On February 16, OpenAI announced Sora, an advanced AI model that generates realistic videos from text descriptions, igniting excitement within the global tech community and prompting discussions on AI's future and its potential to transform industries. This marks the beginning of a new era: text-to-video creation powered by artificial intelligence, reshaping how information is shared and how content is consumed worldwide.
In China, the generative AI startup HiDream.ai has rapidly gained traction in the field of multi-modal AI content creation, significantly impacting image and video generation. Launched in March 2023, HiDream.ai aims to develop the strongest multi-modal visual models and applications in China. By mid-year, the company had completed its seed funding and began building a core team and computational resources. In August, it released its flagship model, Qianxiang 1.0, featuring over 6 billion parameters; by October, this grew to more than 10 billion, coinciding with the introduction of the e-commerce product "E象" in Shanghai.
By December, HiDream.ai secured official approval for its models and algorithms, a pivotal step in advancing its commercialization strategy. The company further expanded its market presence in January 2024, attracting over 50,000 monthly active users and establishing partnerships with more than 20 e-commerce clients and 2,000 small to medium-sized businesses. As industry giants like OpenAI and Google entered the multi-modal landscape, HiDream.ai emerged as one of China's pioneering startups focused on generative image and video technologies.
HiDream.ai’s foundation is built on a comprehensive understanding of AI's future, emphasizing an independently developed multi-modal generative model that fills gaps left by larger firms in visual content creation. Their strategy, termed "one horizontal and one vertical," includes the "Pixeling千象" application, a platform designed specifically for designers. With over 13 billion parameters, the flagship model enables text, image, and video content generation.
The company has introduced two main platforms: Pixeling千象 and E象. Pixeling千象 is a user-friendly AIGC creation platform in Chinese, streamlining design processes through image and video generation and editing. E象 serves as an AI tool for e-commerce sellers, expediting the creation of numerous product images to optimize operations and enhance sales. Both platforms feature API integration, offering developers significant usability.
On the foundational model front, HiDream.ai has successfully trained a 13 billion parameter image diffusion transformer and plans to launch a significant upgrade (Version 3.0) in Q1 2024. Additionally, major upgrades for video generation (Version 2.0) are slated for March and May. HiDream.ai emphasizes key aspects of video generation including visual storytelling, content accuracy, ultra-high-definition quality (4K/8K), and controllability both globally and locally.
Rather than a straightforward text-to-video transition, the team innovatively transforms text into images to create storyboards, which are then extended to incorporate temporal dimensions. This approach enhances the stability, detail, and aesthetic quality of videos, allowing for longer duration outputs. Using a large language model, they automatically generate scripts for each shot, create keyframes through "text-to-image" transformation, convert these into individual video segments, and assemble them into cohesive, multi-shot videos lasting 15 seconds or more.
HiDream.ai’s advanced visualization tools stand as a key competitor to Sora in China, delivering robust visual support for the e-commerce sector while empowering creators with efficient and accessible tools. The company is pioneering commercial applications of AIGC for video, establishing itself as a leader in the dynamic video AI landscape in China. The emergence of the text-to-video era heralds a transformative era of intelligent solutions.
Ranking
