8B model plus a reasoning-driven prompt agent. | 中 | SV011 |
| CV011 | The Hugging Face organization snapshot shows HiDream’s O1 and I1 model pages carrying meaningful public attention, indicating real developer distribution beyond the company website. | 低 | SV012 |
| CV012 | The HiDream-I1 paper describes a 17B image foundation model with Full, Dev, and Fast variants and says the stack extends into HiDream-E1 image editing. | 中 | SV013 |
| CV013 | Together AI lists HiDream-I1-Full as a serverless model priced at $0.009 per megapixel, showing third-party commercialization and distribution. | 中 | SV014 |
| CV014 | S&P Global says 2026 GenAI investor sentiment remains bullish but much more discriminating, with value shifting from model training toward inference and applications. | 中 | SV015 |
| CV015 | S&P Global says many of the 2024 GenAI front-runners stalled, were recapitalized, or were hollowed out, showing that private AI valuation momentum can reverse quickly. | 中 | SV015 |
| CV016 | Aventis says the median 2024 AI Series C pre-money valuation was about $795.2 million. | 中 | SV016 |
| CV017 | Aventis says the median AI revenue multiple in its large AI financing and M&A sample was 24.2x, with fundraising rounds commonly pricing around 25x to 30x revenue while public SaaS is closer to about 6x revenue. | 中 | SV016 |
| CV018 | Adobe’s August 2026 market cap of $108.5 billion against 2025 revenue of $23.76 billion implies about 4.6x revenue. | 中 | SV017, SV018 |
| CV019 | C3 AI’s August 2026 market cap of $1.61 billion against 2025 revenue of $0.30 billion implies about 5.4x revenue. | 中 | SV019, SV020 |
| CV020 | Duolingo’s August 2026 market cap of $6.41 billion against 2025 revenue of $1.03 billion implies about 6.2x revenue. | 中 | SV021, SV022 |
| CV021 | Kuaishou’s August 2026 market cap of $23.53 billion against 2023 revenue of $16.01 billion implies about 1.5x revenue. | 中 | SV023, SV024 |
| CV022 | Shutterstock’s August 2026 market cap of $0.21 billion against 2025 revenue of $0.98 billion implies about 0.2x revenue. | 中 | SV025, SV026 |
| CV023 | Getty Images’s August 2026 market cap of $0.18 billion against about $0.98 billion of trailing revenue implies about 0.18x revenue. | 中 | SV027, SV028 |
| CV024 | The selected public comp set therefore spans roughly 0.2x to 6.2x revenue, far below the private AI fundraising multiples described by Aventis. | 中 | SV016, SV017, SV018, SV019, SV020, SV021, SV022, SV023, SV024, SV025, SV026, SV027, SV028 |
| CV025 | Runway’s June 2023 Series C extension raised $141 million at a $1.5 billion valuation. | 中 | SV029, SV030 |
| CV026 | Synthesia’s January 2026 Series E raised $200 million at a $4 billion valuation. | 中 | SV031 |
| CV027 | Stability AI’s recent public funding history shows roughly $225 million to $231 million raised, while independent summaries still emphasize legal exposure and weak revenue relative to capital intensity. | 中 | SV032, SV034 |
| CV028 | Midjourney’s estimated $500 million of 2025 revenue with no outside investors shows that the strongest image-AI benchmark is capital-efficient monetization rather than fundraising alone. | 低 | SV033 |
| CV029 | HiDream’s mix of state-backed and film-industry capital may add strategic channel value that a generic AI multiple comparison would miss. | 中 | SV003, SV004, SV006 |
| CV030 | The public record still does not disclose audited revenue, ARR, gross margin, burn, retention, or customer concentration for HiDream itself. | 中 | SV001, SV003, SV005 |
| CV031 | Because those core operating inputs are absent, a conventional discounted cash-flow model cannot be defended from public evidence alone. | 中 | SV005, SV015, SV016 |
| CV032 | At a $1 billion valuation floor, HiDream would need roughly 20x revenue at $50 million of revenue, 10x at $100 million, and 6.7x at $150 million, showing how sensitive the mark is to an undisclosed number. | 中 | SV003, SV016 |
| CV033 | Relative to the median AI Series C benchmark, HiDream’s unicorn floor already prices some combination of strategic scarcity, vertical relevance, and expected revenue acceleration. | 中 | SV003, SV016 |
| CV034 | The strongest pro-valuation case is that HiDream combines benchmark credibility, open-source reach, enterprise API proxies, and strategic media capital in a category where sovereign and vertical AI capital still matters. | 中 | SV001, SV002, SV003, SV009, SV010, SV015 |
| CV035 | The strongest anti-thesis is that benchmark wins and activity proxies can coexist with weak monetization quality or heavy compute burn in a market increasingly focused on earnings visibility. | 中 | SV006, SV007, SV014, SV015 |
| CV036 | A cautious base-case band of roughly $0.9 billion to $1.3 billion is the most supportable range from public evidence alone. | 低 | SV003, SV015, SV016, SV017, SV018, SV019, SV020 |
| CV037 | A bull case above $1.5 billion requires verified high-quality recurring revenue, better disclosure on margins and retention, and proof that strategic partnerships convert into durable monetization. | 中 | SV003, SV006, SV029, SV031 |
| CV038 | A bear case below $1 billion becomes plausible if pricing pressure intensifies, economics remain undisclosed, or capital markets re-rate multimodal media AI toward public creative-platform multiples. | 中 | SV015, SV016, SV021, SV022, SV025, SV026, SV027, SV028 |
