Video as Code: Should AI Generate Pixels, or Write Programs?

About this event

Media is becoming generative: made on demand, for one person and one question.

There are two ways to make it:

- A video model generates the pixels directly. - A language model writes a program, and software runs it to make the media: a Remotion project, a Manim scene, a Blender script, a shader, an HTML page, a language of your own. We think the two routes are orthogonal. Some media is best generated directly, and some is best written as code. Both will be huge, and nobody knows the split yet. Both are open research.

This evening is about the second route. When media is a program, you can read it, diff it, render it again at any resolution, and change one number instead of starting over. And because it is code, an LLM can write it.

We say video because it is the part of generative media that gets the most attention right now. The same idea applies to images, diagrams, interfaces, and more.

The evening

We meet in person in San Francisco, and the whole evening is on Zoom too.

5:45 PM · Doors open

6:15 PM · Academa

For the last 18 months, we have built media as code, video and GUIs, and all of it for one use case: technical education.

On September 28, we started a new series of films on Instagram. Ten days later, our account had grown from 25 followers to over 15,000, with 1.8 million views. We will talk about what we have built, what broke along the way, and where we th...

Read the full listing on Luma