Kanata is a multi-screen video feed that is comprised of a matrix of animated faces that slide to the right.

Kanata uses two pieces of AI:

  1. a classifier selects the relevant YouTube videos autonomously – this classifier is trained on videos that are hand-picked by humans.
  2. a face detector finds and extracts the faces from within the selected videos.

Kanata is written in python and bash and heavily relies on open-source software.

Kanata is programmatically restricted to only consuming videos that are licensed as Creative Commons – the assets generated by Kanata are licensed as Creative Commons as well.

Kanata is live at kamangir.net/Kanata.


  1. Zhang, K., Zhang, Z., Li, Z., and Qiao, Y. (2016). Joint face detection and alignment using multitask cascaded convolutional networks. IEEE Signal Processing Letters, 23(10):1499–1503 – https://github.com/ipazc/mtcnn
  2. YouTube channels that provide access to their content through Creative Commons linceses: list of channels coming soon… 🚧