| CV039 | The exact post-money valuation, preference stack, and any insider secondary overhang are not public in the reviewed source set. | 低 | SV005 |
| CV040 | The most supportable recommendation is research-more rather than buy or avoid, because the company appears credible while the price support remains incomplete. | 中 | SV003, SV006, SV015, SV016 |
| CV041 | Confidence in that recommendation is only medium because product proof and investor appetite are visible while valuation support and cash-economics evidence are not. | 中 | SV001, SV003, SV015, SV016 |
| CV042 | Risk rating should be high because the valuation depends on undisclosed economics in a market already punishing AI stories with weak pricing power or weak earnings visibility. | 中 | SV015, SV016, SV021, SV022, SV025, SV026, SV027, SV028 |
| CV043 | Valuation stance should be stretched at an unspecified unicorn-plus price because the public record proves category relevance more clearly than monetization quality. | 中 | SV003, SV015, SV016 |
| CV044 | Exit readiness is incomplete because any IPO-style or crossover process would still need filing-grade disclosure on revenue quality, margins, governance, and cap-table terms. | 中 | SV005, SV015, SV016 |
| CV045 | Strategic film and media partnerships may deepen HiDream’s moat in entertainment workflows, but public evidence has not yet quantified revenue conversion from those ties. | 中 | SV003, SV006 |
| CV046 | Open-source release and third-party hosting widen adoption, but they can also erode scarcity unless HiDream converts that distribution into sticky enterprise monetization faster than peers. | 中 | SV011, SV012, SV014, SV032 |
| CV047 | HiDream’s current valuation should be treated as a financing outcome rather than proof of intrinsic value until private diligence closes the economic disclosure gap. | 中 | SV003, SV015, SV016 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | HiDream.ai | About - HiDream.ai | Founded in March 2023, the company is dedicated to developing multimodal foundation models and AI applications. |
| SO002 | HiDream.ai | HiDream.ai - Where AI Meets Creativity | HiDream.ai possesses the only global base model that simultaneously supports four modalities (text, image, video, 3D models). |
| SO003 | Qichacha | 北京智象未来科技有限公司 | 成立日期 2023-03-02;注册地址 北京市海淀区苏州街16号(北京神州数码大厦)5层-02。 |
| SO004 | Aiqicha | 北京智象未来科技有限公司 - 智象未来 - 爱企查 | 北京智象未来科技有限公司是一家科技型中小企业(2024)、高新技术企业(2025)。 |
| SO005 | Baidu Baike | 北京智象未来科技有限公司 | 创始人梅涛为加拿大工程院外籍院士,曾任京东集团副总裁。 |
| SO006 | Baidu Baike | Tao Mei | He is currently the founder and CEO of HiDream.ai. |
| SO007 | IEEE Xplore | Tao Mei - IEEE Xplore Author Profile | States Tao Mei was Vice President at JD.COM, Senior Research Manager at Microsoft Research, and Founder/CEO of HiDream.ai. |
| SO008 | The SaaS News | HiDream.ai Raises RMB 1.5B Series C | This latest funding, combined with two earlier rounds completed in the past three months, brings the company's total capital raised during this period to over RMB 2.1 billion. |
| SO009 | Media OutReach Newswire | HiDream.ai Raises RMB 1.5 Billion Series C to Advance Native Omni-modal World Models #HiDreamAI | HiDream.ai’s products currently serve users in more than 100 countries and regions, including over 50 million professional users and more than 40,000 enterprise customers. |
| SO010 | Realtime AI News | Hefei-Backed AI Unicorn HiDream.ai Raises $290M in Three Months, Signals Multimodal Shift | With over 2.1 billion yuan raised across the three rounds, HiDream.ai is now valued at over $1 billion. |
| SO011 | 36Kr | "Zhixiang Future" has completed its Series C round of financing. | So far, HiDream has completed three rounds of financing in nearly three months, with the total accumulated financing exceeding 2.1 billion RMB. |
| SO012 | Seedtable | HiDream.ai Raises 221.4M USD in Series C Funding | HiDream.ai builds native omni-modal world models and the AI agent products that sit on top of them. |
| SO013 | InforCapital | HiDream.ai - AI Infrastructure, $73M Raised | InforCapital | In April 2026 HiDream.ai closed a CNY 500M+ (~USD 70M) financing round. |
| SO014 | NetEase | 上影股份战略投资智象未来,布局“内容+AI”新增长空间 | 上海电影股份有限公司与AI企业智象未来宣布达成战略投资合作。 |
| SO015 | Jiemian | 从战略投资到产业共建,华策影视与智象未来联合布局AI视听新生态 | 华策影视拟对智象未来进行战略投资。 |
| SO016 | 科技日报 | 智象未来图像生成模型评分位列全球第二 | HiDream-O1-Image-1.5在...榜单上,位列全球第二,评分仅次于OpenAI。 |
| SO017 | 36Kr | Chinese Startups Claim Global Top Spot Twice in Half a Month, Shaking Up AI Image Generation Arena | the open-source model HiDream-O1-Image-Dev-2604 ... topped the global list of open-source models in the text-to-image leaderboard. |
| SO018 | GeekPark | 原生全模态架构首次跑通 智象未来 HiDream-O1-Image 获 AA 榜开源模型第一 | HiDream-O1-Image 以 8B 参数规模,在超过 3000 个样本对比中取得 1187 ELO,登顶开源模型第一。 |
| SO019 | Artificial Analysis | Text to Image Leaderboard - Top AI Image Models | HiDream-O1-Image-1.5 appears at #11 with 1,227 Elo and $80 /1k imgs in the fetched August 2026 snapshot. |
| SO020 | China Biz Insider | HiDream.ai Overtakes Google in 2026 AI Race | Its e-commerce marketing agent, HiBurst, has secured a position among TikTok’s top five official service providers. |
