This LoRa allows generating Image to Video with Haigure.

FastWan was used in the examples

During training there were minimal hints with the following structure:

haigure, [squats and moves his arms/moves only arms]

Also in the hints there was the following, but I'm not sure how exactly it worked:

[camera movement/static camera], woman talking

In some videos from the dataset the camera moved and the woman spoke.

The dataset consists of 40 square videos with sides of 480/512/768 and length of 121 frames with real women, but LoRa also works well with anime.

LoRa was trained with ostris/ai-toolkit