Does your webvid dataset have a watermark, I think there is a watermark, isn't that a data bias?

[py85252876]

Hi, For T2V generation tasks, the model only takes text as input. For I2V generation tasks, the input images are all of the same quality. Therefore, the detection and source tracing models still need to use the "model's unique signatures" to make predictions.