Comment by DrAwdeOccarim

Does anyone know how this actually was done? Like, did they export every frame as a PNG and then run them each one by one through the model? Or did they somehow "load" the video into the model directly (which then internally somehow steps through each frame?)