Recently, the Tsinghua-affiliated company Shengshu Technology introduced an exciting feature that allows anyone to easily generate videos with various backgrounds using just a single image, as if possessing "video magic."
This feature is known as the "Subject Reference" function, which stands out by enabling any subject in the video—whether it's a person, an animal, or a fictional character—to maintain a consistent appearance across different scenes.
Key Features Include:
Supports consistent and controllable appearance of single characters, including face, upper body, and full body.
To maintain consistency in the subject's facial appearance, simply capture a clear image of the single subject's face.
To maintain consistency in the subject's upper body (face + upper body attire), capture a clear image of the single subject's upper body.
To maintain consistency in the subject's full body (full body features), capture a clear image of the single subject's full body.
Tang Jiayu, CEO of Shengshu Technology, noted that traditional video generation methods often fail due to insufficient detail stability, whereas this new technology offers creators more freedom and control.
For example, if AIbase wants to change the scene of a previously generated image of a little girl, they can simply click the "Reference Video" option, upload the image, select the subject, and briefly describe the new scene:
For instance:
The little girl wearing a helmet soaring through the sky.
This results in a video like this:
If you only want to reference the character's appearance without considering clothing or other features, simply select the head area when choosing the subject.
Then, AIbase can easily place the little girl on a grassy field through a simple description. Despite a slight glitch with her hands, the overall effect is quite impressive.
Of course, this feature works just as well with real people and pet images. For example, using a photo of a beautiful woman, you can generate a vivid video by simply describing the background:
In addition to people, anime characters, and pets, Vidu's Subject Reference also supports products. Here, AIbase tested a previously generated image of a diamond ring:
The overall camera movement is quite smooth, though it's a bit blurry. Adding some professional prompts should enhance the effect.
It's important to note that this feature currently supports the generation of a single subject only. If your image contains multiple people or objects, the system will require you to select one subject for processing. This ensures that each subject's representation in the video remains consistent, even though it cannot process multiple objects simultaneously.
Interested users can experience it here: https://top.aibase.com/tool/vidu