Titles and animations in code

What is programmatic video rendering?

Programmatic video rendering means a video is written as code, and a renderer draws every frame from that code. Everyframe makes its titles, captions, cards, charts and the way they move this way. Each animation is a short program in TypeScript, a common programming language, that places layers on a screen and moves them over time. No image or video model takes part, so every word and number on screen is the one the program wrote.

It already runs in production. Each of the 1,410 award videos for GZS was written as code from one entry's data and rendered with nobody editing, as our render fleet page shows.

Checked before it is drawn

Before any frame is drawn, the build runs the program to the end and measures every line of text with the same code the renderer later draws it with. What is left is a plain description of the animation. A three-second example comes to 1,776 bytes and covers 90 frames.

Everyframe checks that description before it draws a frame, so the agent finds most problems without watching the video. A headline too small to read on a phone, for example, fails the build. The same programs size a name to fit its card and set one title in three languages.

A new renderer for these programs is built and tested, and it is being rolled out to the machines Everyframe runs on. Until it lands, Everyframe draws its motion graphics through FFmpeg, with a browser-based fallback for a few effects. You can see what the new renderer draws on renderbox.dev/motion.

Character animation from a rig

The new renderer also animates characters. A character is drawn once as parts that turn on their own pivots, with each mood and expression kept as a variant of the same drawing. A scene is then a list of named moves on a timeline, such as a wink, a hop or a celebration, and every frame is drawn from that one rig. So the character stays on model, in its brand's colours, in every shot and every video.

A generative model draws a character fresh in every shot, so its face, proportions and colours drift from one video to the next. Everyframe Characters, which we are designing now, starts from the rig. Munch belongs to StorySell, one of our own products.

A brand brings its mascot in once, the agent rigs it, and a person checks the rig on a preview. From then on the agent animates the mascot in scenes from a brief in words. It is not in Everyframe yet.