Parallax Video Generator via Real-Time Multiplane Image Network (RT-MPINet)

We use a smaller 256x256 model for faster inference on CPU instances.

Notes:

  1. Use a higher number of frames (>80) and loops (>4) to get a smoother video.
  2. The default uses 60 frames and 4 camera loops for fast video generation.
  3. We have 3 models available (larger the model, slower the inference):
    • Small: 6.6 Million parameters
    • Medium: 69 Million parameters
    • Large: 288 Million parameters (Not available in this demo due to storage limits, you need to download this model and run locally)
    • Please refer to our Project Page for more details
Video Type
0.001 0.1
10 180
1 10
Model Type
Input Resolution