Hey there,

This is my weekly recap, where I share what I’ve been working on in AI and behind the scenes of my business.

This week I experimented with Fable 5.2 making photorealistic videos entirely with code. That means videos that look very real, without using an AI video generator. It’s really incredible what it can do, and it makes you wonder where all of this could go as the models improve.

I also joined the Marketing Against the Grain podcast to spill some sauce about my video editing workflow.

Let’s dive in.


Thanks to the sponsor of this week’s newsletter, Firecrawl. I’ve been using them for years across different projects, and I’m really happy they reached out to sponsor me.

The web wasn't built for AI agents. Firecrawl makes it work for them. Your agents can search and scrape the web while tapping into Firecrawl Alexandria's library of 100M+ indexed sources and 400+ data providers, all through one connection. Install the Firecrawl plugin in one click for ChatGPT, Claude, Cursor and many other agents.


1. My experiments with Fable 5.2

This week I discovered that Fable 5.2 is really impressive at making photorealistic videos with only code.

Earlier in the week, I saw a short video from the point of view of a cat running through a garden. The footage looked very real, and the interesting thing was that it had been made without an AI video model. Here’s the original post on X.

This got me curious, so I wanted to recreate it. With one prompt, Fable built me a robot vacuum driving through a virtual living room.

The whole room is written in C++, the sound is synthesized in Python, and even the camera is simulated, including the lens, focus, motion blur, auto exposure and grain.

The scene is made of 4.27 million triangles, and the rug alone has about 900,000 individual fibres 🤯 Rendering it involved tracing about 120 billion light paths. It took about 2 hours from start to finish on my Mac.

And again, all of that was made with code, without an AI video model or Blender.

I then made another video breaking down exactly how it works.

The reason I find this interesting is that you can imagine this method basically simulating a real world and then driving a camera through it. With an AI video model, you’re kind of creating the video directly from pixels. Two different ways to arrive at a similar outcome.

Also, if you want to try it yourself, the code and prompt are on GitHub.

I then did a different experiment…

Fable 5.2 built me a flyable, Avatar-like world… in just a single 69 KB HTML file. It’s called Lumen Isles and I share more about how it was made in my X post.

It took about 3 prompts, and Fable worked on it for maybe 30 minutes.

If you want to try it yourself: Fly around Lumen Isles here.

You can also check out the code here.


2. Joined Marketing Against the Grain

I was on HubSpot’s Marketing Against the Grain podcast, and the episode is now out.

We talked about how I’ve automated my video editing with Claude, including a live demo of the workflow.

We also covered how I give feedback to improve the edits, the tools I use, and testing new ideas for short-form content.

Pretty fun episode.


A little life update

We had an event at my son’s kindergarten this week where I had to help him make a lantern. It was going badly. He wanted a leopard, but the lantern looked like a weird human face 😩

In a last-minute attempt to save the day, I asked ChatGPT to fix my design.

Luckily it worked. I was able to fix the lantern, and my kid was happy again 😀

I have several of these moments every day now where I think “How did I ever survive without AI?” It’s so ingrained in my life now. Can be both good and bad, depending on how you see it.

ChatGPT helping redesign a leopard lantern

Have a great weekend everyone!

Moritz

Reply

Avatar

or to participate