A new experiment shows how Google’s artificial intelligence draws

In experiments .withgoogle.com we have Scrying Pen, a new project from Google based on the ability to predict drawings, an example of what a properly trained artificial intelligence can do, capable of completing the picture we started with freehand.

A new experiment shows how Google's artificial intelligence draws, A new experiment shows how Google’s artificial intelligence draws, Optocrypto

A new experiment shows how Google’s artificial intelligence draws

In the same vein as what they did with autodraw.com, the system makes recommendations on how the drawing should end, although in this case, it offers more freedom of action:

Autodraw: we cause a stroke, and Google guesses what we want to draw by providing complete figures at 100%, which will replace what we have done.
Scrying Pen: a new experiment that shows in green lines the recommendation of how we should follow our drawing.

To use it first we have to indicate what we want to draw, something that can be done from the lower right menu. It is based on SketchRNNN, and they explain it in the following way:

[…] the “ink” extends backward into the past and forwards into possible futures. The two sides form a strange loop: the ink of the future influences the way of drawing, which in turn becomes the new “past” ink that influences future watercolors.
This project is open source, it is available on GitHub, and in its web, it is possible to see some examples in animation format so that we can verify its possibilities.