SCAIL-2 Multi-Reference Character Swap

This is an AI video generation application based on SCAIL 2 WanAnimatePlus SAM3 three-person tracking multi-reference image character replacement.

Users only need to upload a three-person driving video, then upload a main reference image of three people, and can continue to supplement with multiple reference images A, B, and a background image. The system will automatically identify the actions of the three characters in the driving video and replace the three characters in the reference images to the corresponding positions in the video.

Unlike ordinary single-image character replacement, this process supports multiple reference images to collaboratively participate in character replacement. The main reference image is used to determine the basic identity of the three characters and their left/middle/right correspondence, while the supplemental reference images enhance facial features, clothing, body proportions, pose angles, and detail stability. The background image can be used to control the final scene style.

The entire process is automatically completed: driving video reading, video dimension unification, three-person main reference image processing, multi-reference image processing, background image processing, SAM3 three-person main body recognition, left/middle/right character mask generation, multi-reference image batching, SCAIL 2 conditional fusion, WanAnimatePlus sampling generation, video decoding, frame interpolation, and final video output.

The final output is a three-person character replacement video that inherits the actions, rhythm, and camera structure of the driving video while striving to maintain the identity, clothing, appearance, and left/right positional relationships of the three characters in the reference images.

It is suitable for applications such as three-person character replacement videos, three-person dance videos, multi-character replacements, group digital human videos, AI team showcases, short drama character replacements, advertisement model replacements, group action transfer, multi-reference character consistency videos, and RunningHub video application packaging.

If packaged as a RunningHub application, the core selling point can be directly expressed as:

Upload a three-person driving video and multiple reference images to generate a stable three-person character replacement video with one click.