HiDream.ai
A vision-first native omni-modal world model builder — Hefei-headquartered, China’s first commercial DiT-architecture multimodal AI unicorn
HiDream.ai (legal entity: HiDream.ai (Hefei) Information Technology Co., Ltd.) is a vision-centric multimodal generative AI company founded in March 2023 in Hefei, Anhui Province, China. In our view, the academic and industry pedigree of founder and CEO Dr. Tao Mei is a core credibility asset for the company, and it is directly relevant to the subsequent discussion of the state-backed, long-duration capital the company has attracted.
Holds a B.Eng. and Ph.D. from the University of Science and Technology of China (USTC). Spent 12 years at Microsoft Research Asia (MSRA), rising to Senior Researcher leading multimedia analytics and vision computing efforts, before serving as Vice President of JD.com and Deputy Dean of JD Explore Academy. He is a Foreign Member of the Canadian Academy of Engineering and holds Fellow status concurrently at IEEE, IAPR, and CAAI. He has authored more than 200 papers with over 30,000 citations, received 15 best-paper awards at international conferences, and holds more than 60 patents. He also serves as an adjunct professor and doctoral advisor at USTC and at the Chinese University of Hong Kong, Shenzhen.
We view it as notable that Dr. Mei is not a purely academic figure: his time at JD.com — a low-margin, high-throughput retail operation with a culture built around disciplined, granular execution — is, by his own account in public interviews, directly reflected in HiDream.ai’s commercialization discipline. More than 90% of the core team hold doctoral or master’s degrees, with prior experience at Microsoft, Baidu, Tencent, and Huawei, which we view as a constructive signal on execution capability.
Information on the founding team’s credentials and career history is sourced from company disclosures and press interviews and has not been independently verified by a third party. Quantitative claims such as “30,000+ citations” originate from the founder’s own public materials, and we treat these as reference data points only.
Since founding, HiDream.ai has built its commercial architecture on the proprietary HiDream family of foundation models, organized under a “1 model + 1 platform + 3 scenario products” (1+1+3) framework. At launch in August 2023, the HiDream foundation model was, by the company’s account, the world’s first online-deployed image and video generation model built on a Diffusion Transformer (DiT) architecture, and parameter scale has since expanded materially.
The operating model rests on three independently-operating, mutually-reinforcing pillars.
A proprietary DiT-based foundation model capable of cross-generation across text, image, video, and 3D modalities. Parameter scale has grown from 6 billion at the 2023 launch to several hundred billion at present, supporting minute-scale marketing video generation for enterprise clients and 5-to-15-second short-form video generation for individual users.
A model-as-a-service layer that exposes HiDream capabilities through standardized APIs to external developers and enterprises. Certain models, such as HiDream E1.1, have been released under an MIT open-source license and natively integrated into ComfyUI, extending the company’s reach into the developer community.
Commercial marketing (brand advertising and product-shoot replacement), film and television creation (script-based multi-shot video generation with 4K resolution and global/local control), and social media content creation. At WAIC 2026, the company unveiled vivago R1, a multimodal creation agent that automates creative planning, model orchestration, and content generation for long-form video generation and editing.
Revenue momentum: Per company disclosure, first-quarter 2026 revenue already exceeded full-year 2025 revenue, which we take as a signal that commercialization has accelerated since the 1+1+3 structure was formalized. That said, we note this figure derives from a private company’s self-reported disclosure (see Data Gap Notice below).
The revenue growth claim (“Q1 revenue exceeded full-year prior-year revenue”) and the latest model’s parameter scale (“200B+”) are both company-reported figures; we have not identified an independent audit or third-party benchmark that cross-validates them. Leaderboard rankings for open-source models (the HiDream image model on Hugging Face; HiDream E1.1 on Artificial Analysis) are comparatively verifiable via public platforms, but performance claims for closed models require separate verification.