| SO021 | Cyberspace Administration of China | 生成式人工智能服务管理暂行办法 | 提供和使用生成式人工智能服务,应当遵守法律、行政法规,尊重社会公德和伦理道德。 |
| SO022 | Regulations.AI | Measures for the Identification of AI-Generated (Synthetic) Content (人工智能生成合成内容标识办法) | The rules require explicit and implicit identifiers for AI-generated synthetic content and supporting traceability. |
| SO023 | Cambridge Forum on AI: Law and Governance | Navigating China’s regulatory approach to generative artificial intelligence and large language models | China’s approach combines support for AI innovation with layered controls over security, content, and accountability. |
| SO024 | Rimon Law | China AI Regulatory Developments | July 2026 Analysis | July 2026 developments expanded compliance expectations around anthropomorphic and synthetic-content systems. |
| SO025 | GitHub | GitHub - HiDream-ai/HiDream-O1-Image | May 14, 2026: We open-sourced HiDream-O1-Image-Dev-2604 with its prompt refiner. |
| SM001 | HiDream.ai | HiDream.ai - Where AI Meets Creativity | HiDream.ai possesses the only global base model that simultaneously supports four modalities (text, image, video, 3D models). |
| SM002 | HiDream.ai | About - HiDream.ai | Founded in March 2023, the company is dedicated to developing multimodal foundation models and AI applications. |
| SM003 | Media OutReach Newswire | HiDream.ai Raises RMB1.5 Billion in Series C Financing to Advance Native Omni-modal World Models | Its product matrix includes HiHarness, HiBurst, Zhenzan, vivago and HiDreamFans, serving over 50 million professional users and over 40,000 enterprise customers in more than 100 countries and regions. |
| SM004 | Grand View Research | Generative AI Market Size, Share & Trends Analysis Report | The global generative AI market size was valued at USD 22.2 billion in 2025 and is projected to grow from USD 29.6 billion in 2026 to USD 324.7 billion by 2033, at a CAGR of 40.8% from 2026 to 2033. |
| SM005 | Fortune Business Insights | Generative AI Market Size, Share & Industry Analysis | The global generative AI market size was valued at USD 103.58 billion in 2025 and is projected to grow from USD 161 billion in 2026 to USD 1,260.15 billion by 2034, exhibiting a CAGR of 29.30% during the forecast period. |
| SM006 | Business Wire / ResearchAndMarkets | Generative AI Market Report 2025 - ResearchAndMarkets | The market value for foundation models reached an estimated US$ 4.1 billion in 2024, while GenAI development platforms reached US$ 17 billion. Meanwhile, GPU-based hardware systems used for GenAI workloads generated revenues of US$ 132.3 billion in 2024. |
| SM007 | Stanford HAI | AI Index Report 2026 | Generative AI hit nearly 53% population-level adoption within three years... Organizational adoption rose to 88%. |
| SM008 | Forbes | Video AI Wars: How Chinese Labs Are Winning The Race OpenAI Abandoned | Chinese labs are winning the mass-market creator-economy and micro-drama segments through price and vertical integration. Google and Runway are positioned to win in the enterprise, agency and Hollywood segments. |
| SM009 | China Biz Insider | Chinese AI Video Models Advance as Kuaishou’s Kling 3.0 and ByteDance’s Seedance 2.0 Intensify Competition | Pricing gaps between domestic and overseas providers remain substantial. Chinese platforms charge roughly US$0.40 for a five-second video, compared with around US$5 for Google’s model and approximately US$2.5 for Sora 2. |
| SM010 | Atlas Cloud | Kling vs Wan vs Seedream: China’s Top AI Video Models in 2026 | Many teams choose to use all three — selecting Kling, Wan, or Seedream based on each task. |
| SM011 | Google DeepMind | Veo | Veo 3 lets you add sound effects, ambient noise, and even dialogue to your creations – generating all audio natively. |
| SM012 | Runway | Runway | An all-in-one cloud-based creative platform that offers endless ways to generate and edit video, images and audio in one workspace... Used by 60m+ creatives around the world. |
| SM013 | Kling AI | Kling AI | API Platform. |
| SM014 | Hailuo AI | Hailuo AI | AI video creation platform. |
| SM015 | OpenAI Help Center | What to know about the Sora discontinuation | The Sora web and app experiences were discontinued on April 26, 2026. The Sora API will be discontinued on September 24, 2026. |
| SM016 | vivago.ai | What is vivago.ai? | vivago.ai aggregates 10+ top-tier video and image models... under one subscription from $7.9/month. |
| SM017 | Artificial Analysis | Text to Image Leaderboard | Leaderboard snapshot for image-generation models including HiDream-O1-Image-1.5. |
| SM018 | 36Kr | 智象未来开源图像模型全球第二 | 36Kr coverage of HiDream image-model benchmark performance and open-source traction. |
| SM019 | Science and Technology Daily | 智象未来发布图像生成大模型HiDream-I1 | HiDream-O1-Image-1.5 ranked second globally on the Artificial Analysis benchmark at 1265 Elo in the June snapshot. |
| SM020 | HiDream-ai GitHub | HiDream-O1-Image | HiDream-O1-Image is a native multimodal image generation model based on a unified image Transformer architecture. |
| SM021 | theSaaSnews | HiDream.ai Raises RMB 1.5B in Series C | Its products are serving over 50 million professional users and 40,000 enterprise customers across more than 100 countries and regions. |
