Following up on yesterday's workshop, I wanted to put together a proper walkthrough of what we showed - because this one's a bit different from our usual real-time video generation.

Episode #5 - What is it about

Two big things in this episode:

1. Scope now supports plugins

This is still in preview (APIs will change, things might break), but it works. The plugin architecture opens up Scope for extensibility - developers can bring new models, new capabilities, and new interactions into the platform. This is a natural step in Scope's evolution, and we're excited to see what people build with it.

2. World models are here

If you've been using Scope with Krea Realtime or LongLive, you know the flow — give it an image or prompt, get real-time video output. World models flip that around. Instead of generating video, you're generating a navigable 3D environment. WASD to move, mouse to look around. It's like being inside an AI-generated video game.

Full video

The scope-overworld plugin brings Overworld's Waypoint-1-Small model into Scope. In the video I walk through the full installation and then explore a few different seed images — including walking on Mars and wandering through a cozy origami Christmas scene I generated with LongLive in a previous episode.

What you can do with this:

• Navigate AI-generated worlds in real-time on consumer hardware
• Record footage for use in other tools
• Stream to Unity or Unreal via Spout
• Build web apps using the WebRTC API

The model was largely trained on first-person shooter environments, so seeds that match that aesthetic tend to work best. And yes, there are artifacts - this is the smallest model, early days. But that's the point of building in public. Things will get better.

Resources

Here are some links and valuable resources to get you started:

• scope-overworld plugin
• Scope GitHub
• Waypoint-1-Small on HuggingFace

If you try this out and create something cool, drop it in the hub - I want to see what you make.