At the end of this year, domestic video companies have gone crazy! Yesterday, Tongyi Wanxiang launched a groundbreaking feature that generates videos with Chinese characters, igniting excitement among AI creators. Today, Hailuo AI introduced a new "Subject Reference" function, taking AI video creation to a new level. Users only need to upload a picture to allow any character to move freely in various scenes, maintaining high fidelity while not losing creative freedom.
From the shares of creators on social media, we can see that whether it's Spider-Man riding a motorcycle or Daenerys from "Game of Thrones" interacting with a wolf in the forest, the character representations consistently maintain high fidelity and stability.
Compared to traditional image-to-video technologies, the "Subject Reference" feature uses a completely different technical approach. It does not simply animate static images but can generate complete video clips based on textual prompts.
The official videos demonstrate that by uploading a single photo of a person, the system can not only preserve their facial features but also allow them to perform any natural and coherent actions in the scene, such as skateboarding, riding a bike, raising eyebrows, frowning, etc., with very vivid and detailed emotional expressions on their faces.
From a technical implementation perspective, Hailuo AI has chosen an image reference-based approach rather than the traditional LoRA technology. This choice allows users to accurately identify and restore character features with just one image, without needing to provide a large amount of material. More importantly, the generation speed has significantly improved, reducing the previous long wait to a response in seconds.
The commercial value of this technology is equally noteworthy. According to Statista, by 2028, the market size for generative AI products in advertising and marketing is expected to reach $107.5 billion. The "Subject Reference" feature perfectly aligns with this market demand, especially in product advertising production, where a single model image can generate multiple versions of video content for different product scenarios.
As a startup established only three years ago, MiniMax has established a leading position in the global AI video generation field with Hailuo AI. Currently, only three companies—Hailuo AI, Keling AI, and Jimeng AI—can truly demonstrate technical strength and development potential, marking the rise of Chinese enterprises in the global AI video generation sector.
In the future, Hailuo AI plans to expand the "Subject Reference" feature to more areas such as multiple people, objects, and scenes, further unleashing creative potential. As AI video generation enters its second phase, the emergence of this technology undoubtedly brings more possibilities for creators, making the vision that "every idea is a blockbuster" gradually become a reality.
Experience it here: https://hailuoai.com/video/create