| SM022 | 36Kr Europe | Hefei-backed AI unicorn HiDream.ai raises $290m in three months | The company hopes to use the funds to advance what it calls 4D world models and commercial products. |
| SM023 | CAC / Cyberspace Administration of China | Interim Measures for the Management of Generative Artificial Intelligence Services | Provides the operating framework for generative AI services in China. |
| SM024 | Regulations.AI | Measures for Labeling AI-Generated Synthetic Content (summary) | The new measures increase requirements around explicit and implicit labeling of AI-generated content. |
| SM025 | Rimon PC | China AI and Data Laws in 2026: What Global AI Companies Need to Know | China’s AI and data-law stack creates obligations on training data, content controls, and cross-border handling. |
| SP001 | HiDream.ai | HiDream.ai - Where AI Meets Creativity | HiDream.ai possesses the only global base model that simultaneously supports four modalities (text, image, video, 3D models). |
| SP002 | GitHub | HiDream-I1 README | HiDream-I1 is a new open-source image generative foundation model with 17B parameters that achieves state-of-the-art image generation quality within seconds. |
| SP003 | GitHub | HiDream-O1-Image | HiDream-O1-Image is a natively unified image generative foundation model built on a Pixel-level Unified Transformer (UiT). |
| SP004 | Hugging Face | HiDream-I1-Full model card | Spaces using HiDream-ai/HiDream-I1-Full 100. |
| SP005 | arXiv | HiDream-I1 paper | HiDream-I1 is a new open-source image generative foundation model with 17B parameters. |
| SP006 | GeekPark | 智象未来 HiDream-O1-Image 完成开源 | HiDream-O1-Image以8B参数规模...在超过3000个样本对比中取得1187 ELO,登顶开源模型第一。 |
| SP007 | Artificial Analysis | Text to Image Leaderboard | Leaderboard snapshot including HiDream-O1-Image-1.5 and peer models. |
| SP008 | Science and Technology Daily | 智象未来发布图像生成大模型HiDream-I1 | HiDream-O1-Image-1.5 ranked second globally on Artificial Analysis at 1265 Elo in the June snapshot. |
| SP009 | vivago.ai | What is vivago.ai? | vivago.ai aggregates 10+ top-tier video and image models ... under one subscription from $7.9/month. |
| SP010 | Apple App Store | vivago AI: AI Image & Video | 4.6 rating from 59 Ratings; Basic-Monthly Auto Renewal $9.99. |
| SP011 | Google Play | vivago.ai | Create stunning AI images and videos with vivago.ai. Powered by top-ranking AI models, including HiDream-O1-1.5. |
| SP012 | Business Wire | vivago 2.0: The All-in-One AI Tool That Makes Creativity Accessible to All | Brands and businesses can also benefit ... create eye-catching promo videos to attract more customers. |
| SP013 | Runway | Runway | An all-in-one cloud-based creative platform... Used by 60m+ creatives around the world. |
| SP014 | Google DeepMind | Veo | Veo 3 lets you add sound effects, ambient noise, and even dialogue ... outputs will be marked with SynthID. |
| SP015 | Kling AI | Kling AI | API Platform. |
| SP016 | Hailuo AI | Hailuo AI | AI video creation platform. |
| SP017 | OpenAI Help Center | What to know about the Sora discontinuation | The Sora web and app experiences were discontinued on April 26, 2026. The Sora API will be discontinued on September 24, 2026. |
| SP018 | Forbes | Video AI Wars: How Chinese Labs Are Winning The Race OpenAI Abandoned | Chinese labs are winning the mass-market creator-economy and micro-drama segments through price and vertical integration. |
| SP019 | China Biz Insider | Chinese AI Video Models Advance as Kuaishou’s Kling 3.0 and ByteDance’s Seedance 2.0 Intensify Competition | A five-second 720P video costs roughly RMB 4 on Kling and RMB 2.3 on Seedance; Google’s model about US$5 and Sora 2 about US$2.5. |
| SP020 | Atlas Cloud | Kling vs Wan vs Seedream: China’s Top AI Video Models in 2026 | Many teams choose to use all three — selecting Kling, Wan, or Seedream based on each task. |
| SP021 | NetEase / 163 | 上海电影与智象未来达成战略投资合作 | 上海电影与智象未来达成战略投资合作。 |
| SP022 | Jiemian | 华策影视与智象未来达成战略投资合作 | 华策影视与智象未来达成战略投资合作。 |
| SP023 | Media OutReach Newswire | HiDream.ai Raises RMB1.5 Billion in Series C Financing to Advance Native Omni-modal World Models | HiDream.ai is committed to developing world models that transcend textual understanding and embody a deeper grasp of the physical world. |
| SP024 | Qichacha | 北京智象未来科技有限公司 | Registry record for the Beijing operating entity. |
| SP025 | Aiqicha | 北京智象未来科技有限公司 - 爱企查 | Aiqicha lists high-tech enterprise and patent information. |
| SP026 | DeepWiki | HiDream-I1 overview | DeepWiki provides browsable documentation and structure around the HiDream-I1 repository. |
| SI001 | HiDream.ai | HiDream.ai - Where AI Meets Creativity | HiHarness serves 100+ key accounts and 500+ billion API calls, with 200+ APIs. |
| SI002 | Qichacha | 北京智象未来科技有限公司 | Registry record for Beijing Zhixiang Future Technology Co., Ltd. |
| SI003 | Aiqicha | 北京智象未来科技有限公司 - 爱企查 | Aiqicha lists the company as a high-tech enterprise and shows insured-person snapshots. |
| SI004 | Media OutReach Newswire | HiDream.ai Raises RMB1.5 Billion in Series C Financing to Advance Native Omni-modal World Models | Serving over 50 million professional users and over 40,000 enterprise customers in over 100 countries and regions. |
| SI005 | theSaaSnews | HiDream.ai Raises RMB 1.5B in Series C | Serving over 50 million professional users and 40,000 enterprise customers across more than 100 countries and regions. |
