In the AI era, "seeing is believing" is being given a new meaning. On October 14, reporters visited the HiDream.ai in the Grand Union of Innovation and observed AI‑generated video clips that—whether everyday office scenes or cinematic camera language—were virtually indistinguishable from filmed footage. These vividly lifelike images all come from the company’s independently developed multimodal large language model.

"Future creators will only need to convert their ideas into textual descriptions, and AI can generate the required visual content in real time, whether static images or dynamic video," said Zhai Yuechong, head of brand communications at HiDream.ai. "What you describe becomes what is drawn" has moved from science fiction into practical application.

As a global leader in multimodal generative AI, HiDream.ai's self‑developed "HiDream Multimodal Large Language Model" is a multi‑billion‑parameter-scale model that has one of the industry’s richest multimodal licensed corpora. It possesses strong capabilities for generating and understanding visual content and is among the first multimodal large models to complete dual filings for both model and algorithm. Its model generates highly coherent video content with natural, smooth visuals, supports 4K high definition, and uses cinematic camera language with outstanding expressiveness, giving it leading advantages in AIGC technology and digital creativity.

In April 2025, HiDream’s self‑developed open‑source image generation model HiDream‑I1 reached the top of the international authoritative ranking Artificial Analysis within 24 hours, becoming the first Chinese self‑developed generative AI model to enter the global top tier, and resetting industry records across three dimensions: image quality, semantic understanding, and artistic expression.

On the technical innovation front, HiDream has achieved multiple global breakthroughs: its text‑to‑image capability ranks first among global open‑source models; it is first in open‑source community trend rankings; it launched the world’s first purely autoregressive image editing model, developed 15‑second multi‑shot video generation technology, produced a commercial DiT model, and pioneered a hybrid diffusion–autoregressive model architecture.

"Competition in AI is intense; we must continue to break through in model development, technical innovation, and commercial applications," said Mei Tao, founder and CEO of HiDream. The outstanding global ranking of the company’s self‑developed open‑source models demonstrates China’s technical strength in AI.

"Our collaboration with Anhui is a two‑way commitment," Mei added. Hefei’s well‑developed AI industry ecosystem complements the company’s multimodal technologies, and the region’s abundant talent pool and targeted policy support provide fertile ground for growth. "We will deepen our roots in Hefei and leverage local advantages to take our self‑developed AI technologies from Anhui to the world."

Source: anhuinews.com