Meet the Self-driving videos

Or how we create videos with no hands.

No, I’m not talking about automatic cars, or the future of transportation. Neither about video selfies or a not-very-choosy automated tinder account.

Even though they would be worth speaking about…

I’m meaning videos that drive themselves. 
Videos that decide on their own how and when to be produced, that gets created without hands, and published automatically.

Last year I’ve written an article about software eating the world, and in particular video. In there I collected all the first attempts & projects made by creative studios and tech giants that I thought are worth mentioning.

If you didn’t read it, it can be summarised with 3 simple statements:

1. Video is the most powerful type of internet content.
2. The internet needs more video.
3. Videos today are too expensive and slow to produce.

To solve the issue, lots of people have very different ideas. Some even think we should close the internet.

Ok, Donald Trump does not count.

The solution, to me, is making videos speak with live data, with APIs.

So, that’s what we did.
We’ve built ALGO, a cloud-based software that we, at ILLO, use to produce live video campaigns.
And now, I’m happy to share the first big client project we did with it. 
A data-driven video campaign for Serie A Tim, the Italian Football League.

Here you can check a rough campaign wrap-up.

When TIM, the biggest Italian telco owning the official league app, approached us they were already pushing to users the goal’s highlights, but wanted more video content that could increase user’s activity in app.

We proposed them to tell the story of every single match with short, infographic videos.

380 matches throughout the year, mostly happening in the weekends.
Imagine being briefed with this project.

A potential nightmare.

But this came in with perfect timing. Algo was in the making, and just needed a first cool project.

We sat down to analyse the requirements.
We needed to produce 10 videos per match, before, during and after the match. The videos needed to be created and delivered in the shortest timeframe possible, and by shortest I mean just a few minutes. And needed to be super light in weight (when delivering to hundreds of thousands of football fans, every bit is a cost).

We first got access to the data. Tons of data about every action, every kick, every tackle.

Raw data organised for other purposes, that we needed to convert to serve our own purpose.

Then we worked on the content, to help fans understand how the match is going, whether they’re watching it or not.
We identified 10 different templates: Attack, Defence, Fair-play and so on.

In function of how the match is going, we set up an algorithm that generates an index corresponding to each template and decides which is the current most relevant topic to talk about in the video. The index is made of many different data in the specific theme, normalised on the past historic average of that index.

To increase variety, once a video template has been rendered in a match, it gets pulled out, so that it won’t render twice.

At the right time of the match, Algo pulls the match data and uses them to customise the template selected. We customise everything: team’s logos, the names of the players, every number or index shown. Many of the animations are based on the data too, so that users can consume the data visually, in a blink of an eye.

To highlight diversity even more, each video colour palette is customised randomly, choosing from 5 different palettes.

This customisation process lasts just a few seconds. It’s done in After Effects running in the cloud and it’s done by a script written in ExtendScript Toolkit. The After Effect project is then transferred to a render machine that does the heavy-but-dumb job.
We render squared videos, 15 seconds long, instagram-styled. The rendering takes around 1 minute on average.

The resulting video file is then updated directly to the client’s app where it becomes available to football fans in less than two minutes and in an average weight of 1.5Mb.

We produce 10 videos per match,
there are 10 matches per day
and 38 match-days

That means 3800 videos.

Not bad for a studio that used to deliver ~120 videos per year.

The process is completely automated and currently running for another few months without any human interaction needed.

Have a look at some of them!

DEFENCE
LINE-UPS
EFFICIENCY

For the geeks among you, if you want to have a look at a technical overview, here’s approximately how the process works.

And if you want to actually experience a data-driven video campaign,
you should download the Serie A Tim app for iOS and Android or WP, and check it during the weekend.
The first 5 videos are free, then you have to subscribe (the client made our content premium).

You can find some other info about this project on our website.

The project is inspired by a bunch of other projects made by great people in this area of automated video research & development.

And now, some incredibly selfish Q&A:

So, is ALGO a product? Or a service?
It’s a common core product that we completely customise for each project, because any new project relies on custom design, animation and coding.

I get it, it’s a SaaS!
Well not exactly, you can’t access it through your browser, at the moment. But it could be.

Is this the final version of what you have in mind for Algo?
Of course not, the future is beautiful.

It definitely is.

Thanks for reading it all. We’re flattered you were intrigued by the project.
If you feel it’s worth it, give it a ❤

I’m Luca Gonnelli, co-founder & motion director at ILLO.

We’re a design shop creating videos for innovative projects.
We work for startups, agencies and brands worldwide.

Get in touch at illo.tv
or luca@illo.tv