| SI006 | Apple App Store | vivago AI: AI Image & Video | Basic-Monthly Auto Renewal $9.99; Plus $29.99; Pro $99.99; Purchase 1000 Credits $12.99. |
| SI007 | Google Play | vivago.ai | Powered by top-ranking AI models, including HiDream-O1-1.5. |
| SI008 | Business Wire | vivago 2.0: The All-in-One AI Tool That Makes Creativity Accessible to All | Brands and businesses can use AI to generate visuals and promo videos while cutting photoshoot costs. |
| SI009 | vivago.ai | What is vivago.ai? | The key differentiator: vivago.ai bundles Sora 2, Veo 3.1, Kling Video O1, and its own model — starting from $7.9/month. |
| SI010 | InforCapital | HiDream.ai company profile | HiDream.ai has raised $73M across 3 funding rounds since 2023. |
| SI011 | Caplight | HiDream.ai company page | Sparse public secondary-market information for HiDream.ai. |
| SI012 | PitchBook (reader snapshot) | HiDream.ai profile | Reader snapshot is too sparse to support a usable pricing or valuation view. |
| SI013 | Tracxn (reader snapshot) | HiDream.ai funding and investors | Reader snapshot is minimal and not sufficient for clean underwriting. |
| SI014 | Baidu Baike | 北京智象未来科技有限公司 | Company profile summarizing products, milestones, and investors. |
| SI015 | China Biz Insider | Chinese startup HiDream.ai upends generative AI hierarchy, overtaking Google and ByteDance | HiBurst generated over one million videos annually and supported GMV exceeding RMB 100 million; UiT compressed training costs to 10%-20% of industry average. |
| SI016 | vivago.ai | vivago.ai home | vivago AI Agent | Turn Ideas into Video & Image AI Masterpieces |
| SI017 | vivago.ai | Privacy policy | vivago.ai privacy page for global users. |
| SI018 | 36Kr Europe | Hefei-backed AI unicorn HiDream.ai raises $290m in three months | HiDream.ai raised around $290 million across three months. |
| SI019 | 36Kr | 智象未来完成15亿元人民币C轮融资 | 36Kr funding flash for the Series C round. |
| SI020 | Seedtable | HiDream.ai funding snapshot | Database-style funding snapshot for the Series C. |
| SI021 | Science and Technology Daily | 智象未来发布图像生成大模型HiDream-I1 | Benchmark coverage illustrating competitive quality context behind commercialization. |
| SI022 | Business Wire / ResearchAndMarkets | Generative AI Market Report 2025 - ResearchAndMarkets | Foundation models reached an estimated US$4.1B in 2024 while GPU-based GenAI hardware reached US$132.3B. |
| SI023 | Forbes | Video AI Wars: How Chinese Labs Are Winning The Race OpenAI Abandoned | Video generation at scale is a money-losing business if your only revenue model is subscriptions from individual creators. |
| SI024 | OpenAI Help Center | What to know about the Sora discontinuation | The Sora web and app experiences were discontinued on April 26, 2026. |
| SI025 | China Biz Insider | Chinese AI Video Models Advance as Kuaishou’s Kling 3.0 and ByteDance’s Seedance 2.0 Intensify Competition | Combined ARR across major video model companies totaled less than US$1 billion as of January 2026. |
| SI026 | TechCrunch | Almost 40 New Unicorns Have Been Minted So Far This Year. Here They Are. | Coverage of newly minted unicorns in 2026 reflects abundant private capital attention around AI winners. |
| SI027 | Crunchbase News | Global Unicorn Counts Rise On AI, Robotics And Chips In H1 2026 | AI remained a major driver of new-unicorn creation in the first half of 2026. |
| SI028 | TechRound | 2026 Unicorn Tracker: Your Real-Time Guide to This Year’s Newly Minted Unicorns | Tracker of newly minted unicorns in 2026. |
| SE001 | HiDream.ai | HiDream.ai - Where AI Meets Creativity | HiDream.ai possesses the only global base model that simultaneously supports four modalities (text, image, video, 3D models). |
| SE002 | HiDream.ai | About - HiDream.ai | Founded in March 2023, the company is dedicated to developing multimodal foundation models and AI applications. |
| SE003 | GitHub | HiDream-I1 README | HiDream-I1 is a new open-source image generative foundation model with 17B parameters that achieves state-of-the-art image generation quality within seconds. |
| SE004 | GitHub | HiDream-O1-Image | HiDream-O1-Image is a natively unified image generative foundation model built on a Pixel-level Unified Transformer (UiT). |
| SE005 | arXiv | HiDream-I1 paper | HiDream-I1 is constructed with a new sparse Diffusion Transformer structure... with dynamic Mixture-of-Experts architecture. |
| SE006 | Hugging Face | HiDream-I1-Full model card | HiDream-I1 is a new open-source image generative foundation model with 17B parameters... Best-in-Class Prompt Following ... released under the MIT license. |
| SE007 | Hugging Face | HiDream-I1-Dev Space | HiDream I1 Dev - a Hugging Face Space by HiDream-ai ... Running. |
| SE008 | Hugging Face | HiDream-E1-Full Space | HiDream-E1-Full space exists on Hugging Face as a public demo surface. |
| SE009 | Hugging Face | HiDream-O1-Image Space | HiDream O1 Image - a Hugging Face Space by HiDream-ai ... Running. |
| SE010 | Hugging Face | HiDream-O1-Image-Dev Space | HiDream O1 Image Dev - a Hugging Face Space by HiDream-ai ... Running. |
| SE011 | DeepWiki | HiDream-I1 overview | DeepWiki provides browsable documentation around the HiDream-I1 repository. |
