Automated Story Synthesis From Disney

Mubbasir Kapadia from Disney Research in Zurich have reported on CANVAS, a tool for composing “stories” [1]. The idea is to streamline the storyboarding process, with an interface that makes it easier to create a story. Being Disney, the system also generates a 3D animation that illustrates the story.

The result is a tool that enables “untrained users to create complex narrative-driven animations within minutes” (p. 200) Tres cool!

Technically, the system tries to balance the role of the human author with the automated system, keeping the latter largely hidden. This work builds on earlier approaches, mapping author specified points in the plot into “a constraint satisfaction problem with multiple, possibly contradicting goal constraints”, which is not in any way trivial to solve.

The system is similar to many AI planning systems, except with a pretty clever interface–and automatic movie generation!. Given a library of domain knowledge, actors and relationships, the author constructs points in a story with a graphical interface. Then you can push the “complete” button to fill in gaps in the story, and “play” to see a 3D animation of the story. Wow!

Aside from the instant automation (which would make a great product just for home use), I was intrigued by their demo scenario: the story of a bank robbery. This made me wonder if this technology would actually be useful for planning a real caper.

Thinking about it I suspect that, with the right domain knowledge including accurate intelligence and simulation about defenses, this might be a useful planner for this kind of escapade. In fact, the ability to test out alternative narratives would be extremely useful for planning contingencies and back up plans.

Does this mean we might see black market planning software, with libraries of specific domain knowledge for sale? Maybe. “Make your own movie” software, with premium mods to output more than just the movie: it might output a “parts list”, scripts, and briefing materials, so you could act out the caper in real life.

With the instant planning an replanning, you might push out the “script” to the “actors” in real time using mobile devices, monitoring their progress in your 3D simulation. The actual behavior of the actors would be fed back in as new plot points, forcing recalculation of the “story” in real time. Would that really work? I dunno.

And, of course, experts could sell their knowledge over the dark net. Want to know how to break in to a specific bank? Maybe you can buy a mod on the net, filled with expert local knowledge and intelligence.

Pretty neat, buh? A push button heist.

On the other hand…

I’d say this approach would be quite risky, and not just because the input data might be incomplete or wrong. The system would be a tempting target for hacking. Imagine the mischief generated by manipulating such a complex simulation, if the user was crazy enough to actually try it. You would need pretty solid security against hackers and also against leaks because possession of the simulation would be pretty damning evidence against you.

I can imagine white hats collecting intelligence in the form of these libraries of domain knowledge off the dark web, and possible baiting traps with tempting honey—false or misleading input data, treacherous behavior modules, and sneaky tell tale units that report back to the white hats, and lead you into a trap.

I would note that the entire idea might be misguided. While complex, precisely planned crimes make great stories, there is much to be said for simplicity and fluid improvisation. Fantastically complex stories are fragile and tend to fail catastrophically, whereas a simple, straightforward raid might be much more robust in the face of real life events.

…Sorry, I seem to have wandered way off topic.

This technology looks really neat, and I’d love to be able to buy it.

I don’t think it would lead to clever crimes. But it might lift the quality of web animations and game stories. Your average teen ager could make a pretty decent movie with something like this.

That would be neat.

  1. Kapadia, Mubbasir, Seth Frey, Alexander Shoulson, Robert W. Sumner, and Markus Gross, CANVAS: computer-assisted narrative animation synthesis, in Proceedings of the ACM SIGGRAPH/Eurographics Symposium on Computer Animation. 2016, Eurographics Association: Zurich, Switzerland. p. 199-209.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s