HiDream.ai has followed a comparatively fast-paced capital-raising trajectory from Seed through Pre-A, A, A+, B, and C rounds. Notably, between roughly May and July 2026 the company closed three consecutive rounds within about three months, taking cumulative proceeds above RMB 2.1 billion and, by media accounts, pushing its valuation above USD 1 billion — the threshold for unicorn status.
Incorporated in Hefei and closed a seed round. In August 2023 the company launched HiDream foundation model 1.0 (6 billion parameters), claiming the world’s first online-deployed DiT-architecture image and video generation model. In September–October 2023 the company launched its own product, “Zhixiang AI 1.0,” expanding parameter scale to 10 billion.
Cumulative commercial-photography image generation surpassed 1.5 million instances, and API calls surpassed 2.5 million, demonstrating early commercial traction. Alpha Founders Capital and iFlytek are understood to have led early rounds around this period.
Dunhong Capital, a specialist culture-and-technology investor, led this round. Dunhong went on to participate in each of the A, A+, B, and C rounds, effectively serving as an anchor investor across three consecutive rounds.
Led by Hefei Chanantou (Hefei Industrial Investment Group), with participation from the Anhui Province AI Mother Fund, Hubei Yangtze River Film Group, and Huafu Jiaye. Proceeds were directed toward HiDream model R&D and product-line iteration. As of end-November 2024, the company had reportedly served more than 40,000 enterprises and attracted more than 10 million users across over 100 countries.
JD.com Group and the Jinhua Jinwu Empowerment Fund jointly participated. At the end of 2025, JD.com added to its position as a strategic investor; we expect JD’s extensive business scenarios across logistics, retail, healthcare, and industrial operations to serve as a real-world testbed for HiDream.ai’s multimodal technology. Given Dr. Mei’s prior tenure as a JD.com vice president, we view the strategic fit as strong.
Existing investor Oriental Fortune Capital continued to participate. From this point, the company formally articulated its “one HiDream model + one Token Hub platform + three scenario product lines (commercial marketing, film/TV, social media)” structure. Management stated that first-quarter 2026 revenue already exceeded full-year 2025 revenue.
The company closed a new financing round concurrently with the unveiling of a native omni-modal world model exceeding 200 billion parameters, reaffirming its “model plus agent” dual-engine strategy alongside the 1+1+3 structure.
Deal structure: Co-led by the National Social Security Fund’s Sichuan Revitalization Sci-Tech Fund (marking the Fund’s first-ever investment in a multimodal foundation model company), ICBC Capital (its first investment in the foundation model sector), Hongyi Asset Management, and Dunhong Capital. Follow-on participants included Xiamen Guomao Capital, Shanghai Film New Vision Fund, Hubei Yangtze River Industrial Investment Group, Huace Film & TV, Hangyuan Capital, Chuangyunhai Capital, Huafu Investment, Yuhang Jinkong Holdings, Bank of Communications Capital, and Ruosong Fund. Existing shareholders Hefei Chanantou, Oriental Fortune Capital, Jinpu Investment, Jinhua Jintou, Zhongzhe Chuang, and Caixin Capital all added to their positions.
Significance of investor composition: The simultaneous convergence of national-level long-duration capital (the social security fund), policy-bank-affiliated capital (ICBC Capital), multiple regional state-owned funds, and film/TV industrial capital (Shanghai Film, Huace) is, in our view, best read as conviction-driven “patient capital” backing a technology roadmap rather than momentum-driven capital chasing valuation.
Cumulative proceeds and valuation: Across roughly three consecutive rounds between May and July 2026, cumulative proceeds exceeded RMB 2.1 billion, and the resulting valuation is understood to have crossed USD 1 billion, formally placing the company in unicorn territory.