| SE012 | GeekPark | 智象未来 HiDream-O1-Image 完成开源 | HiDream-O1-Image以8B参数规模...在超过3000个样本对比中取得1187 ELO,登顶开源模型第一。 |
| SE013 | Science and Technology Daily | 智象未来发布图像生成大模型HiDream-I1 | HiDream-O1-Image-1.5 ranked second globally on the Artificial Analysis benchmark at 1265 Elo in the June snapshot. |
| SE014 | Artificial Analysis | Text to Image Leaderboard | Leaderboard snapshot including HiDream-O1-Image-1.5 and peer models. |
| SE015 | Hugging Face | ArtificialAnalysis Text-to-Image-Leaderboard Space | ArtificialAnalysis / Text-to-Image-Leaderboard space exists on Hugging Face. |
| SE016 | vivago.ai | What is vivago.ai? | vivago.ai aggregates the world's top video and image AI models ... alongside 300+ viral templates, HiDreamClaw AI Agent, and AI Chat Agent. |
| SE017 | Business Wire | vivago 2.0: The All-in-One AI Tool That Makes Creativity Accessible to All | vivago 2.0 comes with 300+ ready-to-use templates, helping users create professional-quality content in just three seconds. |
| SE018 | Apple App Store | vivago AI: AI Image & Video | Basic-Monthly Auto Renewal $9.99 ... Plus $29.99 ... Pro $99.99. |
| SE019 | Google Play | vivago.ai | Powered by top-ranking AI models, including HiDream-O1-1.5. |
| SE020 | NetEase / 163 | 上海电影与智象未来达成战略投资合作 | 上海电影与智象未来达成战略投资合作。 |
| SE021 | Jiemian | 华策影视与智象未来达成战略投资合作 | 华策影视与智象未来达成战略投资合作。 |
| SE022 | Qichacha | 北京智象未来科技有限公司 | Qichacha lists the Beijing operating entity and shareholder structure. |
| SE023 | Aiqicha | 北京智象未来科技有限公司 - 爱企查 | Aiqicha lists the company as a high-tech enterprise with patent and software-copyright records. |
| SE024 | China Law Translate | Interim Measures for the Management of Generative Artificial Intelligence Services | These measures apply to ... services to the public in the PRC for the generation of text, images, audio, video, or other content. |
| SE025 | Regulations.AI | Measures for Labeling AI-Generated Synthetic Content (summary) | Operationally, service providers that generate synthetic content must embed explicit and implicit labeling mechanisms. |
| SE026 | vivago.ai | vivago global privacy policy | vivago global privacy policy page. |
| SE027 | GitHub | HiDream-I1 repository | HiDream-I1 repository root on GitHub. |
| SE028 | GitHub | HiDream-O1-Image prompt_agent.py | Prompt agent source file in the HiDream-O1-Image repository. |
| SU001 | Sina Finance | 蓝色光标与智象未来达成战略合作 | 蓝色光标与多模态生成式人工智能企业智象未来达成战略合作。 |
| SU002 | Sina Finance | 上影股份战略投资智象未来,布局“内容+AI”新增长空间 | 上海电影股份有限公司与AI企业智象未来宣布达成战略投资合作。 |
| SU003 | 36Kr | 智象未来与华策影视正式签署战略合作协议 | 智象未来与浙江华策影视股份有限公司正式签署战略合作协议。 |
| SU004 | NetEase / 163 | 上海电影与智象未来达成战略投资合作 | 上海电影与智象未来达成战略投资合作。 |
| SU005 | Jiemian | 华策影视与智象未来达成战略投资合作 | 华策影视与智象未来达成战略投资合作。 |
| SU006 | HiDream.ai | HiDream.ai - Where AI Meets Creativity | HiHarness serves 100+ key accounts and 500+ billion API calls. |
| SU007 | Media OutReach Newswire | HiDream.ai Raises RMB1.5 Billion in Series C Financing to Advance Native Omni-modal World Models | Its products serve over 50 million professional users and over 40,000 enterprise customers in more than 100 countries and regions. |
| SU008 | theSaaSnews | HiDream.ai Raises RMB 1.5B in Series C | Its products are serving over 50 million professional users and 40,000 enterprise customers across more than 100 countries and regions. |
| SU009 | Sina portal / Media OutReach | HiDream.ai Raises RMB 1.5 Billion Series C to Advance Native Omni-modal World Models | HiDream.ai’s products currently serve users in more than 100 countries and regions, including over 50 million professional users and more than 40,000 enterprise customers. |
| SU010 | TMTPost | HiDream.ai Secures 1.5 Billion Yuan Series C Round to Accelerate ... | Funding will support product and commercial ecosystem expansion. |
| SU011 | Apple App Store | vivago AI: AI Image & Video | 4.6 rating from 59 Ratings; Basic-Monthly Auto Renewal $9.99. |
| SU012 | Google Play | vivago.ai | Powered by top-ranking AI models, including HiDream-O1-1.5. |
| SU013 | vivago.ai | Vivago AI Agent home | Create AI Videos & Images by Chat | vivago AI Agent |
| SU014 | vivago.ai | What is vivago.ai? | Web, Android (10M+), and iOS. |
| SU015 | vivago.ai | vivago global privacy policy | vivago global privacy policy page. |
| SU016 | Business Wire | vivago 2.0: The All-in-One AI Tool That Makes Creativity Accessible to All | vivago 2.0 also features a vibrant creative community that connects millions of users. |
| SU017 | vivago.ai | What is vivago.ai? (AI route) | vivago.ai aggregates 10+ top-tier video and image models. |
| SU018 | 36Kr Europe | Hefei-backed AI unicorn HiDream.ai raises $290m in three months | Funds support world models and commercial products. |
| SU019 | China Biz Insider | Chinese startup HiDream.ai upends generative AI hierarchy, overtaking Google and ByteDance | HiBurst generated over one million videos annually and supported GMV exceeding RMB 100 million. |
| SU020 | Qichacha | 北京智象未来科技有限公司 | Qichacha lists the Beijing operating entity and shareholder. |
| SU021 | Aiqicha | 北京智象未来科技有限公司 - 爱企查 | Aiqicha lists the company as a high-tech enterprise. |
| SU022 | InforCapital | HiDream.ai company profile | HiDream.ai is an AI infrastructure startup based in Beijing, China. |
