HiDream's world-model bet: 50M users, ¥2.1B raised, unicorn in three years

HiDream.ai is a Chinese generative-AI startup building self-developed multimodal foundation models and products (vivago.ai, HiBurst, HiFans, Token Hub) for image/video generation, film and TV creation, and marketing content.

That visual generation — image, video, 3D — leads to world models, so a small team could beat giants by pairing open models with fast, outcome-based commercialization.

Cumulative funding over ¥2.1B (roughly $290M) by July 2026: 2024 A round led by Hefei state capital, end-2025 JD Group strategic investment, April 2026 round of over ¥500M, and the July 2026 ¥1.5B C round with 20+ investors.

Founded 2023 in Hefei by Mei Tao, a Canadian Academy of Engineering foreign member and ex-Microsoft Research scientist who later ran JD's Discovery Institute; his core team had written one of the world's first text-to-video papers (TGANs-C, 2017). His thesis: multimodal generation is the most likely path to general AI, and it monetizes earlier than pure language.

HiDream open-sourced HiDream-I1 and E1 in April 2025, topping Artificial Analysis within 24 hours. 2025 revenue passed ¥100M; vivago.ai added ~10M users in January 2026 and entered Google Play's video-editing top 10 in 100+ countries. In April 2026 it raised over ¥500M and Q1 revenue exceeded all of 2025; in May 2026 HiDream-O1-Image became the #1 open-source text-to-image model on Artificial Analysis; in July 2026 it closed a ¥1.5B C round with 20+ investors including social-security-fund capital, ICBC and film-industry groups.

Reached unicorn status in three years (cumulative funding over ¥2.1B, valuation over $1B) with 50M+ users, 1M+ paying users and 40K+ enterprise clients. The bet that visual generation leads to world models is still open, but HiDream has turned it into a funded, revenue-generating business.

Visual generation monetizes earlier than pure language models, so a startup could show revenue before the frontier race ended.

Open-sourcing top-ranked models gave HiDream global mindshare that closed competitors could not easily buy.

A native full-modality (UiT) architecture differentiated HiDream from later-stage multimodal stacks and attracted world-model believers.

RaaS (revenue-share) pricing aligned HiDream with client outcomes instead of token prices, supporting fast enterprise adoption.

State and industrial capital (Hefei, JD, film groups) provided patient funding plus distribution into real content businesses.

Don't fight for the biggest language model: pick the modality that monetizes fastest, and make the bet public — open-sourcing the model bought mindshare while RaaS revenue funded the race.

As of 2026-09-02, HiDream.ai remains privately held and scaling: cumulative funding over ¥2.1B with a post-money valuation above $1B after its July 2026 C round, and a completed shareholding reform that renamed the entity 智象未来（合肥）科技股份有限公司. Its products cover 50M+ users and 40K+ enterprise clients in 100+ countries; vivago.ai has 1M+ paying users, and HiBurst is a top-five TikTok AI partner while HiFans powers 10,000+ physical stores (PEdaily, April 2026). No IPO filing or further funding was verified for this case.