We do not hold a single one of our sample films as a video file

We do not hold a single one of our sample films as a video file. The KAZENA Motion page we put up on 8 October carries 11 sample films of 15 seconds each, and 10 of them are not MP4s. They are written as HTML, CSS and JavaScript, and they are drawn on the spot inside the browser of whoever opens the page. The remaining one is the trailer for NINJA-極-, our own original anime, and that is the only one that is real footage, edited down to 15 seconds. Apart from NINJA-極- and our own product KAZENA Books, every brand in the samples is fictional, made for a concept piece.

There were 2 reasons for building them this way. One is that we did not want to send anything heavy to a visitor. A page with 11 videos lined up on it runs to tens of megabytes before anything else is loaded. The other is that the samples themselves are the evidence for how we work. If we say we build films in code and then put nothing on the page but exported video, the words and the thing do not match.

Here is how it is put together. Each film is a single function that takes a point in time and returns the frame at that point. Pass 0.0 seconds and you get the first frame; pass 7.5 seconds and you get the frame at that instant. Nothing in the drawing uses the current clock, a random number or a timer. Pass the same time and you get the same frame, however many times you ask.

Keeping that rule turns a film into something that can be inspected. For every film we had a machine capture 10 frames on a desktop screen and 10 on a phone. The browser's clock is swapped for a fake one, set to the second we are aiming at, and only then is the frame taken. Instead of watching it and saying it looks about right, we can pull the same frame out again and again and put them side by side.

Here is what those frames found. On a phone screen the price text was cut off. A heading in LUMA wrapped in a place we had not intended. Text placed on a map was too small to read. The car in VOLTA, and the performance figures in the phone layout, were clipped at the top edge. A place name in Teluk Nusa was clipped at the right edge. In 1 spot the frame's height had collapsed to 0, so the picture inside it was not appearing at all.

All of it was fixed on the day we found it. With a video file, moving 1 piece of text means exporting again, swapping the file and putting it back in place. Here every one of them was a 1 line change. When fixing a difference is cheap, you fix more of them.

We changed how sizes are written too. Every dimension is based on the width of the container rather than the width of the screen. Because of that, the same film sits correctly in the large frame at the top of the page, in the narrow band on the home page and in the small thumbnail in the sample list. We do not build 1 film 3 times for 3 sizes.

On a tall screen the same function switches to a different composition. We are not cropping a wide picture; it carries a separate layout rebuilt for the vertical shape.

What a customer receives is an MP4 for each ratio (H.264, 30fps) together with the full production source. The source is HTML, CSS and JavaScript with a README, and we include a list of the fonts and audio we used. The point is that next year, when a price figure or 1 sentence needs replacing, it can be changed without making the film again.

This is not all upside, so let me be honest about what we gave up instead. Our deliverable does not contain an After Effects or Premiere project file. That means we cannot hand it straight to a production company or an in-house designer who works on the tools the video industry commonly uses. Even with the source, the person who can open it and change it is someone who writes code. Small changes such as a price or 1 word can be made without rebuilding, but a larger change to the structure needs new production work. We tried not to leave the phrase you can re-edit it vague about where it stops.

A failure as well. The first thing we built this way was a single 20-second promotional film. Looking at the finished thing, I judged that the motion we had taken as our model was simply not good enough. Nobody would want to buy from it. There was no arguing with that, so we threw the whole thing away and rebuilt it as 11 sample films of 15 seconds. Being able to build in code guarantees nothing about the quality of what gets built. Choosing the tool did not stand in for the work.

There is another difference that only showed itself later. An exported video file answers no questions about its own contents. A year from now, when only a price figure needs changing, opening the file tells you nothing about where that figure is written. Code can be read. Whether the thing is left in a readable state is what decides the cost of changing it next year. It is the same thinking as the records we keep of our own work every time.

A note on where this stands now. KAZENA Motion is not taking orders, quotes or payments yet. We are in the middle of adding the video-production KBLI, the business-field licence, to the company, and until it is in place this page is only an introduction. Every price shown is a planned price, set to be reviewed after we see the first 90 days of real projects. If you want to be told when we start taking work, you can ask from the page. I am writing about the licence only as our own paperwork; I am not a lawyer or a tax accountant, so please confirm the registrations for your own business with a professional.