| SU023 | China Biz Insider | Chinese AI Video Models Advance as Kuaishou’s Kling 3.0 and ByteDance’s Seedance 2.0 Intensify Competition | The AI video sector remains at an early commercial stage and combined ARR across major video model companies totaled less than US$1 billion as of January 2026. |
| SU024 | Forbes | Video AI Wars: How Chinese Labs Are Winning The Race OpenAI Abandoned | Chinese labs are winning the mass-market creator-economy through price and vertical integration. |
| SU025 | vivago.ai | What is vivago.ai (pricing and community details) | The page details plans, templates, agent access, and creator testimonials. |
| SR001 | CAC / Cyberspace Administration of China | Interim Measures for the Management of Generative AI Services | The measures apply to services using generative AI to provide text, images, audio, and video content to the public in China. |
| SR002 | China Law Translate | Interim Measures for the Management of Generative AI Services (translation) | Providers must use data and base models with lawful sources and respect intellectual property. |
| SR003 | Regulations.AI | China rules for labelling AI-generated content | China issued measures requiring AI-generated content to be labeled. |
| SR004 | CAC / Cyberspace Administration of China | 中华人民共和国网络安全法 | Network operators must take technical and other necessary measures to ensure secure and stable operation. |
| SR005 | CAC / Cyberspace Administration of China | 中华人民共和国数据安全法 | Important-data processors must conduct periodic risk assessments and report them to relevant authorities. |
| SR006 | CAC / Cyberspace Administration of China | “十五五”开局之年推进算力网建设观察 | National integrated computing-network construction is accelerating as AI token demand surges. |
| SR007 | vivago.ai | vivago global privacy policy | vivago global privacy policy page. |
| SR008 | vivago.ai | vivago terms of service | vivago.ai: Bring Every Moment To Life. |
| SR009 | vivago.ai | vivago terms of service (English variant) | vivago.ai: Bring Every Moment To Life. |
| SR010 | HiDream.ai | HiDream.ai homepage | HiHarness serves 100+ key accounts and 500+ billion API calls. |
| SR011 | GitHub | HiDream-I1 LICENSE | HiDream-I1 LICENSE. |
| SR012 | GitHub | HiDream-O1-Image LICENSE | HiDream-O1-Image LICENSE. |
| SR013 | Hugging Face (Wayback snapshot) | README.md · HiDream-ai/HiDream-I1-Full at main | Released under the MIT license... Generated images can be freely used for personal projects, scientific research, and commercial applications. |
| SR014 | GitHub | HiDream-O1-Image prompt agent | Prompt agent code is exposed in the public repo. |
| SR015 | GitHub | HiDream-I1 repository | Public repository for HiDream-I1. |
| SR016 | Media OutReach Newswire | HiDream.ai Raises RMB1.5 Billion in Series C Financing... | Its products serve over 50 million professional users and over 40,000 enterprise customers. |
| SR017 | theSaaSnews | HiDream.ai Raises RMB 1.5B in Series C | The company develops native omni-modal foundation models. |
| SR018 | TMTPost | HiDream.ai Secures 1.5 Billion Yuan Series C Round ... | Funding will support product and commercial ecosystem expansion. |
| SR019 | Sina Finance | 蓝色光标与智象未来达成战略合作 | BlueFocus and HiDream reached strategic cooperation around AI content production and multimodal marketing. |
| SR020 | Sina Finance | 上影股份战略投资智象未来,布局“内容+AI”新增长空间 | Shanghai Film and HiDream announced strategic investment cooperation around IP and immersive experiences. |
| SR021 | 36Kr | 智象未来与华策影视正式签署战略合作协议 | Huace and HiDream formally signed a strategic cooperation agreement. |
| SR022 | China Biz Insider | Chinese AI Video Models Advance as Kling 3.0 and Seedance 2.0 Intensify Competition | The AI video sector remains at an early commercial stage and combined ARR across major video model companies totaled less than US$1 billion as of January 2026. |
| SR023 | Forbes | Video AI Wars: How Chinese Labs Are Winning The Race OpenAI Abandoned | Chinese labs are winning the mass-market creator economy through price and vertical integration. |
| SR024 | Artificial Analysis | Text to Image Arena | Artificial Analysis text-to-image arena. |
| SR025 | Science and Technology Daily | 智象未来图像生成模型评分位列全球第二 | HiDream-I1 topped several benchmarks after open-sourcing. |
| SR026 | Qichacha | 北京智象未来科技有限公司 | Qichacha lists the Beijing operating entity and shareholder information. |
| SR027 | Aiqicha | 北京智象未来科技有限公司 - 爱企查 | Aiqicha lists the company as a high-tech enterprise. |
| SR028 | Rimon Law | China AI and Data Laws in 2026: What Global AI Companies Need to Know | China has layered obligations across cybersecurity, data, personal information, and AI-specific rules. |
| SR029 | Cambridge Forum on AI Law and Governance | Navigating China’s Regulatory Approach to Generative AI and LLMs | China’s approach emphasizes both development and governance of generative AI. |
| SR030 | Sina portal / Media OutReach | HiDream.ai Raises RMB 1.5 Billion Series C ... | Products serve users in more than 100 countries and regions. |
| SV001 | HiDream.ai | 智象未来 - 解锁 AIGC 无限可能 | HiHarness ... 200+ APIs ... 100+ Key Accounts ... 500+ Billion API Calls. |
| SV002 | HiDream.ai | 免费AI图片视频生成与4K编辑 | HiDream.ai | 200+ APIs ... 100+ Key Accounts ... 500+ Billion API Calls. |