• Rounds closed: 3 consecutive rounds (Series B follow-on → hundred-million-RMB-class round → Series C)
• Cumulative proceeds: RMB 2.1 billion+ (including the RMB 1.5 billion Series C)
• Valuation: Above USD 1 billion (unicorn status, per media reports)
• Use of proceeds: Native omni-modal world model R&D, foundational infrastructure build-out, industrial ecosystem expansion
• New investor characteristics: First-ever social security fund participation, first-ever policy-bank-affiliated capital, new film/TV industrial capital additions
Cumulative proceeds (RMB 2.1B+) and the “unicorn valuation ($1B+)” figure are sourced from multiple Chinese tech media outlets (PEdaily, NetEase Tech, ZOL, among others); the company has not officially disclosed a specific post-money valuation figure. Exact amounts for the Pre-A and A+ rounds are likewise undisclosed; we cite media characterizations such as “several hundred million RMB” as reported rather than presenting them as confirmed figures.
China’s vision-generation AI market is densely competitive, populated by well-capitalized platform-affiliated players such as Kuaishou’s Kling and MiniMax, as well as independent startups such as ShengShu Technology’s Vidu, alongside overseas players such as OpenAI’s Sora and Google’s Veo. We see HiDream.ai’s differentiation resting on the following pillars.
The August 2023 launch of what the company describes as the world’s first online-deployed DiT-architecture image and video generation model has translated into tangible validation within the open-source ecosystem. The HiDream image model ranks third on the Hugging Face open-source model leaderboard, and its image-editing model, HiDream E1.1, has reached the top tier of the Artificial Analysis image-editing agent leaderboard (released under an MIT license, with native ComfyUI integration).
The National Social Security Fund’s first-ever investment in the multimodal foundation model space, ICBC Capital’s first investment in the foundation model sector, and the simultaneous participation of numerous regional state-owned funds and listed film/TV capital together represent an unusual fundraising profile among domestic AI startups. We read this as “patient capital” reflecting long-term conviction in the technology roadmap rather than momentum capital chasing near-term valuation gains, and it may signal further inflows from state-affiliated LPs in future rounds.
The multi-layer revenue structure spanning the model (HiDream), the platform (Token Hub), and three scenario product lines (commercial marketing, film/TV, social media) reduces exposure to demand swings in any single industry vertical. In particular, we view the equity participation of film/TV industrial capital such as Shanghai Film New Vision Fund and Huace Film & TV as a potential customer-acquisition channel in content production, IP development, and AI marketing.
While competitors such as Kling, MiniMax, and Vidu compete primarily on video-generation quality and speed, HiDream.ai has articulated a medium-term goal of transitioning from “vision generation” to “world generation” — a native omni-modal world model capable of understanding spatial, temporal, and physical relationships. This opens optionality toward adjacent markets such as 3D interaction and embodied intelligence, though we view execution against this roadmap as still unproven.
| Company | Core positioning | Notes |
|---|---|---|
| HiDream.ai | DiT-based omni-modal world model, 1+1+3 structure | RMB 1.5B Series C, Jul 2026; unicorn status |
| Kling (Kuaishou) | Platform-affiliated video generation AI | Deployed at scale via parent-company traffic and capital |
| MiniMax | Multimodal foundation model, overseas-market focus | Highly valued independent startup |
| Vidu (ShengShu Technology) | Video-generation-specialist model | Domestic video-generation track competitor |
| OpenAI Sora / Google Veo | Overseas video-generation AI | Global market benchmark |
Investor risk factors: (i) As a mainland Chinese-incorporated private company (rather than an offshore Cayman-style holding structure), the company carries equity-structure and cross-border capital-flow regulatory risk relevant to non-domestic investors; (ii) revenue and model-performance metrics rely heavily on company self-disclosure, with limited independent verification infrastructure; and (iii) cumulative proceeds of roughly RMB 2.1 billion (approximately USD 300 million) are relatively modest against better-capitalized competitors such as Kling and MiniMax, which could leave the company at a capital disadvantage in future large-scale compute competition. That said, we view the qualitative composition of the investor base — anchored by national-level long-duration capital and industrial capital — as a constructive signal.