LiblibAI's model-supermarket bet: $130M B round, 4M DAU — China's top AI-app raise of 2025

LiblibAI is a Chinese AI-creation platform: a web 'model supermarket' where users generate images and video with open and closed models, train their own LoRAs, and publish or sell art and workflows, monetized through compute credits and subscriptions.

That Chinese creators would gather around an open-model hub — aggregated models, LoRA training, shared workflows — paying for compute, not closed apps like Midjourney.

Oct 2025 Series B of $130M led by Sequoia China, CMC Capital and a strategic investor; four rounds closed in its first year (2024–2025), including a Giant Network-led round reported Feb 2025.

Founded May 2023 (Beijing Qidian Xingyu Tech) by Chen Mian, a former business lead at Mobike, MissFresh and ByteDance. The insight: China's open-source AI-image boom had no integrated home — users juggled desktop Stable Diffusion installs, scattered model sites and no marketplace, payment rails or pro tools.

LiblibAI closed four funding rounds in its first year. By Feb 2025 it reported 20M+ creators, 500K+ user-trained models and workflows, and 500M+ generated images, with enterprise clients like Luckin Coffee and Tsinghua University. In Oct 2025 it raised $130M in a Series B — China's largest AI-app round of 2025 — and shipped LiblibAI 2.0 with aggregated video models (Kling, Hailuo, Vidu, WAN) and professional workflows; overseas design agent Lovart reached about 200K DAU with a ~$30M annualized revenue estimate.

Still scaling and still loss-making (per First Financial), but it became China's largest multimodal-model creation community: 4M DAU, ~25M users, and the record B round — a signal that capital had moved from frontier foundation models to application-layer AI platforms (the round was larger than Manus's $75M April 2025 raise).

Open-source image generation exploded in China, but creators had no integrated platform to generate, train and monetize.

A creator community is a moat: 20M+ creators and 500K+ trained LoRAs made LiblibAI the default home for Stable Diffusion artists.

Aggregating open and closed models let LiblibAI ride every model improvement (Midjourney, Qwen, Kling, WAN) without building its own foundation model.

The record $130M round showed capital rotating from foundation models to the application layer in China's AI market.

The model is the commodity, the community is the moat: letting creators train, share and sell LoRAs turned China's Stable Diffusion scene into a 25M-user platform and 2025's biggest AI-app round.

As of 2026-09-02, LiblibAI remains privately held and loss-making while scaling: 4M DAU and roughly 25M total users as of October 2025, with the $130M Series B earmarked for global expansion and continued creator incentives. Its overseas design agent Lovart had about 200K DAU with an estimated ~$30M annualized revenue (First Financial). No later funding, IPO or shutdown events were verified for this case.