| SV003 | Media OutReach Newswire | HiDream.ai Raises RMB 1.5 Billion Series C to Advance Native Omni-modal World Models #HiDreamAI | The round brings HiDream.ai’s total financing over the past three months to more than RMB 2.1 billion and marks its entry into unicorn status. |
| SV004 | The SaaS News | HiDream.ai Raises RMB 1.5B Series C | This latest funding, combined with two earlier rounds completed in the past three months, brings the company's total capital raised during this period to over RMB 2.1 billion. |
| SV005 | Seedtable | HiDream.ai Raises 221.4M USD in Series C Funding | An RMB 1.5bn Series C in July 2026 — its third round in three months, taking total funding past RMB 2.1bn — made it a unicorn. |
| SV006 | TMTPOST | HiDream.ai Secures 1.5 Billion Yuan Series C Round to Accelerate Multimodal AI Development | Proceeds from the financing will be directed toward scaling continuous compute infrastructure, advancing multimodal generative foundation models, and deepening commercial partnerships across digital media, entertainment, and industrial design sectors. |
| SV007 | Qichacha | 北京智象未来科技有限公司 - 企查查 | 成立日期 2023-03-02;注册地址 北京市海淀区苏州街16号(北京神州数码大厦)5层-02。 |
| SV008 | Aiqicha | 北京智象未来科技有限公司 - 爱企查 | 该公司成立于2023年03月02日,位于北京市海淀区苏州街16号(北京神州数码大厦)5层-02。 |
| SV009 | GeekPark | HiDream-O1-Image 正式开源 | HiDream-O1-Image 以 8B 参数规模,在超过 3000 个样本对比中取得 1187 ELO,登顶开源模型第一。 |
| SV010 | Artificial Analysis | Image Arena - Top AI Image Models | HiDream HiDream-O1-Image-1.5 ... rank 11 in the fetched leaderboard snapshot. |
| SV011 | GitHub | HiDream-ai/HiDream-O1-Image | May 8, 2026: We've open-sourced HiDream-O1-Image (8B) ... together with the Reasoning-Driven Prompt Agent. |
| SV012 | Hugging Face | HiDream-ai/HiDream-O1-Image · Hugging Face | HiDream-O1-Image ... 14.2k ... HiDream-I1-Full ... 18.3k in the fetched organization snapshot. |
| SV013 | arXiv | HiDream-I1: A High-Efficient Image Generative Foundation Model with Sparse Diffusion Transformer | HiDream-I1 is constructed ... with 17B parameters ... we provide HiDream-I1 in three variants: Full, Dev, and Fast. |
| SV014 | Together AI | HiDream-I1-Full API | Together AI | Price $0.009 / MP. |
| SV015 | S&P Global Market Intelligence | Generative AI funding: A sober retrospective and the trends shaping 2026 | Investor sentiment remains bullish, but more discriminating. |
| SV016 | Aventis Advisors | AI Valuation Multiples in 2026 | The median revenue multiple for AI companies stood at 24.2x. |
| SV017 | CompaniesMarketCap | Adobe (ADBE) - Market capitalization | As of August 2026 Adobe has a market cap of $108.50 Billion USD. |
| SV018 | CompaniesMarketCap | Adobe (ADBE) - Revenue | In 2025 the company made a revenue of $23.76 Billion USD. |
| SV019 | CompaniesMarketCap | C3 AI (AI) - Market capitalization | As of August 2026 C3 AI has a market cap of $1.61 Billion USD. |
| SV020 | CompaniesMarketCap | C3 AI (AI) - Revenue | In 2025 the company made a revenue of $0.30 Billion USD. |
| SV021 | CompaniesMarketCap | Duolingo (DUOL) - Market capitalization | As of August 2026 Duolingo has a market cap of $6.41 Billion USD. |
| SV022 | CompaniesMarketCap | Duolingo (DUOL) - Revenue | In 2025 the company made a revenue of $1.03 Billion USD. |
| SV023 | CompaniesMarketCap | Kuaishou Technology (1024.HK) - Market capitalization | As of August 2026 Kuaishou Technology has a market cap of $23.53 Billion USD. |
| SV024 | CompaniesMarketCap | Kuaishou Technology (1024.HK) - Revenue | In 2023 the company made a revenue of $16.01 Billion USD. |
| SV025 | CompaniesMarketCap | Shutterstock (SSTK) - Market capitalization | As of August 2026 Shutterstock has a market cap of $0.21 Billion USD. |
| SV026 | CompaniesMarketCap | Shutterstock (SSTK) - Revenue | In 2025 the company made a revenue of $0.98 Billion USD. |
| SV027 | CompaniesMarketCap | Getty Images (GETY) - Market capitalization | As of August 2026 Getty Images has a market cap of $0.18 Billion USD. |
| SV028 | CompaniesMarketCap | Getty Images (GETY) - Revenue | The company's current revenue (TTM) is $0.98 Billion USD. |
| SV029 | TechCrunch | Runway, a startup building generative AI for content creators, raises $141M | The Series C extension — which values Runway at $1.5 billion, a source familiar with the matter tells TechCrunch — brings the company’s total raised to $237 million. |
| SV030 | Reuters via Yahoo Finance | UPDATE 1-AI company Runway valued at $1.5 bln in latest funding - source | AI company Runway has been valued at $1.5 billion in its latest round of funding ... after it raised $141 million. |
| SV031 | Synthesia | Synthesia raises $200 million Series E at $4 billion valuation to shape the future of work | Synthesia ... has raised a $200 million Series E funding round at a $4 billion valuation. |
| SV032 | Sacra | Stability AI revenue, funding & news | Sacra | Stability AI raised approximately $80M in June 2024 ... bringing total funding to $225M since its founding. |
| SV033 | Axis Intelligence Research | Midjourney Statistics 2026: Revenue, Users, Market Share & Growth Data | Midjourney generated an estimated $500 million in revenue during 2025 ... No outside investors. |
| SV034 | Owler | Stability AI Funding - Owler | In total Stability AI has raised $231.0M. Stability AI's last funding round was on Mar 2025. |