On July 24, the multi-image reference model and function of Keling AI video were upgraded. After a blind test, the effect of the new model was improved by a full 102% compared to before. After the upgrade, the videos generated by the multi-image reference function have more consistent characters, significantly improved dynamic quality, more reasonable and natural multi-subject interactions, and smoother motion performance. At the same time, the picture quality has also been improved across the board, and the overall effect is more refined. In addition, the multi-image reference function has been upgraded to support users to only refer to local content/specific areas of the picture, such as allowing the model to more accurately refer to specific subjects, faces or clothing in the picture, avoiding interference from irrelevant elements, and easily obtaining more perfect results.